BREAKING DOWN THE BAYESIAN METHOD
A note on the report’s somewhat unusual methodology. WSO compiles data and ranks firms by percentile using Bayesian probability, so each chart — with a few exceptions, such as actual compensation figures and average weekly hours worked — shows the same series of 30 percentages in descending order. For example, BCG ranked third in Overall Employee Satisfaction at 95.8%, while Bain & Co. ranked third in Career Advancement Opportunities at an identical 95.8%. How is this possible? As Curtis explains, percentiles simply show where a company ranks out of a certain number of companies — in this case, 71. When WSO asks an employee a certain question, consulting firms get different scores based on the responses, Curtis says, but that doesn’t change the total number of firms involved in the dataset.
The reason the percentiles are exactly the same in the example above is because for that specific metric, they were both ranked No. 3 out of 71 total firms — in this case the percentile formula is (71 – 3)/71. Another way to look at it: The average score of metric 1, Overall Employee Satisfaction, is 3.6 stars. If a company gets one 5-star review on that metric, “we don’t want to rank that ahead of a company with 10 reviews and an average ranking of 4.9 stars,” Curtis says.
Why do it this way? “The Bayesian stuff,” he says, “is just a way to deal with companies where there are only a few observations. So, what Bayesian stats allows you to do is pull the companies with less reviews closer to the average of the entire group, so that as more votes come in they are pulled more toward their actual average and away from group average. (This gives you) higher confidence that the score is a true reflection versus just one or two votes (and) prevents companies with only a few great or bad reviews from ranking above or below companies with many great or bad reviews.”