proposal research employee turnover

Predicting and explaining employee turnover intention

• Regular Paper
• Open access
• Published: 23 May 2022
• Volume 14 , pages 279–292, ( 2022 )

Cite this article

You have full access to this open access article

• Matilde Lazzari 1 ,
• Jose M. Alvarez   ORCID: orcid.org/0000-0001-9412-9013 2 , 3 &
• Salvatore Ruggieri   ORCID: orcid.org/0000-0002-1917-6087 3  

29k Accesses

24 Citations

7 Altmetric

Explore all metrics

Turnover intention is an employee’s reported willingness to leave her organization within a given period of time and is often used for studying actual employee turnover. Since employee turnover can have a detrimental impact on business and the labor market at large, it is important to understand the determinants of such a choice. We describe and analyze a unique European-wide survey on employee turnover intention. A few baselines and state-of-the-art classification models are compared as per predictive performances. Logistic regression and LightGBM rank as the top two performing models. We investigate on the importance of the predictive features for these two models, as a means to rank the determinants of turnover intention. Further, we overcome the traditional correlation-based analysis of turnover intention by a novel causality-based approach to support potential policy interventions.

Similar content being viewed by others

Reporting reliability, convergent and discriminant validity with structural equation modeling: A review and best-practice recommendations

Gordon W. Cheung, Helena D. Cooper-Thomas, … Linda C. Wang

Work-Life Balance: an Integrative Review

M. Joseph Sirgy & Dong-Jin Lee

Authoritarian leadership styles and performance: a systematic literature review and research agenda

Elia Pizzolitto, Ida Verna & Michelina Venditti

Avoid common mistakes on your manuscript.

1 Introduction

Employee turnover refers to the situation where an employee leaves an organization. It can be classified as voluntary , when it is the employee who decides to terminate the working relationship, or involuntary , when it is the employer who decides [ 33 ]. Voluntary turnover is divided further into functional and dysfunctional [ 26 ], which refer to, respectively, the exit of low-performing and high-performing workers. This paper focuses on voluntary dysfunctional employee turnover (henceforth, employee turnover) as the departure of a high-performing employee can have a detrimental impact on the organization itself [ 62 ] and the labor market at large [ 33 ].

It is important for organizations to be able to retain their talented workforce as this brings stability and growth [ 30 ]. It is also important for governments to monitor whether organizations are able to do so as changes in employee turnover can be symptomatic of an ailing economic sector. Footnote 1 For instance, the European Commission includes it in its annual joint employment report to the European Union (EU) [ 14 ]. Understanding why employees leave their jobs is crucial for both employers and policy makers, especially when the goal is to prevent this from happening.

Turnover intention, which is an employee’s reported willingness to leave the organization within a defined period of time, is considered the best predictor of actual employee turnover [ 34 ]. Although the link between the two has been questioned [ 13 ], it is still widely used for studying employee retention as detailed quit data is often unavailable due to, e.g., privacy policies. Moreover, since one precedes the other, the correct prediction of intended turnover enables employers and policy makers alike to intervene and thus prevent actual turnover.

In this paper, we model employee turnover intention using a set of traditional and state-of-the-art machine learning (ML) models and a unique cross-national survey collected by Effectory Footnote 2 , which contains individual-level information. The survey includes sets of questions (called items ) organized by themes that link an employee’s working environment to her willingness to leave her work. Our objective is to train accurate predictive models, and to extract from the best ones the most important features with a focus on such items and themes. This allows the potential employer/policy maker to better understand intended turnover and to identify areas of improvement within the organization to curtail actual employee turnover.

We train three interpretable (k-nearest neighbor, decision trees, and logistic regression) and four black-box (random forests, XGBoost, LightGBM, and TabNet) classifiers. We analyze the main features behind our two best performing models (logistic regression and LightGBM) across multiple folds on the training data for model robustness. We do so by ranking the features using a new procedure that aggregates their model importance across folds. Finally, we go beyond correlation-based techniques for feature importance by using a novel causal approach based on structural causal models and their link to partial dependence plots. This in turn provides an intuitive visual tool for interpreting our results.

In summary, the novel contributions of this paper are twofold. First, from a data science perspective:

we analyze a real-life, European-wide, and detailed survey dataset to test state-of-the-art ML techniques;

we find a new top-performing model (LightGBM) for predicting turnover intention;

we carefully study the importance of predictive features which have causal policy-making implications.

Second, method-wise:

we devise a robust ranking method for aggregating feature importance across many folds during cross-validation;

we are the first work in the employee turnover literature to use causality (in the form of structural causal models) for interventional (causal) analysis of ML model predictions.

The paper is organized as follows. First, we review related work in Sect.  2 . The Global Employee Engagement Index (GEEI) survey is described in Sect.  3 . The comparative analysis of predictive models is conducted in Sect.  4 , while Sect.  5 studies feature importance. Section  6 investigates the causal inference analysis. Finally, we summarize the contributions and limitations of our study in Sect.  7 .

2 Related work

We present the relevant literature around modeling and predicting turnover intention. Given our interdisciplinary approach, we group the related work by themes.

Turnover determinants . The study of both actual and intended employee turnover has had a long tradition within the fields of human resource management [ 45 ] and psychology [ 34 ]. The work has focused mostly on what factors influence and predict employee turnover [ 27 ]. Similarly, a complementary line of work has focused on job embeddedness, or why employees stay within a firm [ 42 , 60 ]. A number of determinants have been identified for losing, or conversely, retaining employees [ 56 ], including demographic ones (such as gender, age, marriage), economic ones (working time, wage, fringe benefits, firm size, carrier development expectations) and psychological ones (carrier commitment, job satisfaction, value attainment, positive mood, emotional exhaustion), among others. The items and themes along with employee contextual information reported in GEEI capture these determinants.

Most of this literature has centered on the United States or on just a few European countries. See, for instance, [ 56 ] and [ 57 ], respectively. Our analysis is the first to cover almost all of the European countries.

Modeling approaches . Traditional approaches for testing the determinants of employee turnover have focused largely on statistical significance tests via regression and ANOVA analysis, which are tools commonly used in applied econometrics. See, e.g., [ 27 , 56 ]. This line of work has embraced causal inference techniques as it works often with panel data, resorting to other econometric tools such as instrumental variables and random/fixed effects models. For a recent example see [ 31 ]. For an overview on these approaches see [ 5 ].

There has been a recent push for more advanced modeling approaches with the raise of human resource (HR) predictive analytics, where ML and data mining techniques are used to support HR teams [ 46 ]. This paper falls within this line of work. Most ML approaches use classification models to study the predictors of turnover. See, e.g., [ 2 , 20 , 24 , 36 ]. The common approach among papers in this line of work is to test many ML models and to find the best one for predicting employee turnover. However, despite the fact that some of these papers use the same datasets, there is no consensus around the best models. Using the same synthetic dataset, e.g., [ 2 ] finds the support vector machine (SVM) to be the best-performing model while [ 20 ] finds it to be the naive Bayes classifier. We note, however, that similar to [ 24 ] we find the logistic regression to be one of our top-performing models. This paper adds to the literature by introducing a new top-performing model to the list, the LightGBM.

Similarly, this line of work does not agree on the top data-driving factors behind employee turnover either. For instance, [ 2 ] identifies overtime as the main driver while [ 24 ] identifies it to be the salary level. This paper adds to this aspect in two ways. First, rather than reporting feature importance on a final model, we do so across many folds for the same model, which gives a more robust view on each feature’s importance within a specific model. Second, we go beyond the limited correlation-based analysis [ 3 ] by incorporating causality into our feature importance analysis.

Among the classification models used in the literature and from the recent state-of-the-art in ML, we will experiment with the following models: logistic regression [ 35 ], k-nearest neighbor [ 53 ], decision trees [ 11 ], random forests [ 10 ], XGBoost [ 12 ], and the more recent LightGBM [ 37 ], which is a gradient boosting method [ 23 ]. Ensemble of decision trees achieve very good performances in general, with few configuration parameters [ 16 ], and especially when the distribution of classes is imbalanced [ 9 ], which is typically the case for turnover data. Recent trends in (deep) neural networks are showing increasing performances of sub-symbolic models for tabular data (see the survey [ 7 ]). We will experiment with TabNet [ 6 ], which is one of the top recent approaches in this line. Implementations of all of the approaches are available in Python with uniform APIs.

Modeling intent . A parallel and growing line of research focuses on predicting individual desire or want (i.e., intent or intention) over time using graphical and deep learning models. These approaches require sequential data detailed per individual. The adopted models allow to account for temporal dependencies within and across individuals for identifying patterns of intent. Intention models have been used, for example, to predict driving routes for drivers [ 55 ], online consumer habits [ 58 , 59 ], and even for suggesting email [ 54 ] and chat bot responses [ 52 ]. Our survey data has a static nature, and therefore we cannot directly compare with those models, which would be appropriate for longitudinal survey data.

Determining feature importance . Beyond predictive performance, we are interested in determining the main features behind turnover. To this end, we build on the explainable AI (XAI) research [ 28 ], in particular XAI for tabular data [ 49 ], for extracting from ML models a ranking of the features used for making (accurate) predictions. ML models can either explain and present in understandable terms the logic of their predictions (white-boxes) or they can be obscure or too complex for human understanding (black-boxes). The k-nearest neighbor, logistic regression, and decision trees models we use are white-box models. All the other models are black-box models. For the latter group, we use the built-in model-specific methods for feature importance. We, however, add to this line of work in two ways. First, we device our own ranking procedure to aggregate each feature’s importance across many fold. Second, following [ 63 ] we use structural causal models (SCM) [ 47 ] to equip the partial dependence plot (PDP) [ 22 ] with causal inference properties. PDP is a common XAI visual tool for feature importance. Under our approach, we are able to test causal claims around drivers of turnover intention.

Turnover data . Predictive models are built from survey data (questionnaires) and/or from data about workers’ history and performances (roles covered, working times, productivity). Given its sensitive information, detailed data on actual and intended turnover is difficult to obtain. For instance, all of the advanced modeling approaches previously mentioned either use the IBM Watson synthetic data set Footnote 3 or the Kaggle HR Analytics dataset Footnote 4 . This paper contributes to the existing literature by applying and testing the latest in ML techniques to a unique, relevant survey data for turnover intention. The GEEI survey offers a level of granularity via the items and themes that is not present in the commonly used datasets. This is useful information to both employers and policy makers, which allows this paper to have a potential policy impact.

Causal analysis. We note that this is not the first paper to approach employee turnover from a causality perspective, but, to the best of our knowledge, it is the first to do so using SCM. Other papers such as [ 25 ] and [ 48 ] use causal graphs as conceptual tools to illustrate their views on the features behind employee turnover. However, these papers do not equip their causal models with any interventional properties. Some works, e.g., [ 4 , 21 , 61 ], go further by testing the consistency of their conceptual models with data using path analysis techniques. Still, none of these three papers use SCM, meaning that they cannot reason about causal interventions.

3 The GEEI survey and datasets

Effectory ran in 2018 the Global Employee Engagement Index (GEEI) survey, a labor market questionnaire that covered a sample of 18,322 employees from 56 countries. The survey is composed of 123 questions that inquire contextual information (socio-demographic, working and industry aspects), items related to a number of HR themes (also called, constructs), and a target question. The target question (or target variable , the one to be predicted) is the intention of the respondent to leave the organization within the next three months. It takes values leave (positive) and stay (negative). The design and validation of the GEEI questionnaire followed the approach of [ 18 ]. After reviewing the social science literature, the designers defined the relevant themes, and items for each theme. Then they ran a pilot study in order to validate psychometric properties of questions to assess their internal consistency, and to test convergent and discriminant validity Footnote 5 of questions.

Contextual information is reported in Table  1 , together with type of data encoded – binary for two-valued domains (male/female gender, profit/non-profit type of business, full/part time work status), nominal for multi-valued domains (e.g., country name), and ordinal for ranges of numeric values (e.g., age band) or for ordered values (e.g., primary/secondary/higher education level).

Items refer to questions related to a theme. The items for the Trust theme are shown in Table  2 . There are 112 items in total Footnote 6 . Each item admits answers in Likert scale. A score from 0 to 10 is assigned to an answer by a respondent as follows:

Strongly agree \(\rightarrow \) 10

Agree \(\rightarrow \) 7.5

Neither agree nor disagree \(\rightarrow \) 5

Disagree \(\rightarrow \) 2.5

Strongly disagree \(\rightarrow \) 0

The direction of the response scale is uniform throughout all the items [ 50 ]. Table  3 shows the list of all 23 themes. For a respondent, a score from 0 to 10 is also assigned to a theme as the average score of the items of the theme.

From the raw data of the GEEI survey, we constructed two Footnote 7 tabular datasets, both including the contextual information. The dataset with also the scores of the themes is called the themes dataset . The dataset with also the scores of the items is called the items dataset . The datasets are restricted to respondents from 30 countries in Europe. The GEEI survey includes 303 to 323 respondents per country, with the exception of Germany which has 1342 respondents. We sampled 323 German respondents stratifying by the target variable. Thus, the datasets have an approximately uniform distribution per country. Also, gender is uniformly distributed with 50.9% of males and 49.1% of females. These forms of selection bias do not take into account the (working) population size of countries. Caution will be mandatory when making conclusions about inferences on those datasets. Finally, Fig.  1 shows the distribution of respondents by age and gender.

Distribution of respondents by Age and Gender

Target variable by Country

In summary, the two datasets have a total of 9,296 rows each, one row per respondent. Only 51 values are missing (out of a total of 1.1M cell values), and they have been replaced by the mode of the column they appear in. The positive rate is 22.5% on average, but it differs considerably across countries, as shown in Fig.  2 . In particular, it ranges from 12% of Luxemburg up to 30.6% of Finland.

4 Predictive modeling

Our first objective is to compare the predictive performances of a few state-of-the-art machine learning classifiers on both the datasets, which, as observed, are quite imbalanced [ 9 ]. We experiment with interpretable classifiers, namely k-nearest neighbors (KNN), decision trees (DT) and ridge logistic regression (LR), and with black-box classifiers, namely random forests (RF), XGBoost (XGB), LightGBM (LGBM), and TabNet (TABNET). We use the scikit-learn Footnote 8 implementation of LR, DT, and RF, and the xgboost Footnote 9 , lightgbm Footnote 10 , and pytorch-tabnet Footnote 11 Python packages of XGB, LGBM, and TABNET. Parameters are left as default except for the ones set by hyper-parameter search (see later on).

AUC-PR of logistic regression on a single theme. Bars show mean ± stdev over \(10 \times 10\) cross-validation folds

We adopt repeated stratified 10-fold cross validation as testing procedure to estimate the performance of classifiers. Cross-validation is a nearly unbiased estimator of the generalization error [ 40 ], yet highly variable for small datasets. Kohavi recommends to adopt a stratified version of it. Variability of the estimator is accounted for by adopting repetitions [ 39 ]. Cross-validation is repeated 10 times. At each repetition, the available dataset is split into 10 folds, using stratified random sampling. An evaluation metric is calculated on each fold for the classifier built on the remaining 9 folds used as training set. The performance of the classifier is then estimated as the average evaluation metric over the 100 classification models (10 models times 10 repetitions). An hyper-parameter search is performed on each training set by means of the Optuna Footnote 12 library [ 1 ] through a maximum of 50 trials of hyper-parameter settings. Each trial is a further 3-fold cross-validation of the training set to evaluate a given setting of hyper-parameters. The following hyper-parameters are searched for: (LR) the inverse of regularization strength; (DT) the maximum tree depth; (RF) the number of trees and their maximum depth; (XGBoost) the number of trees, number of leaves in trees, the stopping parameter of minimum child instances, and the re-balancing of class weights; (LightGBM) minimum child instances, L1 and L2 regularization coefficients, number of leaves in trees, feature fraction for each tree, data (bagging) fraction, and frequency of bagging; (TABNET) the number of shared Gated Linear Units.

AUC-PR of logistic regression on a single item. Bars show mean ± stdev over \(10 \times 10\) cross-validation folds

As evaluation metric, we consider the Area Under the Precision-Recall Curve (AUC-PR) [ 38 ], which is more informative than the Area Under the Curve of the Receiver operating characteristic (AUC-ROC) on imbalanced datasets [ 15 , 51 ]. A random classifier achieves an AUC-PR of 0.225 (positive rate), which is then the reference baseline. A point estimate of the AUC-PR is the mean average precision over the 100 folds [ 8 ]. Confidence intervals are calculated using a normal approximation over the 100 folds [ 19 ]. We refer to [ 8 ] for details and for a comparison with alternative confidence interval methods.

