Employee Attrition Analysis for Divis Labs

Submitted by
Abhishek
Shashank
 Contribution
Data Type Observations
in %
Description 0 - No turnover 1245 85%

1 - Retirement 27 2%
• Dataset has 76 variables with 19
categorical and 57 continuous.
2 - Voluntary Resignation 105 7%
• Type variable is the target
variable indicating whether an 3 - Involuntary 54 4%
employee is currently active in
company or not. 4 - Job Termination, Employee
is Fired 39 3%
Exploratory Data
Analysis
Distribution of event type

• We decided to plot the

distribution of different event
types in order to identify the event
type corresponding to maximum
number of employees leaving the
company. We found that all event
types are in different proportion
with respect to employees and
maximum number of employees
belongs to Voluntary resignation
event type
Business Travel Vs Turnover Types:

• We plotted this graph to identify

the relation between variables
turnover types and Business
Travel. We found that in all the
four event types - Involuntary, Job
Termination, Resignation and
Voluntary retirement, people who
are travelling frequently are more
likely to leave the company.
Job satisfaction Vs Turnover
Type

• We has created a new variable “Satisfied” and

plotted SGPLOT in order to check the effect of
job satisfaction and found Employees who are
not satisfied are leaving more among
voluntary resignation type. This might be one
of the main reason for employees to leave the
company. However, in all the other three types
majority of people are satisfied with their jobs
though they are leaving the company.
Education Field Vs Turnover Type

• We found that education field is

other factor affecting the attrition
rate in all the event types. All
Employees who belongs to Life
Sciences and medical field are
more likely to leave the company
in among other education fields.
Overtime Vs Turnover Type

• In this scenario By plotting

Overtime Vs Turnover types, we
could be able to find that
employees who are doing
overtime are resigning the
company in the types (voluntary,
Involuntary & Retirement). Those
who are not doing overtime, they
are being terminated from the
company.There seems a possibility
of overtime as one of the key
performance indicator in the
company which is main factor for
work pressure.
Gender Vs Turnover Type

• Gender is one of the key variable in order to

understand the development of a turnover
decision. To test this fact, we have plotted
Gender Vs Turnover type and found that men
has greater rates of actual turnover than
women in each event type. But this may be
because of the proportion of females is less in
comparison to men in this data.
Frequency table for Gender
Vs Turnover Type

• We have generated frequency table and

found that males are more in number
than females leaving company and
turnover is independent of Gender.
Frequency
• From the frequency table results of all event
types, we can say that frequencies are not same
for all event types.
• Out of 1470 employees in Divis 225 are leaving
the organization because of Retirement,
Voluntary, Involuntary and Job Termination.
These are attiring at different frequencies, so all
these event types are different.
• We did further analysis to say that
47%(105/225) employees leaving the company
voluntarily and the other employees leaving the
company due to retirement, job termination and
Involuntary and this has become a concern for
the organization.
Insignificant Variables

Variable Reason for removal

We removed it because it just shows a
Employee Count numeric value of 1 for all the rows.

It does not contribute to the model because

Over18 this variable has a character ‘Y’ for all the
rows in the dataset.
Significant Variables

Distance From Job Marital Years

Home Involvement Status At Company
Years There were few variables which
Daily Rate Job Level Stock Option Level Since
were very significant and directly
Last
Promotion
contribute towards the target
variable “Type”.
Environment Job Role Total Working Years
Satisfaction
Gender Job Satisfaction Work Life Balance
Hazard rate is
not same for
all event types
Hazards for event type together or Separately- P Value
Comparison
• By looking at the p-values of
Wilcoxon test, we can say that all
event types are significant and
different from each other except
Involuntary Resignation and Job
termination because these two
event types are not significant and
there seems a possibility of
combining these two event types.
Survival estimates for different turnover Types
 • Cox Proportional Model is much like logistic regression,
where it enables us to know the hazard ratio between two
groups considering the covariates.
• The hazard is the probability of employees leaving the
company(Turnover) to the employees who are not leaving
the company(No Turnover) at a given point of time.

Cox • The primary reasons to choose Cox Proportional Model is it

does not need any distribution to be known to run the
model.
Proportional • We have considered all the four different turnover types and

Model modelled the data separately for all the turnover types
together and individually.
• We have built different hypothesis for four different types
focusing to find the significant covariates which affects the
types of turnovers in the company, which will help the
organization to focus on these attributes to reducing the
attrition rate.
Log-ratio test shows that all event types need to be handled separately
 PHreg model
for involuntary
resignation and
job termination

Though we created individual models for involuntary

resignation and termination event types, after fit statistics
check we observed that a nested model with both involuntary
and termination is doing better than the individual models.
We created a Model with involuntary Resignation and
termination as events and all other types as censored.
We have got this model which is significantly way different and
much better than the Null model, we can see that the fit
statistics also differing between model with and without
covariates
Hazard Ratio Estimate
for involuntary
Resignation and Job
termination
Log-ratio test
 • Frequency plot of all event types indicates frequency for all turnover types are totally different from each
another which implies each attrition type is at different rate thus we can say that all turnover types are
different.
• We have plotted LLS graphs to observe the dimensionality between turnover types and found that
Involuntary Resignation and Job termination are linearly related. In order to go in detail further, we have
implemented proportional hazards model and found that as per Wilcoxon test ,that their p-values are
not significant, and they are not different thus we used Involuntary Resignation and Job termination

Effect of •
together.
We also performed log ratio test and observed that p-value for all the combined event type is 0 which
means all events types are different and need to handle individually.

Covariates & • We performed log Ratio test on combined model of Involuntary Resignation and Job termination and
found p-value as 0.193 which is not significant .hence we can use the nested model in this case.

Conclusions • For Retirement – Age, Job Involvement, Business Travel, Bonus and Interaction Variable Years in current
role are significant. For Voluntary Resignation-Stocks, Overtime, Business Travel, Environment
Satisfaction, Job satisfaction and Years in current role are significant.
• For Involuntary Resignation and Job termination – Environment Satisfaction, Distance from home, Stocks
and Interaction variable -Years in current roles and number of companies worked are significant.
• We found that Employee Bonus is a significant factor for employees who are retiring from the company
but has not significant effect on the other event types.
Recommendations
▪ Management should focus on overtime of the employees with
alternative approaches.
▪ Management should try reducing the Business travel as we
observe employees who are travelling more are more likely to
leave the company.
▪ We have observed a strong positive trend between stocks and
the employee turnover, employees holding own stocks in the
company are less likely to leave.
▪ By creating work life balance for employees with good
leadership can lead to a better job satisfaction and involvement
levels. This might reduce the turnover rate.
▪ We found that Employee Bonus is a significant factor for
employees who are retiring from the company but no significant
effect on the other event types.