Predictive Modeling for Employee Satisfaction - Data Science

This project focuses on building a predictive model to determine employee job satisfaction scores. The analysis begins with descriptive statistics and correlation analysis to understand the relationships between various factors such as salary, work hours, and project load, and the satisfaction level. Linear regression models are developed, with the initial model showing that only a small percentage of the variability in the satisfaction score is explained by the input variables. The final model, using only last evaluation and number of projects as predictors, also has a low explanatory power, indicating that other variables like work environment, policies, and relationships with colleagues could significantly impact job satisfaction. The project provides an executive summary, model details, discussion on model adequacy, and suggestions for the company to improve employee satisfaction. The project uses the five-step methodology and includes the Minitab results.

Running Head: PREDICTIVE MODELLING FOR JOB SATISFACTION SCORE
PREDICTIVE MODELLING FOR JOB SATISFACTION SCORE
Name of the Student:
Name of the University:
Author Note:

1
PREDICTIVE MODELLING FOR JOB SATISFACTION SCORE
Executive Summary
The main objective of this study is to predict the job satisfaction rating of employees in a
company. In other words, it is required to find a model that can predict the satisfaction score
perfectly. For this study, first some descriptive studies are done on the variables. Then a
correlation matrix is formed which showed that the satisfaction level is negatively correlated
with number of projects ,salary of an employee whereas it is positively correlated with last
evaluation, average working hours and company tenure. The initial model was built up taking
satisfaction score as dependent and salary, average working hours, company tenure, last
evaluation and number of projects as independent variables. The model shows that only 8.95%
variability in predicted variable can be explained by the model inputs. It shows that only the
effect of number of projects and last evaluation are significant at 5% level. Hence, the final
model is built up taking these variables as predictors. The R square value shows that 8.36%
variability in satisfaction score can be explained by the predictors. However the adjusted R
squared value is increased which indicates that the final model gives a better fit to the data. The
model fails to explain 91.64% variability in Y. If some relevant factors like fair policies,
creativity in job, good environment, work hour flexibility, fair relationship with colleagues and
supervisors are taken into account, then it might increase the job satisfaction of employee.
Hence, it is advisable to the company that they might consider these factors in order to increase
working satisfaction of their employees.

2
PREDICTIVE MODELLING FOR JOB SATISFACTION SCORE
Discussion
Data Description
The main objective is to build a predictive model using five steps methodology. The
independent and dependent variables are-
 Dependent variable- Job Satisfaction Level of an employee
 Independent variables-
o Salary of a worker
o Last evaluation
o Average monthly work hours
o Number of projects
o Company Tenure
o Excessive work hours( Yes/No)
o Promotion in last 5 years (Yes/No)
o Left company( Yes/No)
o Has done for than 5 projects? (Yes/No)
o Time greater than 4 (Yes/No)
o Department
o Accident at work (Yes/No)
Descriptive Study
The descriptive summary for the variables (salary, last evaluation, average work hours,
number of projects, tenure, satisfaction level) are shown below.

3
PREDICTIVE MODELLING FOR JOB SATISFACTION SCORE
The table shows that on average the satisfaction score is 0.62 with standard deviation
0.25. Median shows that 50% cases have score less than 0.66. The level of satisfaction 0.44 and
0.46 have highest frequency. Q1 and Q3 shows that 25% cases have score less than 0.44 and
greater than 0.83 respectively. Skewness value shows that the distribution is negatively skewed
and kurtosis shows that it is platykurtic.
Last evaluation of satisfaction has an average 0.72 with s.d. 0.18. The median value
indicates that 50% cases have score less than 0.75 and highest frequency obtained at 0.5(mode).

4
PREDICTIVE MODELLING FOR JOB SATISFACTION SCORE
Correlation
The correlation matrix shows that the strongest positive correlation is found between
number of projects and last evaluation score. Moreover average monthly hours is strongly
positively correlated with last evaluation and number of projects. The satisfaction level is
negatively associated with number of projects and salary of an employee. However the
satisfaction shows a positive association with last evaluation, company tenure and average
monthly working hours respectively.
Initial Model:
Now a linear regression model is constructed taking satisfaction level as dependent and
others as independent variables1. The regression outputs are shown below.
1 Hoffmann, John P., and Kevin Shafer. Linear regression analysis. Washington, DC: NASW Press, 2015.

5
PREDICTIVE MODELLING FOR JOB SATISFACTION SCORE

6
PREDICTIVE MODELLING FOR JOB SATISFACTION SCORE
Interpretation:
 The regression equation shows that the satisfaction level will give a constant value 0.465
when other variables are not taken into consideration.
 One unit increase in last evaluation will increase the value of satisfaction level by 0.456
unit.
 If the number of projects increase by one unit, then the satisfaction score will decrease by
0.07 unit.
 If company tenure increases by one unit, then the score will increase by 0.0028 unit.
 Increase in salary and average monthly working hours do not have any impact on the
satisfaction score.
 The R square value shows that 8.95% variability in satisfaction level can be explained by
the variables.
 The normal plot shows that the residuals approximately follow normal distribution.
 The residual vs fitted value plot shows no specific pattern. This means that there are no
outliers. The residuals bounce randomly around 0 line which means that the linear
relationship is valid and their variances are equal.

7
PREDICTIVE MODELLING FOR JOB SATISFACTION SCORE
 It can be observed that the p value for the variables last evaluation and number of
projects are less than 0.05. Therefore effect of these variables are significant at 5% level
of significance.
Final Model:
Now a regression model is built up taking satisfaction score as dependent and last
evaluation and number of projects as independent variables. The necessary outputs are shown
below.

8
PREDICTIVE MODELLING FOR JOB SATISFACTION SCORE
Interpretation:
 Satisfaction level will show a constant value 0.4967 if the effect of the independent
variables are null.
 If the last evaluation score increases by one unit then satisfaction score will increase by
0.475 unit.
 One unit increase in number of projects, then the satisfaction score decreases by 0.06
unit.

9
PREDICTIVE MODELLING FOR JOB SATISFACTION SCORE
 The R square value shows that 8.36% variability of satisfaction score can be explained
by the independent variables. It can be noted that the R square value decreases as
variables are removed from the model. However it can be seen that the value of the
adjusted R square is increased which only happens when deletion or inclusion of
variables improves the model. Therefore it can be concluded that the final model
provides a better prediction for job satisfaction of an employee in terms of last
evaluation and number of projects done by him.
 The normal plot for residuals show that they are approximately normally distributed.
 The plot for fitted value vs residuals shows that the residuals do not follow any pattern
which indicates the absence of outliers, existence of linear relationship and also implies
that the variances of error terms are equal2.
The R square value is very low for the final model. It fails to explain 91.64% variability of
the dependent variable in terms of predictors. Hence the model fitted is not so much appropriate
for the given dataset. It can be noted that there are many factors like working environment,
policies, relation with co-worker and supervisor, work hour flexibility, job creativity play
significant roles for job satisfaction. Hence, it can be suggested to involve these variables in the
dataset in order to obtain an appropriate predictive model for job satisfaction rating.
2 Chatterjee, Samprit, and Ali S. Hadi. Regression analysis by example. John Wiley & Sons, 2015.

10
PREDICTIVE MODELLING FOR JOB SATISFACTION SCORE
Bibliography
Hoffmann, John P., and Kevin Shafer. Linear regression analysis. Washington, DC: NASW
Press, 2015.
Chatterjee, Samprit, and Ali S. Hadi. Regression analysis by example. John Wiley & Sons, 2015.