All Collections
People Analytics
Retention Insights Overview
Significant Turnover in Retention Insights
Significant Turnover in Retention Insights

Guidance on how to identify groups with significant turnover and address critical turnover risks proactively in Retention Insights.

Jessie Walsh avatar
Written by Jessie Walsh
Updated over a week ago

What can I learn from this page?

How to identify groups with significant turnover and address critical turnover risks proactively.

Who is this guide for?

Account Admins

Significant turnover

In this view under Retention Insights, select the Show significant turnover slider to toggle highlighting of demographic groups that have statistically significant turnover.

When this feature is toggled-on a new column, “Effect size”, appears in the table and becomes the default sort (descending). Demographic groups with the highest effect size will appear at the top of the results. The remaining groups will be sorted by Exited (descending) by default.

What is significance and how do you test for it?

Statistical significance examines whether a finding or result is likely to occur due to chance (e.g if Department A is experiencing a turnover of 15% and Department B is experiencing a turnover of 10%, is this difference of 5% meaningful?)

Whilst significance will observe whether findings are due to chance, effect size shows the magnitude of differences found between groups.

Significance is tested using the following inputs of each group:

  • Headcount (12 months prior to the last complete calendar month)

  • Exits (over 12 months only including people who were there 12 months ago).

These are compared to the same inputs for the company overall, across the same timeframe.

For any given group, the following situations mean we do not test it for significance:

  • If the group does not include more than 25 people (Headcount as above)

  • If the current headcount (latest complete month) is 0

For those groups where significance could not be calculated, an icon will appear in the Effect size column, with a hover-over descriptor. This indicates the significance test was not performed due to group size constraints.

What does effect size mean in relation to significance?

Effect size is a popular measure amongst statisticians when looking to understand if there is a meaningful difference between two groups.

Statistical significance is sensitive to the number of respondents (sample size) and increasing sample sizes will result in more and more correlations being deemed significant even where the effect size is minimal.

For this reason, effect size ranking has been chosen as the primary way to highlight which groups are experiencing higher than expected turnover within your organisation.

Effect size shows the magnitude of differences found between groups and will return one of five results:

  1. Very high (if effect size is greater than 2)

  2. High (if effect size is greater than 1 but less than 2)

  3. Medium (if effect size is less than 1)

  4. -

How we calculate significant turnover

The statistical test used to see if the turnover in a particular demographic is significantly different from expected is a hypergeometric test, or Fisher's exact test.

The inputs into this test are the following:

  1. The number of employees in the reference group. The default reference group will be the Company overall. Applying a filter to the page will update the reference group (e.g filtering by a specific location).

  2. The number of employees who left the reference group in the 12 month period.

  3. The number of employees in the demographic group we are testing for significance.

  4. The number of employees who left in the demographic group we are testing for significance.

With these 4 pieces of information the statistical analysis determines if the number of people who left in the demographic group we are assessing is greater than what would be expected given the number of people who left in the reference group. Significant turnover assessed with a two-tail test at a 95% level of significance. A two-tailed test means we assess if the turnover within the group is significantly above or below the expected rate compared to the reference group, however, we only report when it is significantly above the reference group rate.

Significance however is not a very good indicator of how meaningful an insight is, as it gets easier to achieve significance (even for very small differences) as the sample size increases. To ensure we report the most meaningful and significant groups we perform an additional effect size computation on the demographic groups found to be significant. The effect size calculation used is Glass’s delta. Glass’s delta counts the number of standard deviations away the observed number of employees who left in the demographic group is relative to the reference group. This computation is very similar to the popular Cohen’s D effect size computation, but is relative to the reference group.


Why isn’t a particular demographic highlighted even though turnover seems higher than the company overall?

Significance is determined based on a set of rules as described in our Significant Turnover definitions table.

Why do I see lots of exits in a group, but low, or no turnover?

Because the people who exited have joined this group within the Last 12 months, and were not included in the headcount used for turnover, retention, or significance testing.

While it is important to have visibility over large exit events, the real power of this feature is to help highlight significant turnover over time, which is often harder to detect.

Why do I see only obvious groups highlighted as significant?

Groups with obviously high turnover will show up, certainly at a company Level, but these groups can still tell a powerful story.

Select the table rows to understand what drives high turnover for the people who remain in these groups.

For a deeper analysis, add the groups to filters. This will set a new reference population and highlight significant turnover across demographics relative to the newly filtered group.

What if no groups are found to have significant turnover?

That's good! You can confidently say that the rate people are leaving your company, at least across your demographics, is within the expected statistical probability.

As above though, try adding a group to filters and see if any significant turnover can be found when setting a new reference population.

See also:

Did this answer your question?