Are there hidden patterns in the gender pay gap? The gap exists despite regulation, hype, and preoccupation with the subject. The extra mile for HR and People Analytics professionals on this topic is related to analytics skills and critical thinking. I’ll bind the two ends, presenting a short case study with practical advice and an opportunity to challenge and use your critical thinking. (The article was based on my lecture at the People Analytics World, April 2022, where I offered out-of-the-box thinking on this traditionally unsolved issue. Also, read my list of Public Speaking)

In this article, I’ll answer four questions. First, why are we failing to see the hidden patterns in the gender pay gap? Second, what are some hidden patterns based on data? Third, how can HR professionals work better with data scientists to spot the hidden patterns? And lastly, how can we generalize this case study for up-skilling and re-skilling in critical thinking and an analytical mindset?

Why are we failing to see the hidden patterns in the gender pay gap?

I’m sure you heard a lot about the gender pay gap, at least in the media in your countries. For example, in Israel, my country, although employers must report the gender pay, data reveals that for 1 NIS a man earns, a woman makes 68 cents. Why is that happening? What can we do about it?

As a citizen and a professional in People Analytics, I consider the gender pay gap more than a compliance issue. It’s a mission to help women succeed at work and in life, which can influence families, communities, and society.

I believe this is a common goal or aspiration for most of you and many HR professionals. However, I’m not sure we necessarily share a common perspective on how to start making an impact.

My perspective comes from data and analytics. The data in your organization is where you should start, shed light on the current situation, understand its factors, direct your intervention, and guarantee that your insights are discussed in a broader context of the business and workforce markets.

However, many HR professionals start elsewhere – with programs, being confident about the organizational development point of view. Even when data is their starting point, it often takes the form of reporting and not exploring.

What is the difference between reporting and exploring? To explore data, you must have an analytical mindset. It enables you to analyze information and identify patterns in the data to solve problems. You use your curiosity by asking the question, “why?”.

I don’t expect HR professionals to become data scientists and run advanced statistics to identify patterns in the data. Still, I’m sure that being a better inner client of data professionals and solutions is essential, and a key to your success is asking “why?”. It will enable you to tell a clear story, impact any topic related to people, track improvement and progress, and certainly contribute to closing the gender pay gap.

What are some hidden patterns based on data?  

Let’s take the extra mile beyond reporting and dashboarding on the one hand, and don’t jump to recommended interventions and programs on the other hand. Instead, we will focus on exploration.

What is exactly that extra mile beyond reporting and dashboarding? Dashboards enable us to present different metrics and KPIs and answer the questions: Did we reach our goals? How far are we from achieving our goals? However, by using dashboards, we can’t answer the question: Why? Instead, we need to analyze the factors that drive those KPIs presented on our dashboards.

Regarding the gender pay gap, we will explore the data beyond finding differences between men and women in compensation. Instead, we will ask “why?” to explore how those differences occurred, implying what we should do about it.

While reading the rest of this article, I suggest you imagine that you are in a meeting with a data scientist that supports your work as an HR leader. Let’s assume that your academic background is not higher than a bachelor’s degree in social science. Trust me, your common sense and curiosity are good enough to lead the conversation. Don’t worry. This short case study has no advanced analytics, only statistics suited to your hypothetical background. (But if you’d like to explore the R code that generated the following visualizations and results, visit my GitHub profile).

My source and inspiration for this case study was a dataset of employee salaries in a municipal authority organization. For public transparency, this organization shared a few years ago its dataset, which contained almost 10 thousand records. The open data included annual salary information and some demographics. Although People Analytics in your organization probably involves the integration of additional data sources into such analysis, from different platforms, like recruitment, learning, and performance, this simple dataset is sufficient for our purpose.

First, I created an anonymized and randomized version of the dataset. So, it would be impossible to point to individuals or even recognize the organization from the following findings. But I guarantee that the dataset I used is realistic. Then, I ran some Inferential Statistics. I used only binary gender categories in the analysis, men vs. women, since that was a classification in the dataset. Some organizations, however, may use more gender categories, but that would be beyond our scope. Of course, like any public organization, I assumed that the contributing organization was subjected to strict regulations regarding equal pay. But only going beyond the basic comparison between men and women enabled me to spot other patterns and reach some insights. So, without further ado, let’s explore the findings.

Do men and women on average earn the same in this public sector organization? Well, almost. Women’s and men’s annual salaries were 73K vs. 77K dollars per year. So, for every 1$ a man in this organization earns, a woman makes 95 cents. It is not a huge gap, at least not compared to the pay gap reported in my country. But it is worth exploring the annual salary distributions.

When you explore only the averages, you lose information, e.g., outliers. Therefore, we want to explore both indications for center and dispersion of the earning distribution without losing information. So, I placed here two visualizations, a boxplot near a histogram with a density plot, and set the genders vertically, one on top of the other, so that the comparison would be easy for the bare eye.

You typically won’t see such charts on your dashboards, but this is a common way to start your exploration of the data, so I suggest you get to know and leverage these visualizations. Notice a slight difference between men and women in the deviation of histograms on the left from the shared normal approximation curve. Which gender deviates at the lower part of the distribution or the higher part? I bet you can see the pattern. Also, look at the boxplot’s centers, which represent the medians. We’ll further examine the sources of variance in salaries to understand how men earn more than women.