Let us first concentrate on the case of the themes dataset. As a feature selection pre-processing step, we run a logistic regression for each theme, with the theme as the only predictive feature. Fig.  3 reports the achieved AUC-PRs (mean ± stdev over the \(10 \times 10\) cross-validation folds). It turns out that the top three themes (Retention factor, Loyalty, and Commitment) include among their items a question close or exactly the same as the target question. For this reason, we removed these themes (and their items, for the item dataset) from the set of predictive features. The nominal contextual features from Table 1 , namely Country, Industry, and Job function, are one-hot encoded.

(Unweighted) items dataset: Critical Difference (CD) diagram for the post hoc Nemenyi test at 99.9% confidence level [ 17 ]

The performances of the experimented classifiers are shown in Table 4 (top). It includes the AUC-PR (mean ± stdev), the 95% confidence interval of the AUC-PR, and the elapsed time Footnote 13 (mean ± stdev), including hyper-parameter search, over the \(10 \times 10\) cross-validation folds. AUC-PRs for all classifiers are considerably better than the baseline (more than twice the baseline even for the lower limit of the confidence interval). DT is the fastest classifier Footnote 14 , but, together with KNN, also the one with lowest predictive performance. LGBM has the best AUC-PR values and an acceptable elapsed time. LR is runned up, but it is almost as fast as DT. RF has a performance close to LGBM and LR but it slower. XGB is in the middle as per AUC-PR and elapsed time. Finally, TABNET has intermediate performances, but it is two orders of magnitude slower than its competitors.

The statistical significance of the difference of mean performances of classifiers is assessed with two-way ANOVA if values are normally distributed (Shapiro’s test) and homoscedastic (Bartlett’s test). Otherwise, the nonparametric Friedman test is adopted [ 17 , 32 ]. For the theme dataset, ANOVA was used. The test shows a statistically significant difference among the mean values (family-wise significance level \(\alpha = 0.001\) ). The post hoc Tukey HSD test shows a no significant difference between LGBM and LR. All other differences are significant, as shown in Table  4 (top).

As a natural question, one may wonder how the performance would change if the datasets were weighted to reflect the workforce of each country. We collected the employment figures for all the countries in our training dataset for 2018, which was when the survey was carried out. The country-specific employment data was obtained from Eurostat Footnote 15 (for the EU member states as well as for the United Kingdom) and from the World Bank Footnote 16 (for Russia and Ukraine). The numbers correspond to the country’s total employed population between the ages of 15 and 74. For Russia and Ukraine, however, the number corresponds to the total employed population at any age. We assigned a weight to each instance in our datasets proportional to the workforce in the country of the employee. Weights are considered both in training of classifiers and in the evaluation metric (weighted average precision). The weighted positive rate is 20%. Table  4 (bottom) shows the performances of the classifiers over the weighted dataset. The mean AUC-PR is now smaller for most classifier, the same for LGBM, and slightly better for RF. Standard deviation has increased in all cases. The post hoc Tukey HSD test now shows a small significant difference between LGBM and LR.

Let us now consider the items dataset. Figure  4 shows the predictive performances of single-feature logistic regressions. Table  5 reports the performances of classifiers on all features for both the unweighted and the weighted data. Overall performances of each classifier improve over the theme dataset. Elapsed times also increase due to the larger dimensionality of the dataset. Differences are statistically significant. LGBM and LR are the best classifiers for both the unweighted and the weighted datasets. Figure  5 shows the critical difference diagram for the post hoc Nemenyi test for the unweighted dataset following a significant Friedman test. An horizontal line that crosses two or more classifier lines means that the mean performances of those features are not statistically different. In summary, we conclude that the LR and LGBM classifiers have highest predictive power of the turnover intention.

Weighted theme dataset: CD diagram for the post hoc Nemenyi test at 99.9% confidence level for the top-10 LR feature importances

Weighted item dataset: CD diagram for the post hoc Nemenyi test at 99.9% confidence level for the top-10 LR feature importances

Weighted theme dataset: CD diagram for the post hoc Nemenyi test at 99.9% confidence level for the top-10 LGBM feature importances

Weighted item dataset: CD diagram for the post hoc Nemenyi test at 99.9% confidence level for the top-10 LGBM feature importances

5 Explanatory factors

We examine the driving features behind the two top-performing models found in Sect.  4 : the LGBM and the LR. We use each model’s specific method for determining feature importance and aggregate the importance ranks over the 100 experimental folds. This novel approach yields more robust estimates (a.k.a., lower variance) of importance ranks than using a single hold-out set. We do so for the weighted version of both the theme and item datasets.

For a fixed fold, feature importance of the LR model is determined as the absolute value of the feature’s coefficient in the model. The importance of a feature in the LGBM model is measured as the number of times the feature is used in a split of a tree in the model. We aggregate feature importance using their ranks, as in nonparametric tests statistical [ 32 ]. For instance, LR absolute coefficients \((|\beta _1|, |\beta _2|, |\beta _3|, |\beta _4|) = (1, 2, 3, 0.5)\) lead to the ranking (3, 2, 1, 4).

The top-10 features w.r.t. the mean rank over the 100 folds are shown in Fig.  6 to Fig.  9 for the theme/item datasets and LR/LGBM models. For the theme dataset (resp., the item dataset), LR and LGBM share almost the same set of top features with slight differences in the mean ranks. For example, the Sustainable Employability , Employership , and Attendance Stability themes are all within the top-five features for both LR and LGBM. For the item dataset, we observe Time in Company , Satisfied Development , and Likelihood to Recommend Employer to Friends and Family to be among the top-five shared features. Interestingly, Gender , a well-recognized determinant of turnover intention, is not among the top features for both datasets. Also, no country-specific effect emerges.

The Friedman test shows significant differences among the importance measures in all four cases in Fig.  6 to Fig.  9 .

Further, the figures show the critical difference diagrams for the post hoc Nemenyi test, thus answering the question whether there is any statistical difference among them. An horizontal line that crosses two or more feature lines means that the mean importances of those features are not statistically different. In Fig.  8 , for example, the Motivation , Vitality , and Attendance Stability themes are grouped together.

Statistical significance of different feature importance is valuable information when drawing potential policy recommendations as we are able to prioritize policy interventions. For example, given these results, a company interested in employee retention could focus on improving either motivation or vitality, as they strongly influence LGBM predictions and, a fortiori , turnover intention. However, the magnitude and direction of the influence is not accounted for in the feature importance plots of Fig.  6 to Fig.  9 . This is not actually a limitation of our (nonparametric) approach. Any association measure between features and predictions (such as the coefficients in regression models) does not allow for causal conclusions. We intend to overcome correlation analysis, as a means to support policy intervention, thought an explicit causal approach.

6 Causal analysis

In Sect.  4 , we found LGBM and LR to be the best performing models for predicting turnover intention, and in Sect.  5 we studied the driving features behind the two models. Now we want to assess whether a specific theme T has a causal effect on the target variable, written \(T \rightarrow Y\) , given the trained model b (as in b lack-box) and the contextual attributes in Table  1 . We use \(T^*\) to denote the set of remaining themes and \(\tau \) to denote the set of all themes, such that \(\tau = \{T\} \cup T^*\) . Establishing evidence for a direct causal link between T and Y would allow our model b to answer intervention-policy questions related to the theme scores. Given our focus on T , in this section we work only with the theme dataset.

We divide all contextual attributes into three distinct groups based on their level of specificity: individual-specific attributes, I , where we include attributes such as Age and Gender ; work-specific attributes, W , where we include attributes such as Work Status and Industry ; and geography-specific attributes, G , where we include the attribute Country . Footnote 17 We summarize the causal relationships across the contextual attributes, a given theme’s score T , the remaining themes \(T^*\) , and the target variable Y using the causal graph \({\mathcal {G}}\) in Fig.  10 . The nodes on the graph represent groupings of random variables, while the edges represent causal claims across the variable groupings. Within each of these contextual nodes, we picture the corresponding variables as their own individual nodes independent from each other but with the same causal effects with respect to the other groupings. Footnote 18

Causal graph \({\mathcal {G}}\) showing the three groups of contextual attributes (individual I , geographic G , and working W ), the collection of themes ( \(\tau \) ) and the target variable Y . We are interested in the edge going from \(\tau \) into Y

A more detailed look into \(\tau \) (dashed black-rectangle) where we can see the distinct edges going from T and \(T^*\) into Y . The three incoming edges represent the information flow from W , G , and I into \(\tau \) . Here, for illustrative purposes, we have ignored those same edges going into Y

Notice that in Fig.  10 two edges go from \(\tau \) to Y . This is because we have defined \(\tau = \{T\} \cup T^*\) and are interested in identifying the edge between T and Y (marked in red), while controlling for the edges from \(T^*\) to Y (marked in black as the rest). This becomes clearer in Fig.  11 where we detailed the internal structure of \(\tau \) . Here, we assume independence between whatever theme is chosen as T and the remaining themes in \(T^*\) . Footnote 19 Further, as with the contextual nodes representing the variable groupings, \(T^*\) represents the grouping of all themes in \(\tau \) but T where each theme is its own node and independent of each other while have the same inward and outward causal effects. Footnote 20

Under \({\mathcal {G}}\) , all three contextual attribute groups act as confounders between T and Y and thus need to be controlled for along with \(T^*\) to be able to identify the causal effect of T on Y . Otherwise, for example, observing a change in Y cannot be attributed to changes in T as G (or, similarly, I or W ) could have influenced both simultaneously, resulting in an observed association that is not rooted on a causal relationship. Therefore, controlling for G , as for the rest of the contextual attributes insures the identification of \(T \rightarrow Y\) . This is formalized by the back-door adjustment formula [ 47 ], where \(X_{C} = I \cup W \cup G \cup T^*\) is the set for all contextual attributes:

In ( 1 ), the term \(P(X_{C}=x_{C})\) is thus shorthand for \(P(I=i, W=w, G=g, T^*=t^*)\) . The set \(X_{C}\) satisfies the back-door criterion as none of its nodes are descendants of T and it blocks all back-door paths between T and Y [ 47 ]. Given \(X_{C}\) , under the back-door criterion, the direct causal effect \(T \rightarrow Y\) is identifiable. Further, ( 1 ) represents the joint distribution of the nodes in Fig.  10 after a t intervention on T , which is illustrated by the \( do \) -operator. If T has a causal effect on Y , then the original distribution P ( Y ) and the new distribution \(P(Y| do (T:=t))\) should differ over different values of t . The goal of such interventions is to mimic what would happen to the system if we were to intervene it in practice. For example, consider a European-wide initiative to improve confidence among colleagues, such as providing subsidies to team-building courses at companies. Then the objective of this action would be to improve the Trust theme’s score to a level t with the hopes of affecting Y .

Pairwise Conover-Iman post hoc test p-value for Trust theme vs Country in a clustered map. The map clusters together countries whose score distributions are similar

The structure of the causal graph \({\mathcal {G}}\) in Fig.  10 is motivated both from the data and from expert knowledge. Here we argue that I , W , and G are potential confounders of T and Y . For instance, consider the Country attribute, which belongs to G . It is sensible to picture that Country affects T as employees from different cultures can have different views on the same theme. Similarly, Country can affect Y as different countries have different labor laws that could make some labor markets more dynamic (reflected in the form of higher turnover rates) than others. We also observe this in the data. In particular, the Country attribute is correlated to each of the themes: the nonparametric Kruskal–Wallis H test [ 32 ] shows a p-value close to 0 for all themes, which means that we reject the null hypothesis that the scores of a theme in all countries originate from the same distribution. Consider the Trust theme. To understand which pair of countries have similar/different Trust score distributions, we run the Conover-Iman post hoc test pairwise. The p-values are shown in the clustered map of Fig.  12 . The groups of countries highlighted Footnote 21 by dark colors (e.g., Switzerland, Latvia, Finland, Slovenia) are similar among them in the distribution of Trust scores, and dissimilar from the countries not in the group. Such clustering shows that the societal environment of a specific country has some effect on the respondents’ scores of the Trust theme. Similar conclusions hold for all other themes.

Further, both G and I have a direct effect also on W . We argue that country-specific traits, from location to internal politics, will affect the type of industries that developed nationally. For example, countries with limited natural resources will prioritize non-commodity-intensive industries. Similarly, individual-specific attributes will determine the type of work that an individual can perform. For example, individuals with higher education, where education is among the attributes in I , can apply to a wider range of industries than an individual with lower levels of educational attainment.

To summarize thus far, our goal in this section is to test the claim that a given T causes Y given our model b and our theme dataset. To do so we have defined the causal graph \({\mathcal {G}}\) in Fig.  10 and defined the corresponding set \(X_{C}\) that satisfies the back-door criterion that would allow us to test \(T \rightarrow Y\) using ( 1 ). What we are missing then is a procedure for estimating ( 1 ) over our sample to test our causal claim.

For estimating ( 1 ) we follow the procedure in [ 63 ] and use the partial dependence plot (PDP) [ 22 ] to test visually the causal claim. The PDP is a model-agnostic XAI method that shows the marginal effect one feature has on the predicted outcomes generated by the model [ 43 ]. If changing the former leads to changes in the latter, then we have evidence of a partial dependency between the feature of interest and the outcome variable that is manifested through the model output. Footnote 22 We define formally the partial dependence of feature T on the outcome variable Y given the model b and the complementary set \(X_C\) as:

If there exist a partial dependence between T and Y , then \(b_{T}(t)\) should vary over different values of T , which could be visually inspected by plotting the values via the PDP. If \(X_C\) satisfies the back-door criterion, [ 63 ] argues, then ( 2 ) is equivalent to ( 1 ), Footnote 23 and we can use the PDP to check visually our causal claim. Under this scenario, the PDP would have a stronger claim than partial dependence between T and Y , as it would also allow for causal claims of the sort \(T \rightarrow Y\) . Footnote 24 Therefore, we could assess the claim \(T \rightarrow Y\) by estimating ( 2 ) over our sample of n respondents using:

Using ( 3 ), we can now visually assess the causal effect of T on Y by plotting \({\hat{b}}_T\) against values of T . If \({\hat{b}}_T\) varies across values of t , i.e. \({\hat{b}}_T\) is indeed a function of t , then we have evidence for \(T \rightarrow Y\) [ 63 ].

However, before turning to the estimation of ( 3 ), we address the issue of representativeness (or lack thereof) in our dataset. One implicit assumption used in ( 3 ) is that any j element in \(X_C^{(j)}\) is equiprobable. Footnote 25 This is often assumed because we expect random sampling (or, in practice, proper sampling techniques) when creating our dataset. For example, the probability of sampling a German worker and a Belgian worker would be same. This is a very strong assumption (and one that is hard to prove or disprove), which can become an issue if we were to deploy the trained model b as it may suffer from selection bias and could hinder the policy maker’s decisions.

To account for this potential issue, one approach is to estimate \(P(X_C=x_c)\) from other data sources such as official statistics. This is why, for example, we created the country weighted versions of the theme and item datasets back in Sect.  4 . Here it would be better to do the same not just for country, but to weight across the entirety of the complementary set. Footnote 26 However, this was not possible. The main complication we found for estimating the weight of the complementary set was that there is no one-to-one match between the categories used in the survey and the EU official statistics. Therefore, it is important to keep this in mind when interpreting the results beyond the context of the paper. By using the (country-)weighted theme dataset, we can rewrite ( 3 ) as a country-specific weighted average:

where \(\alpha _j\) is the weight assigned to j ’s country, and \(\alpha = \sum _{j=1}^n \alpha ^{(j)}\) . Under this approach, we are still using the causal graph \({\mathcal {G}}\) in Fig.  10 .

We proceed by estimating the PDP using ( 4 ). We define as T our top feature from the LGBM model in the weighted theme dataset, which was the Motivation theme as seen in Fig.  8 . We then use the corresponding top LGBM hyper-parameters and retrain the classifier on the entire dataset. Footnote 27 Finally, we compute the PDP for Motivation theme as shown in Fig.  13 . We do the same for the LR model for comparison.

PDP for the Motivation theme for both LGBM and LR models using the weighted theme dataset

From Fig.  13 , under the causal graph \({\mathcal {G}}\) , we can conclude that there is evidence for the causal claim \(T \rightarrow Y\) for the Motivation theme. For the LGBM model, the theme score ( x-axis ), which ranges from 0 to 10, as it increases the corresponding predicted probabilities of employee turnover decrease, meaning that a higher motivation score leads to a lower employee turnover intention. We see a similar, though smoother, behaviour with the LR model. This is expected as the LGBM can capture non-linear relationships between the variables better than the LR.

