Are Van Rentals a Good Proxy for Migration?
Every Day I’m Myth-Busting
Nicole Kaeding at the Tax Foundation has a new blog post that caught my eye earlier this week with maps purporting to show migration. The post is based on estimates provided by United Van Lines on moving van rentals, and goes on to suggest that taxes substantively impact migration. For today’s post, I’ll discuss the relevance of moving vans for migration. I know that sounds weird, but the truth is that these moving van numbers come up a lot in many different places.
But before I go any further, I should mention that (1) I used to work at the Tax Foundation, and worked specifically on migration issues, (2) I know Nicole outside of the Interwebs, and (3) she cites my previous work, so, obviously, I’m a sucker for praise, and am not going to disagree with her core point. But the main reason I won’t disagree with her core point is because her core point is correct. I’ll just quote her:
Individuals move for a variety of factors. Climate, job opportunities, family, among others, impact the decision to relocate. Taxes can influence the decision too.
She goes on to offer some very well-fitting examples of different ways taxes may (and may not) impact migration. She’s also more careful than most places I’ve seen to bracket claims about climate, region, and related long-run trends. This is all good. So yeah, give her post a read if you’re interested.
But the post has one key issue that isn’t really her fault, but is a good opportunity to look at a key piece of the public commentary on migration. That fault is very simple: Nicole treats the United Van Lines study as if it’s a meaningful study. That’s frustrating, because the United Van Lines study isn’t a meaningful study.
Moving Van Migration
Just How Many Migrants Are You Measuring?
So let’s look at this moving van data. The Tax Foundation has helpfully put it into a nice, color-coded map that I actually like better than UVL’s original material for its simplicity. So here it is:
The green states get lots of inflows, the purplish states get lots of outflows. Those do seem to roughly match up with what we know about migration. So, that’s cool. Maybe this is a good proxy.
But hold on. If you download the original study you’ll see something weird. UVL only handled 123,047 shipments in 2015, versus total 2015 interstate migration of somewhere in the ballpark of 7.4 million interstate migrants tracked by the ACS. But of course, a single moving van doesn’t mean a single migrant. Let’s say the average moving van is accompanied by 4 migrants: that gets us to about 500,000 migrants. United claims to be the largest moving company; I found at least 2 other major home moving firms. Plus, moving companies need large economies of scale and scope to be viable, so this industry is fairly concentrated. Let’s say that United is 1/4 of all moving vans. That gets us to 2 million migrants.
So our universe of moving-van-migrants is only about 1/4 of all migrants. Even if we assume a moving van means 5 migrants, we only get to 2.5 million. Or we could assume that United Van Lines is in a really diffuse industry with tons of small players. But even if we assume each van has 5 migrants, you’d need United to have just 7–9% of the market to get full coverage.
Now, to be fair, a sample size that includes even 5% of the total population is a fantastic sample in principle. The ACS only samples 1% of the total population. So in some ways, this UVL data seems like it’s got much better coverage than the ACS.
But that large sample is only valuable if it’s a random sample. The ACS is the closest we get to a truly random sample, and so a 1% sample that’s truly random is awesome. But the van rental data is not a truly random sample of migrants, and thus its results are biased.
Selection Bias and Moving Vans
Measuring the Migration of a Specific Population
The reason it’s likely to be a biased sample is simple, and also explains why even a full census of the entire moving van industry wouldn’t be equivalent to a full census of migrants. Most migrants don’t use these shipping services. Migrants tend to be young, lower-income, renters. They’re less likely to be families. This is not the ideal clientele for a moving company.
And that raises bigger concerns. If UVL had a random sample of migrants, it’d be extremely valuable. But it’s not random. UVL has a sample that is systematically biased to exclude the largest group of migrants. Insofar as it measures migration, it is most likely to measure migration of the group it actually samples from, which is basically middle- and upper-class homeowners.
Here’s a fun list of groups UVL almost certainly undersamples:
- Renters — Many have few enough possessions they don’t need to hire a moving company. Renters are the majority of migrants, so undersampling them means you miss the majority of migration, and especially migration tied to high-renter-density areas like, I dunno, urban areas that dominate the big outflow states.
- Students — These folks aren’t even going to be getting a U-Haul, let alone a moving company. Student migration is extremely erratic and totally different from other groups, so excluding them is a big deal.
- Poor People — Likely to have fewer possessions, and also unlikely to spend extra money hiring movers. And we know for a fact poor people have different migration rates and trends.
- DIY-types — Think I’m joking? Try convincing my father-in-law to hire movers. That’s what he’s got sons-in-law for. And such DIY types may also have different lifestyle preferences correlated with different migration patterns.
- Truck-Owners — Again, such a specific dataset from a narrow source has weird gaps in it. Like people who may just be unlikely to hire vans for unusual reasons, such as an ability to haul their own possessions uncorrelated with other factors. Do truck owners have different migration? No idea! But it certainly seems possible.
