Last week, Janelle Jones and I released a new study showing that opening an Amazon fulfillment center does not actually boost overall employment in the county where it opens. Analyzing data for counties in 19 states containing Amazon fulfillment centers, we found that within two years, the opening of such a facility leads to a 30 percent increase in warehouse and storage employment in the surrounding county. However, this does not lead to an increase in overall employment in the county.

In response to our study, Amazon circulated a comment arguing that “Amazon’s investments led to the creation of 200,000 additional non-Amazon jobs” and “counties that have received Amazon investment have seen the unemployment rate drop by 4.8 percentage points on average.”

Both of these claims are misleading. First, although it’s true that unemployment fell during the economic recovery in places that Amazon opened fulfillment centers, unemployment also fell substantially in places without an Amazon presence at all. Nationally, over 2010–2016, unemployment fell by 4.7 percentage points—essentially the same as the 4.8 percentage point fall Amazon touts—even though most areas of the country did not have an Amazon warehouse. Amazon is simply riding the tide of the national economic recovery. Unlike the analysis that Amazon presented, our study actually accounts for employment changes that are happening regardless of Amazon during the recession and recovery, by controlling for national, regional, and state-specific employment shocks.

Amazon’s claim of adding 200,000 non-Amazon jobs to the economy is also problematic because it appears to be a prediction rather an actual examination of what happened in the real world. Although it’s not clear how Amazon calculated an additional gain of 200,000 jobs, it seems to be based on what is called an “input-output” model. These models, by assumption, necessarily predict net job gains if Amazon or any company spends money in a county. The point of our study was to test that claim and see if it’s actually happening in the data, not to simply assume it to be the case by resorting to a model. This is why our study used historical employment data to assess directly whether counties that opened fulfillment centers had different employment outcomes than other counties had, around the time the fulfillment centers opened.

The bottom line: EPI analysis shows that Amazon fulfillment centers don’t raise employment in the counties where they open. Amazon’s claims to the contrary are misleading and don’t look at real-world data.