Employee Retention Analysis Using Survival Modeling in R (2/2)

HR analytics
Statistics
R
Published

October 19, 2025

In the previous post, the fundamentals of survival modeling were introduced, covering how to create a survival object, fit a Kaplan-Meier model, and visualize survival curves.

This follow-up post explains how distinct values of a single variable, as well as combinations of multiple variables, can help uncover differences in retention ratios across employee cohorts.

Loading the required packages and data

library(ggplot2)
library(knitr)
library(survival)
library(survminer)
library(data.table)
df <- fread("WA_Fn-UseC_-HR-Employee-Attrition.csv")
source("custom_themes.R") # import pre-built custom themes


Survival object creation

df[, Attrition := ifelse(Attrition == "Yes", 1, 0)]
s_obj <- Surv(time = df$YearsAtCompany, event = df$Attrition)
head(s_obj)
[1]  6  10+  0   8+  2+  7+



Exploring the BusinessTravel and OverTime variables

For this exercise, the BusinessTravel and OverTime variables are examined to understand how they relate to employee retention.

kable(table(df$BusinessTravel, df$OverTime))
No Yes
Non-Travel 115 35
Travel_Frequently 191 86
Travel_Rarely 748 295


Recoding both variables, so that the chart labels read as plain English rather than as stored values.

df[, OverTime := ifelse(OverTime == "No", "No Over Time", "Over Time")]
df[, BusinessTravel := fcase(BusinessTravel == "Non-Travel", "No travel",
                             BusinessTravel == "Travel_Rarely", "Travels rarely",
                             BusinessTravel == "Travel_Frequently", "Travels frequently")]
kable(table(df$BusinessTravel, df$OverTime))
No Over Time Over Time
No travel 115 35
Travels frequently 191 86
Travels rarely 748 295

The table above also gives the size of each group, which is worth carrying into the last chart: the six combined groups range from 35 to 748 employees.

Fitting Kaplan-Meier models for single variables and their combinations

km_fit_BusinessTravel <- survfit(s_obj ~ BusinessTravel, data = df)
km_fit_OverTime <- survfit(s_obj ~ OverTime, data = df)
km_fit_BusinessTravel_OverTime <- survfit(s_obj ~ BusinessTravel + OverTime, data = df)


Simplifying the strata names to make interpretation easier. All three models are fitted through the data argument, so their strata are named the same way and a single expression cleans all of them.

strip_prefix <- function(fit) {
  names(fit$strata) <- gsub("BusinessTravel=|OverTime=", "", names(fit$strata))
  fit
}
km_fit_BusinessTravel <- strip_prefix(km_fit_BusinessTravel)
km_fit_OverTime <- strip_prefix(km_fit_OverTime)
km_fit_BusinessTravel_OverTime <- strip_prefix(km_fit_BusinessTravel_OverTime)


The model summaries are converted into a clean table to review the survival statistics for each group.

