This interesting blog post by Mike West from One Model describes a data anomaly in “Best Employer” awards. Many of these awards are based on employee engagement surveys, which are consistent and scientific, but susceptible to a subtle sampling bias.
The issue is that engagement is highest for new employees. I have seen this phenomenon in other surveys, and I have pondered why this would be true. It will make sense when you consider your personal experience. When you are first employed, you have recently chosen to work for that employer, you have just been chosen by the manager, and you get the greatest concentration of training and personal attention.
By contrast, years later you might wish you could work elsewhere, even if you have not made an effort to move. You may have changed managers, breaking the personal sense of loyalty and trust. Even under a favorable scenario you will be deemed “fully-performing” …and be neglected as a result. Negative career events occur over the years, and with greater length of service you will have more opportunity for annoyances, defeats, and betrayals. You might leave, and lo and behold the cycle starts all over again!
Mike West notes that growing companies hire more staff into brand new positions. This means a larger fraction of their workforce have less than one year of job tenure, which means a larger fraction of the survey sample will have high engagement. Yes, it is nice to work for a growing company, but growth itself is not what makes people happy.
If you were the only new hire in a company that is stable in size and has low turnover, you might be just as excited as a peer who joined a growing company. But the growing company would get a better score. The article references the constantly-growing Google, often rated the very best employer. Google tends to lose the top spot when they hire fewer people.
So, how do you game the awards? Make email addresses for new staff more readily available. How to correct this anomaly? Companies conducting surveys should report the data on a stratified basis, adjusting for length of service. Or, run a multivariate model which isolates employee culture and adjusts for the length-of-service effect.
But hey, it’s math. It’s all fun and games until someone loses an award.