But before we do so, if you were leading the conversation with a data scientist, how would you criticize these numbers? You would probably raise a question about male and female occupations. The dataset includes some roles with both genders and other positions held by only men or women.

After screening out those male and female occupations, I repeated the analysis and got similar results. The pay gap only slightly increased, to 72K vs. 78K respectively of women vs. men earnings. For gender-diverse roles, women make 92 cents for 1$ men do. In your analysis, you should explore each diverse role and sort roles by gender pay gap to report where the gap is higher. Since we have a few hundred occupations in the dataset, this would be beyond our scope.

If you report the gender pay gap using a dashboard, you may slice the annual salaries of genders by age, tenure, and additional demographics. However, your dashboard slicer won’t point to the interactions of variables. An interaction may arise when you explore the relationship between more than two variables. The effect of one causal variable on an outcome depends on the state of a second causal variable. Do we have such interaction in our dataset? And if we do, what would it tell us about the causes of the gender pay gap?

Let’s explore gender with only one additional variable. What would be your first variable of choice from background variables and demographics? My choice was tenure.

Exploring gender pay averages across tenure ranges reveals that while both genders start at a similar earning level and are promoted while gaining tenure, men are promoted at higher rates, as the different slope indicates. In addition, a statistical procedure called ANOVA analysis of variance (that you may recall from your fundamental statistics learning) reveals that the interaction between gender and tenure variables is significant, meaning that the slopes are not random.

Interestingly, when filtering the diverse roles, we can see that the gap is even more comprehensive as years go by. So, in this specific dataset, it is clear that some explanation of the gender pay gap is related to things that happen along with the careers in this organization. Any intervention should consider something that happens along the way. And we found this hidden pattern only by adding a single variable and analyzing it using multi-variate statistics.

What if we add more variables? What if we use predictive analytics? What would we learn? Let me give you some clues on further exploration.

Assume you integrate compensation data and additional datasets covering performance reviews, promotions, and internal mobility. You would be able to explore, across different roles, how women are compensated and promoted in comparison to men. Additionally, you can study biases in the way women are evaluated compared to men. For example, are they perceived differently regarding performance, self-management, relationships, and potential leadership?

Furthermore, What is the correlation between yearly reviews and promotions for the two genders? Conducting such analysis in other real organizations revealed differences between genders in such perceptions. Do men and women who received similar reviews get a similar promotion? These questions can undoubtedly shed more light on what’s happening during work tenure, explaining the growing compensation gap between genders. They can build a story and a business case for intervention.

HR professionals can work better with data scientists to spot hidden patterns

The practice of data science is multidisciplinary. It encompasses three general skills – the business domain of expertise, statistical modeling, and hacking skills. Therefore, a crucial part of your challenge in People Analytics is the effort to establish communication between different professionals who hold different skills.

You heard a lot about the People Analytics journey that enables HR professionals to become more strategic because they speak the language of the business and impact using the right questions and insights derived from people’s data. But they can support decision-making only when they communicate those questions to data scientists.

I encourage you to be proactive in your conversation with the data scientist that supports your work. Ask, “why?”; Suggest hypotheses; Challenge explanations, and offer alternative descriptions that the data scientist can confirm and disprove. As a domain expert in human resources, organizations, and the workforce, feel free to be creative and lead the exploration of the data. Your domain expertise is invaluable in completing the data scientist skills.

Part of your role in leading and leveraging People Analytics is being a translator, enabling this communication. It would be best to make sure that the data scientists understand the business needs in workforce-related analysis. It would help if you articulated the right business questions, so the research findings yield the best storytelling with data.

Generalizing the case study for up-skilling and re-skilling in critical thinking

This case study demonstrates how you can lead the story to more impact only by adding a single variable to the analysis beyond dashboarding. Your proactivity is critical. You don’t have to be a data scientist. But leveraging an analytical mindset when working with data scientists can move the needle beyond simple metrics.

The gender pay gap is part of a broader topic of Diversity, Equality, and Inclusion. Other groups should be analyzed and handled to dismiss bias and discrimination by ethnicity, age, sexual orientation, disability, and other minorities. But to strengthen our analytical muscles, it is much easier to start with data that the organization continuously collects, as long as it pays its workforce. The pay data is usually in good shape and integrity, the topic is regulated in many countries, and it is easy to find benchmarks across industries and economies.

Do you need particular procurement of compensation software to examine the gender pay gap? Maybe this would help. But I wanted to demonstrate that you can analyze it, for free, with R, maybe with some help from a data scientist.

HR departments establish a People Analytics function with data professionals or external consultants. As people processes are analyzed, the gender pay gap can be explored across recruitment, development, and retention. We focused on pay data in this demo, but I offered some direction to more potential projects.

We also see a demand for the topic in the HR-Tech industry. But we focus on a DIY (Do It Yourself) approach. This approach may help you sharpen your analytical mindset, leverage data that you have immediately, and eventually enable you to be players in the procurement processes of People Analytics tools as you continue your journey.

