Analyzing Employee Turnover - Descriptive Methods

The Leaky Bucket of Turnover

In marketing there’s an analogy that makes a comparison between adding customers to the business and adding water to a leaky bucket.  No matter how much water you pour into the bucket, the bucket never seems to fill up.  This might be due to leaks (customers leaving for other businesses) or evaporation (customers who leave the market). 

Replacing "customer" with "employee" and we have a decent analogy for how recruiting and turnover interact.  Recruiters pour water (employees) into the bucket and the leaks (voluntary turnover) and evaporation (involuntary turnover) are where the water escapes.  If we're seeing a 10% voluntary turnover rate that means that we've got to find that equal number of people on top of whatever we might already need to grow the business.  That employee churn is a struggle for businesses.

Now what if we understood where those holes were in the bucket and could patch them?  Instead of churning through staff and recruiting resources we could get closer to fully employed.  Without an understanding of where (or why?) the turnover is occurring though, this problem is impossible to fix.

Methods of analyzing turnover

In the sections below, I’d like to lay out some basic methods for identifying and analyzing turnover within an organization.  At a high level I’m going to cover descriptive methods of measuring turnover and some intermediate methods that account for changes over time.

Descriptive Methods

Calculating the turnover rate and retention rate is well within the ability of any HR professional.  These are great measures for getting started down the path of analyzing turnover.  There are more advanced techniques to describe turnover and identify factors that predict turnover, but these are good initial measures.

Turnover Rate

This is the HR turnover measure that you'll hear about most often.  The annual turnover rate is defined as "number of employees who have left the business in the past year divided by total headcount". 'Headcount' in this measure is calculated by averaging the total staff at the start of the period and total staff at the end of the period. You may also see this measure calculated for other periods than annual such as monthly or weekly.  The interval for which the rate is calculated should be identified in the measure name.

I hope I haven't lost you yet, but in case I have here's Wikihow to the rescue: "How to Calculate Attrition Rate".  Photo to the left is also from the Wikihow.

An example of where this might be put to a good use is to compare your turnover rate to an industry average. There are lots of benchmarks out there and I've pulled this one (below) from CompensationForce.com. However, cross-industry comparisons could be problematic if you have a turnover/retention strategy that is a differentiator for your business within your industry. 

Once you have identified the turnover rate for the company, drilling down to understand your turnover rate by different demographics or parts of a business is another way to gain insight into how the business operates.  Being able to compare two similar units by their turnover rate can help you focus where (and whether) to intervene to assist a manager with managing (reducing) their voluntary turnover. You may also analyze turnover rate with demographic breakdowns to ensure that there is not a disparate impact to particular groups.

I cover this method because it is a cornerstone metric in turnover research and a starting point for more sophisticated analysis, but it’s a measure that is often only a good start.  It gets the job done of answering retrospectively "how many people have left" at the bare minimum level of satisfaction, but that's usually all it will do.  If you're in a meeting with senior executives and you need to answer questions regarding the “who, what, why, when and how” of turnover, only knowing this number will leave you feeling in the dark.

Calculating the turnover rate doesn't tell you an individual’s likelihood to leave, give hints into who among the population is leaving, tell you at what point they left during their career, how to fix it, or if the turnover is even a problem at all.  For instance, if the underlying driver is a poor onboarding experience which causes employees to leave within their first month, this measure will not be able to help you determine that by itself.  Part of the reason I'm fired up about turnover analysis is that I know we can do so much more than this measure.

Retention Rate

Retention rate is similar to the turnover rate, but has a nuanced benefit. Retention rate captures the loss of current employees over a given period and in addition strips out the noise of employees who join and quit over that same period. Depending on what you’re hoping to assess in the study, this can greatly increase the validity of your study. From the SHRM post "Calculating Retention Rate" -

# of individual employees who remained employed for entire measurement period / # of employees at start of measurement period) x 100

This measure allows you to answer questions about current staff separate from new hires. Similar to turnover rate, you may see this broken out by demographic, function/family, or parts of the business. 

You might want to use the retention rate over the turnover rate if you’re studying how a new engagement program, merger, or management decision has affected the current staff.  It's also a cleaner metric to isolate year-over-year changes in turnover from growth of the company. High growth might mean lots of new hires churning which would affect your turnover rate metric, but would not affect your retention metric.

Intermediate Methods

I call these intermediate because they don't yet rely on advanced statistics and are still descriptive in nature.  However, moving averages and cohort analysis both introduce a time component into the analysis that can help make sense of turnover data.

Moving Averages

Moving averages get a little more exciting.  This is a basic smoothing technique that helps you analyze a data series over time.  This method is borrowed from finance where it is commonly used in the analysis of stock market data. 

