How Far Will Puerto Rico’s Migration Fall?
Lots of People Have Opinions, Nobody Has Answers
Because I write about Puerto Rico, lots of people have asked me to write about the impact of Hurricane Maria. I wrote prospectively about Hurricane Irma, and the same thinking applies to Maria. But people want to know what has happened since.
Here’s the problem: there’s no data yet. Air traffic data won’t be available for weeks or months still, and is very noisy. The Census Bureau will publish 2016–2017 population estimates in December or January, but those reflect July 2017 population, so won’t include Maria: we won’t have any “official” U.S. government estimate of the impact until December 2018 or January 2019. The Puerto Rico Community Survey for 2017, like the ACS for Puerto Rico, won’t be available until October 2018. And even then, its sampling frame is such that about 75% or more of its survey-weight will have occurred before Hurricane Maria. Not until October 2019 will we have a PRCS sample that’s majority-weighted post-Maria. I’ve discussed these timing problems in migration data here and here.
So my plan was simple. I’d sit tight and responsibly write nothing about Puerto Rico until we had some actual data.
However, the pressure to say something is immense, particularly for people and groups who are, well, Puerto Rico-focused. Enter, the Center for Puerto Rican Studies. They published an article suggesting that, well, I’ll let you read it:
Between 114,000 and 213,000 will leave the island annually from 2017 to 2019, with up to 470,000 lost residents total.
Here’s their chart:
Their method is very, very simple:
This is very arbitrary. For my prospective look, I used actual econometric parameters from study of disasters in the US at the county level, with adjustments. My adjustments were, of course, arbitrary and perhaps excessively large. This study… really just assumes that outflows will double or treble. Whereas usually we want a study to econometrically derive an estimate of outflows, this briefing simply declares a level of outflows, and then, having declared what outflows will be, is focused on estimating where those outflows might end up going.
They also assume 2017 inflows into Puerto Rico will be zero.
Let’s interrogate these assumptions.
Consider the PRCS/ACS Sampling Frame
The study authors use the ACS/PRCS sample for their work. Yet they have 2017 fully reflecting the hurricane.
Let’s do something fun.Let’s assume that the weighted monthly estimate of net migration from January to August was equal to 2016, but from September to December is equal to 5 times worse than 2016. In other words, I assume that post-Maria 2017 migration is much worse than the Centro assumes for their paper. However, unlike the Centro, I actually pause to take a second and review how the ACS is conducted: continuous sampling across 12 months and weighting across those samples. Here’s the “annual estimate from monthly values” I’m using:
If we use these estimates and weight them using the ACS sampling frame as well as some very generous adjustments assuming ACS weights more recent data more heavily for migration (I don’t know that this occurs, I just want to be as generous to the Centro as possible), we get the following estimates:
See, the Centro doesn’t seem to be doing any weighting. They assume zero inflows into Puerto Rico in 2017: but did nobody move there before Hurricane Maria? And is zero really plausible even post-Maria? And in assuming a huge spike in outflows, one has to keep in mind that ACS samples the entire year. So Just at a very basic level, the Centro and I show different time trends for ACS-measured migration. I show a similar total volume of migration, but to get it, I actually have to assume a much, much bigger impact of Hurricane Maria (quintupling of monthly outflows, not doubling or trebling) than the Centro estimates. By incorporating the different sampling frames for the ACS, it moderates lots of estimates by spreading out the chronology of measured migration. This, by the way, is why, for many geographies, it’s necessary to match different years of ACS, IRS, and CPS-measured migration: because the timing of their sampling can differ substantially. By neglecting the ACS sampling frame, the Centro’s estimates almost certainly make outflows look more severe than may be responsible to forecast, and may also wrongly forecast the measurement timing.
How Likely Is That Rate?
But there’s another plausibility check we can use. We can look at the historic net migration of Puerto Rico and other states and see if these changes are of a plausible extent. Here’s Puerto Rico’s net migration rate, including the Centro’s forecast, as well as an alternative forecast I think is more responsible to forecast.