I offer the DIY approach in my introductory course to People Analytics, and I encourage HR leaders to be proactive when I support them in research and data science projects. I believe that it eventually enables you to make people’s data valid for informed decisions and employee experience and improve business performance and enhance competitive advantage.

The post Finding Hidden Patterns in Gender Pay Gap Data appeared first on Littal Shemer Haim.

Can you survive boring results in your People Analytics project? Many People Analysts in my network happily share remarkable findings, and obviously, they like to tell an exciting story from their data analysis. But unfortunately, sometimes, their success is mistakenly considered as discovering significant findings.

Insignificant results and storytelling with data

There are cases in People Analytics in which statistically insignificant results build a good story with data. Furthermore, sometimes insignificant results are the desired outcome.

One example is when there shouldn’t be significant differences between groups in the organization, and indeed, you don’t find them when you compare by demographics or organizational background variables. So please don’t exhaust yourself exploring the data until you find some notable differences in such a case.

Instead, stick to the business questions and research objectives, and summarize your analysis with what you already got. You may be surprised to find out that the sponsors of your analysis project embrace the so-called boring story and maybe even thrilled to have it.

Another example is when you gain insights that confirm an already known fact or domain knowledge. In such a case, not-so-exciting results are good results. Furthermore, validating established professional wisdom may say something about your data quality and integrity. So as long as you follow best practices in analytics, embrace the tiresome results and make sure to continue the data maintenance.

Leverage multivariate statistics for new insights

Of course, there is always some value in confirming and validating well-established hypotheses. However, if you do want to bring something new to the table, you can always try to enrich your analysis.

For example, you can integrate data from various sources and explore interactions of variables. Unfortunately, multivariate statistics is not always in the skillset of people analysts, but they can contribute new perspectives to any discussion when it does. Therefore, I recommend going back to your notebooks of statistics fundamentals. Perhaps you’ll find helpful technics in your good-old learning materials.  

In my lecture at the People Analytics World 2022, I demonstrated a case of an organization that was subjected to strict regulations regarding equal pay. Comparing men’s and women’s salaries did not reveal striking differences. However, adding a single variable to the analysis, e.g., tenure, uncovered some hidden patterns of the gender pay gap related to things that happen along with the careers in this organization.

Know how data can play tricks on you

Many statistical fallacies lead to mistakes in data analysis and interpretation. Therefore, I encourage everyone who practices People Analytics in each part of the value chain of HR to explore common errors and be familiar with examples in order to avoid them.

In my introductory course, The People Analytics Journey, I mention the most prominent fallacies based on my experience. However, in this context of surviving boring results in People Analytics, I want to describe only one: Simpson’s Paradox.

Sometimes a statistical relationship that you explored within the entire workforce in your organization could be reversed within subgroups, e.g., gender, tenure, or education level groups. For example, I demonstrated Simpson’s Paradox in a case study where absenteeism was mildly negatively related to overload. However, when I analyzed the relationship within education level subgroups, it was positive among mid and high education levels.

So, when you are surprised by the boring results in your People Analytics project, especially when you expect to find a statistical relationship between variables based on the literature or your own experience, try to dig deeper and explore the statistical relationship among subgroups.

Support from subject matter experts

The pressure to bring exponential value insights is not unique to People Analytics. However, People Analytics practices are special because they still struggle to demonstrate the business case and ROI in many HR departments. But subject matter experts, i.e., HR and OD leaders, have a valuable role in supporting People Analysts. They can help them by preventing them from keeping on tediously analyzing when nothing interesting comes up in the analysis or guide them to dig deeper in the right direction.

It’s more than saving time and resources; it may prevent analysts from feeling incompetent when there is no exciting outcome to their unfruitful efforts. But, of course, to do so, HR and OD leaders must gain an analytical mindset.

On this journey, I always recommend a helping hand of an expert in organizational research and data science. Such engagement enables you to bridge the skillset gap of various players in a People Analytics project, push the entire team to more professional applications, and leverage the opportunity to increase your impact.

The post People Analytics: Survive Boring Findings appeared first on Littal Shemer Haim.

Educating and mentoring HR professionals to embrace the practices of People Analytics is a challenge. There are barriers, and it takes time and effort to overcome them. However, one issue remained unsolved for years: The lack of open HR data to practice on. Although there are many inspiring case studies of People Analytics, obviously, organizations don’t share their people data for the sake of learning. Simulation-based data may be an alternative, though usually it is oversimplified and lacks real or interesting patterns to explore.

 

A Practice with Open Data

In my recent teaching initiatives, e.g., the People Analytics session in Lahav Executive Education at the University of Tel Aviv, I wanted to demonstrate HR managers that their academic background, professional experience, and their common sense, is enough for exploring organizational occurrences and effects based on data. HR managers don’t have to become data scientists in order to conduct People Analytics projects. But they do need to communicate with Data Scientists, bring them business questions to study, and request research outputs. For that reason, I constantly search for open HR data and use it in learning sessions. Fortunately, I could present a case study of Gender Equality, that theoretically and methodological was based on a real project, but the analytics part was conducted on open data that was offered by other organizations.

