Storybook for Employee Attrition and Job Satisfaction

Attrition Analysis:

Storybook for Employee Attrition and Job Satisfaction

Assignment 4

Masroor Dhar

Storybook for Employee Attrition and Job Satisfaction

An issue that every company deals with is attrition. Attrition is defined as the decline in the number of employees or staff in the organization through regular means, such as resignation or retirement. Hence the ability to slice and dice attrition many ways to understand trends and their root-causes can seriously help leadership to make the required changes to build a healthier more performing sales force. Numerically analyzing attrition is a bit tricky. This stems from the fact that the base of employees is continually in flux. Every month new hires join the salesforce, some employees are involuntarily terminated, some voluntarily leave the company and some others go inactive without leaving the company like when they go for a long-term leave of absence. Companies spend a lot of resources in finding good talent to work for them and all that goes in vain when they quit prematurely. Well if the company would know beforehand the risk of losing that resource they would plan ahead to intervene or have a road map of having that resource retain. This is a general HR problem in every company and had been in my mind to solve for a while now.

We will make use of Watson analytics to develop the two display models to address the problem of attrition and job satisfaction. In addition to that we will add twitter analysis to make a precise storybook for our end users.

Dataset Description and Evaluation

The data set used to employee attrition was downloaded from the sample dataset of Watson analytics. The sample data has 1,470 rows and 35 columns (i.e. 1, 4700 instances and 35 variables). Variables include each employee’s age, distance from home, amount of business travel, education level, whether or not the employee left the company, and several other employee characteristics (Appendix, Figure 1.). Some of the important variable details consisting of its data type, measurement level and its description are mentioned below in the table. In addition to encoding the categorical data type variables with their ordinal fields (Appendix, Figure 2.).

The data set has no missing value or any imbalance in influential categories. However, some of the variables like “standard hours” have a constant value of 80 recorded for every field. The data quality score for these fields is 0, and was excluded from the analysis. Also “employee number” and “over18” variables have a unique set of values with data quality score of 0, and were also excluded from the analysis. After processing and transforming your data, we can do some exploratory data analysis (EDA) to gain the insights of the data set and its value addition to your final goal.

In order to get the more detailed overview we will create a data group and hierarchies for the variable distance from home and separate the distance in three groups with the employee that live in radius of less than 10 miles from office. Second group is of the employee living in radius less than 20 but greater than 10 miles. Adding hierarchy to the data will help you navigate through large amounts of data, allowing you to drill up and down as required.
Assembling Display for Attrition

From my last assignment, HR employee attrition data set allow me to find specific group of employee who is most likely to stay with company. We will make the dashboard from the discoveries evaluated earlier. The first slide of the dashboard will be the predictive decision tree model for attrition. Attrition is a categorical target, so we predict the value before developing the model and in this case we will discuss the value of attrition as “No” and gives us the predictive strength of 86% (Appendix, Figure 3.).

The second slide of the dashboard is decision rules, which states that the employee working in the company for more than 2 years, having stock levels greater than 0 and obviously not working for overtime are not going to leave the company(Appendix, Figure 4.).In this case we can see that we 97% sure for the first to decision rules that the employee will not leave the company.

The third slide uses the spiral visualization which shows you the key drivers. The closer the driver is to the center, the stronger the influence that driver has on the target; even combination of fields can be a stronger driver. The predictive strength of the combined variables is 84% and which is a good percentage to predict the outcome of the attrition. Looking at the visualization itself gives an insight that the job level and overtime are the strongest predictors for the attrition of the employees (Appendix, Figure 5.).

Some more visualization were used by comparing it with the distance from the home and gender which clearly depicts that employee from both the gender living far from work place has high rate of attrition. It also about the effect of employee who has worked in less number of companies has the higher rate of attrition (Appendix, Figure 6.).

Assembling Display for Job Satisfaction

We will start our dashboard by using the spiral visualization. Looking at the visualization itself gives an insight that the number of project an employee worked plays an important role, as the employee who worked on less projects are most likely to leave the company. The predictive strength of the combined variables is 38% and a heat map gives us a more detailed look at the problem about the employees that leave seem to either work a lot or below average (Appendix, Figure 7.).

One more slide for the dashboard is the set of decision rules for the employee satisfaction. Since response is continuous target variable, linear regression based approach was used to create the model and then the key insight indicators highlight unusually high or low groups. This model gives us the predictive strength of 44%. The color of the node is based on the average of the target for the measure. The higher the average values of the target for a node, the darker the color (Appendix, Figure 8.).

In addition to that more model like heat map were used to give us more insight about the value for highest average satisfaction for the employees, which is 0.71. Hence those employees will not leave the company. Through our analysis, managing the level of satisfaction is the key to keep employees with the firm. This is especially important for employees who have been around for more than 3 years. Other than that the employee evaluation and number of projects should also be monitored (Appendix, Figure 9.).

Twitter Analysis on Attrition

It will be very helpful to use the data to compare with our assignment data. However, it seems like Twitter does not have the right data to go with the HR employee attrition data set. The analysis was not clear from the Twitter data. For instance, the most relevant one what we searched for is the relation of the two hostages #hr and #attrition with the people who tweeted on the topic with the negative sentiments. All sentiment negative signals have same affection on the hashtags and are evenly distributed (Appendix, Figure 10.). In addition to that most of the responses were neutral, which is quite obvious as most of the people do not write about their professional life on the social media (Appendix, Figure 11.). They only provided the percentage of male and female that has mentioned, but did not be specific of what topic area (Appendix, Figure 12). Twitter data will work well, maybe in different topic area such as, foods, games, and place it provides us a very good report.

Infographic for More Insights

One more assembled dashboard in the form of infographic is made for the storybook. The slides depict the attrition among the different job roles. People who are married and work as research scientist and sales executive have highest job involvement and less attrition. Applying the filters of job satisfaction and setting it to lowest value of 1 and 2, we can easily infer the relationship between the rates of attrition in comparison to the working overtime across the departments more easily. It’s evident that the employees who work overtime are the one with the lowest job satisfaction, hence the rate of attrition is always high (Appendix, Figure 13.).

Storybook for Employee Attrition and Job Satisfaction

After you discovered the new insights, and want to share that with others. You can assemble these curated top findings in a storybook. A storybook is a guided analytic template that targets specific business problems and accelerates your time to achieving insights. We will focus on the most pertinent information by selecting our displays for employee attrition and satisfaction problem, which then helps you interpret results to arrive at insights. To create the storybook we tap on expert storybook to create the new one. Give your storybook a title and add a description to help end users to understand it properly and effectively. For better look of the storybook we need to change the background picture as per the requirement of the project. In this assignment, we developed four different type of assembled displays mentioned above, and inputted them into the storybook.

The contents explained in the first assembled displays defines the decision rules for employee working in the company for more than 2 years, having stock levels greater than 0 and obviously not working for overtime are going to stay with the company (Appendix, Figure 14.). Looking at the next visualization itself gives an insight that the job level and overtime are the strongest predictors for the attrition of the employee’s (Appendix, Figure 15.). It clearly depicts that employee far from work place has high rate of attrition. It also talks about the effect of employee who has worked in less number of companies has the higher rate of attrition (Appendix, Figure 16.).  People working for overtime and with the lower job level are most likely to leave the company with predictive strength of 84% (Appendix, Figure 17.). Employees who do not travel at all and are not working overtime are most likely to leave the company (Appendix, Figure 18.).