PDP for the Adaptability theme for both LGBM and LR models using the weighted theme dataset

We repeat the procedure on a non-top-ranked theme for both models, namely the Adaptability theme (the capability to adapt to changes), to see how the PDPs compare. The results are shown in Fig.  14 . In the case of the LGBM, the PDP is essentially flat and implies a potential non-causal relationship between this theme and employee turnover intention. For the LR, however, we see a non-flat yet narrower PDP, which also seems to support a potential non-causal link. This might be due again to the non-linearity in the data, where the more flexible model (LGBM) can better capture the effects in the changes of T than the less flexible one (LR) that can tend to overestimate them.

To summarize this approach for all themes, we calculate the change in PDP, which we define as:

and do this for all themes across the LGBM and LR models. The results are shown in Table  6 . Themes are ordered based on the LGBM’s deltas. We note that the deltas across models tend to agree: the signs (and for some themes like Motivation even the magnitudes) coincide. This is inline with previous results in other sections where the LR’s behaviour is comparable to the LGBM’s. Further, comparing the ordering of the themes in Table  6 with the feature rankings in Fig.  6 and 8 , we note that some of the theme’s with the largest deltas (such as Sustainable Emp. and Employership ) are also among the top-ranked features. Although there is no clear one-to-one relationship between the two approaches, it is comforting to see the top-ranked themes also having the higher causal impact on employee turnover as it implies some potential shared underlying mechanism.

Table  6 also provides a view on how each theme causally affects employee turnover, where themes with a positive delta cause a decrease in employee turnover. As the theme’s score increases, the probability of turnover decreases. The reverse holds for negative deltas. We recognize that some of these results are not fully aligned with findings by other papers, mainly from the managerial and human resources fields. For example, we find Role Clarity to cause employee turnover to increase, which is the opposite effect found in other studies [ 29 ]. These other claims, we note, are not causal. Moreover, such discrepancies are possible already by taking into account that those findings are based on US data while ours on European data. As we argued when motivating Fig.  10 , we believe the interaction between geographical and work variables (such as in the form of country-specific labor laws or the health of its economy) affect employee turnover. Hence, the transportability of these previous results into a European context was not expected.

Overall, Table  6 along with both Fig.  13 and Fig.  14 can be very useful to inform a policy maker as they can serve as evidence for justifying a specific policy intervention. For example, here we would advised for prioritizing policies that foster employee motivation over policies that focus on employee and organization adaptability. Overall, this is a relatively simple XAI method that could be used also by practitioners to go beyond claims on correlation between variables of interest in their models.

7 Conclusions

We had the opportunity to analyze a unique cross-national survey of employee turnover intention, covering 30 European countries. The analytical methodologies adopted followed three perspectives. The first perspective is from the human resource predictive analytics, and it consisted of the comparison of state-of-the-art machine learning predictive models. Logistic Regression (LR) and LightGBM (LGBM) resulted the top performing models. The second perspective is from the eXplainable AI (XAI), consisting in the ranking of the determinants (themes and items) of turnover intention by resorting to feature importance of the predictive models. Moreover, a novel composition of feature importance rankings from repeated cross-validation was devised, consisting of critical difference diagrams. The output of the analysis showed that the themes Sustainable Employability , Employership , and Attendance Stability are within the top-five determinants for both LR and LGBM. From the XAI strand of research, we also adopted partial dependency plots, but with a stronger conclusion than correlation/importance. The third perspective, in fact, is a novel causal approach in support of policy interventions which is rooted in causal structural models. The output confirms those from the second perspective, where highly ranked themes showed PDPs with higher variability than lower ranked themes. The value added from the third perspective here is that we quantify the magnitude and direction for the causal claim \(T \rightarrow Y\) .

Three limitations of the conclusions of our analysis should be highlighted. The first one is concerned with comparison with related work. Due to the specific set of questions and the target respondents of the GEEI survey, it is difficult to compare our results with related works that use other survey data, which cover a different set of questions and/or respondents. The second limitation of our results consists of a weighting of datasets, to overcome selection bias, which is limited to country-specific workforce. Either the dataset under analysis should be representative of the workforce, or a more granular weighting should be used to account for country, gender, industry, and any other contextual feature. The final and third limitation of our results concern the causal claims. Our analysis is based on a specific and by far non-unique causal view of the problem of turnover intention where, for example, variables such as Gender and Education level that belong to the same group node I are considered independent. The interventions carried out to test the causal claim are reliant on the specified causal graph, which limits our results within Fig.  10 .

To conclude, we believe that further interdisciplinary research like this paper can be beneficial for tackling employee turnover. One possible extension would be to collect country’s national statistics to avoid selection bias in survey data or, alternatively, to align the weights of the data to a finer granularity level. Another extension would be to carry out the causal claim tests using a causal graph derived entirely from the data using causal discovery algorithms. In fact, an interesting combination of these two extensions would be to use methods for causal discovery that can account for shifts in the distribution of the data (see, e.g., [ 41 ] and [ 44 ]). All of these we consider for future work.

Consider, for example, the recent wave of workers quitting their jobs during the pandemic due to burn-out. See “Quitting Your Job Never Looked So Fun” and “Why The 2021 ‘Turnover Tsunami’ Is Happening And What Business Leaders Can Do To Prepare” .

https://www.effectory.com

https://www.kaggle.com/pavansubhasht/ibm-hr-analytics-attrition-dataset

https://www.kaggle.com/c/sm/overview

Two items belonging to a same theme are highly correlated (convergence), whilst two items from different themes are almost uncorrelated (discrimination). See https://en.wikipedia.org/wiki/Construct_validity

As a consequence of construct validity, each item belongs to one and only one theme.

We also experimented with a dataset with both themes and items scores, whose predictive performances were close to the items dataset. This is not surprising, since a theme’s score is an aggregation over a subset of items.

https://scikit-learn.org/

https://xgboost.readthedocs.io/

https://lightgbm.readthedocs.io/

https://github.com/dreamquark-ai/tabnet

https://optuna.org/

Tests performed on a PC with Intel 8 cores-16 threads i7-6900K at 3.7 GHz, 128 Gb RAM, and Windows Server 2016 OS. Python version 3.8.5.

Notice that the implementations of DT and LR are single-threaded, while the ones of RF, XGB, LGBM, and TABNET are multi-threaded.

https://ec.europa.eu/eurostat/web/lfs/data/database

https://databank.worldbank.org/reports.aspx?source=2 &series=SL.TLF.CACT.NE.ZS

Given that we focus only on European countries, the attribute Continent is fixed and thus controlled for. We can exclude it from G .

For example, under the causal graph \({\mathcal {G}}\) , \(I \rightarrow W\) implies the causal relationships \(Age \rightarrow Industry\) , \(Gender \rightarrow Industry\) , \(Age \rightarrow Work \; Status\) , \(Gender \rightarrow Work \; Status\) , but not \(Age \rightarrow Gender\) nor \(Gender \rightarrow Age\) .

We recognize that this is a strong assumption, but the alternative would be to drop all themes except T and fit b on that subset of the data, which would have considerable risks of overestimating the effect of T on Y .

To use the proper causal terminology, all themes have the same parents (the incoming edges from the variables in I , G , and W ) and the same child ( Y ). No given theme is the parent or child of any other theme in \(\tau \) .

The clustered map adopts a hierarchical clustering. Therefore, groups can be identified at different levels of granularity.

This under the assumption that the model that is generating the predicted outcomes approximates the “true” relationship between the feature of interest and the outcome variable. This is way [ 63 ] emphasizes the importance of having a good performing model for applying this approach.

To be more precise, ( 2 ) is equivalent to the expectation over ( 1 ), which would allow us to rewrite ( 1 ) in terms of expectations rather than in terms of probabilities and thus formally derive the equivalence between the two.

Here, again, under the assumption that b approximates the “true” where \(b(T) \rightarrow {\hat{Y}}\) contains relevant information concerning \(T \rightarrow Y\) .

Under this assumption, we can apply a simple average.

For example, by estimating the (joint) probability of being a German worker who is also female and has a college degree.

It is common to use the PDP on the training dataset [ 43 , 63 ] and since we are not interested here in testing performance, we use the entire dataset for fitting the model.

Akiba, T., Sano, S., Yanase, T., Ohta, T., Koyama, M.: Optuna: A next-generation hyperparameter optimization framework. In: KDD, pp. 2623–2631. ACM (2019)

Alduayj, S.S., Rajpoot, K.: Predicting employee attrition using machine learning. In: IIT, pp. 93–98. IEEE (2018)

Allen, D.G., Hancock, J.I., Vardaman, J.M., Mckee, D.N.: Analytical mindsets in turnover research. J. Org Behav 35 (S1), S61–S86 (2014)

Article   Google Scholar  

Allen, D.G., Shanock, L.R.: Perceived organizational support and embeddedness as key mechanisms connecting socialization tactics to commitment and turnover among new employees. J. Org. Behav. 34 (3), 350–369 (2013)

Angrist, J.D., Pischke, J.S.: Mostly Harmless Econometrics. Princeton University Press (2008)

Arik, S.Ö., Pfister, T.: Tabnet: Attentive interpretable tabular learning. In: AAAI, pp. 6679–6687. AAAI Press (2021)

Borisov, V., Leemann, T., Seßler, K., Haug, J., Pawelczyk, M., Kasneci, G.: Deep neural networks and tabular data: A survey. CoRR abs/2110.01889 (2021)

Boyd, K., Eng, K.H., Jr., C.D.P.: Area under the precision-recall curve: Point estimates and confidence intervals. In: ECML/PKDD (3), LNCS , vol. 8190, pp. 451–466. Springer (2013)

Branco, P., Torgo, L., Ribeiro, R.P.: A survey of predictive modeling on imbalanced domains. ACM Comput. Surv. 49 (2), 31:1-50 (2016)

Breiman, L.: Random forests. Mach. Learn. 45 (1), 5–32 (2001)

Breiman, L., Friedman, J.H., Olshen, R.A., Stone, C.J.: Classification and Regression Trees. Wadsworth (1984)

Chen, T., Guestrin, C.: Xgboost: A scalable tree boosting system. In: KDD, pp. 785–794. ACM (2016)

Cohen, G., Blake, R.S., Goodman, D.: Does turnover intention matter? Evaluating the usefulness of turnover intention rate as a predictor of actual turnover rate. Rev. Pub. Person. Adm. 36 (3), 240–263 (2016)

Commission, E.: Joint employment report 2021. https://ec.europa.eu/social/BlobServlet?docId=23156 &langId=en (2021)

Davis, J., Goadrich, M.: The relationship between precision-recall and ROC curves. In: ICML, ACM International Conference Proceeding Series , vol. 148, pp. 233–240. ACM (2006)

Delgado, M.F., Cernadas, E., Barro, S., Amorim, D.G.: Do we need hundreds of classifiers to solve real world classification problems? J. Mach. Learn. Res. 15 (1), 3133–3181 (2014)

MathSciNet   MATH   Google Scholar  

Demsar, J.: Statistical comparisons of classifiers over multiple data sets. J. Mach. Learn. Res. 7 , 1–30 (2006)

DeVellis, R.F.: Scale development: Theory and applications. Sage (2016)

Dietterich, T.: Approximate statistical tests for comparing supervised classification learning algorithms. Neural Comput. 10 , 31895–1923 (1998)

Fallucchi, F., Coladangelo, M., Giuliano, R., Luca, E.W.D.: Predicting employee attrition using machine learning techniques. Computer 9 (4), 86 (2020)

Firth, L., Mellor, D.J., Moore, K.A., Loquet, C.: How can managers reduce employee intention to quit? J. Manag. Psychol. pp. 170–187 (2004)

Friedman, J.H.: Greedy function approximation: A gradient boosting machine. Ann. Stat. 29 (5), 1189–1232 (2001)

Article   MathSciNet   Google Scholar  

Friedman, J.H.: Stochastic gradient boosting. Comput. Stat. Data Anal. 38 (4), 367–378 (2002)

Gabrani, G., Kwatra, A.: Machine learning based predictive model for risk assessment of employee attrition. In: ICCSA (4), Lecture Notes in Computer Science , vol. 10963, pp. 189–201. Springer (2018)

Goodman, A., Mensch, J.M., Jay, M., French, K.E., Mitchell, M.F., Fritz, S.L.: Retention and attrition factors for female certified athletic trainers in the national collegiate athletic association division I football bowl subdivision setting. J. Athl. Train. 45 (3), 287–298 (2010)

Griffeth, R., Hom, P.: Retaining Valued Employees. Sage (2001)

Griffeth, R.W., Hom, P.W., Gaertner, S.: A meta-analysis of antecedents and correlates of employee turnover: Update, moderator tests, and research implications for the next millennium. J. Manag. 26 (3), 463–488 (2000)

Google Scholar  

Guidotti, R., Monreale, A., Ruggieri, S., Turini, F., Giannotti, F., Pedreschi, D.: A survey of methods for explaining black box models. ACM Comput. Surv. 51 (5), 93:1–93 (2019)

Hassan, S.: The importance of role clarification in workgroups: Effects on perceived role clarity, work satisfaction, and turnover rates. Public Adm. Rev. 73 (5), 716–725 (2013)

Heneman, H.G., Judge, T.A., Kammeyer-Mueller, J.: Staffing organizations, 9 edn. McGraw-Hill Higher Education (2018)

Hoffman, M., Tadelis, S.: People management skills, employee attrition, and manager rewards: An empirical analysis. J. Polit. Econ. 129 (1), 243–285 (2021)

Hollander, M., Wolfe, D.A., Chicken, E.: Nonparametric Statistical Methods, 3 edn. Wiley (2014)

Holtom, B.C., Mitchell, T.R., Lee, T.W., Eberly, M.B.: Turnover and retention research: a glance at the past, a closer review of the present, and a venture into the future. Acad. Manag. Ann. 2 (1), 231–274 (2008)

Hom, P., Lee, T., Shaw, J., Hausknecht, J.: One hundred years of employee turnover theory and research. J. Appl. Psychol. 102 , 530 (2017)

Hosmer, D.W., Lemeshow, S.: Applied Logistic Regression, 2 edn. Wiley (2000)

Jain, N., Tomar, A., Jana, P.K.: A novel scheme for employee churn problem using multi-attribute decision making approach and machine learning. J. Intell. Inf. Syst. 56 (2), 279–302 (2021)

Ke, G., Meng, Q., Finley, T., Wang, T., Chen, W., Ma, W., Ye, Q., Liu, T.: LightGBM: A highly efficient gradient boosting decision tree. In: NIPS, pp. 3146–3154 (2017)

Keilwagen, J., Grosse, I., Grau, J.: Area under precision-recall curves for weighted and unweighted data. PLoS ONE 9 (3), 1–13 (2014)

Kim, J.H.: Estimating classification error rate: repeated cross-validation, repeated hold-out and bootstrap. Comput. Stat. Data Anal. 53 (11), 3735–3745 (2009)

Kohavi, R.: A study of cross-validation and bootstrap for accuracy estimation and model selection. In: IJCAI, pp. 1137–1145. Morgan Kaufmann (1995)

Magliacane, S., van Ommen, T., Claassen, T., Bongers, S., Versteeg, P., Mooij, J.M.: Causal transfer learning. CoRR abs/1707.06422 (2017)

Mitchell, T.R., Holtom, B.C., Lee, T.W., Sablynski, C.J., Erez, M.: Why people stay: using job embeddedness to predict voluntary turnover. Acad. Manag. J. 44 (6), 1102–1121 (2001)

Molnar, C.: Interpretable Machine Learning (2019). https://christophm.github.io/interpretable-ml-book/

Mooij, J.M., Magliacane, S., Claassen, T.: Joint causal inference from multiple contexts. J. Mach. Learn. Res. 21 , 99:1–99:108 (2020)

Ngo-Henha, P.E.: A review of existing turnover intention theories. Int J Econ. Manag. Eng. 11 , 2760–2767 (2017)

Nijjer, S., Raj, S.: Predictive analytics in human resource management: a hands-on approach. Routledge India (2020)

Pearl, J.: Causality, 2 edn. Cambridge University Press (2009)

Price, J.L.: Reflections on the determinants of voluntary turnover. Int. J. Manpower 22 (7), 600–624 (2001)

Sahakyan, M., Aung, Z., Rahwan, T.: Explainable artificial intelligence for tabular data: A survey. IEEE Access 9 , 135392–135422 (2021)

Salzberger, T., Koller, M.: The direction of the response scale matters - accounting for the unit of measurement. Eur. J. Mark. 53 (5), 871–891 (2019)

Sato, T., Rehmsmeier, M.: Precision-recall plot is more informative than the ROC plot when evaluating binary classifiers on imbalanced datasets. PLoS ONE 10 (3), e0118432 (2015)

