Project Description and Objectives Human Resources Analytics – also known as people analytics or workforce analytics – is defined as the application of data analytics to improve the way companies identify, attract, acquire, develop, and retain talent. According to Deloitte’s 2017 Human Capital Trends report, 71% of companies see people analytics as a high priority, but only 9% believe they have a good understanding of which talent dimensions drives performance in their organizations1. The increased focus among top leadership on human resources, technological innovation, and the increasingly competitive landscape are all critical factors in driving the adoption of HR analytics across the industries. Successful adoption of HR analytics depends on understanding where and how to apply analytics to identify and implement appropriate initiatives to achieve better organizational and business outcomes. In this case study, we discuss the application of some of the analytical techniques and perform several analyses in R to answer specific questions about a company's workforce. For this purpose, we use the employee dataset produced by IBM scientists that closely resembles real HR data. The overall objective of this case study is to demonstrate use cases of HR Analytics on the sample data to specifically find out a) the best recruiting source for the firm, b) the reasons for low employee engagement, c) whether the current employees are getting paid too little compared to new hires, d) whether the performance ratings are being given consistently, and e) whether employee safety is an issue within the organization. Data Descriptions The data for this case study is based on a simulated firm of 1470 employees with operations in four different locations. The available parameters can be broadly classified into the following thematic areas:
Methodology Generally, any successful HR analytics process is comprised of five distinct steps. The five steps are highlighted in the infographic below. In our case study, we focused on finding the answers to the five specific objectives specified in the ‘project description and objectives’ section above. In order to identify the optimal solutions, we followed an approach that involved a) identifying groups of employees to compare, b) calculating summary statistics such as mean, maximum etc. and (c) compare differences between groups and check whether the differences are statistically significant. Key Findings Identifying best recruiting source for the firm: We focused on two metrics – ‘sales quota attainment’ and ‘attrition rate’ – and examined the ‘average sales quota attainment’ achieved by the employees from each recruiting source. It turns out that the best recruiting source for the firm is through online applications and the worst is search firms. When analyzing based on the ‘attrition rate’, we found that the online applications and the search firms repeated their performance as the best and worst sources of recruiting, respectively. Based on the graphs above, online recruiting sources produced the best hires and search firms produced the worst hires – as measured by these two metrics. Analyzing employee engagement: Engaged employees are those who are involved in, enthusiastic about, and committed to their work and workplace2. Employee engagement can be measured either through a survey or through employee behavioral data. Measuring employee engagement is crucial because high levels will result in higher productivity and lower attrition rates. Based on the survey data provided, employee engagement levels are captured on a scale of 1 through 5 - with 5 being the highest level of engagement and disengaged employees represented by a score of 1 or 2. First, we looked at the average engagement scores by department and found that the sales department had the lowest average engagement score and the finance department had the highest. To further explore, we looked at the other variables – mean salary and mean number of vacations days. The above graphs indicate, interestingly, that the department with the highest percentage of disengaged employees also has the fewest average vacation days taken. There is not, however, a large level of observable variation in the salary levels. In order to discover whether there was a significant difference between the groups, we used t-test and chi-squared tests. We can say that a difference between two groups is “significant” only when we are confident that the two groups are not samples from the same population. Though the employees in the sales department take fewer vacation days on average than the rest of the company, we tested whether this observation is statistically significant as well. We found the p-value to be very low, so we reject the hypothesis that the employees in the sales department took fewer vacation days, on average, than the rest of the company. As such, we cannot conclude whether the low number of vacation days is causing employees to be disengaged or being disengaged is causing the employees to take fewer vacation days. Analyzing wage parity between new hires and existing employees: It is important to perform routine benchmarking of the salaries of both current employee and new hires. A considerable wage gap between existing and new hires might lead to job dissatisfaction and attrition among the existing employees. The focus, as part of this analysis, was to find out whether new hires are paid significantly more than current employees. We found that the new hires have a higher average salary than the more tenured employees and the difference is statistically significant with a p-value of 0.019. However, it is important to check whether the analysis is prone to omitted