For the Analysts and Data Science enthusiasts among my readers, it is worth mentioning that although it is not the first time I demonstrate People Analytics practices based on open data, this time my objective is a bit different. I did not use practical Machine Learning in this case study. The analysis process was based on research methodology and Statistics that a Bachelor of Social Science, i.e., someone with a B.A. degree, should understand and can comfortably communicate. Nevertheless, I used R for my analysis, because I believe that HR people who may not have learned or used R and manage to receive analytics from an inner supplier or an outsource service, should have a grasp on how a desktop of a Data Scientist looks like, and what in the functionality of R Studio makes it so popular.

My source and inspiration for the dataset was Montgomery County Maryland’s employee salaries in 2017. The open data included annual salary information such as gross pay and overtime pay for all active, permanent employees, and some demographics. The reason for opening this dataset to the public is the Digital Government Strategy of Montgomery County Maryland which aims to serve residents, employees, and other partners better. In this case, it serves the purpose of education, in an open-source community of People Analytics students, professionals, and enthusiasts. However, the dataset used is anonymized and randomized.

 

Gender Pay Gap

Pay transparency is among Global Talent Trends in 2019, according to LinkedIn. But “Transparency isn’t the goal. The goal is paying everyone fairly”, as Anil Dash, CEO at Glitch was wisely quoted in the report. Transparency forces Organizations to make sure they keep the compensation balanced across genders and other groups’ characteristics. Although people share salaries on sites like Glassdoor and LinkedIn, only 27% of companies are transparent about pay. The first step to establishing pay transparency, as recommended in LinkedIn’s report, is to conduct an internal audit, and explore how the company’s pay compares to competitors and whether it has a major pay gap across gender, race, and those in similar roles. If significant inequities are found, a detailed plan to fix them is recommended.

A pay gap audit or exploration may be a People Analyst’s task. However, in the People Analytics project, descriptive statistics is not enough. We need to go deeper into understanding the reasons for our findings and the directions for a solution. In the following analysis, I included some diagnostics and Inferential Statistics, to understand the reasons for the patterns in pay data. I assumed that as any American public organization, Montgomery County Maryland is subjected to some kind of strict regulation regarding equal pay. But only going beyond the basic descriptive statistics enabled me to find some interesting patterns. So, without further ado, let’s explore the findings.

Gender Pay in Montgomery County Maryland

“Telling a story with data” is almost a cliché in our field. Nevertheless, there is no substitute for the exploration of data visually, before moving on to test the hypothesis. There are plenty of visual tools out there. The great thing about R, however, apart from its price (free!), is the flexibility it enables in creating the story and reproduce it again and again as the data is updated. In the following description of my analysis, I did not explain every term in statistics, since I assume the readers learned them on their undergraduate studies. But “no one remembers”, right? So, the links in every statistical term may walk you through a “memory refreshment experience”, if you choose to follow them. 

I started my exploration, as shown in Figure 1, with the pay distributions. I intended to present, in a single slide, both common and separated gender pay distributions. I also wanted to explore both indications for center and dispersion, without losing information about outliers. So, I placed a boxplot near a histogram with a density plot and ordered the genders vertically, one on the top of the other, so the comparison would be easy for the bare eye.

If you look closely in Figure 1, you’ll notice a little difference between men and women, both in the deviation of histograms from the shared distribution, i.e., that normal approximation curve, and the center of the boxplot, which represent the median. Running t-test resulted in a p-value below 0.05, which means that on average, the pay differences between men and women are statistically significant. This significant result is impacted by a large number of cases in the dataset (about 9400 employees). The average yearly pay gap is about 4.5k US$. (I repeated the visualization and t-tests for all pay variables I had in my dataset, but for the purpose of simplicity, let’s remain with only one variable).

 

Figure 1: Gender Pay Distributions

Obviously, the average pay gap is not the whole story. Additional variables should be added, to deeply understand the source of the gap. Adding background variables, e.g., full vs. part-time job and tenure may change the story. For the analysis presented in Figure 2, I had to create new variables based on the raw data. I mention it because it is important to take into consideration that, usually, the data you download from your systems won’t be ready for analysis. A significant part of the Data Scientist time will be invested in cleaning, mounting, and preparing the data for the analysis.

Exploring gender pay averages across tenure ranges reveals that while both genders are promoted while gaining tenure, men are promoted with higher rates, as the different slope indicates. Running ANOVA reveals that the interaction between the gender and tenure variables is significant, meaning that the different slopes are not a random occurrence. Such interaction was not found between gender and full/part-time. However, we do witness full-time employees promoted at a higher rate, in comparison to part-time employees, as slops indicate. This interaction, between full/part-time and tenure, is also significant.

 

Figure 2: Gender effect, Tenure effect, Full/part-time effect

 

But who holds most of the part-time jobs? Apparently, the proportion of part-time employees in Montgomery County Maryland is significantly higher among women (18%), in comparison to men (3%). In other words, the accumulative gap between men and women throughout their careers, as they gain tenure, may stem from their assignment in full and part-time jobs. In a Linear regression model that explains the annual salary by gender, assignment, and tenure, the gender is not a significant predictor, as opposed to the other variables: tenure and assignment. Together these variables explain 37% of the variance of annual pay, which is a fair result, but still, other factors impact it too. Positions and occupations may be among those factors.