Schuurmans, J., Frasincar, F., Cambria, E.: Intent classification for dialogue utterances. IEEE Intell. Syst. 35 (1), 82–88 (2020)

Seidl, T.: Nearest neighbor classification. In: Encyclopedia of Database Systems, pp. 1885–1890. Springer (2009)

Shu, K., Mukherjee, S., Zheng, G., Awadallah, A.H., Shokouhi, M., Dumais, S.T.: Learning with weak supervision for email intent detection. In: SIGIR, pp. 1051–1060. ACM (2020)

Simmons, R.G., Browning, B., Zhang, Y., Sadekar, V.: Learning to predict driver route and destination intent. In: ITSC, pp. 127–132. IEEE (2006)

Sousa-Poza, A., Henneberger, F.: Analyzing job mobility with job turnover intentions: an international comparative study. J. Econ. Issues 38 (1), 113–137 (2004)

Tanova, C., Holtom, B.C.: Using job embeddedness factors to explain voluntary turnover in four European countries. Int. J. Human Res. Manag. 19 , 1553–1568 (2008)

Wang, S., Hu, L., Wang, Y., Sheng, Q.Z., Orgun, M.A., Cao, L.: Intention nets: Psychology-inspired user choice behavior modeling for next-basket prediction. In: AAAI, pp. 6259–6266. AAAI Press (2020)

Wang, S., Hu, L., Wang, Y., Sheng, Q.Z., Orgun, M.A., Cao, L.: Intention2basket: A neural intention-driven approach for dynamic next-basket planning. In: IJCAI, pp. 2333–2339. ijcai.org (2020)

William Lee, T., Burch, T.C., Mitchell, T.R.: The story of why we stay: A review of job embeddedness. Annu. Rev. Organ. Psych. Organ. Behav. 1 (1), 199–216 (2014)

Wunder, R.S., Dougherty, T.W., Welsh, M.A.: A casual model of role stress and employee turnover. In: Academy of Management Proceedings, vol. 1982, pp. 297–301 (1982)

Wynen, J., Dooren, W.V., Mattijs, J., Deschamps, C.: Linking turnover to organizational performance: the role of process conformance. Public Manag. Rev. 21 (5), 669–685 (2019)

Zhao, Q., Hastie, T.: Causal interpretations of black-box models. J. Bus. Econ. Stat. 39 (1), 272–281 (2021)

Download references

Acknowledgements

The work of J. M. Alvarez and S. Ruggieri has received funding from the European Union’s Horizon 2020 research and innovation programme under Marie Sklodowska-Curie Actions (grant agreement number 860630) for the project “NoBIAS - Artificial Intelligence without Bias”. This work reflects only the authors’ views and the European Research Executive Agency (REA) is not responsible for any use that may be made of the information it contains.

Open access funding provided by Università di Pisa within the CRUI-CARE Agreement.

Author information

Authors and affiliations.

Effectory Global, Amsterdam, Netherlands

Matilde Lazzari

Scuola Normale Superiore, Pisa, Italy

Jose M. Alvarez

University of Pisa, Pisa, Italy

Jose M. Alvarez & Salvatore Ruggieri

You can also search for this author in PubMed   Google Scholar

Corresponding author

Correspondence to Salvatore Ruggieri .

Ethics declarations

Conflict of interest.

M. Lazzari declares that she is an employee of Effectory Global. J. M. Alvarez and S. Ruggieri declare that they have no conflict of interest.

Additional information

Publisher's note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ .

Reprints and permissions

About this article

Lazzari, M., Alvarez, J.M. & Ruggieri, S. Predicting and explaining employee turnover intention. Int J Data Sci Anal 14 , 279–292 (2022). https://doi.org/10.1007/s41060-022-00329-w

Download citation

Received : 11 July 2021

Accepted : 15 April 2022

Published : 23 May 2022

Issue Date : September 2022

DOI : https://doi.org/10.1007/s41060-022-00329-w

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Employee turnover
• Predictive models
• EXplainable AI (XAI)
• Structural causal models
• Find a journal
• Publish with us
• Track your research

An official website of the United States government

The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

• Publications
• Account settings

Preview improvements coming to the PMC website in October 2024. Learn More or Try it out now .

• Advanced Search
• Journal List

Factors impacting employee turnover intentions among professionals in Sri Lankan startups

Lakshmi Kanchana

1 SLIIT Business School, Sri Lanka Institute of Information Technology, Malabe, Sri Lanka

2 Ceyentra Technologies, Panadura, Sri Lanka

Ruwan Jayathilaka

3 Department of Information Management, SLIIT Business School, Sri Lanka Institute of Information Technology, Malabe, Sri Lanka

Associated Data

All relevant data are within the paper and its Supporting Information files ( S2 Appendix . Data File).

Employee turnover is one of the topical issues worldwide. The impact of factors affecting employee turnover varies occasionally and new factors are considered. Many countries have examined various factors that affect employee turnover. The main objective of this research is to consider psychographics and socio-demographic factors in one study and analyse the impact on employee turnover. A Probit regression model through the stepwise technique was used to analyse the collected data. Using ventures in Sri Lanka as a case study, this study demonstrates that employee turnover occurs in different stages and independent factors impact differently in each stage. The study population was professionals who have been a key part of Sri Lankan startups, which involved 230 respondents. Data analysis was performed through a forward stepwise technique through STATA. The results verified that job satisfaction and co-worker support negatively impact employee turnover, whereas leader member exchange positively impacts employee turnover. This study also proved a significant positive relationship between male employees in their thirties and high employee turnover. This study’s findings help to identify the areas management should focus on to minimise employee turnover to retain experienced and skilled employees.

Introduction

Having the right combination of human resources/employees can assist firms to be effective in driving change, boosting business performance, as well as to achieving and sustaining a competitive edge. Companies need to give high priority to employee development and predict employee behaviour [ 1 ]. Organisations spend more time and take much effort to identify the good fit employees for the company. Companies invest in many ways for employees, as they are one of the organisation’s valuable assets [ 2 ]. Organisations conduct workshops for employees, buy online tutorials, evaluate employee performance, and provide feedback to them, which are some common types of investments in human resources. These processes sharpen employees’ skills and capabilities, directly affecting the organisation’s success. However, some organisations are weak in strategy adoption while not focusing constantly on these processes or employee voice. As such, these employees suddenly quit the company resulting in increased employee turnover. The issue of employee turnover is considered as one of the global obstacles for organisations worldwide, which directly and adversely affects strategic plans and opportunities of gaining competitive advantages [ 3 ]. As such, this issue can have massive effects on a company’s performance, especially for new businesses and startups. Therefore, it is essential to identify the factors that affect employee retention, which is also a topical issue worldwide. This type of approach enables businesses to achieve its strategic goals while retaining satisfied and skilful employees.

Many variables influence employee turnover intentions [ 4 – 6 ]. Previous studies imply that job satisfaction, work-life balance, trust, and management support are the critical factors that impact employee retention [ 7 – 9 ]. Further, promoting employee well-being leads to decrease employee turnover [ 10 ]. Providing psychological and social support through counselling promotes the quality of work-life [ 11 ]. With time, newly considered factors such as leader member exchange, workplace culture, happiness, joy in the workplace, career management, innovative work behaviour and employee delight are equally important and have been identified. As such, it is important to focus on these factors and build relationships between employees and the organisation.

Firm performance reflects the ability of an organisation to use its human resources and other material resources to achieve its goals and objectives. Firm performance belongs to the economic category, and it should consider the use of business means efficiently during the production and consumption process [ 12 ]. Employee retention is defined as encouraging employees to remain in the organisation for a long period or the organisation’s ability to minimised employee turnover [ 13 ]. Turnover intention is the intention of the employee to change the job or organisation voluntarily [ 14 ].

Sri Lankan business firms were chosen as a case study to examine this resarch problem. In Sri Lanka, over 1 million (Mn) businesses operate. By 2018, 10,510 new businesses had been registered in Sri Lanka. Among these companies, startup companies play a key role in the Sri Lankan economy. Startups come up with radical innovations and changes, and these disrupt the existing market with new products and services. Furthermore, Sri Lanka has a middle rank of ease of doing business. With these favourable conditions and educational and family backgrounds, many people like to apply their new idea and fill the market gap. The new generation in Sri Lanka are interested/are keen on innovations at work and being a part of unique products or services. Currently, most startups are technology-driven and do not have geographical limitations.

Startups are expanding day by day. These businesses are in different stages as ideation, traction, break-even, profit, scaling and stable. According to the “Sri Lanka Startup Report 2019” issued by PricewaterhouseCoopers (PWC), “55% of startups responded are in the growing revenue or expansion stage, 29% of respondents reported an annual revenue of more than Sri Lankan Rupees (LKR) 10 Mn, 40% are still in the less than LKR 1 Mn revenue category and 61% of respondents reported being profitable”. In this setting, employee turnover can be a setback for most startups yet to reach business stability.

Most startups are relatively new. According to (PWC) [ 15 ], 36% of the businesses have operated for less than a year, 44% have been in operation for 1–3 years and only 8% have operated for more than five years. These are still growing and in the early stages of executing their strategies. In this situation, most companies are willing to expand their staff strength. PricewaterhouseCoopers [ 15 ] evidenced that 82% of companies were willing to do so in the next year.

Studies conducted in Asian countries on this subject are assumably similar to the situation of Sri Lanka [ 4 , 5 , 16 ]. This study aims to create a model with critical and newly identified independent factors (job satisfaction, work-life balance, happiness, management support, career management, innovative work behaviour, leader member exchange, and co-worker support) influencing employee turnover in Sri Lankan startups.

Based on their knowledge and the existing literature, authors have considered widely used factors to investigate the employee turnover issue. Therefore, job satisfaction, happiness, work-life balance, career management, management support, innovative work behaviour, leader member exchange and co-worker support were selected based on previous literature findings [ 4 – 6 , 8 , 17 – 19 ]. As in the previous papers and along with the current study’s results, authors identified both positive and negative impacts on employee turnover among Sri Lankan startups.

This study aims to analyse the impact of job satisfaction, happiness, work-life balance, career management, management support, innovative work behaviour, leader member exchange, and co-worker support on employee turnover in startups in Sri Lanka. The present study’s scientific value can be elaborated by comparing it with previous studies. This study’s contribution can be explained in five ways. Firstly, the most critical and newly considered factors were identified together with the support of past literature. Secondly, the present study was classified into different levels of employee turnover. As such, by considering the various levels, the micro-level changes and probabilities of the impact on employee turnover can be better identified. Further, this study helps to reduce the methodological gap. Thirdly, the Sri Lankan context has been selected as the case study. This is because, to the best of the authors’ knowledge, there was no previous research done by local researchers that includes all the widely measured variables investigating the combined effect on employee turnover. Fourthly, the analysis results can be used to identify the strengths and weaknesses of startups in Sri Lanka. Finally, this study identifies the challenges faced by startups and identifies how policy modifications can strengthen the startup ecosystem.

The upcoming sections of this paper are structured as follows. Section 2 discusses the literature review, and section 3 explains data and methodology, Section 4 contains results and discussion highlighting how the research objectives are achieved. Section 5 marks the conclusion, with implications, research limitations and future research directions.

Literature review

As employee turnover is one of the most critical indicators for an organisation, many studies have been conducted on this topic with dissimilar demographical and geographical samples. The existing literature adds theoretical or methodological improvements to this topic. Accoridngly, this study included most variables that significantly impact employee turnover, summarising the independent variables that affect employee retention.

This study is based on the initially defined 47 journal articles through advanced filtration. Reputed journal databases, such as Emerald insight, Science Direct, Taylor & Francis, SAGE journals, ResearchGate, Sabinet, IEEE Xplore and Google Scholar were referred. Fig 1 below describes the literature search flow. Thirteen articles were excluded due to overlapping, insufficient information and irrelevant to the topic. The selected articles have been sorted according to the independent variables.

Source: Based on authors’ observations.

S4 Appendix contains the literature summary of the above presented literature search flow diagram. The following sections present the details of each category.

Job satisfaction

Job satisfaction refers to the employee’s positive emotions, feeling and attitudes on the job and workplace. Positive emotional experiences directly affect higher job satisfaction [ 7 ]. Kim, Knutson [ 7 ] found that satisfaction significantly affects employee turnover regardless of the generation of the employee. Gen Y employees do not easily build loyalty toward the organisation unlike older employees. Turnover intentions seem significantly higher in new generations compared to older generations. New generations are impatient with their organisation and older generations are more patient with it. However, even the new generation of employees tends to stay in their organisation if their level of satisfaction is acceptable. They found that newer the generation of employee, satisfaction level and loyalty is lower than the older generation. This shows that employee turnover is higher in newer generations. Feedback obtained from most employees in generations Y and Z in startups supports this finding.

Da Camara, Dulewicz [ 20 ] found that organisational emotional intelligence has a larger effect on employee satisfaction. Further, this study has discovered that organisational emotional intelligence helped improve job satisfaction and commitment, which reduced turnover intentions significantly. However, organisational commitment and satisfaction describe only 19% of the total intention to leave. Moreover, the descriptive statistics found a high level of job satisfaction and the intention to leave was at the mid or average level of the scale. Camara further stated that job satisfaction clearly implies the feeling about their job. But some research findings can be contradictory. Some employees are fully satisfied with the job and still want to leave the organisation for various reasons. However, this research focused only on charity workers. As such, it is important to gather many indicators that affect employee turnover and thereafter, one can analyse the real situation and generalise the findings.

Satisfaction also depends on the number of employees at the same level. When it gets higher, job satisfaction increases and reduces the intention to leave [ 8 ]. This study found that female employees are more satisfied with their jobs, while older employees are more likely to leave the organisation. However, this study focused only on online-level employees and supervisors.

Oosthuizen, Coetze and Munro studied the relationship between job satisfaction and turnover intention in the IT industry. Oosthuizen, Coetzee [ 6 ] revealed that job satisfaction significantly predicted employee turnover. The study also found that the work-home life balance has a major effect on job satisfaction. Predicting turnover intention based on overall work-life balance is a tough task. The findings further proved that white employees show less job satisfaction compared to black employees. However, they didn’t observe any significant interaction between overall work-life balance and job satisfaction in predicting employee turnover intention. With these results, this indicator must be examined further.

Considering the Asian context, Pakistan IT professionals’ turnover intentions were studied in a similar research [ 21 ]. Recruitment & section, team & management support, performance & career management, salary & compensation, employee commitment, job security, recognition, organisational demographics, and personal demographics have an effect on job satisfaction. However, this study suggested adding more factors, such as work-life balance and employee engagement, which may significantly impact employee retention. This means that human resource management has a significant influence on job satisfaction.

The study by Zeffane and Bani Melhem [ 22 ] investigated the turnover intention of public and private sector employees in the United Arab Emirates. Here, the researchers revealed that government employees are more satisfied with their job and are most unlikely to leave than private sector employees. The turnover intentions of private sector employees are not significantly affected by job satisfaction, whereas the public sector is almost affected by it. Kaur and Randhawa [ 16 ] examined the turnover intention of Indian private school teachers. It revealed that job satisfaction has a direct link with the civil status of the teachers, explaining that married teachers tend to have less job satisfaction. However, for unmarried teachers, there is more intention to leave organisations. Supervisor’s influence had indirect impacts on turnover intentions. However, this research limited the sample to private school female teachers. Here, the study highlighted the importance of having more influencing variables on employee retention and recommended considering these for a comprehensive analysis. Only then the model can be near to the real situation.

Thomas A. Wright [ 2 ] discovered that the employee’s well-being moderates the relationship between satisfaction and turnover intention. Satisfaction had a strong negative relationship with turnover intention, while well-being remained low. The study by Nae and Choi [ 23 ] evidenced the direct relationship between job satisfaction and employee turnover. However, this also pointed out that employee well-being moderates the indirect relationship between job satisfaction and turnover. However, this moderator was significant only for a few specified occasions, such as employees having a highly secure attachment, and low counter-dependent and over-dependent attachment styles.

As per the literature, job satisfaction is an important factor in determining the impact on employee turnover. Accordingly, hypothesis one has been developed.

Work-life balance

Work-life balance can be identified as the satisfactory co-existing of an employee’s work-life and personal life. On one hand. this led to a positive influence on both employees and the organisation. On the other hand, negative work-life-balance has harmful effects on employees. Most employees had abuse alcohol due to this issue in the hospitality industry, which indirectly influences the organisation’s productivity. Additionally, most women have suffered from depression due to poor work-life balance in the hospitality industry. Besides, burnout, exhaustion, and stress are common among employees with poor work-life balance. Therefore, the employee’s commitment heavily depends on work-life balance, an essential requirement for employee retention [ 24 ]. This study states that it can be developed by adding more independent variables such as commitment and job satisfaction.