Those undersamples are more than enough reason to think that the United Van Lines data is at best a really eccentric measurement of migration. But there are two other biases that I can’t really call “undersamples” as much as just structural problems with this kind of “convenience sample.”
- Geographic Coverage — All the major van lines have pretty much national coverage. But not always at the same density. It’s very likely that market share varies by locality. I don’t know how to get such data, but it seems unimaginable to me that UVL has the exact same share of the market in every county. Now if the distribution of coverage rates is random, this isn’t necessarily a problem, unless that randomness happens to give us a very eccentric random result. But coverage gaps probably aren’t random, but rather relate to geography, pricing, and advertising, all factors that may correlate to individuals with different migration preferences.
- Begging the Question — It’s possible UVL’s data is, in a sense, cooked, but in a non-malicious way. If UVL is smart, they will advertise and organize themselves in such a way as to capitalize on existing migration flows. But that means that their distribution of services and advertising will be biased by their available information about migration trends. That means that whatever trend UVL thinks exists is likely to lead them to provide services that align with that trend. This can lead to a self-fulfilling prophecy where they market to places they already believe migration is occurring, causing their study’s results to be biased in favor of “popular narratives” of migration, even if they’re not generally true.
- Non-Residential Moving — UVL handles household and corporate moving. The study is supposed to cover only household moving. But it’s not clear how rigid that divide actually is. Plus, some people may rent a van without moving help, and use it for a business move, miscategorizing themselves. Insofar as non-residential moves sneak into the data, it could reflect additional moves that are not truly a form of migration.
Using Bad Data Well
Know Where It Fits
If you want to use the UVL data to show that there is corroborating evidence that migration to a given state is more positive or more negative for homeowners, families, and the middle- or upper-middle class, that’s probably reasonable. But it’s not stand alone evidence even for that population due to its many unmeasurable biases. It’s not evidence you can apply to all migration. It’s just one very broken piece of the puzzle.
That said, it is really cool to have natural data like this. And UVL has made maps of the data available going back to 1978, which means that it’s one of the longest-running consistent annual estimators of net migration we have. So that’s also a really good feature of the map. A time series of the UVL data for given states could be valuable and interesting, if we had control variables for UVL’s coverage, market share, advertising strategies, relative pricing, etc.
Furthermore, conveniently, the sample of homeowning middle-class families probably is relevant to Nicole’s topic of taxes since those people are more likely to be bearing a heavy tax burden than young, poor renters. So this measure may be more reliable for directly tax-motivated migration, though I wouldn’t lean too heavily on that.
What is helpful though is that the data does basically fit with other sources. I won’t bore you with maps you’ve all seen before, but look at those gainer and loser states: they really are more-or-less the ones that we know from IRS SOI, ACS, PEP, and the Census are losing people to domestic migration. Because this less-reliable data is corroborated by more reliable sources, it makes a fairly interesting and useful “extra check.” And because it seems fairly reliable, it also means we can view the historic data as being in about the same ballpark as what ACS or PEP would have shown did they reach back to 1978.
So if you want to use the van data, you should either use it as (1) a ballpark estimate for historic years or (2) corroborating evidence justifying estimates like ACS, PEP, or IRS. Unfortunately, that’s not how the Tax Foundation uses it: the data is taken as truly representative of migration generally when, uh-oh, it totally isn’t.
Know the Data You Cite
Moving van data is not a very reliable indicator of migration generally, and commentators and policymakers would be well-served leaning more heavily on more statistically meaningful sources. That said, the moving van data provided has some real benefits in terms of its longevity and usefulness as a large, corroborating sample for specific populations. Altogether, Tax Foundation’s use of the data was really not that bad, I just wish they’d also mentioned that, hey, this source corroborates what the actual data tell us about migration. More importantly, Tax Foundation’s central claim, that taxes have a substantial indirect and some direct effect on migration, was an example of reasonable, justified policy claims about migration. More migration commentators could learn from that kind of well-hedged, moderate language.
For more details on how taxes impact migration, see my post here.
For details on how taxes and migration impact inequality together, see here.
If you like this post and want to see more research like it, I’d love for you to share it on Twitter or Facebook. Or, just as valuable for me, you can click the recommend button at the bottom of the page. Thanks!
Follow me on Twitter to keep up with what I’m writing and reading. Follow my Medium Collection at In a State of Migration if you want updates when I write new posts. And if you’re writing about migration too, feel free to submit a post to the collection!
I’m a graduate of the George Washington University’s Elliott School with an MA in International Trade and Investment Policy, and an economist at USDA’s Foreign Agricultural Service. I like to learn about migration, the cotton industry, airplanes, trade policy, space, Africa, and faith. I’m married to a kickass Kentucky woman named Ruth.
My posts are not endorsed by and do not in any way represent the opinions of the United States government or any branch, department, agency, or division of it. My writing represents exclusively my own opinions. I did not receive any financial support or remuneration from any party for this research. More’s the pity.