The challenges of counting the nation’s unauthorized immigrants
For Updated Methodology
A more recent discussion of our methodology for estimating the unauthorized immigrant population is available here.
In a report released today, the Pew Research Center estimates that there were 11.7 million unauthorized immigrants in the United States in 2012, compared with an estimated 11.5 million in 2011. But there’s a lot more to those numbers than meets the eye, and they’re not comparable to previously issued Pew Research estimates. To help explain how the new numbers were derived and how to interpret them, we spoke with senior demographer Jeffrey S. Passel, who’s been researching the unauthorized-immigrant population for more than three decades. An edited transcript follows:
Explain your basic approach to estimating the number of unauthorized immigrants.
The basic method we used is the same as before: We make an estimate of how many immigrants are in the country legally, and we have a government survey that measures how many immigrants total are in the country. Then we subtract the legal immigrants from the total.
Well, it’s a bit more complicated than that. We get numbers on legal immigrants from two federal agencies: First, we get a count of people admitted legally to the U.S. with so-called “green cards” for permanent residence from the Department of Homeland Security. Before the DHS was created, the numbers came from the Immigration and Naturalization Service. We know what countries those immigrants are from, how old they are and a lot of other demographic characteristics. The other group of people the U.S. admits are refugees; we get these counts from the Office of Refugee Resettlement, which is part of the Department of Health and Human Services. Then we use those counts to develop a population estimate, using standard demographic techniques that account for deaths, departures and new arrivals each year.
The estimates of the total number of immigrants come from two Census Bureau surveys: the Current Population Survey or CPS and the American Community Survey or ACS. Both ask people where they were born, whether they’re citizens and so forth — but not whether they’re here legally. From those, we get a number for what’s called the “foreign-born” population. The big difference between this report and our previous ones is that before we used only CPS data; this time we’re using the ACS data for 2005 through 2011 and CPS data for 2012 and before 2005.
Why the change?
The CPS (which, by the way, provides the data used to calculate the monthly unemployment rate) is a large survey by the standards of the survey world — normally about 55,000 households a month. We use the CPS from March of each year, when they increase the sample size to 80,000 and expand the questionnaire to include questions that get at income and poverty and health-insurance coverage. But when you restrict it to just the immigrants and then look only at some immigrants, it can have pretty sizable margins of error.
So we started using the ACS, which is a very large survey: they end up interviewing more than 2 million households a year. The Census Bureau only started the ACS in 2005, to replace the old “long form” from the decennial census. This is the first time we’ve incorporated the ACS data into our estimates. We get access to a sample that works out to 1% of the entire country, or about 3 million people. Since the ACS is much, much bigger than the CPS, we get much smaller margins of error, so we can have greater confidence in measuring year-to-year changes in the unauthorized population.
But the 2012 ACS wasn’t available when we were putting together this report — the Census folks only released it last week, and we won’t get access to the individual-level data we need for another month or two. But data from the March CPS was available to us, so we used that for the 2012 estimate. At some point we’ll have a 2012 estimate based on the ACS that will replace the one in the report based on the CPS, with a smaller margin of error, and we’ll have a 2013 estimate based on the CPS with a large margin of error.
Using the ACS data wasn’t the only change you made this time around, was it?
No, we also re-weighted prior years’ data. You see, the population numbers you get from a survey are tied to annual population estimates the Census Bureau puts out. What they do is work from the previous census and update the estimates each year; then, every 10 years, they get a new benchmark with the latest census. Last year we didn’t have the 2010 Census as a benchmark, but now we do, so we have a much better idea of what the population looked like in, for example, 2009 than we had before when the estimates were based on the 2000 Census.
However, Census doesn’t go back and re-estimate the CPS and ACS data from past years when it gets the new decennial benchmarks. So we went back and re-estimated the data for every year from 2001 through 2009, based on the 2010 Census data. We do that because we think it’s important that our surveys be consistent over time to talk about year-to-year changes.
Similarly, CPS data before 2000 was not at all comparable to what came afterward, so this time we also went back and re-weighted all the data from the 1990s as well. So this is the first time we’ve released estimates for the 1990s.
In the graphs from the report it sure looks like the line showing the unauthorized-immigrant population is going up from the recent low of 11.3 million in 2009. So why do you say only that “the number may be rising again”?
These numbers come from a sample, as all survey estimates do. Surveys don’t interview everyone in the country; they only interview a sample of the population. The sample results may be different from what we would get if we interviewed everyone. Not only that, but the results can vary from sample to sample. That’s what we mean when we talk about a “margin of error.”
In the case of these estimates, the margin of error is pretty large relative to the year-over-year changes. What that means is that what appears to be a change could be just randomness related to the characteristics of the sample. We need to see a big enough change before we can decide that it’s real. For example, when we saw the first signs of a decline in the unauthorized population in 2007 and 2008, we couldn’t say for sure that there was a drop — we had to wait till we had the 2009 data because the CPS sample wasn’t large enough to detect the 2007-08 change.
With the ACS, though, we can say that the 2007-08 change was “statistically significant,” or outside the margin of error. A better reading of the 2011-12 change will have to await the 2012 ACS.
In last year’s report Pew Research estimated the number of unauthorized immigrants at 11.1 million in 2011. Doesn’t that mean there was a statistically significant increase between 2011 and 2012?
No. Because of all the changes in data sources and methodology this year, last year’s estimates have been superseded and aren’t comparable with the current ones.
Drew DeSilver is a senior writer at Pew Research Center.