The highly negative work-life interference has amplified the turnover intentions of IT employees in Pakistan. They also found that the organisation that invested heavily in creating proper work-life balance recorded the lowest turnover among other organisations in the IT industry in Pakistan. Oosthuizen, Coetzee [ 6 ] revealed that the overall work-life balance had no clear influence on the satisfaction of an employee’s current job. Gender was a primary separation point of work-life balance variation among employees. Female employees looked more satisfied with their work-life balance than male employees [ 6 ]. In this light, work-life balance is one part of quality work life other than career opportunities and job characteristics. Organisational embeddedness has a positive and strong relationship with work-life balance. Positive work-life balance has a negative relationship with turnover intention [ 25 ]. However, the sample of this research was based on two healthcare firms. Since the whole world is tech-driven, it is realistic to focus on the IT industry too for generalisability of findings.

According to this study, superiors’ influence on work-life balance highly impacts job satisfaction. Supportiveness and the supervisor’s flexibility on subordinates’ help achieve the desired work-life balance for employees. As noted before, the employee turnover intention is heavily dependent on work-life balance. As such, a study on work-life balance can predict the turnover intention of an employee accurately compared to other factors. Work-life balance can be measured and categorised into three. Interference of work on personal life, work and family conflict and facilitation of work and family are those categories that the researcher suggested. The sample for the study of Kaur and Randhawa [ 16 ] was Indian private school teachers. The researcher suggested that formulating teacher-friendly policies to enhance work-life balance will reduce teachers’ turnover intentions. The researcher also suggested that the imbalance workload on employees supports increasing employee turnover intentions. However, most of the employees in this study were females.

Organisations that focused on employees’ proper work-life balance have recorded better efficiency, innovation, and talent retention [ 26 ]. Employee engagement and life satisfaction have been significantly mediated by the work-life balance of restaurant employees in Nevada, USA [ 27 ]. However, there are not sufficient recent researchers in Sri Lanka on work-life balance and employee retention. Therefore, taking up this study as an opportunity to research is essential. According to the above literature, hypothesis two has been constructed; work-life balance has a negative impact on employee turnover.

Employee happiness is a psychological feeling they have with the workplace. This is an essential factor in maintaining a successful and profitable organisation.Wright and Cropanzano [ 17 ] described happiness as phycological well-being. Personal well-being is one better way to explain employee retention. By moderating this factor, firms can achieve better employee turnover.

The workplace must be a source of happiness for employees. Unhappy employees in a workplace tend to increase employee turnover, absenteeism, low productivity, and time wasted deadlines. Creating happiness within the workplace is not a simple process. It is a comprehensive and continuous process. Happy employees generally have a fair idea of the organisation’s vision, mission and values. Employees in each department should have a clear idea about their goals [ 28 ]. However, happy employees are not always productive. But they can guide and explore things without organisations forcing them. Those employees required proper career management and support to be productive.

The workplace’s physical environment plays a major role in employee happiness and cheerfulness and friendliness of the physical environment are fundamentals. Employee’s attitude also has a more significant effect on happiness. Gratitude, appreciation, servant leadership from the organisation, hope and interpersonal connection are the main factors that affect the employee’s positive attitude. Humour, fun and games also play a major role in keeping employees happy. Other than those factors, wellness activities, celebrations and compensation are the minor factors affecting employee happiness. Based on the above cited literature, hypothesis three can be developed; employee happiness has a negative impact on employee turnover.

Management support

Management support is a must in the move from a good to a great company. Management stands by employees and supports them mentally and physically. Van den Heuvel, Freese [ 29 ] conducted research from the data of 699 employees at three divisions within the Dutch subsidiary of a multinational organisation. Management increased employee autonomy by supporting them to work from anywhere at any hour. This positively affected employee engagement and was negatively related to employee retention. Trust in management is a critical factor in employee turnover.

A cross-sectional survey has been conducted for front-line healthcare staff in China by Li, Mohamed [ 30 ] to measure the impact of organisational support on employee turnover intention. This study’s results could verify that organisational support negatively affected employee turnover intention. Saoula and Johari [ 31 ] studied this area and determined a negative relationship between organisational support and employee turnover intention. As both of the above explained research have been conducted in non-Western countries, the findings help to complete the theoretical framework for the current study in the Sri Lankan context.

Wong and Wong [ 5 ] researched the world’s most populous county, China, to identify the relationship between perceived organisational support and employee turnover. The findings suggested that trust, job security and distributive justice negatively impact employee turnover. However, China is an Asian country, and these similarities may apply to specific research findings in the Sri Lankan context.

Employee perception of management support for employee health is a factor in employee retention. Xiu, Dauner [ 32 ] studied this area with employees’ data from a public university, which was the first empirical examination of organisational support for employee health and retention. This kind of approach leads to building trust with employees. Moreover, these findings are essential to human resource managers who are willing to promote employee well-being at the workplace. Hypothesis four has been developed based on above discussed literature.

Career management

Initiatives must carry out different strategies for old and young employees because their priorities are different. Digest [ 18 ] discloses that young employees are impressed by flexible working opportunities, career advancement, positive working relationships and inclusive management forms. Young employees are more likely to be talented, leading to an organisation’s success and they can also become key workers in the company.

Saoula and Johari [ 31 ] researched the effect of personality traits (big five) on employee turnover intention. The researchers state that the relationship between the big five personality traits and turnover intention will support early prediction of employee turnover intentions. Identifying employee’s personalities and helping them to find the most suitable job role is a long-term process, though it will be highly advantageous for both employees and the organisation.

Rawashdeh and Tamimi [ 33 ] focused on the latest management developments of leading organisations worldwide. They state that there is a strong relationship between the availability of training and supervisor support for training and organisational commitment. Further, they proved that there is a strong negative association between organisational commitment and employee retention. These research findings verify the social exchange theory [ 34 ]. However, the research suggested that the above study can improve by adding new factors like motivation and co-worker support for training. Hypothesis five has been developed by concluding the above explained literature.

Innovative work behaviour

Innovative behaviour is a leading factor in gaining a competitive advantage. Shih, Posthuma [ 35 ] investigated the negative impacts of innovative work behaviour on employee turnover and conflict with co-workers. According to the studies, there is a positive relationship between innovative work behaviour and employee turnover. Further, it found that perceived distributive fairness can negatively moderate this relationship. However, the writer has suggested extending the research to different geographical locations and industries.

The organisational learning culture is a key factor for innovative work behaviour. Saoula, Fareed [ 36 ] conducted research in Malayasia, a developing country in Asia to examine the relationship between organisational learning culture and employee turnover intention. The organisational learning culture improves learning capability, supports sustainable development, and affects organisation’s positive changes. As organisational learning culture and employee turnover intention have a negative relationship, the result helps to identify the impact of innovative work behaviour. According to the existing literature, limited studies have been conducted on this topic.

Agarwal, Datta [ 4 ] conducted research with managerial employees in India to examine the relationship between innovative work behaviour and employee turnover. This study asserted that the variables have an inverse relationship. As innovative work behaviour examinations in an Asian county country like India, it is important to consider this variable in this model. With the presence of the above mentioned literature, hypothesis six has been formulated.

Leader member exchange

As per many leadership methods, leader member exchange depends on the leadership style. Tobias M. Huning [ 37 ] conducted research to identify the effect of servant leadership on employee turnover. Servant leadership supports employee empowerment, development, interpersonal acceptance, and courage. This study found that servant leadership negatively impacts employee turnover. However, this leadership style does not directly affect employee retention. Gyensare, Kumedzro [ 38 ] studied the impact of transformational leadership on employee turnover. This type of leadership supports work engagement of the employee, and it negatively relates to employee retention. Considering both aspects, the study found that increasing work engagement is vital to curtail employee retention.

Leader support is an indirect factor in employee retention. According to the studies, employee engagement and work-life balance act as mediation for perceived supervisor support and employee turnover relationship [ 16 ]. The supervisor supports the career success of employees and it affects both directly and indirectly the career success of the employee and retention one year later [ 9 ]. Therefore, this study shows that co-worker support has a significantly positive impact on employee turnover. However, these results maintained the diversity of the sample. As this has been examined in India, a South Asian country, the same results can apply to the Sri Lankan context. Based on the above-mentioned literature, hypothesis seven has been developed; leader member exchange has a negative impact on employee turnover.

Co-worker support

Co-worker support will be in both formal and informal ways and in two different forms, emotional support, and instrumental support. The support of co-workers enhances the confidence level of the employee. Further, it helps to accept challenges in the work environment. Kmieciak [ 19 ] has worked on research to identify the effect of co-worker support on employee retention in the IT industry. However, a significantly negative impact was not evident on co-worker support. As this is a recently published research paper, the results are more valuable to the current research. The researcher has investigated more about the impact of subordinates’ support. Here, the analysis has been done only with 118 employees from a Polish software company. Considering the above limitations enables researchers to further study this topic with a larger sample size for generalisability of findings.

Abugre and Acquaah [ 39 ] researched in Ghana to identify the relationship between co-worker relationships and employee retention. The findings of this research imply that co-worker support is negatively associated with employee turnover. It further stated that cynicism of the employee is positively associated with employee turnover. The speciality of this research is identifying the importance of encouraging co-worker support rather than employee cynicism. These newly published research results can be used along with all other variables that affect employee turnover. According to the above literature, hypothesis eight has been constructed.

These studies have a common limitation in gathering more independent variables and analysing the impact. Therefore, a need exists to measure the effect of job satisfaction, work-life balance, happiness, management support, career management, innovative work behaviour, leader member exchange, and co-worker support together on employee turnover.

In Sri Lanka, no research has so far considered all eight factors affecting employee turnover in one study. With the above-mentioned literature findings, this study assists the government in identifying the impact of every factor on employee turnover in startups in Sri Lanka.

Data and methodology

This study was reviewed and approved by Sri Lanka Institute of Information Technology Business School and the Sri Lanka Institute of Information Technology ethical review board. Data were collected through a questionnaire using both online and manual channels. Each individual in this study gave verbal consent prior to the formal interview. The data was collected from August to September 2022 ( S1 Appendix ). The authors directly distributed the questionnaire. Moreover, authors could contact management in startups and distribute the questionnaire in their organisation. The questionnaire is composed of ten (10) sections. The first part of the questionnaire was designed to collect the demographic characteristics of the correspondents. The second to ninth sections focused on independent variables, job satisfaction, work-life balance, happiness, management support, career management, innovative work behaviour, leader member exchange, and co-worker support. Finally, the tenth section was designed to identify employee turnover indicator. A minimum of four questions was added under each indicator. The researchers facilitated anonymously answering all the questions in the questionnaire. The participants should be a part of startup and he/she should consider the behaviour and culture of that startup when answering the questions. All nine indicators were covered by Likert scale questions from 1 to 5 rating scale, depicting (1) strongly disagree to (5) strongly agree to collect respondents’ attitudes and opinions. Each respondent took about 10–15 minutes to complete answering the questionnaire and took approximately 5–7 minutes to fill out the questionnaire. Furthermore, the average values were calculated to measure the value given by respondents for each indicator. The data file used for the study is presented in S2 Appendix .

PricewaterhouseCoopers [ 15 ] statistics determined the study’s population and it explained the total number of elements to be focused on in this study. The researchers applied a random sampling method, mainly employees who are a part of or have been a part of the startup. This sampling technique was appropriate because it was free of bias. The sample size was selected by referencing the Krejcie and Morgan sampling table and Calculator.net [ 40 ] with a confidence level of 95% and 7% of margin of error. The calculation results indicated a minimum of 171 professionals. A stepwise ordered probit analysis method was used as the selected variables are widely used indicators for employee turnover; therefore, a micro-level analysis was required to study how these variables impact. A pilot survey was conducted to identify whether the purpose of the questions was clear to the respondents.

The data used for the estimation include 83 low employee turnover, 79 moderate employee turnover and 68 high employee turnovers of employees in Sri Lankan startups. The initial estimation results are presented in Table 1 .

Source: Authors’ compilation based on survey data.

The mean values of all independent variables are greater than 2.5. Respondents were further grouped as per demographic and geographic characteristics. The respondents’ gender identity ratio is nearly 1: 2. When considering the age groups, most are in 20–30 years. Many employees in startup companies are in their twenties and are graduates. The respondents represent all the districts in Sri Lanka, most of which are from Kalutara, Colombo, Galle and Matara districts.

Research framework and hypothesis

The conceptual framework was developed with the literature review and existing knowledge, as illustrated in Fig 2 . This model was developed with the combination of eight hypotheses. These independent variables have been identified as critical factors that impact employee turnover.

Source: Authors’ compilation.

The following hypotheses have been developed in line with the research framework.

• Hypothesis 1 : Job satisfaction has a negative impact on employee turnover in startups in Sri Lanka.
• Hypothesis 2 : Work-life balance has a negative impact on employee turnover in startups in Sri Lanka.
• Hypothesis 3 : Happiness has a negative impact on employee turnover in startups in Sri Lanka.
• Hypothesis 4 : Management support has a negative impact on employee turnover in startups in Sri Lanka.
• Hypothesis 5 : Career management has a negative impact on employee turnover in startups in Sri Lanka.
• Hypothesis 6 : Innovative work behaviour has an impact on employee turnover in startups in Sri Lanka.
• Hypothesis 7 : Leader member exchange has an impact on employee turnover in startups in Sri Lanka.
• Hypothesis 8 : Co-worker support has a negative impact on employee turnover in startups in Sri Lanka.

Methodology

This study focuses on the demographical variables that affect employee turnover. For this, the present study’s authors considered employee feedback concerning Sri Lankan startups. The ordered probit regression determines the significant variables [ 41 ]. The probit model is an estimation technique for equations with dummy dependent variables that avoids the unboundedness problem of the linear probability model by using a variant of the cumulative normal distribution [ 42 ]. Further, this study examines the likelihood of three types of employee turnover. Accordingly, employee turnover is divided into three categories, considering the equality of data for each category based on employee turnover.

• Group 1 (y = 1): low = mean value of the employee turnover less than 1.50
• Group 2 (y = 2): moderate = mean value of the employee turnover greater than 1.5 and less than or equal to 2.25
• Group 3 (y = 3): high = mean value of the employee turnover greater than 2.25 and less than or equal to 5

The following equation represents the general form of the ordered probit model.

The y i value represents i th value of the dependent variable, employee turnover and x i represents the i th common independent variable. The β value is a vector parameter and ℇ i considered as the normally distributed random error term with a zero mean. The following ordered probit model has been developed by detailing the general equation.

Table 2 indicates the variables explained in previous literature and definitions of the previously mentioned equation that affects employee retention. The forward stepwise regression model has been used to analyse the data set.

Results and discussions

It is mandatory to test the internal consistency reliability before data analysis. The most common measure of reliability is Cronbach’s alpha (α) value, which determines whether the internal instruments are constant [ 43 ]. The reliability results for each indicator are presented in Table 3 . As all the Cronbach alpha values are greater than 0.6 scale reliability coefficients, all variables in this study are acceptable.

Analytical sample (N = 230)

Source: Authors’ calculation based on survey data

In the first step, the initial ordered probit model was executed, and this model explained 73% of the variation in employee retention by the variation in independent variables. S3 Appendix contains the table of the initial ordered probit regression model. The ordered probit model forwarded with the forward stepwise technique to identify the exact number of variables that impact employee turnover. A forward stepwise technique was adopted for the variable selection in each specification. Here, the new variables for selection were considered with a p-value < 0.20 and the previously selected variable for removal with a p-value ≥ 0.25. Three different model diagnostic criteria were considered in assessing the reliability of the results. The forward stepwise methodology suggested that the significance of the existing variables could be increased by adding more variables to the model. Marginal effects were separately calculated for low, moderate, and high employee turnover. Table 4 presents the final estimation results of the ordered probit model and illustrates the substantive effects of the independent variables. Here, 71.74% of the variation in employee turnover is explained by the variation in job satisfaction, LMX and co-worker support, considering the sample size and independent variables.

*** significant at the 1% level

** significant at the 5% level and * significant at the 10% level.

Source: Authors’ calculation based on surveying data.

Looking at the signs of the marginal effects in Table 4 , overall, high employee turnover is negatively associated with job satisfaction, co-worker support, and innovative work behaviour, whereas high employee turnover is positively associated with leader member exchange.

To control for the potential effect on different levels of employee turnover, the age factor was also included in the model, the coefficient of which implies that high employee turnover is 0.20 points and 0.60 points for the 20–30 years age range and 31–40 years age range, respectively. Employee turnover in 31–40 years age range employees is higher than that of other age ranges.

