Why People Use IRS Migration Data By Lyman Stone Much migration commentary focuses on data derived from the Internal Revenue Service s Statistics of Income publication (this data referred to as IRS SOI henceforth). This data is published annually and stretches back to the 1980s. The popularity of IRS SOI data derives from four features of the data. First, it is a longer time series than is available for most other measures of domestic migration. Second, it is not a survey-based method, but a direct measure of actual migration. Third, the data is accompanied by information about Adjusted Gross Income, which many commentators use to estimate economic impact of migration. Fourth, recent improvements to the data since 2011 allow more detailed demographic break-outs than are available from Census Population Estimates (PEP), or, in some cases, from publicly available pre-packaged American Community Survey (ACS) data products. All four of these features amount to legitimate reasons to favor the use of IRS SOI data. ACS data only stretches back to 2000, 2005, or 2007 depending on the exact topic and geographic level and, although it is very detailed, it has a statistical margin of error due to being a survey. PEP data has no margin of error, but domestic migration data does not stretch back as far, and PEP data is subject to frequent revisions. PEP data also does not provide any demographic break-outs of migration, or gross migration estimates, only net migration. Meanwhile, now-discontinued migration data from the long-form decennial Census had extensive demographic detail and was more than just a survey, and stretches back to either 1910 or 1950 depending on the topic, but is not available in an annual format, and ended with the 2000 long form Census, now replaced by the ACS. However, even as IRS SOI data is probably the most widely cited and used migration data, it is also probably the most widely misused migration data. Common Misuses of IRS Migration Data There are four common misuses of IRS SOI data. In order to avoid unnecessary antagonism, this article will not cite specific cases of each type of misuse, but undoubtedly readers familiar with the state of migration commentary will recognize these misuses. IRS SOI Data Doesn t Measure Migration of Income It is common for migration commentators to treat the AGI of IRS SOI migrants as migration of money. This is an egregiously wrong use of the data. The IRS SOI user guide makes clear that this is not a viable interpretation of the data, and thus those who read the data this way have either failed to perform the most basic due diligence by looking at the manual, or else actively mislead their readers.
There are four reasons why AGI associated with migration is not equivalent to migration of money. First, the AGI associated with a given migrant return is the total AGI for the filer after migration. It is not how much money they earned prior to departure, it is how much money they earned after arrival (although of course, AGI is not quite equivalent to earnings, a point only worth belaboring for tax experts). Thus the AGI reported has no necessary connection to the departure state; it cannot be considered an outflow. Second, many migrants experience job transitions, moving out of one job to another. This is one of the most common reasons for migration. When a migrant moves from one state to another as a result of a job transition, it is likely that they move from a job that will still exist after their departure, to a job that existed before their arrival. In other words, the AGI they had before departure is likely to more-or-less remain in the previous state (minus transition costs as the employer fills the vacancy), while their new AGI in the arrival state likely existed prior to their arrival (again, minus transition costs related to employment vacancy). Third, insofar as migration of money is a real thing, any economist will tell you that it refers to capital mobility. Capital mobility is a radically different phenomenon than labor mobility. Unfortunately, figures on capital mobility within the United States are less available than figures on labor or population mobility, that is, migration. Conflating labor mobility (migration) with capital mobility is entirely incorrect, as anyone who s taken even basic macroeconomics can explain. Labor migrants may shift their investments alongside migration, especially residential investment if they build a new home, and investment generally may respond to population changes, but the phenomena remain distinct and should not be confused by sloppy terminology. Finally, some income, such as rental, ownership, or investment income may migrate with migrants. But that income type is not separated out of AGI in the IRS SOI migration files, and the largest share of investment income, namely capital gains, is realized in very lumpy and tax-sensitive patterns that may be correlated with migration across taxing jurisdictions. This money may move, but, depending on the exact nuances of the economic question being asked, it may be more appropriate to count capital gains on an accrual rather than a realized basis, in which case even separating investment income from AGI would not give an accurate depiction of money migration. In sum, labelling the AGI associated with IRS SOI migration as migration of money is unjustifiable. IRS SOI Data Doesn t Measure All Migration There are several reasons IRS SOI data may fail to capture all migration, and why that may be significant. But those reasons can be lumped into two key categories: sample bias and procedural problems. These factors together suggest that although IRS SOI data is a valuable indicator of migration, it is almost certainly a less complete indicator than ACS, PEP, or especially old Census data. Sample bias relates to the manner in which the IRS collects its data. Non-filers will not show up in the data. This un-counted population is disproportionately likely to include the young, low-income, homeless, illegal residents, felons, students, and even some retirees. These groups are extremely likely