km_fit_summary_BusinessTravel <- as.data.table(summary(km_fit_BusinessTravel)[c("strata","time", "n.risk", "n.event", "n.censor", "surv", "std.err")])
km_fit_summary_OverTime <- as.data.table(summary(km_fit_OverTime)[c("strata","time", "n.risk", "n.event", "n.censor", "surv", "std.err")])
km_fit_summary_BusinessTravel_OverTime <- as.data.table(summary(km_fit_BusinessTravel_OverTime)[c("strata","time", "n.risk", "n.event", "n.censor", "surv", "std.err")])
kable(km_fit_summary_BusinessTravel, digits = 3)
strata time n.risk n.event n.censor surv std.err
No travel 0 150 1 6 0.993 0.007
No travel 1 143 4 14 0.966 0.015
No travel 2 125 1 5 0.958 0.017
No travel 3 119 1 15 0.950 0.019
No travel 4 103 1 8 0.941 0.021
No travel 5 94 2 19 0.921 0.024
No travel 10 38 2 42 0.872 0.041
Travels frequently 0 277 7 4 0.975 0.009
Travels frequently 1 266 17 8 0.912 0.017
Travels frequently 2 241 7 23 0.886 0.019
Travels frequently 3 211 6 14 0.861 0.021
Travels frequently 4 191 6 17 0.834 0.023
Travels frequently 5 168 4 22 0.814 0.025
Travels frequently 6 142 4 15 0.791 0.027
Travels frequently 7 123 3 13 0.772 0.028
Travels frequently 8 107 3 14 0.750 0.030
Travels frequently 9 90 2 17 0.733 0.032
Travels frequently 10 71 5 20 0.682 0.037
Travels frequently 11 46 1 4 0.667 0.039
Travels frequently 13 39 1 7 0.650 0.042
Travels frequently 18 26 1 11 0.625 0.047
Travels frequently 21 15 1 9 0.583 0.059
Travels frequently 23 8 1 3 0.510 0.086
Travels rarely 0 1043 8 18 0.992 0.003
Travels rarely 1 1017 38 90 0.955 0.006
Travels rarely 2 889 19 72 0.935 0.008
Travels rarely 3 798 13 79 0.920 0.009
Travels rarely 4 706 12 66 0.904 0.010
Travels rarely 5 628 15 134 0.882 0.011
Travels rarely 6 479 5 45 0.873 0.012
Travels rarely 7 429 8 60 0.857 0.013
Travels rarely 8 361 6 49 0.843 0.014
Travels rarely 9 306 6 43 0.826 0.015
Travels rarely 10 257 11 75 0.791 0.018
Travels rarely 11 171 1 20 0.786 0.018
Travels rarely 13 139 1 23 0.780 0.019
Travels rarely 14 126 2 12 0.768 0.021
Travels rarely 15 112 1 11 0.761 0.022
Travels rarely 16 100 1 10 0.754 0.023
Travels rarely 17 89 1 8 0.745 0.024
Travels rarely 19 73 1 13 0.735 0.026
Travels rarely 20 66 1 19 0.724 0.028
Travels rarely 22 37 1 20 0.704 0.033
Travels rarely 24 25 1 3 0.676 0.042
Travels rarely 31 11 1 11 0.615 0.070
Travels rarely 32 9 1 0 0.546 0.090
Travels rarely 33 8 1 2 0.478 0.101
Travels rarely 40 1 1 4 0.000 NaN 
kable(km_fit_summary_OverTime, digits = 3)
strata time n.risk n.event n.censor surv std.err
No Over Time 0 1054 6 24 0.994 0.002
No Over Time 1 1024 31 85 0.964 0.006
No Over Time 2 908 12 80 0.951 0.007
No Over Time 3 816 6 82 0.944 0.007
No Over Time 4 728 11 69 0.930 0.008
No Over Time 5 648 7 132 0.920 0.009
No Over Time 6 509 4 48 0.913 0.010
No Over Time 7 457 3 66 0.907 0.010
No Over Time 8 388 3 53 0.900 0.011
No Over Time 9 332 6 57 0.884 0.013
No Over Time 10 269 8 80 0.857 0.015
No Over Time 11 181 1 23 0.853 0.016
No Over Time 13 146 2 29 0.841 0.018
No Over Time 14 126 2 16 0.828 0.020
No Over Time 17 89 1 25 0.818 0.022
No Over Time 18 82 1 9 0.808 0.024
No Over Time 20 65 1 27 0.796 0.026
No Over Time 22 35 1 19 0.773 0.034
No Over Time 23 24 1 1 0.741 0.045
No Over Time 32 8 1 15 0.648 0.095
No Over Time 33 6 1 1 0.540 0.127
No Over Time 40 1 1 3 0.000 NaN
Over Time 0 416 10 4 0.976 0.008
Over Time 1 402 28 27 0.908 0.014
Over Time 2 347 15 20 0.869 0.017
Over Time 3 312 14 26 0.830 0.019
Over Time 4 272 8 22 0.805 0.020
Over Time 5 242 14 43 0.759 0.023
Over Time 6 185 5 19 0.738 0.024
Over Time 7 161 8 13 0.702 0.026
Over Time 8 140 6 18 0.672 0.028
Over Time 9 116 2 17 0.660 0.028
Over Time 10 97 10 22 0.592 0.033
Over Time 11 65 1 7 0.583 0.033
Over Time 15 50 1 13 0.571 0.035
Over Time 16 43 1 5 0.558 0.036
Over Time 19 32 1 8 0.540 0.039