The marginal effects of the psychographic variables reveal that a 1% increase in job satisfaction increases the probability of low employee turnover by 0.47 percentage points. Similarly, 1% increase in job satisfaction decreases the probability for high employee turnover by 0.43 percentage points. With this observation, it can be stated that improving job satisfaction will highly affect to reduce high employee turnover. These results verify the existing statements indicating that job satisfaction has the highest significant and negative estimate value.

The estimated marginal effect of low employee turnover is 0.47 percentage points higher for employees in Sri Lankan Startups with a 1% increase in leader member exchange. High employee turnover is associated with leader member exchange increasing probability by 0.43. However, this study reflects similar findings to those of Tymon, Stumpf [ 9 ]. The reason behind the positive relation is employees learn fast and get qualified with the support of their leaders and then quit the company within the next few years.

Both leader member exchange and co-worker support are significant at the 99% level of employee turnover in the Sri Lankan context. When considering the independent variables for employee turnover in startups in Sri Lanka, co-worker support is a critical factor in determining the level of employee turnover. The 1% increase in co-worker support will also increase the probability of low employee turnover by 0.40 percentage points. But concurrently, change in co-worker support will negatively impact high employee turnover. The results ensure that encouraging co-worker support is crucial rather than employee cynicism.

Innovative work behaviour is one of the most critical factors in employee turnover. With a 1% increase in innovative work behaviour, the estimated marginal effect of high employee turnover is 0.27 percentage points lower for employees in Sri Lankan startups. The results of Shih, Posthuma [ 35 ] indicate a positive relationship exists between innovative work behaviour and employee turnover. However, this study concludes by emphasising the importance of retaining the innovative employees to remain competitive in the industry. For this, startups need to improve and enhance employees’ innovative behaviour and, concurrently, to prevent such employee retention.

Entrepreneurs are the founders of startups. Employees’ entrepreneurial dreams positively affect employee intention to startups. Employees in the startups also will have an ideation to start their own business. According to the study by Li, Li [ 44 ] the mediating role of employees’ entrepreneurial self-efficacy and the moderating role of job embeddedness in the influence of entrepreneurial dreams on employees’ turnover intention to startup.

The main objective of this research is to analyse the impact of critical and newly identified factors on employee turnover in one study. This issue occurs when employees leave the company by giving short notice or quitting unexpectedly. The analysis found that gender and age impact employee turnover in startups in Sri Lanka. In startups, many employees are in the 20 to 30 years age range. Employees between 31 and 40 years show a higher tendency to leave the startups. In Sri Lanka, only 8% of startups have been in operation for more than five years [ 15 ], indicating that the businesses are not stabilised and are still in its early stages. To prevent employee turnover, startups must improve employee job satisfaction. As per the findings, increasing job satisfaction has a significant impact on reducing employee turnover. For most employees in startups, it is their first job. During this time, employees gain work experience and become experts in the field. The leaders allocate much time to train their human resources and the company should gain strategic benefits from this investment. The results of the study prove that leader member exchange has a positive impact on employee turnover, as verified by Tymon, Stumpf [ 9 ] too about this relationship. To overcome this situation, as managers, it is vital to discuss with employees about their career paths, employee interests and company’s business plans while improving their technical skills and experience. This way, the mutual interest of both the employee and the company can be identified and handled. It also builds trust between the company and the employees. Regular support environment and ease of doing business is 66% highly important factor for the success of Sri Lankan startups [ 15 ]. This environment can be easily created with the level of co-worker support to the employee. Employee turnover can be more costly than a startup can imagine, with disruptions to business operations when their employees’ suddenly quit jobs. Therefore, it is must to attain above discussed facts. These results and discussions can be taken as insights to better understand and curtail employee turnover. This study will assist Sri Lankan startups where their skilled employees, who are also experts plausibly remain, enabling the businesses to expand to new markets. Usually, issues relevant to profit-making and business performance, such as a drop in sales and manufacturing are identified by startups. However, employee turnover is generally not identified as an organisational issue.

Theoretical implications

The current study empirically investigated the impact of job satisfaction, innovative work behaviour, co-worker support and leader member exchange on employee turnover. According to the authors’ knowledge, no prior studies were conducted considering the combined impact of all the independent variables on employee turnover. Therefore, this study strengthens the literature by demonstrating how job satisfaction, innovative work behaviour, co-worker support and leader member exchange impact employee turnover in Sri Lankan startups.

The findings reveal that job satisfaction has a negative impact on employee turnover. This finding is consistent with the previous study, job satisfaction significantly predicted employee turnover [ 6 ]. This study consolidates past findings that male employees have higher turnover intention than female employees. Female employees have comparatively higher-level job satisfaction [ 8 ]. This study implies that employees age 31 to 40 years have high employee turnover intention. The research findings are similar to Lu, Lu [ 8 ]; the older employees have high intentions to leave the company.

Practical implications

The study’s findings illustrate the importance of job satisfaction, innovative work behaviour, co-worker support and leader member exchange in affecting employee turnover in startups. This study provides managerial insights on lowering employee turnover in Sri Lankan startups. First, startups need to be aware that experienced employees in startups can be easily taken by well-established companies because, later, they have hand on experience and skills. Therefore, it is important to implement strategies for a solid career development plan, career growth, personal status, and employee recognition. As job satisfaction can predict employee turnover, it is a must to measure those indicators and maintain a favourable level at all times.

Innovative work behaviour is increasingly becoming a significant factor in employee retention. As good startups are a blend of creativity and competitive advantage, it is a must to focus on the IWB of the employee. LMX is a turning point for expanding the business. More importantly, healthy LMX can boost employees’ work engagement. This healthy level can maintain by conducting regular meetings, training programs and informal mentorship with employees’ immediate supervisors [ 8 ]. Further, management can allow employees at all levels to present their fresh ideas and incorporate them to influence organisation’s decision making process. These processes can lower employee hierarchy and build strong relationships while recognising them in the company.

It is important to retain trained and skilled employees who started their career paths in the organisation. Such employees can drive the organisation to success. While measuring employees’ job satisfaction, managers nee to conduct standard ways on performance and improvements of the organisation. It is better if companies can create their key performance indicators because it will help protect the organisation’s core values while expanding the company. Furthermore, having a flexible approach to work in an organisation culture will increase the trust between employees and the organisation. Giving the freedom to take risks and not allowing them to feel alone during work will give value to employees. Finally, all the above actions will strongly impact reducing employee intention to leave the organisation.

Research limitations and future research directions

Further research can improve the study as follows. First, this research includes feedback from 230 employees. More than one-third of these employees are from the IT industry. Since Sri Lankan startups are technology-driven, this ratio is more reliable. However, this research can be generalised by obtaining employees’ feedback from other industries. Secondly, in this questionnaire, the minimum number of questions for independent factors is four. This is to minimise the possibility of demotivating the employee by giving a lengthy and complex questionnaire. Therefore, in future, researchers can design questionnaires incorporating more questions to cover a wider range of independent factors, including open-ended ones. Thirdly, in this sample, many employees were in their twenties, and most hadn’t worked for more than two companies (i.e. employers). As such, it is reasonable to assume that participants’ response is somewhat limited to obtain the broader picture of the research problem. Future researchers can focus on different age groups and analyse the same factors concerning employee retention. Finally, new research can be executed by adopting a case study approach (including case studies representing various types of industries etc), such as employees in multinational companies.

Supporting information

S1 appendix, s2 appendix, s3 appendix, s4 appendix, acknowledgments.

The authors would like to thank Ms. Gayendri Karunarathne for proof-reading and editing this manuscript.

Funding Statement

The authors received no specific funding for this work.

Data Availability

• PLoS One. 2023; 18(2): e0281729.

Decision Letter 0

21 Dec 2022

PONE-D-22-30684How are employee turnover intentions created in Sri Lankan Startups?PLOS ONE

Dear Dr. Ruwan Jayathilaka,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

==============================Reviewer#1Abstract: Rewrite the abstract after manuscript correction and provide picture of whole study.

In the first paragraph of introduction used only this (2016) citation. This citation not justify the paragraph.

Introduction paragraph is not justifying the problem and bag-round of study. Revised the introduction and use the recent citations to justify and logically make connection with them.

 In the introduction (second paragraph) , the contribution of study is confused with variable relationships; why are these relationships a contribution of study? Need strong justification.

 Overall, I suggest a major rewrite of the introduction. It should provide an overview of and focus on one issue with recent citations.

Revised all literature variables and link with variables with new citation

In the literature, justify these hypotheses with literary support.

In literature, justify the conceptual model and theoretical gap.

Where is the total population? How did you choose the sample size? And how did you choose which method, unit of analysis, and research technique to use? Provide justification. Why is this method appropriate for this data set?

General: identifying flaws in the study's design (revised methodology section) and justifying technique

Write the theoretical contribution related to a model. Reviewer#2

Mention the scope of the study, the population, simple size, data collected from…..

Mention the analysis technique/ tool used in the study

Introduction:

The introduction is not clear and very less literature is used. Follow these instruction: The introduction should briefly place the study in a broad context and highlight why it is important. It should define the purpose of the work and its significance.

The current state of the research field should be reviewed carefully and key publications cited. Briefly mention the main aim of the work and highlight the main conclusions. Keep the introduction comprehensible to scientists working outside the topic of the paper.

What is the main research focus?? Firm performance? Employee retention? Employee turnover intention?

Focus more on the main issue of the study

Make the theoretical and practical gaps more clear ?

Why Sri Lanka?

Why stratup s in Sri Lanka? Why employee working in startups in Sri Lanka

What was the key motivation behind focusing on factors affecting employee turnover intention in stratups in Srilanka?

Please, properly justify why the selected variables are included in the model. How did you derive the 08 variables ??

As ,many studies conducted in the world and in Sri Lanka about this topic, what us the main contribution of your study?

The paper should incorporate a more solid argumentation that allows to justify the reason that allows to select the explanatory variables that are considered in the empirical analysis.

Literature and hypotheses development"

Improve the argumentation of hypothesis. Whether, the hypotheses are formulated separately or after the literature review of each section, it should be properly argued.

Each hypothesis should be formulated at the end of a literature section of the each variable presenting the different findings that have been made throughout the literature. With these arguments a reasoning should be developed in a certain direction and the conclusion of that reasoning should be the formulated hypothesis. In the current version of this manuscript the authors are including different aspects of previous literature, but it does not exist any convincing storyline in any direction.

Highlight controversial and diverging hypotheses when necessary.

Researcher should include a summary table / review on studies conducted on Employee Turnover Intention in Sri Lanka to support the literature and arguments.

Below papers has some interesting implications and understanding of concepts and relations that you could discuss in your introduction and literature review and how it relates to your work:

-Li, M., Li, J., Chen, X. “Employees’ Entrepreneurial Dreams and Turnover Intention to Start-Up: The Moderating Role of Job Embeddedness”, 2022, International Journal of Environmental Research and Public Health 19(15),9360

- Saoula, O.,Johari, H, “The mediating effect of organizational citizenship behavior on the relationship between perceived organizational support and turnover intention: A proposed framework” International Review of Management and Marketing, 2016, 6(7), pp. 345–354

- Saoula, O., Johari, H., Bhatti, M.A “The mediating effect of organizational citizenship behaviour on the relationship between personality traits (Big Five) and turnover intention: A proposed framework”, International Business Management, 2016, 10(20), pp. 4755–4766.

Zito, M., Emanuel, F., Molino, M., Cortese, C. G., Ghislieri, C., & Colombo, L. (2018). Turnover intentions in a call center: The role of emotional dissonance, job resources, and job satisfaction. PloS one, 13(2), e0192126.

- Saoula, O., Johari, H., Fareed, M, “A conceptualization of the role of organisational learning culture and organisational citizenship behaviour in reducing turnover intention”, Journal of Business and Retail Management Research, 2018, 12(4), pp. 126–133

- Saoula, O., Fareed, M., Ismail, S.A., Husin, N.S., Hamid, R.A, “A conceptualization of the effect of organisational justice on turnover intention: The mediating role of organisational citizenship behaviour”, International Journal of Financial Research, 2019, 10(5), pp. 327–337.

Poku, C. A., Alem, J. N., Poku, R. O., Osei, S. A., Amoah, E. O., & Ofei, A. M. A. (2022). Quality of work-life and turnover intentions among the Ghanaian nursing workforce: A multicentre study. PloS one, 17(9), e0272597.

- Saoula, O., Fareed, M., Hamid, R.A., Al-Rejal, H.M.E.A., Ismail, S.A, “The moderating role of job embeddedness on the effect of organisational justice and organisational learning culture on turnover intention: A conceptual review”, Humanities and Social Sciences Reviews, 2019, 7(2), pp. 563–571

-Li, Q., Mohamed, R., Mahomed, A., & Khan, H. (2022). The Effect of Perceived Organizational Support and Employee Care on Turnover Intention and Work Engagement: A Mediated Moderation Model Using Age in the Post Pandemic Period. Sustainability, 14(15), 9125.

- Amin, M., Othman, S.Z., Saoula, O, “The Effect of Organizational Justice and Job Embeddedness on Turnover Intention in Textile Sector of Pakistan: The Mediating Role of Work Engagement” Central Asia and the Caucasus, 2021, 22(5), pp. 930–950

Methodology:

How experiment was conducted?

How participants were recruited?

What are the instructions of experiment?

How much was time given to each participant?

What are the theoretical implications of the study ?

Practical implications needs further discussion.

Add/ involve more recent citations/studies where necessary

Please submit your revised manuscript by Feb 04 2023 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at  gro.solp@enosolp . When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

• A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.
• A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.
• An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols . Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols .

We look forward to receiving your revised manuscript.

Kind regards,

Muhammad Fareed, Ph.D

Academic Editor

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at 

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and 

https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. In the ethics statement in the Methods, you have specified that verbal consent was obtained. Please provide additional details regarding how this consent was documented and witnessed, and state whether this was approved by the IRB.

3. Please include your full ethics statement in the ‘Methods’ section of your manuscript file. In your statement, please include the full name of the IRB or ethics committee who approved or waived your study, as well as whether or not you obtained informed written or verbal consent. If consent was waived for your study, please include this information in your statement as well.

Additional Editor Comments:

The paper is generally well written and structured. However, I believe that paper has some shortcomings in terms of

Abstract: Rewrite the abstract after manuscript correction and provide picture of whole study.

In the introduction (second paragraph) , the contribution of study is confused with variable relationships; why are these relationships a contribution of study? Need strong justification.

�Overall, I suggest a major rewrite of the introduction. It should provide an overview of and focus on one issue with recent citations.

�Write the theoretical contribution related to a model.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Partly

Reviewer #2: Partly

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: I Don't Know

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #2: No

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #2: Yes

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: Abstract:

Reviewer #2: Revised topic after correction

 The paper is generally well written and structured. However, I believe that paper has some shortcomings in terms of

6. PLOS authors have the option to publish the peer review history of their article ( what does this mean? ). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy .

Reviewer #1:  Yes:  Oussama Saoula

Reviewer #2:  Yes:  Munwar Hussain Pahi

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool,  https://pacev2.apexcovantage.com/ . PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at  gro.solp@serugif . Please note that Supporting Information files do not need this step.

Author response to Decision Letter 0

16 Jan 2023

Point by point response to editor and reviewers

Dear editor and the reviewers,

We would like to express our profound appreciation to the editor and the reviewers for the valuable comments and suggestions made on our manuscript which were very helpful in revising and improving it.

Please note that the line numbers referred in this document is aligned with the revised manuscript which has track changes.

Reviewer 1 Comment 1: Abstract: Mention the scope of the study, the population, simple size, data collected from…

Authors’ Response: Thank you very much for the valuable comment. The suggestions have been incorporated in the revised manuscript from lines 30 to 32.

“…The study population was professionals who have been a key part of Sri Lankan startups, which involved 230 respondents. …”

Reviewer 1 Comment 2: Abstract: Mention the analysis technique/ tool used in the study

Authors’ Response: Thank you very much. Your comment is well noted. The analysis technique was added in the abstract of the revised manuscript from lines 32 to 33.

“…Data analysis was performed through a forward stepwise technique through STATA …”

Reviewer 1 Comment 3: Introduction: The introduction is not clear and very less literature is used. Follow these instructions: The introduction should briefly place the study in a broad context and highlight why it is important. It should define the purpose of the work and its significance.

Authors’ Response: Thank you very much for the detailed comment. This helps to strength the introduction with recent literatures, better argument, and justifications. The following literature were added in the introduction section with citation nos. 10, 11, 1, and 3 of the revised manuscript.