variable bias (whether the mix of job levels for new hires is different than that of the current employees). We found that new hires are less likely to be hourly employees, and more likely to be salaried or managers. So, to remove this bias, we further analyzed the average salaries of new hires and current employees taking job level into account. As seen from the above graphs, when job level is taken into account, the average salaries of new hires and existing employees is almost the same except for a small difference in hourly workers. However, at a 0.05 level with a p-value of 0.08, we found that the pay difference exists between hourly new hires and hourly current employees is not significant. So, it can be concluded that the new hires are being paid about the same as current employees when job level is considered. Analyzing performance ratings: The employee evaluation/performance appraisal process is inherently prone to bias and is often subjective. This is because it is difficult to quantity the output produced by an employee - especially in the services sector. Bias based on age, race, or other demographic traits may exist and it is important to root it out in any organization. In order to look for evidence of this bias in the simulated company, we analyzed the performance ratings and checked whether one gender was more likely to be a considered a ‘high performer’ - an employee was identified as a ‘high performer’ if they secured a rating of 4 or 5. The graph below shows the distribution of ratings between men and women. Through a chi-square statistical test, we found that the women in this dataset are significantly less likely to be labeled high performers. Furthermore, to check for omitted variable bias, we plotted the distribution of high performers by both gender and job level. Looking at the graph below, we further tested whether women or men have statistically different job level distributions. Using logistic regression, accounting for job levels, we found that men are statistically more likely to be recognized as high performers. Analyzing employee safety: Workplace safety can impact business costs, employee performance, and employee turnover. The focus of this exercise was to see if there was an increase in workplace accidents compared to the previous year and to see what the drivers of this increase might be, if any. We started by calculating the mean accident rate of the two years and tested whether the accident rate was higher in either year. Further, we calculated the increase in the accident rates from the year 2016 to 2017 across the locations. Based on the graphs below, we can observe that the Southfield’s accident rate had the largest percentage increase from 2016 to 2017. To understand why the accident rate at Southfield increased, we checked whether there are any other variables that changed significantly between 2016 and 2017. We observed that employee disengagement changed and that the change was statistically significant. To figure out why there was an increase in accident rate, while taking disengagement into account, we used a regression analysis and observed a statistically significant connection between employee disengagement and accident rates at Southfield. However, without additional information, we cannot know for certain whether employee disengagement is leading to a higher accident rate or a higher accident rate is leading to higher employee disengagement. Key findings from the above analyses are highlighted below:
Based on the above findings, we recommend the following:
Challenges Some of the challenges we faced while undertaking the analyses include the following:
Further, based on our experience, we would like to highlight two crucial bottlenecks for organizations to initiate HR Analytics:
Conclusion This case study highlighted a few use cases of HR analytics. HR analytics can be a powerful way to make a difference in employees’ lives by discovering answers to several workforce related questions. Questions such as whether employees are being paid fairly or not, whether employees are happy, engaged and feel included at workplace, where, when, and how the company should source new talent etc. To generate insights and identify good business recommendations, it is imperative to have high quality, complete data. Further, advanced predictive HR analytics can be deployed to make reliable predictions at every stage of the talent life cycle. With predictive analytics, managers could try to pre-emptively prevent employee turnover or suggest better employee-hiring methods based on known predictors of high performance. Analyzing internal emails, public news articles, social media, and review sites such as Glassdoor can give insight as to what employees may love or hate about the company. This will help management leaders keep a pulse on their organization and respond appropriately with agility. Also, organizational network analysis could inform managers how work really gets done, whether inefficiencies should be fixed, and which employees are crucial connectors or experts on the fringe who the company should try to retain and promote. In conclusion, HR analytics is about solving business problems that can positively impact employees’ wellbeing. Critical thinking and problem-solving skills are the key to this, and data analytic tools/techniques are only a means to solve these business problems. The success lies in asking the right questions and framing problems correctly, capturing and using the right set of data, and the application of the relevant analytical techniques to generate actionable insights that drive meaningful business outcomes. Reference: Human Resources Analytics: Exploring Employee Data in R |