Indeed, a critical reader may raise a question about the male’s and female’s occupation. The dataset includes some occupations with both genders and other occupations with only men or women. I repeated the whole analysis after screening out those male and female occupations, and I got similar results. Yes, analysis within each occupation is also needed. However, there are 390 occupations in this dataset, so I prefer to leave this task to People Analysts in Montgomery County Maryland. (For dynamic charts of this case study, by departments for example, please visit my GitHub)

 

Additional thoughts

The gender pay gap analysis in this article is straightforward. Most HR managers with a B.A. education can handle it, with a little help from a data scientist on some occasions. I encourage HR practitioners who start their journey in People Analytics to practice this analysis. The data is available, and the insights may be vital. According to Gartner’s Digital Employee Experience Survey in 2018, #1 in the top ten memorable experiences that affect employee experience is “Being discriminated against at work”.  No doubt that transparency and closing the pay gap is crucial for employee engagement and indirectly to employer branding.

My last note may be the most important. Women still don’t get their fair share, according to an analysis by Visier. Data from this People Analytics platform reveals that the gender pay gap widened in 2017 rather than becoming smaller: In 2016, women made 81 cents to the dollar a man-made, but in 2017, women made 78 cents to the dollar, according to Visier data. Organizations still have a long way to go to close the gender pay gap, so why don’t you start by analyzing the situation in your organization?

(To explore the R code used in this article, check my GitHub).

Related Course

The People Analytics Journey

An overview of future role of HR leaders in improving business performance by informed decisions about people based on data. People Analytics transforming HR; The Role of People Analytics Leader; Case Studies and Simulations; Emerging trends of HR tech.

The Syllabus

The post Gender Pay Gap and People Analytics: A Practice with Open Data appeared first on Littal Shemer Haim.

Gender diversity in tech

It is a well-known fact that women are a minority in tech, especially among programmers. LinkedIn, for instance, studied the professional gender gap and explored the rate at which men and women have been hired across 12 industries worldwide. The research took a detailed look at leadership positions and software engineers. The findings include some depressing statistics for people who care about gender diversity in tech: In 2016, the rate of women’s new hiring was 18% in software engineering and 30% in leadership roles. And this trend might get even worse: A study from Accenture and “Girls Who Code” warns that, unless action is taken now, the percentage of women in the computing workforce will shrink over the next 10 years to 22%.

This forecast is not surprising since, according to researchers from the University of Wisconsin-Madison, almost 40% of women with engineering degrees either quit or never enter the tech industry. Indeed, women comprise only a small percentage of the biggest tech companies, based on diversity reports released by organizations such as Apple (20% of tech, 35% of non-tech, 28% of leadership jobs), Microsoft (29% of the workforce, 17% of tech, 23% of leadership roles), and Twitter (10% of tech, 21% of leadership positions).

The consequences of women being a minority

What is it like to be the only woman in a tech company team? How does a woman’s work in a man’s world influence her daily routine, colleague relations, work-life balance, job performance, manager reviews, promotion, and compensation?

Some recently published findings are truly disturbing. According to the “Elephant in the valley” survey, 60% of the women working in Silicon Valley have experienced unwanted sexual advances. About two-thirds of the women surveyed said that these advances were from their superior. Moreover, 90% of women interviewed had witnessed sexist behavior at company off-site events or industry conferences, and about 87% of them had heard demeaning comments from their male colleagues.

Although serious in and of itself, sexual harassment is not the whole story: 40% of the women interviewed felt that they ought to talk less about their families in order to be taken more seriously and about 52% of those that took maternity leave, cut it short so that it would not hurt their career. Slightly less than half (47%) of the women had been asked to do lower-level tasks that were not expected of their male colleagues, such as taking notes or ordering food. Additionally, two-thirds of women felt excluded from networking opportunities because they were women. In short, many women experience distressing workplace situations while most men are simply unaware of the issues facing women in the tech workplace.

This “bro culture”, this immature frat-boy behavior, is only one part of the sad story. At a much more fundamental level, the tech industry offers lower salaries to women in comparison with their male colleagues. According to data released by Joint Venture Silicon Valley, men in Silicon Valley earn up to 61% more than their female peers. Women are also offered fewer opportunities for advancement.

Companies actually lose

The women’s minority status in tech and the disturbing organizational culture for women in some companies are not merely social issues. It can impact negatively on company performance. Studies have shown that gender diverse teams are more successful: A research summary published by the National Center of Women and Information Technology (NCWIT) reveals that gender diversity at top management levels improves financial performance and that gender-diverse work teams demonstrate superior team dynamics and productivity. Likewise, companies that are at the top quartile of gender diversity are 15% more likely to outperform financially than those at the bottom quartile, according to McKinsey and Co. Researchers at Carnegie Mellon University have found that including women increases the collective IQ of teams and makes gender diverse teams smarter.

As Josh Bersin nicely summarized it “…companies that build a truly inclusive culture are those that will outperform their peers.” Thus, there is a clear economic incentive for technology companies to do something about gender diversity. But what concrete steps can they take?

Simple steps forward for HR

The first step is to identify where the company’s diversity gaps are. HR leaders can easily use analytics to look at the current employee population and examine headcount by gender. Once the baseline metrics are known, HR leaders can work with business leaders to determine gender diversity goals and allocate budget for these initiatives. Data-driven organizational processes, e.g., data-driven recruiting, enable continuous monitoring of metrics to see whether diversity increases or decreases as people move inbound, outbound, or within the organization.

