I
t’s a common task for anyone who uses labor market data. Combine multiple occupations—or multiple regions—and look at the the median hourly wage for the group of occupations or regions.
This is fine on its own. But if you’re grouping occupations with far different earnings curves—think CEOs and janitors—or regions with major differences in an average worker’s wage, average or weighted-average earnings can be misleading.
This is why we’re excited to introduce a new and better method for calculating occupational percentile wages for groups of occupations and geographies. Nothing in our calculation of earnings for individual occupations has changed. However, we’ve improved aggregate percentile earnings when combining two or more occupations and regions.
Here’s how.
What’s Changed and Why Is It Better?
In the past, Emsi used weighted averages to combine percentile earnings for groups. This works well in typical applications, where users of our labor market data software select reasonably similar occupations across reasonably similar geographies. It does not work as well when aggregating occupations or geographies with highly disparate percentile earnings, like retail salespersons and general and operations managers.
The weighted average (purple) falls between retail salespersons and general and operations managers and doesn’t represent either well.
Emsi’s new methodology, which is coming soon to our data tools, plots all the percentile earnings points for all occupations in the group on a single line and uses an algorithm to fill in estimated responses for jobs not represented by a percentile. Along that new line, we plot the new percentile points (10th, 25th, etc.) for the group of all workers represented in all the occupations. This line represents the new wage curve and percentile earnings that you’ll see when analyzing percentile earnings for groups of occupations. The below chart shows the new curve for the two occupations.
The first half of the new curve (purple) closely follows the earnings of retail salespersons, the lower-paying occupation. As the curve moves out to the right, however, it is pulled up by the higher earnings of general and operations managers, who are much more prevalent in the upper half of the new curve.
Testing This Approach
We tested our old and new grouping methods against disclosed Occupational Employment Statistics (OES) percentile earnings from the BLS for groupings of occupations. The new methodology consistently came closer to OES estimates than the old methodology. The change also returns much better results when disparate occupations are combined.
An Important Note
The practice of combining occupations with far different earnings to try to generate earnings figures that apply to both has never been considered a good idea. This is because an average is not the best representative measure of two widely divergent numbers. The fact that the weighted average method does not accurately represent contrasting occupations is not a poor reflection on weighted averages; rather it illustrates why such disparate occupations (or geographies) should only be combined with caution.
Weighted averaging is a common methodology for combining percentile earnings, so it should be noted that Emsi’s methodology change is not a correction of an error but a modest improvement when comparing occupations with similar earnings and a major improvement when comparing occupations with vastly different earnings.
This approach provides a buffer for data users who might not realize the risk involved in misrepresenting disparate occupations by averaging them.
For more on industry data and workforce trends, subscribe to our newsletter. Read about Emsi data here. Contact Josh Wright via e-mail (jwright@economicmodeling.com) and Twitter (@ByJoshWright).