Reviewer 1 Comment 4: Introduction: The current state of the research field should be reviewed carefully, and key publications cited. Briefly mention the main aim of the work and highlight the main conclusions. Keep the introduction comprehensible to scientists working outside the topic of the paper.

Authors’ Response: Well noted your comment. A major update has done in the introduction section of the study.

Reviewer 1 Comment 5: Introduction: What is the main research focus?? Firm performance? Employee retention? Employee turnover intention?

Authors’ Response: Thank you very much for your comment. This study focuses/ aims to analyse the impact of job satisfaction, happiness, work-life balance, career management, management support, innovative work behaviour, leader-member-exchange, and co-worker support on employee turnover in startups in Sri Lanka. New paragraph has been added in the revised manuscript to explain more about firm performance, employee retention, employee turnover intention from lines 74 to 80.

“Firm performance reflects the ability of an organisation to use its human resources and other material resources to achieve its goals and objectives. Firm performance belongs to the economic category, and it should consider the use of business means efficiently during the production and consumption process [12]. Employee retention is defined as encouraging employees to remain in the organisation for a long period or the organisation’s ability to minimised employee turnover [13]. Turnover intention is the intention of the employee to change the job or organisation voluntarily [14].”

Reviewer 1 Comment 6: Introduction: Focus more on the main issue of the study

Authors’ Response: Thank you for your valuable comment. More priority was given to discuss the main issue of the study. In the first paragraph of the introduction section have updated to highlight the main issue with latest literature. New content has been added from lines 55 to 58.

“…The issue of employee turnover is considered as one of the global obstacles for organisations worldwide, which directly and adversely affects strategic plans and opportunities of gaining competitive advantages [3].…”

Reviewer 1 Comment 7: Introduction: Make the theoretical and practical gaps more clear

Authors’ Response: Thank you very much for the comment. The revised manuscript has been updated by pointing out the existing research gaps. New content has been added from lines 128 to 131.

“…to the best of the authors’ knowledge, there was no previous research done by local researchers that includes all the widely measured variables investigating the combined effect on employee turnover…”

Reviewer 1 Comment 8: Introduction: Why Sri Lanka?

Authors’ Response: Thank you for the comment. Sri Lanka has selected as the case study because to the best of the authors’ knowledge, no any previous research has been done by local researchers considering all the widely affected eight variables together. It leads to improve the introduction part of the paper. Suggestions have been incorporated in the revised manuscript from lines 127 to 130.

“…Sri Lankan context has been selected as the case study. This is because, to the best of the authors’ knowledge, there was no previous research done by local researchers that includes all the widely measured variables investigating the combined effect on employee turnover…”

Reviewer 1 Comment 9: Introduction: Why stratup s in Sri Lanka? Why employee working in startups in Sri Lanka

Authors’ Response: Thank you very much for the comment and this is well noted. Suggestions have been incorporated in the revised manuscript from lines 85 to 89.

“…Sri Lanka has a middle rank of ease of doing business. With these favourable conditions and educational and family backgrounds, many people like to operate/apply their new idea and fill the market gap. The new generation in Sri Lanka are interested/are keen on innovations at work and being a part of unique products or services…”

Reviewer 1 Comment 10: Introduction: What was the key motivation behind focusing on factors affecting employee turnover intention in stratups in Sri Lanka?

Authors’ Response: Thank you very much for the valuable comment. The key motivation of focusing on factors affecting employee turnover intention was to gather widely affected factors together and measure the impact of each indicator at the micro level. The idea was added in revised manuscript from line 110 to 111.

“Based on their knowledge and the existing literature, authors have considered widely used factors to investigate the employee turnover issue …”

Reviewer 1 Comment 11: Introduction: Please, properly justify why the selected variables are included in the model. How did you derive the 08 variables?

Authors’ Response: Thank you for the comment. According to the past literature authors have selected widely used indicators for employee turnover and among these eight variables have been selected. The justification has included in the revised manuscript from line 110 to 117.

“Based on their knowledge and the existing literature, authors have considered widely used factors to investigate the employee turnover issue. Therefore, job satisfaction, happiness, work-life balance, career management, management support, innovative work behaviour, leader member exchange and co-worker support were selected based on previous literature findings [4-6, 8, 17-19]. As in the previous papers and along with the current study’s results, authors identified both positive and negative impacts on employee turnover among Sri Lankan startups..”

Reviewer 1 Comment 12: Introduction: As, many studies conducted in the world and in Sri Lanka about this topic, what us the main contribution of your study?

Authors’ Response: Well noted your comment. Thank you! The contribution of the study has highlighted in the revised manuscript from lines 120 to 133.

“…The present study’s scientific value can be elaborated by comparing it with previous studies. This study’s contribution can be explained in five ways. Firstly, the most critical and newly considered factors were identified together with the support of past literature. Secondly, the present study was divided/classified into different levels of employee turnover. As such, by y considering the various levels, the micro-level changes, and probabilities of the impact on employee turnover can be better identified. Further, this study helps to reduce the methodological gap. Thirdly, the Sri Lankan context has been selected as the case study. This is because, to the best of the authors’ knowledge, there was no previous research done by local researchers that includes all the widely measured variables investigating the combined effect on employee turnover. Fourthly, the analysis results can be used to identify the strengths and weaknesses of startups in Sri Lanka. Finally, this study identifies the challenges faced by startups and identifies how policy modifications can strengthen the startup ecosystem.”

Reviewer 1 Comment 13: Introduction: The paper should incorporate a more solid argumentation that allows to justify the reason that allows to select the explanatory variables that are considered in the empirical analysis.

Authors’ Response: Well noted your comment. Thank you! In the revised manuscript a paragraph was added to present the justification to select the variables in the empirical analysis from lines 110 to 117.

“Based on their knowledge and the existing literature, authors have considered widely used factors to investigate the employee turnover issue. Therefore, job satisfaction, happiness, work-life balance, career management, management support, innovative work behaviour, leader member exchange and co-worker support were selected based on previous literature findings [4-6, 8, 17-19]. As in the previous papers and along with the current study’s results, authors identified both positive and negative impacts on employee turnover among Sri Lankan startup.”

Reviewer 1 Comment 14: Literature and hypotheses development: Improve the argumentation of hypothesis. Whether the hypotheses are formulated separately or after the literature review of each section, it should be properly argued.

Authors’ Response: Thank you very much for your comment. The paper has been updated with the improved argument in literature review. The hypotheses have been formulated at the end of each sub section of literature review. New contents have incorporated as per the below line numbers.

(Line numbers 238 and 240)

“As per the literature, job satisfaction is an important factor in determining the impact on employee turnover. Accordingly, hypothesis one has been developed.”

(Line numbers from 283 to 284)

“…According to the above literature, hypothesis two has been constructed; work-life balance has a negative impact on employee turnover.”

(Line numbers from 306 to 308)

“…Based on the above cited literature, hypothesis three can be developed; employee happiness has a negative impact on employee turnover.”

(Line numbers from 336 to 337)

“…Hypothesis four has been developed based on above discussed literature.”

(Line numbers from 356 to 357)

“…Hypothesis five has been developed by concluding the above explained literature.”

(Line numbers 378 and 379)

“…With the presence of the above-mentioned literature, hypothesis six has been formulated.”

(Line numbers from 397 to 399)

“…Based on the above-mentioned literature, hypothesis seven has been developed; leader member exchange has a negative impact on employee turnover.”

(Line numbers from 417 to 419)

“…These newly published research results can be used along with all other variables that affect employee turnover. According to the above literature, hypothesis eight has been constructed.”

Reviewer 1 Comment 15: Literature and hypotheses development: Each hypothesis should be formulated at the end of a literature section of each variable presenting the different findings that have been made throughout the literature. With these arguments a reasoning should be developed in a certain direction and the conclusion of that reasoning should be the formulated hypothesis. In the current version of this manuscript the authors are including different aspects of previous literature, but it does not exist any convincing storyline in any direction.

Authors’ Response: Thank you very much for this detailed comment. The revised version has strengthened the formulation of hypothesis. Hypothesises formulations has been incorporated at the end of each sub section in the literature review and the storyline has been built. New contents have been incorporated as per the below line numbers.

Reviewer 1 Comment 16: Literature and hypotheses development: Highlight controversial and diverging hypotheses when necessary.

Authors’ Response: Thank you for your valuable comment. This leads to build a discussion in literature review independent variables sub section. New contents have been included as per the below line numbers.

(Line numbers from 187 to 191)

“…Moreover, the descriptive statistics found a high level of job satisfaction and the intention to leave was at the mid or average level of the scale. Camara further stated that job satisfaction clearly implies the feeling about their job. But some research findings can be contradictory. Some employees are fully satisfied with the job and still want to leave the organisation for various reasons…”

(Line numbers from 205 to 207)

“…However, they didn’t observe any significant interaction between overall work-life balance and job satisfaction in predicting employee turnover intention. With these results, this indicator must be examined further.”

(Line numbers from 414 to 415)

“…It further stated that cynicism of the employee is positively associated with employee turnover…”

(Line numbers from 405 to 406)

“…However, a significantly negative impact was not evident on co-worker support.…”

Reviewer 1 Comment 17: Literature and hypotheses development: Researcher should include a summary table / review on studies conducted on Employee Turnover Intention in Sri Lanka to support the literature and arguments

Authors’ Response: Thank you very much for your comment, Literature summary table was added as an appendix, and it was cited in the revised manuscript from line numbers 168 to 169.

“Appendix S4 contains the literature summary of the above presented literature search flow diagram. The following sections present the details of each category.”

Reviewer 1 Comment 18: Literature and hypotheses development: Below papers has some interesting implications and understanding of concepts and relations that you could discuss in your introduction and literature review and how it relates to your work:

- Zito, M., Emanuel, F., Molino, M., Cortese, C. G., Ghislieri, C., & Colombo, L. (2018). Turnover intentions in a call center: The role of emotional dissonance, job resources, and job satisfaction. PloS one, 13(2), e0192126.

- Poku, C. A., Alem, J. N., Poku, R. O., Osei, S. A., Amoah, E. O., & Ofei, A. M. A. (2022). Quality of work-life and turnover intentions among the Ghanaian nursing workforce: A multicentre study. PloS one, 17(9), e0272597.

Authors’ Response: Thank you very much for the detailed comment and sharing the latest literature related to this paper. New literature has been incorporated in the introduction, literature review and results and discussion sections the paper as per the below line numbers.

(Line numbers from 45 to 46)

“…Companies need to give high priority to employee development and predict employee behaviour [1]…”

(Line numbers from 55 to 58)

“…The issue of employee turnover is considered as one of the global obstacles for organisations worldwide, which directly and adversely affects strategic plans and opportunities of gaining competitive advantages [3]…”

(Line numbers from 66 to 68)

“…Further, promoting employee well-being leads to decrease employee turnover [10]. Providing psychological and social support through counselling promotes the quality of work-life [11]...”

(Line numbers from 318 to 325)

“A cross-sectional survey has been conducted for front-line healthcare staff in China by Li, Mohamed [30] to measure the impact of organisational support on employee turnover intention. This study’s results could verify that organisational support negatively affected employee turnover intention. Saoula and Johari [31] studied this area and determined a negative relationship between organisational support and employee turnover intention. As both of the above explained research have been conducted in non-Western countries, the findings help to complete the theoretical framework for the current study in the Sri Lankan context.”

(Line numbers from 344 to 349)

“Saoula and Johari [31] researched the effect of personality traits (big five) on employee turnover intention. The researchers state that the relationship between the big five personality traits and turnover intention will support early prediction of employee turnover intentions. Identifying employee’s personalities and helping them to find the most suitable job role is a long-term process, though it will be highly advantageous for both employees and the organisation.”

(Line numbers from 365 to 373)

“The organisational learning culture is a key factor for innovative work behaviour. Saoula, Fareed [36] conducted research in Malaysia, a developing country in Asia to examine the relationship between organisational learning culture and employee turnover intention. The organisational learning culture improves learning capability, supports sustainable development, and affects organisation's positive changes. As organisational learning culture and employee turnover intention have a negative relationship, the result helps to identify the impact of innovative work behaviour. According to the existing/available literature, limited studies have been conducted on this topic.”

(Line numbers from 608 to 612)

“Entrepreneurs are the founders of startups. Employees’ entrepreneurial dreams positively affect employee intention to startups. Employees in the startups also will have an ideation to start their own business. According to the study by Li, Li [44] the mediating role of employees’ entrepreneurial self-efficacy and the moderating role of job embeddedness in the influence of entrepreneurial dreams on employees’ turnover intention to startup.”

Reviewer 1 Comment 19: Methodology: How experiment was conducted?

Authors’ Response: Thank you for the comment. The flow of methodology could improve with the help of the next three comments, including this. The experiment was conducted using both online and manual channels.

Accordingly, the revised manuscript is updated as follows in lines 432 and 435.

“…The authors directly distributed the questionnaire. Moreover, authors could contact management in startups and distribute the questionnaire in their organisation.…”

Reviewer 1 Comment 20: Methodology: How participants were recruited?

Authors’ Response: Duly noted with thanks! The participants were selected by random sampling method. Authors could contact the management of respective organisations to reach the respondents.

The methodology part has been written in descriptive manner in the revised document. From lines 432 to 435 and lines 457 to 458 were newly added.

“…The authors directly distributed the questionnaire. Moreover, authors could contact management in startups and distribute the questionnaire in their organisation….”

“…The researchers applied a random sampling method, mainly employees who are a part of or have been a part of the startup …”

Reviewer 1 Comment 21: Methodology: What are the instructions of experiment?

Authors’ Response: Thank you very much for the comment and well noted.

Instructions for the experiments were.

• Participants should be a part of the startup

• He/she should consider the behaviour and culture of that startup when answering the questions

• Respondent should answer all the questions

The instructions given in the questionnaire has included in the revised manuscript from lines 444 to 449.

“…A minimum of four questions was added under each indicator. The researchers facilitated anonymously answering all the questions in the questionnaire. The participants should be a part of startup and he/she should consider the behaviour and culture of that startup when answering the questions. All nine indicators were covered by Likert scale questions from 1 to 5 rating scale, depicting (1) strongly disagree to (5) strongly agree to collect respondents’ attitudes and opinions…”

Reviewer 1 Comment 22: Methodology: How much was time given to each participant?

Authors’ Response: Thank you very much for the comments. This helps to build the story line in methodology part. 15 minutes time were given to answer the questionnaire. New content has added from lines 449 to 452

“…Each respondent took about 10-15 minutes to complete answering the questionnaire and took approximately 5-7 minutes to fill out the questionnaire…”

Reviewer 1 Comment 23: What are the theoretical implications of the study?

Authors’ Response: Well noted your comment. Thank you very much! In revised manuscript has added new sub section to discuss theoretical implications from lines 648 to 654.

“The current study empirically investigated the impact of job satisfaction, innovative work behaviour, co-worker support and leader member exchange on employee turnover. According to the authors’ knowledge, no prior studies were conducted considering the combined impact of all the independent variables on employee turnover. Therefore, this study strengthens the literature by demonstrating how job satisfaction, innovative work behaviour, co-worker support and leader member exchange impact employee turnover in Sri Lankan startups.

The findings reveal that job satisfaction has a negative impact on employee turnover. This finding is consistent with the previous study, job satisfaction significantly predicted employee turnover [6]. This study consolidates past findings that male employees have higher turnover intention than female employees. Female employees have comparatively higher-level job satisfaction [8]. This study implies that employees age 31 to 40 years have high employee turnover intention. The research findings are similar to Lu, Lu [8]; the older employees have high intentions to leave the company.”

Reviewer 1 Comment 24: Practical implications need further discussion.

Authors’ Response: Thank you very much for your valuable comment. Practical implications section was improved with further discussion. New contents have been added from lines 665 to 671, 674 to 680, 686 to 690.

“…This study provides managerial insights on lowering employee turnover in Sri Lankan startups. First, startups need to be aware that experienced employees in startups can be easily taken by well-established companies because, later, they have hand on experience and skills. Therefore, it is important to implement strategies for a solid career development plan, career growth, personal status, and employee recognition. As job satisfaction can predict employee turnover, it is a must to measure those indicators and maintain a favourable level at all times.”

“…More importantly, healthy LMX can boost employees’ work engagement. This healthy level can maintain by conducting regular meetings, training programs and informal mentorship with employees’ immediate supervisors [8]. Further, management can allow employees at all levels to present their fresh ideas and incorporate them to influence organisation’s decision-making process. These processes can lower employee hierarchy and build strong relationships while recognising them in the company.”

“…Furthermore, having a flexible approach to work in an organisation culture will increase the trust between employees and the organisation. Giving the freedom to take risks and not allowing them to feel alone during work will give value to employees. Finally, all the above actions will strongly impact reducing employee intention to leave the organisation.”