In the chart below you'll see the raw stock market daily return (choppy & hard to interpret), 15 day moving average, and the 50 day moving average.  From the raw data, it's hard to see if there's been a significant change in price from all of the fluctuation, but if you compare that with the 50 day moving average it starts to become clear that the stock price is rising. The spreadsheet clip below (also from Investopedia) shows how to calculate a 10 day rolling average. 

Applying to turnover, this allows us to more clearly visualize how turnover changes over time.  Of note when, applying this measure, from what I've seen there is typically a combination of management practices, traditions, and psychology which seems to make turnover more likely on Mondays and Fridays and at the start or end of a month.  People don't like to quit on the Wednesday before payday.  Tracking a 14 day (pay period) moving average or a 30 day monthly average should smooth out this potential clumping.

Moving averages gives you information to answer the question "Did the turnover rate change?"  In the event of layoffs or change management plans this measure could be useful to monitor voluntary turnover.  When combined with statistical process control techniques, moving averages can also be an early method to test if a particular business unit or function is losing employees at a higher rate than others after a change in business process. 

Side note: there's a fascinating article by Charlie Trevor and Anthony Nyberg (two excellent turnover researchers) exploring the correlation between layoffs and spikes in voluntary turnover.  

Sample size can be a concern here. If you don't see enough turnover within your organization or subset of your organization, the sample may not be large enough to be effectively measured as a rate. While this is particularly true for moving averages, it’s also true of most measures in this review. The use of moving averages is a step up from the single point-in-time measures, but we still have a lot more we can learn about why turnover occurs. 

Cohort Analysis

Cohort analysis is a technique that is seen most often in marketing analytics to track customer churn.  They want to know when customers will stop using their service much in the same way HR wants to know when employees will quit.  By segmenting the employees into cohorts based on when they joined the firm, you can then track their retention rate by that period cohort.  By looking at the cohorts together you can start to see a shape to the customer life-cycle. 

In the image above (pulled from the Kissmetrics blog), you can replace "customers" with "employees" and get to the same insights as a marketer.  Looking at the Jan 13 cohort of 80 "employees" going horizontally you can see how many are retained from month to month.  Much like the retention rate, a benefit of this measure is that it strips away some of the noise from a mixed sample of different tenured employees.

Averaging the dataset down by column, we can start to make some (rough) assumptions about the rate at which any cohorts of employees are retained once they've entered the company.  If the above table were employee data and we brought on 100 people this month, we could estimate (roughly) that if we didn’t change our practices 80 % of those 100 would still be with us at the end of 9 months (bottom right cell in the table above). I emphasize roughly because this analysis is more heuristic than grounded in statistical analysis.  But it should give more insight than a guess or gut feeling.

The root of cohort analysis is that the life-cycle of the employee affects their likelihood of turnover.  Research has confirmed that newer employees (not necessarily younger) quit at a different rates than more tenured employees (Hom et al 2008).  That sounds like common sense, but we ignore that premise when we use aggregate basic measures like the turnover rate or retention rate and try to apply their insights as a predictive measure to the entire company.  Cohort analysis allows us to start getting insights into turnover likelihood based on where someone is in the employee life cycle.

While cohort analysis gets us closer to understanding some of the principles underlying turnover, there is a still a lot left unexplained.  First, if your historic records are not clean or frankly kept at all, it can be difficult to pull data set from far enough back to make this work well.  To get 9 samples of 1st month turnover it requires 9 months of turnover data.  With only 9 months, the number of samples that you’re averaging together drops steadily the further away you get from the 1st month until, in the image above, in determining how the cohorts will perform in month 9, you’re relying on only 1 sample of data which is not going to give you an accurate representation of employee behavior. 

While you cannot get individual level turnover predictions with this measure, you can start to see the benefits of this for workforce planning on an aggregate level.  A good use of cohort analysis might be to get a better estimate on how many new hires will make it to the end of their training period.  For additional information or examples of Cohort Analysis check out the aptly named CohortAnalysis.com.

Closing

I hope this has introduced you to some of the basic techniques for studying turnover.  For those new to the topic, I'd love to hear your thoughts if this was helpful. There’s a lot more I want to cover on turnover, but frankly these posts are getting long and I thought it best to break them up.  Be on the lookout next week for a follow-up on advanced methods.  I want to tackle some of the more sophisticated methods such as Logistic Regression and Survival Analysis to give you a taste of how we can go about predicting turnover.

- Richard Rosenow

• LinkedIn
• Twitter

Continue on to Part 2 - Analyzing Employee Turnover - Predictive Methods  

 Other articles I've published on Linkedin:

• Analyzing Employee Turnover - Predictive Methods
• HR Analytics Starter Kit Part 1 - Intro to HR analytics
• HR Analytics Starter Kit - Part 2 - Intro to R programming
• HR Analytics Starter Kit - Part 3 - Podcasts
• HR has Last Mover Advantage in HR Analytics
• In Defense of Middle Measures: The Use of Constructs in HR Analytics