But monitoring diversity metrics is not enough. If improving workforce diversity is a business objective, it is essential to keep track of performance metrics and financial metrics by gender groups. Here too workforce analytics can easily pinpoint gender differences, show rates of success within different groups, indicate bias, and keep an eye on promotions. The results may not only move the company forward in terms of gender inclusion, but it may also attract high potential employees – of both genders – in the long run.

Case study: Women in Taboola

One tech company that followed these steps and seriously studied the status of its women employees is Taboola. Taboola provides a web discovery platform, serving up 360B recommendations to over 1B unique visitors every month on some of the web’s most innovative publisher sites including USA Today, Business Insider, Chicago Tribune, and The Weather Channel.

The research, which aimed to explore the status of women among ‘Taboolers’, was conducted in 2016 by Neomi Farkash, global head of HR, and myself, Littal Shemer Haim, a people analytics consultant. Based on Taboola’s employee reviews and HR financial data, four comparisons were made between women and men:

1. Organizational Distribution:
What is the gender distribution within units, locations, roles, etc.?

2. Compensation and Promotion:
Across different roles, how are women compensated and promoted in comparison to men?

3. Performance Review:
How are women evaluated in comparison to men? Are they perceived differently in general and regarding performance, self-management, relationships, potential leadership, etc.?

4. Evaluations and Promotion:
What is the correlation between yearly reviews and promotions for the two genders? Do men and women who received similar reviews get a similar promotion?

Neomi Farkash affirms that ”the research was essential in order to start a discussion, which contributed, among many other activities, to address and reduce any gender gaps in the workplace.”

Creating awareness is an easy first step that every tech company can undertake. The data needed for such research exists and is accessible to HR. No complex analysis is necessary, and any HR analyst can handle it by simply comparing the four factors described in the Taboola case study. I hope that many HR leaders will take this initiative and thus make their small but important contribution to enhance gender diversity in the tech industry. The benefits will be reaped not only by our own daughters and sons when they join the workforce but by society as a whole.

The post Gender diversity in tech: Simple steps forward appeared first on Littal Shemer Haim.

We are all familiar with the story of David and Goliath, a shepherd who has defeated a mighty warrior, and the allegory of the underdog beating the giant. Is this story applicable to “People Analytics”?

In our fictional battle today, the giant would be R, the super-power open-source programming language. The underdog would be DMWay, the “new kid in town”: An Israeli start-up that develops an AI approach to predictive analytics, and claims to enable faster and better predictive models. Will their combat end as the old myth? Which of the two rivals enables us to build better predictive models? What can we learn from their contest about “People Analytics” practices? Let’s begin the fight in the arena of predicting employee attrition.
(To explore the R code used in this article, check my GitHub).

Table of Contents

• Employee attrition data
• The cost of employee attrition
• How to predict attrition?
• Logistic regression: R vs DMWay
• The next step: Deployment
• And the winner is…?
• Infographics

 

Employee attrition data

The reason why our first round in this fictional battle is chosen to be employee attrition lies in the sad reality of HR open data. In a previous post, I mentioned how great it would be to practice analysis and coding based on HR open data. Indeed, it was really encouraging for me to stumble into an employee attrition case, and yet, it was the only open data I found. Nonetheless, employee attrition is a severe problem for many organizations, as I specify below, so both competing technologies, R and DMWay, may offer valuable analysis.

The dataset for the following comparison between R and DMWay was downloaded from Kaggle, a platform for data science competitions, where data scientists can help organizations to solve problems by accurate algorithms based on real data. But not all datasets on Kaggle are real. Datasets published by users on the open data platform are different from those associated with competitions. Some users publish simulated data just for fun or for practice. Unfortunately, that is the case with this employee attrition data.

The dataset titled Human Resources Analytics includes some variables from the realm of HR: numeric variables, e.g., employee satisfaction, employee evaluation, average monthly hours, tenure, and amount of projects, and categorical variables, e.g., work accidents, promotion in last 5 years, department, and salary level. All of these variables may predict the outcome of employee attrition: voluntarily leaving the company. The case study addresses a specific question: “Why are our best and most experienced employees leaving prematurely?” Answering this question would enable this fiction organization to take some actions in order to eliminate or decrease the undesirable outcomes.

At first sight, or rather, by exploring the variables, it seems that the dataset was well simulated as if it was derived from a real HR database. The codebook available is not detailed enough for deeply understanding the meaning of all variables’ units. However, you may assume that some variables would have been re-coded or calculated (e.g., salary level), and others were excluded (e.g., demographics), for the sake of confidentiality. In addition, the data is well structured and clean, in a way you would not expect in case of real data extracted from the HR information system. Nevertheless, let’s continue, assuming that this tidy data is real and complete.

Data exploration reveals that employees who left the company are actually better, in comparison to those who stayed. As shown in Figure 1, although less satisfied and rewarded, employees who left are better evaluated, are involved in more projects, work more, have longer tenure, and are less involved in accidents. These results imply the enormous costs of employee attrition.