Reviewer 1 Comment 25: Add/ involve more recent citations/studies where necessary

Authors’ Response: Thank you very much for the comment and this is well noted. New citations were added in revised manuscript in introduction, literature review and results and discussions sections as per the below line numbers.

(Line numbers from 612 to 616)

Reviewer 2 Comment 1: Abstract: Rewrite the abstract after manuscript correction and provide picture of whole study.

Authors’ Response: Thank you very much for your comment. Abstract has been rewritten after doing the manuscript corrections.

Reviewer 2 Comment 2: In the first paragraph of introduction used only this (2016) citation. This citation does not justify the paragraph.

Authors’ Response: Thank you very much for the comment. The 1st paragraph of introduction section has upgraded with recent literatures with the citation nos. 1, and 3.

New contents have included from lines 45 to 46, and lines 55 to 58.

Reviewer 2 Comment 3: Introduction paragraph is not justifying the problem and bag-round of study. Revised the introduction and use the recent citations to justify and logically make connection with them.

Authors’ Response: Thank you so much for the comment. The introduction section was upgraded with recent literatures with the citation nos. 10, 11, 1, and 3 and justifications.

Reviewer 2 Comment 4: In the introduction (second paragraph), the contribution of study is confused with variable relationships; why are these relationships a contribution of study? Need strong justification.

Authors’ Response: Thank you very much for your valuable comment. The content has been updated with recent citations in line 64.

“Many variables influence employee turnover intentions [4-6]…”

Reviewer 2 Comment 5: Overall, I suggest a major rewrite of the introduction. It should provide an overview of and focus on one issue with recent citations.

Authors’ Response: Thank you very much, the comment well noted. New literatures have added in revised manuscript and highlighted the main issues and the research gaps. Every paragraph of the introduction has updated according to the reviewers’ comments.

Reviewer 2 Comment 6: Revised all literature variables and link with variables with new citation.

Authors’ Response: Thank you very much for the comment and well noted. After adding new literatures, revised all literature variables and linked with variables with new citation.

Reviewer 2 Comment 7: In the literature, justify these hypotheses with literary support.

Authors’ Response: Thank you very much for the comment. A storyline was developed on hypothesis formulation. New contents have been incorporated as per the below line numbers.

Reviewer 2 Comment 8: In literature, justify the conceptual model and theoretical gap.

Authors’ Response: Well noted your comment. Thank you! Conceptual model and theoretical gap have justified in the revised manuscript from lines 420 to 424.

“These studies have a common limitation in gathering more independent variables and analysing the impact. Therefore, a need exists to measure the effect of job satisfaction, work-life balance, happiness, management support, career management, innovative work behaviour, leader member exchange, and co-worker support together on employee turnover.”

Reviewer 2 Comment 9: Where is the total population? How did you choose the sample size? And how did you choose which method, unit of analysis, and research technique to use? Provide justification. Why is this method appropriate for this data set?

Authors’ Response: Thank you very much for this valuable comment.

Total population was 1300 and sample size identified by referring calculator.net online sample size calculator. A stepwise ordered probit analysis method was used as the selected variables are widely used indicators for employee turnover therefore authors required to do a micro level analysis for these variables.

The details of sampling have added in revised manuscript from lines 459 to 465.

“…The sample size was selected by referencing the Krejcie and Morgan sampling table and Calculator.net [40] with a confidence level of 95% and 7% of margin of error. The calculation results indicated a minimum of 171 professionals. A stepwise ordered probit analysis method was used as the selected variables are widely used indicators for employee turnover; therefore, a micro-level analysis was required to study how these variables impact. A pilot survey was conducted to identify whether the purpose of the questions was clear to the respondents.”

Reviewer 2 Comment 10: General: Identifying flaws in the study's design (revised methodology section) and justifying technique.

Authors’ Response: Well noted your comment. Thank you! Methodology section has been updated in revised manuscript from line 520 to 522.

“…The probit model is an estimation technique for equations with dummy dependent variables that avoids the unboundedness problem of the linear probability model by using a variant of the cumulative normal distribution [42]…”

Reviewer 2 Comment 11: Discussion: Write the theoretical contribution related to a model.

Authors’ Response: Thank you very much for your valuable comment.

The revised manuscript has been updated with a new sub section to discuss theoretical implications from lines 648 to 661.

Reviewer 2 Comment 12: Revised topic after correction

Authors’ Response: Well noted your comment. The topic has been updated in lines 1 to 2 and 19 to 20, and the new topic is,

“Factors impacting employee turnover intentions among professionals in Sri Lankan startups”

Submitted filename: Response to the Reviewers.docx

Decision Letter 1

31 Jan 2023

PONE-D-22-30684R1

Dear Dr. Lakshmi Kanchana,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/ , click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at gro.solp@gnillibrohtua .

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact gro.solp@sserpeno .

Additional Editor Comments (optional):

Dear Author/s,

Thank you for making all the corrections.

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #1: All comments have been addressed

Reviewer #2: All comments have been addressed

2. Is the manuscript technically sound, and do the data support the conclusions?

Reviewer #2: (No Response)

3. Has the statistical analysis been performed appropriately and rigorously?

4. Have the authors made all data underlying the findings in their manuscript fully available?

5. Is the manuscript presented in an intelligible fashion and written in standard English?

6. Review Comments to the Author

Reviewer #1: The authors have adequately addressed the comments raised in a previous round of review and I feel that this manuscript is now acceptable for publication

7. PLOS authors have the option to publish the peer review history of their article ( what does this mean? ). If published, this will include your full peer review and any attached files.

Acceptance letter

Dear Dr. Jayathilaka:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact gro.solp@sserpeno .

If we can help with anything else, please email us at gro.solp@enosolp .

Thank you for submitting your work to PLOS ONE and supporting open access.

PLOS ONE Editorial Office Staff

on behalf of

Dr. Muhammad Fareed

Academia.edu no longer supports Internet Explorer.

To browse Academia.edu and the wider internet faster and more securely, please take a few seconds to  upgrade your browser .

Enter the email address you signed up with and we'll email you a reset link.

• We're Hiring!
• Help Center

RESEARCH PROPOSAL (moffat Pihelo) HIGH TURNOVER (3)

Related Papers

Tapaswini panigrahi

Irish Interdisciplinary Journal of Science & Research (IIJSR)

IIJSR Journal

This studies targets to apprehend the reasons of worker turnover and retention techniques in a business enterprise. Key studies findings suggest that personnel have numerous motives to go away their workplaces, including activity strain, activity pride, activity security, motivation, wages, and rewards. Furthermore, worker turnover has a large effect on a business enterprise because of the prices related to worker turnover and may negatively affect the productiveness, sustainability, competitiveness, and profitability of a business enterprise. However, the business enterprise need to apprehend the wishes of its personnel, with the intention to assist agencies, undertake positive techniques to enhance worker overall performance and decrease turnover. Thus, imposing techniques will growth activity pride, motivation and the productiveness of people and agencies that may lessen employment issues, absenteeism, and worker turnover. In a HR setting, turnover or work turnover is the rate at which a business gains and loses representatives. Basic method for portraying it are "the way lengthy workers will generally remain" or "the pace of traffic through the rotating entryway." Turnover is estimated for individual organizations and for their industry all in all. Assuming a business is said to have a high turnover comparative with its rivals, it implies that representatives of that organization have a more limited typical residency than those of different organizations in a similar industry. High turnover might be hurtful to an organization's efficiency in the event that gifted labourers are much of the time leaving and the specialist populace contains a high level of fledgling labourers. Unreasonable turnover can be an expensive issue, one with a significant effect on efficiency. One firm had a turnover pace of over 120% each year. It cost the organization $1.5 million a year in lost efficiency, expanded preparing time, expanded worker determination time, lost work productivity, and other roundabout expenses. Yet, cost isn't the main explanation turnover is significant. Extended preparing times, hindered plans, extra time, botches, and not having educated representatives set up are a portion of the disappointments related with over the top turnover. Turnover rates normal around 16% each year for all organizations, yet 21% each year for IT companies.54 Computer organizations normal higher turnover on the grounds that their representatives have numerous potential chances to change occupations in a "hot" industry. Many investigations show that organizations with low turnover rates are very representative situated. Representative situated associations request info and contribution from all workers and keep a valid "open-entryway" strategy. Workers are given open doors for progression and are not obsessively hovered over. Workers accept they have a voice and are perceived for their commitment.

International Journal of Information Research and Review

ARTICLE INFO ABSTRACT Employee is the main resource for organization. Recently, there were many concerns of staff resignation within industries. High staff turnovers cause increase of costs of hiring workforces. Owing this issue, " Assessment of Factor of Employee's Turnover " was proposed for research with the objective of examination of factors causing staff resignation from MVi. In total 26 staff both women and men who resigned 2016 and 2017 was selected for interview. Quantitative data was used. Three main steps were done including research questionnaires and material development, sampling technique and criteria and data analysis and reporting. There were four main parts: the degree of staff satisfaction of each of the variables (1 to 12), the correlation between each of variable (1 to 12) with overall satisfaction, the causes of staff resign of the variable (13-27) and the correlation between each of the staff resign causes (variable 13 to 27). All those data have been transformed into non-parameter with the Spearmen Rho. MVi staff members were appreciated working with MVi. The satisfaction of each variable were that the staff scored high for most factors like staff leave, communication with colleagues, staff capacity development opportunities and roles and responsibilities. Former staff who worked in 2016 and 2017 was most appreciated with the factors above. Three factors which scored by respondents were very low in comparison with other factors. Those are work overload, job security and salary. All those factors were located between score 3 (neutral and score 4 (satisfied). Among 12 satisfaction variables, three of them were most significantly positively related to staff motivation. For internal factors, the average score was 3.19 which represents mostly true of the statement above. The variables which were significantly related to staff resignation were communication with superiors, finding higher position with new agencies better salaries and benefits from new agency. The recommendations were produced including awarding for staff, review salary scale, tasks, maintaining staff. Qualitative research to explore what job security, salary, and work-overloads are not significantly related to staff satisfaction should be done. The qualitative research on factors of staff resign should be done.

Laura Mamuli

High staff turnover affects the smooth running of institutions. This study established the effect of staff turnover on performance of work in Masinde Muliro University of Science and Technology (MMUST). Specific objectives of the study were: to identify effects of staff turnover on administrative work and to identify financial/economic effects of staff turnover. A conceptual framework formed the basis of this study. Correlational research design was used in this study. Cluster random sampling procedure was used to collect data. Questionnaires, interviews, document analysis and observation were blended to capture authentic and exhaustive data. A randomly selected sample of 25 departments was used in this study. A total of 152 respondents participated. Data were analyzed using inferential and descriptive statistics.. The study established that economically, staff turnover in increases work for the remaining staff, leads to customer dissatisfaction, brings about decreased income due to...

Ntebogang Moroke

There is a general consensus regarding the effects of high staff turnover on the smooth running of various institutions. The purpose of this study is to establish the effect of staff turnover on performance of employees in the North West Provincial Department of South Africa. Questionnaires and document analysis were blended to capture authenticity and exhaustiveness of the data. Participants included the 70 employees in the said department who all filled and returned the questionnaire. Both inferential and descriptive statistics were used to present the results. A chi-square analysis was used as a method for data analysis in this study. Descriptive statistics were also used to describe the profiles of employees. The findings showed that the majority of employees are dissatisfied due many reasons and this causes lots of voluntary resignations among employees. Low productivity in the department is as a result of employee dissatisfaction borne as a result of management’s ignorance. Th...

Siti Aida Samikon

This research aims to understand the causes of employee turnover and retention strategies in an organization. Key research findings indicate that employees have several reasons to leave their workplaces, such as job stress, job satisfaction, job security, work environment, motivation, wages, and rewards. Furthermore, employee turnover has a huge impact on an organization due to the costs associated with employee turnover and can negatively impact the productivity, sustainability, competitiveness, and profitability of an organization. However, the organization must understand the needs of its employees, which will help organizations, adopt certain strategies to improve employee performance and reduce turnover. Thus, implementing strategies will increase job satisfaction, motivation and the productivity of individuals and organizations, which can reduce employment problems, absenteeism, and employee turnover.

African Journal of Business Management

Archils Oburu

neelu maharjan

Abdullah Al-khrabsheh , Islam Bourini

This study aims to examine the relationship between turnover intention and some organisational factors among professional academics at Jordanian Government Universities. Namely the organisational factors include job satisfaction, work exhaustion, occupational health and safety management and organisational culture. A sample of 250 participants was extracted from different Jordanian universities. The participants were limited to academics in Jordan who are working in any government Jordanian University. Statistical analysis was conducted by using SPSS 23. Previous literature was also used to design a structured questionnaire. A total of 250 questionnaires were given out and 250 questionnaires were 165 collected back. The study then conducted correlation and regression analysis to determine the relationship between the independent and the dependent variables. The models for multiple regression offer support for the relationship between turnover intention and organizational factors. The results revealed that the all the exogenous variables had a significant effect on the endogenous variable. Based on these results, the study implies that managers need to acknowledge the importance of examining the factors that reduce the turnover intentions of the employees and improve the commitment level for their employees.

The assets of an organization can be categorized as the fixed and the current. Whereas infrastructure and machinery constitute the fixed assets, human resources and material inventories make up the current assets. Whilst most of the organizational assets tend to depreciate with the passage of time, conversely, its human resources tends to appreciate progressively. Thus, it is the manpower component that has a decisive bearing on an organization's performance, which in turn determines the state of its functioning, so much so that it is not the technological advancement that determines the viability, no to speak of the profitability, of an enterprise as the level of the managerial and operative skills available therein. Hence, high employee turnover is disquieting to any business venture in that it tends to drain its organization of its cream, as it is the more talented and enterprising among the employees that generally contemplate change and find new employers. An attempt is made here to examine the factors that unleash an abnormal turnover and to suggest possible measures to help check the same. These factors can be broadly divided into organizational, personal, ethical, and political. Structural Constraints The size of an organization has a direct bearing on the extent of employee turnover. Smaller organizations suffer most on account of their inability to keep pace with the employees' expectations of financial gain and personal growth. In a way, the small and medium ventures become training centers of manpower in that those they train will eventually be pirated by bigger organizations with attractive offers. The growth rate of an organization also influences its employee turnover for people tend to stick to a growing organization in the hope of personal advancement. Thus, in a stagnant setup, as the expectations of personal growth are dimmed in proportion to the level of its stagnation, there ensues a disproportionate employee turnover. Also, the location in which an organization is situated plays a vital role not only in attracting talent but also in retaining it. Generally speaking, remote and ill-developed areas do not offer the possibilities for a way of life that the more enterprising employees tend to seek. Since the quality of life on offer in the so-called backward areas is far below the normal social want, employee turnover in the units situated therein is inevitable, notwithstanding their general organizational strengths. Just as an example, the lack of adequate educational facilities in such places makes the employees seek greener pastures for providing better avenues for their progeny. The associated prestige of an organization, measured on the scale of public recognition, too tends to influence the turnover ratio. Normally, one tends to associate himself with reputed and well known entities as that would lend him tangible social status thereby augmenting his egotistic

RELATED PAPERS

Journal of Sound and Vibration

Carlos Adolfo Rossit

Zoltan Mako

Nataniel Gomes

Hubert Palus

Lee Wen Chiat

Nader Ale Ebrahim نادر آل ابراهیم

Zeynep Kızıltepe

Proceedings of the 7th International Conference on Properties and Applications of Dielectric Materials (Cat. No.03CH37417)

S. Hettiwatte

Marianna Moravecz

CRISTOBAL MANUEL ARAVENA ORELLANA

Leandro Luiz Marcuzzo

BioNanoScience

Polina Skvortsova

Nephrology Dialysis Transplantation

Salman Hussain

Annemarie Groot

Frontiers in Marine Science

alessandro mannini

Journal of Inborn Errors of Metabolism & Screening

Luis Zepeda

Psychoneuroendocrinology

Amy Desantis

International Journal of Technology

Renanto Handogo

BMC Medical Research Methodology

David Nunan

Research Square (Research Square)

George Lohay

Tamara Floricic

arXiv: General Physics

Jacob Fokkema

Retrovirology

Florencia Pereyra

IEEE Transactions on Energy Conversion

Antonio J. Marques Cardoso

Computers and Electronics in Agriculture

Satyabrata Maiti

RELATED TOPICS

•   We're Hiring!
•   Help Center
• Find new research papers in:
• Health Sciences
• Earth Sciences
• Cognitive Science
• Mathematics
• Computer Science
• Academia ©2024