The Lindy Effect in People Analytics
The relationship between tenure and turnover decisions, and why people are like frogs
Let’s play a game. Look at the description of the three employees below. Who do you think is most likely to resign in the next 6 months?
You might have guessed the right answer if you’re a people analytics pro, but I’ve found that for managers, the direction and strength of the relationships of some of these variables and turnover risk are counterintuitive. For example, often when discussing an employee’s flight risk, leaders have said something to the effect of “they’re not going to leave, they just started here!”
Tenure and the Lindy Effect
If you’ve done any sort of predictive turnover modeling, you know that being new to the organization is one of (if not THE) largest predictor of turnover. Some of this can be explained simply through the notion that “past performance predicts future performance”, i.e., workers who have stayed with an organization in the past are naturally likely to continue to stay. This is not wrong, but the reason behind this pattern is more specific and nuanced, it’s called the Lindy Effect.
The Lindy Effect, which is a statistical tendency for things that have survived for longer in the past to continue longer into the future, can explain everything from the lifespan of Broadway plays to the reign of world leaders. The geometrician Benoit Mandelbrot named the effect after an old diner in New York where comedians theorized that the life expectancy of a comedian was proportional to the total amount of their exposure to the medium. Despite its history, the concept has only recently exploded in popularity, being discussed in Nicholas Nassim Taleb’s book Antifragile, and even spurring a new strain of wooish self-help.
It should be noted that the Lindy Effect does not apply to human lifespans, because risk of death increases with age. Traditionally, this distinction has been described in terms of things that are “perishable” (living things) versus “nonperishable”, like ideas and technology. However, a recent very interesting paper from Toby Ord, a professor of philosophy at the university of Oxford, challenges this notion and seeks to broaden the scope of the overall concept of the Lindy Effect.
Ord points out that in contrast to humans, the lifespans of some species do exhibit the Lindy Effect. Frogs for example, have extremely high hazard ratios towards the beginning of their lifespan, but older frogs who have made it to adulthood are more likely to survive at any given time. In organizations, worker data typically show a similar pattern, with the vast majority of turnover happening at earlier stages of tenure, and the hazard declining thereafter. Because the relationship between worker tenure and decisions to leave exhibits the Lindy Effect, in the absence of other information, if someone has been at their current organization for 5 years, your best guess for how long they will end up staying is 10 years!
But why exactly do workers follow this pattern? This is entirely speculative, but here are some possibilities:
Escalation of Commitment / Sunk Costs: It’s psychologically intuitive that people will continue to invest in something that they have already spent a lot of time / effort on. Escalating commitment bias (also known as “sunk cost fallacy”) has been well documented and could explain why tenured workers “sticky” are despite having potential alternatives.
Firm-Specific vs. General Human Capital: When labor economists look at how workers accrue skills, they differentiate between firm-specific skills and general skills. The greater extent that employees develop firm-specific skills, they may pay a penalty for leaving, because outside organizations may not value their skills as highly. Note this idea is distinct from “sunk costs” – this is a tangible incentive, not a psychological bias.
Selection Effects: As opposed to the “costs vs. benefits” perspective, viewing turnover decisions as being part of a larger sorting process can be illuminating. Despite heterogeneity in recruiting processes, people and organizations don’t know a lot about each other before they form a relationship. Tenured workers may be tenured because they happened to be a good “fit” for the organization for reasons that were opaque from the beginning. The “survivors” may have certain traits, which has implications for (or be a reflection of) organizational culture.
For my money, the Lindy Effect of tenure and turnover is probably primarily caused by selection effects or “winnowing”, as Ord calls it. From his paper:
“…it can also be a kind of winnowing process, where entities with defects fail quickly, leaving behind a greater proportion of those which were more robust all along.
…A book that has been in print longer has survived a market test, demonstrating its enduring interest. A building that has stood longer has passed the test of time, demonstrating its superior materials (or its enduring desirability to those who might maintain it).”
As Bayesian thinkers might argue, we know the least about people when we first meet them, and update our priors with more exposure. There’s good reason to think this extends to talent selection and retention.
Practical Takeaways
For People Analytics: Recognize and model tenure as a variable that has a power-law (as opposed to linear) relationship with turnover. There are a variety of approaches to handle this, one could be using log transformation, but evidence suggests that decision tree models, which will account for this, are among the best for predicting turnover.
For Talent Management: Include tenure as the first variable to educate manager ratings of flight risk. For critical or strategic new hires, do not make the mistake of thinking that low tenure means low risk. Paying special attention to the onboarding experience / new hire integration will pay dividends.
For Workforce Strategy: Think in terms of “employee lifetime value”. Because tenured employees are increasingly “sticky”, and have a lot of firm-specific knowledge, employee tenure itself is an asset. The challenge for CHROs and Strategic Workforce Planning leaders is how to weigh this asset against gaps in skills as they look toward the future.
Answer to the turnover risk game:
Worker 1’s flight risk: 37.9%
Worker 2’s flight risk: 19.3%
Worker 3’s flight risk: 17.8%
Data sourced from the IBM Employee Attrition dataset. Predictions created using logistic regression in R.