The Great Census Bureau eCommerce Data Debate

On Jan. 18, MPD CEO Karen Webster wrote a column titled “Massive Online Retail Data Error Uncovered” that pulls together how the Census undercounts eCommerce sales. 

Taking from the research of her colleagues at MPD, economists Drs. David S. Evans and Richard Schmalensee, and data analytics specialist Scott Murray, Webster discussed how data that Forbes and other like-minded financial publications had relied on from the U.S. Census Bureau was simply wrong. 

And here’s why that matters most, according to Webster: “The Census Bureau is the source for reporting physical and retail sales in the U.S. It’s the gold standard. Companies file reports with the Census Bureau according to strict guidelines. Depending on what the Census is covering, companies send this data in monthly, quarterly, or yearly. It’s why everyone uses them as a source.”

But what Evans and Schmalensee discovered is that something about how that Census data was being reported as it relates to the impact of online sales to brick-and-mortar sales didn’t quite add up.

“All of their field research showed that physical retail sales were cratering, but then the Census data said not much was happening,” Webster wrote. “Then they made a big discovery. One of the biggest online retailers, Walmart.com, wasn’t in the Census figures on online sales for 2013. The Census had clearly missed more than $6 billion of online sales.”

And guess what? The Census Bureau had a serious problem with this report. But more on that later.

It took six months of analysis and some emails with the Census people, and lo and behold, the economists uncovered what Webster called a “data bombshell.”

Online sales as a percentage of all retail sales has been undercounted — by a lot. In fact, what was discovered, Webster wrote, is that “Online sales as a percentage of physical retail sales has been undercounted by about one-third each and every year for at least the last five.”

What that means is that in 2014, eCommerce as a percent of sales wasn’t the 6.4 percent that the Census reports, it was actually 8.2 percent. Ignoring sales by gas stations, which currently aren’t selling anything online, online retail sales will hit 10.3 percent of all retail sales.

“The Census Data for 2015 aren’t all in yet, but we’re pretty confident that they will report something in the neighborhood of about 7.3 percent. Correcting for the one-third mistake, Evans et al. think that right number is more like 9.3 percent,” Webster wrote.

And if you look into the Census data, it appears that physical retailers weren’t participating in the online push, which is entirely untrue. What the findings show is the the Census data failed to account for “a massive amount of online activity by the online divisions of physical retailers themselves.”

In reality, the amount of online sales by those physical retailers is pegged to be 23 times higher than what the Census indicates, and that’s according to the economists’ conservative estimates.

And perhaps it’s easy to see where the errors were made, since Webster does point out that the Census Bureau relied on data supplied by retailers. As Evans points out, the Census can’t actually say anyone is missing. But everything his team looked at, and the feedback from the Census, made it clear, for example, that Walmart.com wasn’t anywhere in the Census’ online data.

When the economists finished their digging and added in other missing physical retailers, they found that the Census had missed about $83 billion in online sales in the U.S. in 2013.

As you can imagine, the Census had a bit of a problem with MPD’s findings. So much that on Jan. 21, it released a statement specifically about “how it categorizes eCommerce sales in its retail sales data sets.”  They contacted Webster and asked her to correct her story.

Here’s the statement they provided:

“The U.S. Census Bureau’s retail sales program provides statistics on all forms of retail sales, including eCommerce sales through the Monthly Retail Trade Survey and the Annual Retail Trade Survey.”

“The Census Bureau categorizes eCommerce divisions of companies with physical storefronts as part of the electronic shopping and mail-order houses industry (NAICS 4541) as long as they do not primarily fulfill eCommerce orders from their stores. Companies provide separate information on their brick-and-mortar stores and their eCommerce divisions. For example, if an electronics store has online and brick-and-mortar sales, the online sales would be tabulated as part of the electronic shopping and mail-order houses industry and the brick-and-mortar sales would be tabulated in electronics and appliance stores (NAICS 443). This is similar to how companies would split reporting between two distinct brick-and-mortar divisions, such as a company that owns grocery stores and department stores.”

“The Census Bureau never discloses information about specific companies. If companies do not provide data for a given month or year, the Census Bureau uses statistical methods to generate imputed sales and eCommerce sales values.”

Well, Evans had a few choice words about that news release. “Yes, I see they say that, and it is a verbatim transcription of a FAQ they have on their site. Unfortunately, it doesn’t square with what we found in looking at their technical documentation in detail and corresponding with a senior Census official responsible for the retail trade data. Census’ statement isn’t consistent with what their technical documentation says they do, what their published tables claim, what data their survey forms request, or what a Census official confirmed.”

Evans went on to point out that since the Census doesn’t actually share their black box with anyone, it is always possible they did something different than their technical documentation and reports say they did. He said he and his co-authors are hoping the Census data analysts, rather than the PR folks, will contact them to help reconcile the conflicting statements from the Census Bureau. According to Evans, “Our pens are poised to revise and correct if need be — but based on technical details, and data, not terse press releases.”

Altogether, it’s estimated that the Census’ retail online undercount wound up being just shy $100B.

The study from three economists provides a detailed analysis. 2013 data found that the Census Bureau missed at least $61.5 billion of sales by physical retailers, like Walmart. Walmart sold around $6 billion through walmart.com that year in the U.S. The Census data also missed more than $21.8 billion of retail sales by manufacturers, like Apple, with huge online retail sales. All told, the Census retail figures didn’t pick up more than $83.3 billion of sales in 2013. The Census data is widely used to assess the impact of online retail on the sector and as a source for other online retail estimates.

And here’s why that all maters.

“When the Census reports that 94 percent of all retail sales still happen in brick-and-mortar locations, everyone feels better. When that number is really 89.7 percent now, the storyline turns sour. And for retailers sitting in a category that is a lot lower than the ‘average,’ it’s downright scary,” Webster concluded.