Employee Turnover

Employee turnover usually refers to all “leavers” of an organization – those who leave voluntarily or involuntarily. It includes those who resign, are made redundant, take retirement, or exit for any other reason. In this article we are only concerned with voluntary turnover and its possible causes.

We will measure the turnover rate at the individual and the team levels. Then, we will explore differences in the turnover rate sacross different countries using the dataset of a multinational organization.

Individual turnover is more simple to measure – either a person has quit or has not; however, it’s not quite simple to model and predict. As for the team level turnover rate (or called separation rate), usually represented as a percentage, can be calculated with the following formula:

Total number of leavers / average total number of employees over the period * 100

We have a dataset which lists all employees’ leaving or stay statuses in categorical data type – 0 or 1 (0 means leaving; 1 means stay). We can use Chi-square to explore regional differences (UK, United States, Canada, and Spain) in individual staff turnover. The p-value (> 0.05) in the result shows us that there is no significant difference between what we would expect in each region and what was observed.

> chisq.test(emp.table, simulate.p.value = TRUE)
 Pearson's Chi-squared test with simulated p-value (based on 2000 replicates)
data:  emp.table
X-squared = 14.509, df = NA, p-value = 0.09245

We have another dataset that represents the team level turnover rate in percentage. In this scenario, the turnover value is a continuous data type; therefore, we will use One-way ANOVA to analyze our question: are there country differences in the team engagement and turnover?

One-way ANOVA is a powerful method for testing the significance of the difference between sample means where two or more categorical groups are compared. In our case, four countries’ turnover means are compared. The null hypothesis is that there is no significant pattern of variance found in the samples; however, if it is, further testing (called “post-hoc testing”) is needed to determine which samples are different and by how much. Since employee engagement is related to turnover, and it’s discussed in the previous post, we will include it in the following analysis.

Levene’s test and One-way ANOVA Welch’s test are applied to both turnover (team separation) and engagement datasets to verify the variance among the samples and the equality of the sample means. We learn that the p-values of Levene’s Test for both are significant (0.001 and 0.000). Welch’s F-statistic for team turnover is 3.45 with 3 degree of freedom and 0.02 of p-value. Welch’s F-statistic for team engagement is 29.26 with 3 degree of freedom and 0.000 of p-value. Therefore, we can say that there is a significant effect of “country” on the team engagement levels and the turnover values.  To identify which countries differ, we need to look at post-hoc tests.

> leveneTest(TeamSeparation ~ Country, data=teamTurnOver, center="median")
Levene's Test for Homogeneity of Variance (center = "median")
 Df F value Pr(>F) 
group 3 5.3681 0.001403 **
 208 
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

> oneway.test(TeamSeparation ~ Country, data=teamTurnOver) #Welch test
 One-way analysis of means (not assuming equal variances)
data:  TeamSeparation and Country
F = 3.4458, num df = 3.000, denom df = 98.758, p-value = 0.01961
> leveneTest(Engagement ~ Country, data=teamTurnOver, center="median")
Levene's Test for Homogeneity of Variance (center = "median")
 Df F value Pr(>F) 
group 3 10.805 1.255e-06 ***
 208 
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

> oneway.test(Engagement ~ Country, data=teamTurnOver) # Welch test
 One-way analysis of means (not assuming equal variances)
data:  Engagement and Country
F = 29.256, num df = 3.000, denom df = 93.477, p-value = 1.985e-13

Because the group variances are not equal, for the “post-hoc test”, we will choose Games-Howell’s test. From the following results we can see that “country” has impact on both engagement and turnover is mainly due to how different Spain is compared to the other countries in our dataset. We can say that Spain has significantly lower engagement than all the other countries; it also has significantly higher turnover than both UK and the United Sates.

> posthocTGH(teamTurnOver$TeamSeparation, teamTurnOver$Country,
+ method = "games-howell", # or Tukey
+ #conf.level = 0.95, 
+ digits=3, 
+ formatPvalue = TRUE)
              n    means  variances
UK            52   0.159  0.0414
United States 73   0.160  0.0185
Canada        32   0.189  0.0165
Spain         55   0.267  0.0503

                     diff     ci.lo    ci.hi  t      df     p     p.adjusted
United States-UK     0.00148 -0.083492 0.0865 0.0458 82.7   1.000 1.000
Canada-UK            0.03041 -0.064583 0.1254 0.8395 81.9  .835   1.000
Spain-UK             0.10789 -0.000101 0.2159 2.6083 104.8 .050   .252
Canada-United States 0.02893 -0.044235 0.1021 1.0436 62.5  .725   1.000
Spain-United States  0.10641  0.016835 0.1960 3.1144 83.3  .013   .079
Spain-Canada         0.07748 -0.021584 0.1765 2.0496 85.0  .178   .713

> posthocTGH(teamTurnOver$Engagement, teamTurnOver$Country,
+ method = "games-howell",
+ digits=3, 
+ formatPvalue = TRUE)
              n  means variances
UK            52 82.7  71.2
United States 73 86.6  111.7
Canada        32 81.7  314.7
Spain         55 69.5  109.4

                    diff     ci.lo    ci.hi t    df     p    p.adjusted
United States-UK      3.92   -0.511   8.36  2.30 121.3 .103 .308
Canada-UK            -1.00   -9.979   7.97  0.30 39.8  .990 .990
Spain-UK             -13.20  -17.987 -8.42  7.20 102.5 <.001 <.001
Canada-United States -4.93   -13.956  4.10  1.46 41.0  .469 .938
Spain-United States  -17.13  -22.015 -12.24 9.13 117.1 <.001 <.001
Spain-Canada         -12.20  -21.380 -3.01  3.55 43.8  .005 .020

Recommendations may be to investigate possible causes of the low employee engagement and high turnover – using predictive models such as regression analysis. It may be a combination of factors, such as job market condition, length of service, appraisal rating. It is the topic we will cover in the next post.

Complete data file and source code in Github

Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

w

Connecting to %s