As you can see, both of us are forecasting truly historic rates of outmigration. But the scale is very different. I’m saying Puerto Rico’s net migration might be as bad as -4%, which is nearly double the previous worst-ever record. The Centro is saying it might get as high as -7%, three times the previous historic record.
But how plausible is that? Well, we don’t have a large sample of event studies of class 5 hurricanes striking small islands. But we can look at all the other US states, 1910-present, and see if any have year-over-year changes in net migration of a similar scale as the Centro is forecasting. We’ll use each state’s standard deviation from its series-length average as our metric. The Centro is estimating a YoY migration change of -30.14 z-score (yes, stats nerds, you’re reading correctly); I’m estimating -10.5. Now, in fairness, my historic estimate for migration in Puerto Rico is a bit smoothed, so the standard deviations are suppressed a bit, resulting in artificially high z-scores. But let’s say that we want to see cases where states had a Z-score of 3/-3 or greater. How many such cases have there been since 1910?
As you can see, there have been some big migration shocks in many states! Mostly during massive wartime mobilizations, however. Non wartime shocks include Alaska in 2015 thanks to a natural resource boom-bust cycle, New Hampshire in 1991 for Lord-knows-what, and Louisiana for Hurricane Katrina in 2006 and 2007. Note, please, that Hurricane Katrina struck Louisiana in 2005, but shows up in 2006 migration data, not 2005!
Even Louisiana’s shock was just -4.25 in 2006, and then they had a big spike up in 2007. The trick is that Puerto Rico’s distance, isolation, and the low incomes of its residents all militate against Katrina-scale outflows, but also against post-Katrina-scale-return.
What’s my point? Simple: we should indeed expect something historic. Both the Centro and I are forecasting the most massive shock to state-equivalent-area annual net migration in American history.
But there are right and wrong ways to do this. The right way is to start from the available empirical literature, which suggests displacement of 2,000 to 25,000. But Hurricane Maria may indeed be a worse disaster than most measured in the literature, compounding other ongoing crises, for very migration-prone people. As a result, it is reasonable to substantially expand our range of estimates, especially as reconstruction seems to be proceeding unevenly. I forecast marginal displacement in 2017–2019 at about 120,000 beyond Puerto Rico’s existing migration losses of 50–60k per year. far beyond what the academic literature would suggest, but, nonetheless, I think prudent. This estimate is substantially in excess of what the academic literature would suggest already. The Centro’s implicit estimate of 200–300,000 marginally displaced people beyond baseline outflows seem’s irresponsible to me; there is not yet evidence to support that claim.
Nonetheless, the Centro may end up being right. Sometimes, bold, out-of-the-money calls are right. Sometimes, non-empirically-founded forecasting methods give correct results. Forecasting is hard and fraught with peril. Especially in rapidly-developing crises, the range of uncertainty is very large. I think 200–300k marginally displaced people is unlikely…but not by any means impossible. I am right with the Centro in forecasting a migration situation more extreme than what any other region has ever experienced, by a substantial margin. However, precisely because there is a great deal of uncertainty, it is vital that forecasters double-down on responsibly-developed forecasting methods. When the storm is fiercest, we need the steadiest hand at the tiller. With prominent political figures suggesting that population losses could be in the millions, it is beholden on serious analysts to put in the elbow grease on getting good data and using the best research.
Once we actually have some data, I’ll have more to say. For now, my estimates of population losses have been slowly drifting upwards. I think Puerto Rico is facing a serious demographic step down that will have critical consequences on the island and the mainland, and my view of this step down is not improving. But nonetheless, I do not believe we have enough information yet to make the kinds of forecasts the Centro made.
Check out my Podcast about the history of American migration.
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 native of Wilmore, Kentucky, a graduate of Transylvania University, and also the George Washington University’s Elliott School. My real job is as an economist at USDA’s Foreign Agricultural Service, where I analyze and forecast cotton market conditions. 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.