Figure 1: Employees who left the company in comparison to those who stayed

(Employees who left – “1”, Employee who stayed – “0”)

 

The cost of employee attrition

Employee attrition is a huge issue for organizations in every industry. An organization can’t completely avoid employee turnover and attrition, but the rate of employees walking out the door may determine the organization’s doom.

Employees who leave take a significant value with them: professional knowledge, specific practices and know-how, relations within the organization and outside (with clients, suppliers, business partners, etc.), and more. But the damage does not end with this. There are enormous costs, sometimes up to the sum of few salaries per employee, which are tied to recruitment, onboarding, training, and ramping up of a new employee. Furthermore, an organization must take into account some alternative costs, namely the value of transactions that could have been made by a senior employee who actually left.

Indeed, understanding the attrition cost is recommended as the first step in any predictive analytics project. Modeling cost is an effective way to determine ROI (return on investment) of a predictive analytics project that addresses the issue of employee attrition. No wonder why analysts, in companies like Walmart, take the effort to demonstrate to management that reducing even a 1% attrition rate sometimes saves millions of dollars. Although modeling attrition costs is inapplicable in our simulated data, it is important to keep in mind that in a real project it would be a good practice to start with that. Furthermore, a decrease in attrition rate in this imaginary organization (31%) may not only save the recruitment and onboarding cost but probably reduce the alternative costs, since, in this example, employees who leave are considered to be the better ones.

How to predict attrition

There are many modeling techniques that can be used to explain or predict attrition. However, in this article, I have chosen to cover the logistic regression. There are two reasons for my choice: First, the logistic regression is easier to interpret. It may not be the most accurate model, but it offers a pretty good solution without much effort. I believe that in the HR realm, the reasonable way to progress with predictive analytics is generally through good variable selection, that can be easily explained, and not by showing off with excessive models. Second, the objective of this article is not only to demonstrate the implementation of predictive analytics for employee attrition but rather to compare two technologies: R vs DMWay. Since the innovative solution of DMWay relies on regression models, it is more practical to compare the workflow and results of these two technologies by the same modeling.

When analyzing attrition, the goal is essentially to explain or predict a binary or logical variable (voluntary leaving the company) by various other available variables (in this simulation, all or some of the following: employee satisfaction, employee evaluation, average monthly hours, tenure, number of projects, work accidents, promotion in last 5 years, department, and salary level). There are plenty of resources on how Logistic Regression works, but in a nutshell, logistic regression is suited for examining the relationship between a categorical response variable and one or more categorical or continuous predictor variables. It creates an equation that in effect predicts the likelihood of a two-category outcome using the selected predictors. Each of the predictors is associated with a significance mark (p-value) that indicates if the predictor is useful or not.

The implementation of logistic regression, both in R and DMWay, follows the same recipe: data partition into training and testing sets, using logistic regression to model attrition as a function of other predictors in the training dataset, evaluate the model by predicting attrition in the testing dataset, and analyzing how good the model is in terms of prediction accuracy and predictors’ importance. I followed these exact steps in the two technologies but had a totally different user experience, and even different results.

Logistic regression: R vs DMWay

The R output of the logistic regression model is presented in figure 2. This is the fourth model I created, in an effort to generate the most accurate model, as specified below. In this model, all variables were included, except the department variable. Furthermore, the data were subset to include only highly evaluated, senior employees, who work full time. This is appropriate for the simulation objective: “Why are our best and most experienced employees leaving prematurely?” For the purpose of simplicity, and due to the lack of demographics, e.g., gender, the model does not include interactions of variables.

Figure 2: R output of the logistic regression model

A quick orientation in the model results: The variable names are listed on the far left under “Coefficients”. In the case of categorical variables, the first value (e.g., “0” in work accident) is considered as a baseline, and other values are included in a separate line, indicating their impact relative to the variable baseline. The significance values are provided on the far right under Pr>(|z|). In our case all variables’ estimates have significant value (equal to or lower than 0.05), i.e., they are unlikely to be obtained by pure chance.

Not surprisingly, the intention to leave the company has a negative relation with many variables. Specifically, being promoted, having a medium or high salary, seniority accumulation, and also work accident, decrease the intention to leave. On the other hand, excessive working hours, the number of projects, the last evaluation, and also satisfaction level, have a positive correlation with the intention to leave.

The model output seems straightforward. It implies what the fiction organization should do in order to keep its best employee. But is it accurate? To figure this out, it’s time to use the model to predict the outcomes in the test dataset. The results enable us to build a “confusion matrix”, in which predicted results are compared to observed results. As shown in Figure 3, which is again R output, the prediction is not perfect. The model “was right” in about 87% of the case. The model “specificity”, i.e., its ability to correctly identify those who left is about 97%, whereas the model “sensitivity”, i.e., its ability to correctly identify those who stayed is about 59%.

