Kevin will be providing an analytical and statistical perspective on a range of business and news topics. This is the first of many installments:
Two weeks ago, the US Bureau of Labor Statistics (BLS) released its monthly job report for February 2019 and announced that 20,000 new jobs were created. News organizations like The New York Times and CNBC have commented on how this job growth is much weaker than expected, and how it’s the worst job growth in the past 17 months.
Would your view change if you found out that the US lost jobs? How about if growth were five times higher? There is a reasonable chance that both are possible.
We can start by considering where this job growth figure came from. The BLS didn’t knock on the doors of every single American business and citizen to determine the exact change in US jobs each month, they surveyed a representative sample of US businesses and made an estimate.
An estimate is just that and is almost certainly going to differ from the actual result in what’s called sampling error. The sampling error can also be estimated based on the sample size and the variance in responses received. This estimate of sampling error is called the standard error.
Standard errors are used to build a confidence interval around the estimate, which is a range that we believe will contain the actual result. In general, we can be 95% confident that the actual result will fall within 1.96 standard errors away from the estimate.
Back to the job growth figures, the BLS estimated that February 2019 job growth was 20,000 and that the standard error of the estimate was 65,000 jobs. We can now calculate the lower and upper bounds of a 95% confidence interval:
Lower bound: 20,000 – 65,000 * 1.96 = -107,400
Upper bound: 20,000 + 65,000 * 1.96 = 147,400
Therefore, we can be 95% confident that the actual job growth was between -107,400 and 147,400 jobs. This means that we can’t even be sure whether job numbers have even grown at all! To be more precise, there is a 38% chance that the US actually lost jobs in February 2019, and an 11% chance that growth was five times higher than the estimate.
In summary, I hope you take two things away:
- Estimates always have some degree of standard error, and that standard error could be wide enough to entirely change a conclusion.
- Next time you read about the US monthly changes in jobs, take it with about ~127,400 grains of salt.
Blog post by Kevin Silberberg
Kevin’s project experience includes quantitative retail forecasting and commercial due diligence. Internally, he supports SATOV’s analytics practice by leading machine learning training sessions and developing web scraping tools to enhance research techniques. He holds an HBA from the Ivey Business School at Western University.
Outside of work, he mentors undergraduate students in statistics and math. In his free time, he enjoys exploring the city, diving into new technology and developing models to optimize various aspects of his life.