Making a lasting impact in HR Analytics: Proven comes before simple, but one cannot be without the other!


By Søren Kold,
Head of Analytics, Ennova
Global Employee and Leadership Index

HR analytics has been on the rise for some years now. The world has become more and more digitalized, and more data are gathered on people in organizations. The value of improving Employee Engagement has gained an increased focus in companies across the world, and the key reason being that there’s generally a wide acceptance that engaged employees performs better, are less absent, are less willing to quit their jobs, have fewer safety incidents, and produce happy and loyal customers etc. Sum that up and it’s easy to understand the business case: Engaged employees are essential for the bottom line! Add in an increased understanding within HR, that using analytics as a tool can help you make better and informed decisions. The analytical approach has helped HR succeed in moving closer to becoming a part of the business instead of being a support to the business. So, the days where you make decisions on a gut feeling should be over, right? Well, in some organizations they are, but in others, they aren’t and we still have some road to travel.

While I believe that we eventually are getting there, we still experience different maturity levels on doing HR Analytics in organizations. Therefore, as in any good fairytale, it sometimes seems like we should overcome a few obstacles, temptations you might say before we eventually get to where we want to be.

The need for better, faster and more accurate analytics must not be confused with surveys using stated importance, correlations mistakenly seen as causation and flashy dashboards promising to change the company based on a percentage favorable metric. The output should, by all means, be presented crisply with a great look and feel, easy to understand and engaging for whoever is the recipient, but the subject at matter must be treated with more substance than flash. You shouldn’t stop learning or try new things once you have understood correlations, but continue to build on each and every important aspect of succeeding with your analytics journey.

In this article, I will provide you with examples of how it’s possible to go from complex analytics to a simple interpretation making recommendations based on a proven foundation. Using analytics that can digest a complex world may give you the right results, but if you’re not able to transfer that to an understandable recommendation your analysis is not worth much.

Understanding people and the organizational context

In my job, an important task is to analyze the answers from engagement surveys and provide clients with accurate results and recommendations for improving Employee Engagement. This means that in most organizations it really doesn’t make much sense to make a single model covering everybody in the organization. Why? Because it’s important to accept that not all employees are equal when you analyze your organization. Whenever we discuss organizational behavior, a key aspect to keep in mind are the different psychological constructs that separate us from each other. Each individual has a set of psychological traits, which to some degree define how we fit a job or an organization, but also to some extent could be seen as something that affects our individual drivers of engagement (if you haven’t got the faintest idea of what I’m talking about, try to Google “Big Five”. This is an easy way to start getting into thinking about psychological differences).

Psychology is one thing but the situation people find themselves in could also influence drivers of their engagement. Try to think of a few people you know. They might work in different jobs, in different sectors, have different ages and different experience. Some might have kids while others may not. Could it be, that some of these factors affect the way they think about their jobs, and maybe their different situations make them have different priorities in their working life? One may be driven by the possibility to develop professionally and personally and another is motivated by his salary and bonuses etc. In any case, it’s our commitment to our clients, to tell them how preference structures look for employees in every department. In reality, we are most likely engaged at work by several factors, but they might not be equally important. This is why a good model should take different perspectives into account.

The model we’re using at Ennova has been developed over many years of surveys using the academia on motivational theories and in close collaboration with our clients. More than 3 million employees have been surveyed using this model.

But even though people are different on some parameters, there’s also similarities between us. Problem is, that this information is not always visible to the organization, but may be found via the patterns we discover in the answers provided, which of course is the tricky part, but sometimes there’s not always a simple solution to a complex problem.

Identifying key drivers of Employee Engagement

When we at Ennova analyze answers from Employee Engagement surveys, we do so using advanced statistical methods. Since the days of Herman Wold (Norwegian econometrician and statistician), PLS has been a widely accepted method used for analyzing intangible information such as the human capital perspective. In addition, we’re using a differentiated approach when analyzing the drivers of Engagement. With Structural Latent Class Analysis, we divide people into relevant homogenous subgroups with respect to their preferences. The analysis uses an iterative approach relying on calculated probabilities for every respondent which makes it possible to divide people into subgroups. We let people be regrouped by repeating the process until the analysis converges and respondents are placed in their final segment. For each segment of respondents, we calculate a structural equation model using PLS, which determines the effect structures for each segment. The output is latent scores for each Engagement driver, effect sizes and weights for each respondent, and the method used thus reflects the best possible guess on the true nature of each group of employees. This method gives us improved statistical fit, more precise estimates of personal preferences and an unbiased estimate of effect sizes for the independent variables in our model.

Summed up this enables us to give precise recommendations for each department within the organization based exactly on its own member’s preferences in shaping their Engagement. So rather than having one overall model which fit the complete organization poorly, we use statistical techniques to identify subgroups which in turn reflects the differences of a multi-faceted organization and its members.