Over Time 21 22 1 10 0.516 0.044
Over Time 24 13 1 4 0.476 0.056
Over Time 31 7 1 6 0.408 0.079 
kable(km_fit_summary_BusinessTravel_OverTime, digits = 3)
strata time n.risk n.event n.censor surv std.err
No travel, No Over Time 0 115 1 6 0.991 0.009
No travel, No Over Time 1 108 1 12 0.982 0.013
No travel, No Over Time 2 95 1 3 0.972 0.016
No travel, No Over Time 4 77 1 19 0.959 0.020
No travel, No Over Time 10 28 1 48 0.925 0.039
No travel, Over Time 1 35 3 2 0.914 0.047
No travel, Over Time 3 28 1 3 0.882 0.056
No travel, Over Time 5 23 2 8 0.805 0.073
No travel, Over Time 10 10 1 8 0.724 0.101
Travels frequently, No Over Time 0 191 2 4 0.990 0.007
Travels frequently, No Over Time 1 185 10 4 0.936 0.018
Travels frequently, No Over Time 2 171 4 19 0.914 0.021
Travels frequently, No Over Time 3 148 2 10 0.902 0.022
Travels frequently, No Over Time 4 136 3 15 0.882 0.024
Travels frequently, No Over Time 5 118 2 17 0.867 0.026
Travels frequently, No Over Time 6 99 3 10 0.841 0.029
Travels frequently, No Over Time 7 86 2 12 0.821 0.032
Travels frequently, No Over Time 8 72 1 9 0.810 0.033
Travels frequently, No Over Time 10 47 1 31 0.792 0.037
Travels frequently, No Over Time 13 24 1 11 0.759 0.048
Travels frequently, No Over Time 18 14 1 7 0.705 0.069
Travels frequently, No Over Time 23 4 1 6 0.529 0.161
Travels frequently, Over Time 0 86 5 0 0.942 0.025
Travels frequently, Over Time 1 81 7 4 0.860 0.037
Travels frequently, Over Time 2 70 3 4 0.824 0.041
Travels frequently, Over Time 3 63 4 4 0.771 0.046
Travels frequently, Over Time 4 55 3 2 0.729 0.050
Travels frequently, Over Time 5 50 2 5 0.700 0.052
Travels frequently, Over Time 6 43 1 5 0.684 0.053
Travels frequently, Over Time 7 37 1 1 0.665 0.055
Travels frequently, Over Time 8 35 2 5 0.627 0.058
Travels frequently, Over Time 9 28 2 2 0.582 0.062
Travels frequently, Over Time 10 24 4 4 0.485 0.068
Travels frequently, Over Time 11 16 1 0 0.455 0.070
Travels frequently, Over Time 21 8 1 9 0.398 0.081
Travels rarely, No Over Time 0 748 3 14 0.996 0.002
Travels rarely, No Over Time 1 731 20 69 0.969 0.006
Travels rarely, No Over Time 2 642 7 58 0.958 0.007
Travels rarely, No Over Time 3 577 4 58 0.952 0.008
Travels rarely, No Over Time 4 515 7 49 0.939 0.009
Travels rarely, No Over Time 5 459 5 101 0.928 0.010
Travels rarely, No Over Time 6 353 1 32 0.926 0.011
Travels rarely, No Over Time 7 320 1 49 0.923 0.011
Travels rarely, No Over Time 8 270 2 38 0.916 0.012
Travels rarely, No Over Time 9 230 6 30 0.892 0.015
Travels rarely, No Over Time 10 194 6 59 0.865 0.018
Travels rarely, No Over Time 11 129 1 13 0.858 0.019
Travels rarely, No Over Time 13 107 1 19 0.850 0.021
Travels rarely, No Over Time 14 95 2 12 0.832 0.024
Travels rarely, No Over Time 17 67 1 20 0.820 0.027
Travels rarely, No Over Time 20 51 1 25 0.803 0.031
Travels rarely, No Over Time 22 27 1 15 0.774 0.041
Travels rarely, No Over Time 32 7 1 11 0.663 0.108
Travels rarely, No Over Time 33 6 1 1 0.553 0.135
Travels rarely, No Over Time 40 1 1 3 0.000 NaN
Travels rarely, Over Time 0 295 5 4 0.983 0.008
Travels rarely, Over Time 1 286 18 21 0.921 0.016
Travels rarely, Over Time 2 247 12 14 0.876 0.020
Travels rarely, Over Time 3 221 9 21 0.841 0.022
Travels rarely, Over Time 4 191 5 17 0.819 0.024
Travels rarely, Over Time 5 169 10 33 0.770 0.027
Travels rarely, Over Time 6 126 4 13 0.746 0.029
Travels rarely, Over Time 7 109 7 11 0.698 0.032
Travels rarely, Over Time 8 91 4 11 0.667 0.034
Travels rarely, Over Time 10 63 5 29 0.614 0.039
Travels rarely, Over Time 15 31 1 13 0.594 0.042
Travels rarely, Over Time 16 28 1 5 0.573 0.046
Travels rarely, Over Time 19 18 1 6 0.541 0.053
Travels rarely, Over Time 24 7 1 8 0.464 0.085
Travels rarely, Over Time 31 4 1 3 0.348 0.119