Figure 3: R output for evaluation of logistic regression model

Another way to evaluate this logistic regression model, and to compare it later to the model made by DMWay, is the AUC, which stands for Area Under the ROC Curve. To make a very very long story short, the “ROC curve” helps to figure out the trade-off between true positive rate (e.g., employees predicted and actually stayed in the company) and false positive rate (e.g., employees predicted to stay but actually left the company), in different cutoff points of prediction, ranged from 0 to 1. In our model, the cutoff point was 0.5, meaning that predicted results equal or above it was considered as “left”. But we could choose other cutoff points and maybe gain better results. With different cutoff values, we move along a curve, where at each point we have different true positive and false-positive rates. Higher and steeper ROC curves are desired and are indicated by a higher area under it (AUC). The perfect theoretical model would have an AUC of 1, and a completely non-predictive model would have an AUC of 0.5. What about our model? It is actually pretty good, but not excellent. It has an AUC of 0.78.

Here we end our session in R and move forward to DMWay. How different the UX is in this software? Will we end up the process with a similar model and predictions? It is worth to mention that each step that was done so far in R was involved with coding because that’s what R is all about. In DMWay, however, the whole process is latent, and the user needs only to make some simple selections in only 4 menus. To illustrate how easy it is to start working with DMWay, you can simply watch this 7min video in which a model is generated and deployed. However, in order to understand and evaluate it, you must already be familiar with the process of predictive analytics, though our R session already gave you the general idea.

I used the same simulated data, clicked buttons to run a model, since not even a single line of code is needed in DMWay, and then… Wow! The results were stunning. Take a look at the model ROC curve in figure 4, with AUC as perfect as 0.96! But what is the model behind this impressive chart? Exploring the model, as shown in figure5 reveals interesting points: First, the excluded variable here is the promotion, while in the R session I excluded the department. Second, some other variables that do include in the model appear in a piecewise mode, i.e., each of their values range has a different coefficient, hence, different influence on the predictive outcome. That is the reason why there is no need to take a subset of excellent employees to extract a good model, as I did in the R session. I must admit, though, that at first sight, the variety of lines in the model output makes it a little harder to explain employee attrition in simple words. However, the reds of the negative correlations make this output much more friendly. Furthermore, in order to tell a story, namely to explain what variables are most contributing to employee attrition in this specific model, all we need is a quick glance at figure 6, which presents the variables in descending order of contribution.

Figure 4: DMWay output for ROC in a logistic regression model

 

Figure 5: DMWay output for a logistic regression model

Figure 6: DMWay output for significant variables

The model made by DMWay outperform the logistic regression in R. Perhaps more efforts in feature selection would have yield better results in R. But faster and better predictive models is the whole point of using DMWay. Ronen Meiri – Ph.D., Founder, and CTO of DMWay, explains that “DMWay automated solution is powered by a sophisticated analytic engine that mimics all the steps taken by experienced data scientists during the analytic process.” This may take weeks or months sometimes. “DMWay is led by leading data science experts”, says Meiri, “and encompasses many years of researching automation and simulating the work of a data scientist. While building predictive analytics models in the world of big data is time-consuming, costly, and risky, DMWay offers everyone the ability to build better predictive models in a matter of hours.”

The next step: Deployment

According to the data mining process known as CRISP-DM, the next step after modeling and evaluation is deployment, namely putting the model in real use. While deploying a model generated in R involves excessive work, mainly to translate it to other programming languages used in the organization, DMWay offers an innovative solution: At the end of the model generation process, it can generate the code for deployment, in three programming languages: R, SQL, and Java. This code is ready for export to the organization’s database.

Does it mean that our fiction organization should now hurry to predict which of its best employees is the next to leave? Although it is technically possible, it is not ethically recommended. The only reason that I mention deployment in this context, is for mentioning this additional strength of DMWay. There are many other business outcomes related to employees that it would be wise to predict and deploy, however, in my opinion, pointing the next employee who is at “flight risk” in a certain moment, is not one of them.

The ethical issue of predicting employee attrition has been long discussed, e.g., in the context of civil rights. I would use this specific model to understand employee attrition, in order to reduce it, or to test the impact of some organizational interventions, though. I think that when pointing to an employee that is not 100% intent to leave, there is a chance for different results, not all in favor of that employee. Furthermore, we should all remember that models will be always models, they can’t encompass the whole reality. The British statistician George E. P. Box said it most appropriately: “Essentially, all models are wrong, but some are useful”.

And the winner is…?

To wrap up our test, let’s come back to the story of David and Goliath. As you probably recall, David goes into battle with only a sling – a simple and common tool among shepherds. He walks right up to Goliath and kills him with a single shot to the head. DMWay’s sling is not simple nor common. But for contemporary business analysts, data scientists, domain experts, and also executives, it turns out that it is effective to defeat the giant R, in terms of time-consuming, model accuracy, and deployment. The whole process of this battle is summarized in the following infographics.

Yet, to be honest, our fiction battle ends differently. In contrast to the biblical story, our giant is not defeated but rather lives happily ever after, since he is “open”, i.e., letting others to use his power. In fact, like much contemporary software, DMWay relies on R in the backstage.

So which tool would you pick for predicting employee attrition?
I’ll appreciate sharing your thoughts in a comment.

 

Related Course

The People Analytics Journey

An overview of future role of HR leaders in improving business performance by informed decisions about people based on data. People Analytics transforming HR; The Role of People Analytics Leader; Case Studies and Simulations; Emerging trends of HR tech.

The Syllabus

The post Predicting Employee Attrition: R vs DMWay appeared first on Littal Shemer Haim.