Translate your analyses and make recommendations accordingly

Since the true value of analytics lies in the interpretation and ability to act upon knowledge, we put together the things each manager must focus on maximizing the potential of his team of employees. Remember that a team may consist of different employee types, but with the differentiated approach, we take into account exactly which type of employee that is, and adjust the recommendations accordingly. If we built only one model, we would, by great chance, over or underestimate recommendations for many employees.

One of the purposes of building statistical models on your HR data is to become not only predictive but prescriptive. You must eventually be able to use the knowledge you have provided to look into the future, or at least have a qualified guess. Therefore, in addition to scores and calculated effects on Engagement, we are now ready to leverage many years of data. Having many thousands of departments tracked over time enables us to be specific about how much each Engagement driver in the survey can develop. This is an important aspect in understanding the dynamics of where you should use your resources in order to improve Engagement, and adding this information to the picture enables us to adjust which focus areas each department should work with. A recommendation could be something along these lines: “The focus areas your department will benefit the most from working with are…”

  • Focus area 1
  • Focus area 2
  • Focus area 3

Underlying these recommendations would be information on scores, effects on Engagement and the potential for development. The effect sizes are of course important since they tell us something about how much Engagement changes if we’re able to develop that specific driver. Scores are important because they tell us the current state of an Engagement driver, and more importantly because they set the guideline for how much we can expect that driver to develop. It’s much easier to develop a low scoring driver than a high scoring driver, and at some point, it becomes quite difficult to maintain a high score – even for great managers. But the real perspective is met when we combine these variables: Consider choosing between 2 variables with almost equal effect on Engagement. Which one should you choose? The answer would be the one you can improve the most for the same cost, but and thus receiving the highest ROI based on increased Employee Engagement.

The use of Engagement metrics as early warning: Predicting Voluntary Employee Turnover

Griffith & Hom (2001) identified that the cost of employee turnover can be substantial. They projected the costs at 93-200 percent of each single leaver’s salary depending on his or her skill, the level of responsibility, and how difficult it is to replace them.

Identifying areas that are potentially very costly to the organization is important and establishing early warning metrics can be crucial. If you, in addition, knows what drives this metric, you have put yourself in a good position to come out on top in a highly competitive market.

We asked respondents in our annual survey the Global Employee and Leadership Index (in short GELx. The survey covers 42 countries worldwide) “Are you thinking of getting another job outside the company?” In a group of 5 possible answer choices, respondents could among others choose: “Yes, the sooner the better” or “No, I have no plans of getting a new job”. We used these two poles as proxies for employees leaving and for employees staying at their respective companies, and built a logistic regression model with that question as our dependent variable. The independent variables in our model reflected our Engagement model, consisting of employees’ assessment of the company Image, the Senior Management, their Immediate Manager, Co-operation within the company, their Job Content, their Working Conditions and evaluation of their Remuneration (salary plus benefits). In addition, we added in some important background information on respondents as well: Size of the company they work for, gender, private or public held company, age, and seniority. A total of 15.113 (15,113) respondents were part of the analysis.

The result is a calculated probability for each respondent, and figure 1 is showing the variation within the model at the individual level. When aggregating these probabilities within intervals of Engagement the results provided a clear picture (figure 2): There’s a strong relationship between the level of Engagement and the probability of voluntary employee turnover. The final proof of concept was to let the model be tested on other data, using only the intervals on Engagement as a predictor and calculating the turnover rate. Comparing these two variables the model showed a great ability to forecast what happened (figure 3): A low score in Engagement Survey 2015 equals a high turnover rate in 2016. With this tool, we can forecast coming employee turnover, and before turnover actually happens, we can identify initiatives to improve the key drivers and thereby reducing turnover and saving the company substantial costs.

Figure 1

Figure 2

Figure 3

In this article, I have tried to show, that there’s not necessarily an easy way out when it comes to understanding your business and the people who work there. You need to find the right way to present recommendations to whoever you’re reporting to, be it your manager, CHRO, CEO or a client. Be ready to prove your claims and build your arguments on data and models. They will tell you how things have been, and in some cases help you predict what’s about to happen in the future. But, we all know that the CHRO or CEO have limited time, so be simple and concise whenever you present your recommendations, and always relate your finding to the business!

Data has become the backbone of modern businesses and analytics offers great tools to make you understand, explore and add value. In a recent LinkedIn blog post, Thomas Rasmussen (Vice President, HR Data & Analytics) from Shell is advocating listening to rock music. Building on research from Stanford University he argues, that there might be benefits from listening to music while doing analyses. There is certainly a truth to that, but in addition to listening, we might also benefit from understanding, that the best of rock musicians are desperate men, and like them, you have to have something bothering you all the time! Working with analytics can be a long haul, so you need to have that inner desire and curiosity that keeps you going, but when you succeed it’s all worth it!

” I see my daddy walking through them factory gates in the rain. Factory takes his hearing, factory gives him life. The working, the working, just the working life.”

Bruce Springsteen (Factory), 1978

1 thought on “Making a lasting impact in HR Analytics: Proven comes before simple, but one cannot be without the other!”

Leave a Reply