Visualizing the Kaplan-Meier survival curves

Show the code 
ggsurv_travel <-
  ggsurvplot(fit = km_fit_BusinessTravel,
            data = df,
            title = "\n\nSurvival Probabilities by Business Travel Status",
            xlab = "Years At Company",
            ylab = "Probability of Continued Employment (%)\n",
            break.x.by=1,
            break.y.by=.1,
            conf.int = FALSE,
            censor = FALSE,
            xlim = c(0, max(df$YearsAtCompany)),
            surv.scale = "percent",
            risk.table = TRUE,
            risk.table.title = "Number of employees still at risk",
            risk.table.height = 0.25,
            risk.table.y.text = FALSE,
            fontsize = 3,
            legend.title = element_blank(),
            risk.table.col = "#808080",
            ggtheme = theme_survival,
            legend = c(0.2, 0.2),
            palette = c("#1F497D", "#9BBB59", "#8064A2", "#C0504D", "#4F81BD", "#F79646")
  )

ggsurv_travel$table <- ggsurv_travel$table + theme(axis.title.x = element_blank())
print(ggsurv_travel)

Employees who travel frequently show the lowest retention, with survival probability dropping to around 70% after 10 years. Those who do not travel maintain the highest retention, staying above 85% even after a decade. The separation between the curves can be seen from the early years.


Show the code 
ggsurv_overtime <-
  ggsurvplot(fit = km_fit_OverTime,
            data = df,
            title = "\n\nSurvival Probabilities by Over Time Status",
            xlab = "Years At Company",
            ylab = "Probability of Continued Employment (%)\n",
            break.x.by=1,
            break.y.by=.1,
            conf.int = FALSE,
            censor = FALSE,
            xlim = c(0, max(df$YearsAtCompany)),
            surv.scale = "percent",
            risk.table = TRUE,
            risk.table.title = "Number of employees still at risk",
            risk.table.height = 0.22,
            risk.table.y.text = FALSE,
            fontsize = 3,
            legend.title = element_blank(),
            risk.table.col = "#808080",
            ggtheme = theme_survival,
            legend = c(0.2, 0.2),
            palette = c("#1F497D", "#9BBB59", "#8064A2", "#C0504D", "#4F81BD", "#F79646")
  )

ggsurv_overtime$table <- ggsurv_overtime$table + theme(axis.title.x = element_blank())
print(ggsurv_overtime)

Employees who work overtime have a noticeably steeper decline in survival probability, falling to roughly 60% after 10 years, compared to about 85% for those who do not. The curves diverge within the first few years.


Show the code 
ggsurv_combined <-
  ggsurvplot(fit = km_fit_BusinessTravel_OverTime,
            data = df,
            title = "\n\nSurvival Probabilities by Business Travel and Over Time Status",
            xlab = "Years At Company",
            ylab = "Probability of Continued Employment (%)\n",
            break.x.by=1,
            break.y.by=.1,
            conf.int = FALSE,
            censor = FALSE,
            xlim = c(0, max(df$YearsAtCompany)),
            surv.scale = "percent",
            risk.table = TRUE,
            risk.table.title = "Number of employees still at risk",
            risk.table.height = 0.35,
            risk.table.y.text = FALSE,
            fontsize = 3,
            legend.title = element_blank(),
            risk.table.col = "#808080",
            ggtheme = theme_survival,
            legend = c(0.2, 0.2),
            palette = c("#1F497D", "#9BBB59", "#8064A2", "#C0504D", "#4F81BD", "#F79646")
  )

ggsurv_combined$table <- ggsurv_combined$table + theme(axis.title.x = element_blank())
print(ggsurv_combined)

Combining the two variables shows that their effects compound. Frequent travelers who also work overtime have the lowest retention, with survival dropping below 50% by year 10. Meanwhile, non-traveling employees without overtime keep the highest retention rate, remaining above 90% through the same period. The difference between the groups becomes apparent almost immediately, with the gap widening over time.

Both of those groups are small, which the risk table under the chart makes visible: at year 10 they hold 24 and 28 employees. The two rarely-traveling groups, which carry most of the data, sit between the extremes with 61% and 87%.


Summary and further directions

These two posts covered the Kaplan-Meier method end to end: building a survival object, reading the curve, and splitting it by employee attributes. The same method is used for customer churn, equipment failure and clinical trials. One caution runs through all of it, and the risk tables under the charts are there to make it visible: once a group is down to a handful of employees, a single departure moves its curve, and conclusions drawn from that stretch carry far less weight than the ones drawn from the early years. The natural next step is the Cox proportional hazards model, which estimates the effect of each variable at the individual level instead of splitting the data into groups.