to have different migration patterns than the population on the whole, and in some cases have extremely significant impacts on total migration and the economy. Because the IRS does not count these people, and makes no adjustment to try and represent the whole resident population, it is an incomplete and a biased sample. Broadly speaking, IRS SOI migration understates the number of migrants by between 10% and 30%, depending on the year and geographic unit in question. However, IRS SOI data is improving. This gets to the procedural problems involved in IRS SOI data. Before 2011, the method used for processing tax returns and classifying them as migrants missed a substantial fraction of migrants, especially high-income migrants who tend to file later in the year. Since 2011, IRS has greatly improved the methodology used, and more returns are correctly categorized as migrants. For data prior to 2011 however, true tax-filer migration rates may differ versus what the IRS SOI data shows. Even since 2011, it is possible that some real migrants still slip through the cracks and fail to be classified as migrants, although this problem has been greatly reduced. Unfortunately, this improvement in methodology creates a break in the IRS SOI time series, meaning that current estimates are not exactly comparable to pre-2011 estimates. IRS SOI Data Measures Shadow Migration The IRS SOI estimate of migration is based on changes in mailing address on subsequent tax return years. However, mailing address for tax returns may not be equivalent to residence, the relevant factor for true migration. Some individuals may list a business address. Some may list a preparer s address. Some may use a P.O. Box. There are many reasons why the address to which a tax return should be sent may vary from actual residence. Furthermore, IRS SOI data measures migration between filings. This is not strictly a 12-month period, and before 2011, many late filings were not included. Individuals who file in March one year but June the next will have 15 months in which a migration could be measured rather than 12. It is likely that the average gap is 12 months, but it is possible that individuals with longer or shorter filing gaps are correlated with migratory life changes. Likewise, this different timing of the estimate can lead to IRS SOI data tracking ACS or PEP data with an uneven time delay as exact seasonality of migration has some inconsistency across years. As such, IRS SOI data has an inconsistent time frame, offering possibly inconsistent year-to-year estimates of migration, even as it also may miss some real migrations, or pick up migrations that didn t really occur. These effects are likely to have a fairly small effect on headline migration numbers, but the point is that they are statistical problems that should give commentators pause in citing IRS SOI data as the most authoritative figure for the whole population. Likewise, these problems are likely to be more severe the more specific a geographic or demographic sector of the data is specified. Major Economic Impacts of Migration Are Hard to Measure The largest impacts of migration are not from the static shift of income or investment, but rather the shift of future earning potential and productivity. For example, if a large number of high-earning people move from Illinois to Florida as they retire, then their income may decline, and is probably not likely to
rise. This might not add to Florida s long-run growth very much. On the other hand, if a large number of low income college graduates move from Florida to Illinois, they may have low incomes, but their incomes are likely to rise. This is much more likely to boost long-run economic growth. In other words, the major economic impact of migration is in shrinking or growing the productive and consumptive base of the economy through the movement of labor, not just static AGI measures. Net migration matters because it reflects a changing base of consumption and production in a very straightforward way. But the demographic composition of gross migration matters just as much, because the productive and consumptive base may be radically altered by different economic profiles of in- and out-migrants. Towards a Better Estimate of Illinois Migration There are numerous sources of data about migration: IRS SOI, ACS, PEP, Census, and others. In this section, I ll provide information about what all of these sources suggest about the history of migration in Illinois. Chart 1 below shows all the available sources, as well as my attempt to harmonize them into a single best estimate of net migration. This estimate should be seen as a plausible, but not absolute, baseline for the level of net migration in a given year. Chart 2 reproduces the same data, but presents it as a percent of population instead of the raw number. Chart 1: Net Migration 60000 40000 20000 0-20000 -40000-60000 -80000-100000 -120000 1900 1906 1912 1918 1924 1930 1936 1942 1948 1954 1960 1966 1972 1978 1984 1990 1996 2002 2008 2014 American Community Survey 1- Year Samples Census Populafon Esfmates Series IRS Stafsfcs of Income Migrafon Data Decennial Censuses, 5-Year Net Migrafon Esfmate, Average Net Decennial Census 1950, 1-Year Net Migrafon Esfmate Decennial Census Net Nafve Migrafon 1910-1950, Decadal Average Net Harmonized Esfmate -140000
Chart 2: Net Migration Rate 0.01 0.008 0.006 0.004 0.002 0-0.002-0.004-0.006-0.008 1900 1906 1912 1918 1924 1930 1936 1942 1948 1954 1960 1966 1972 1978 1984 1990 1996 2002 2008 2014 American Community Survey 1- Year Samples Census Populafon Esfmates Series IRS Stafsfcs of Income Migrafon Data Decennial Censuses, 5-Year Net Migrafon Esfmate, Average Net Decennial Census 1950, 1-Year Net Migrafon Esfmate Decennial Census Net Nafve Migrafon 1910-1950, Decadal Average Net Harmonized Esfmate -0.01 The first data source encountered is Decennial Census data on Net Native Migration from 1910-1950. These Censuses reported the net change of native-born Americans due to migration for each state. However, migration of non-native born individuals is also likely significant. I have, in general, assumed that migration among non-natives partially offset migration among natives, and I ve annualized rates roughly based on the rate of population growth. The result is high inflows during the 1910s, but outflows in all other years except 1947, when demobilization may have boosted migration. In the 1950 Census, there is a 1-year net migration estimate, which I take as given. That is the sharp bend right at 1950 in Chart 1. For the 1960-2000 Census years, Census surveyed migration in the previous 5 years. I divide this figure by 5, which is not an exact measure of annual net migration, but should be very close. I then annualized by creating more-or-less smooth trends between these periods. Beginning in 1990, there is reliable PEP and IRS SOI data. I assign IRS SOI data to the first year in its designator (so 2013-2014 data is assigned to 2013). This is pursuant to communications with the migration analyst at IRS and the previously discussed uneven time frame and seasonality of IRS SOI migration data. The result of these issues is that IRS SOI data, especially after 2011, probably best reflects net migration in the first year of its title, although this is not universally the case. I take a weighted average of these sources and the decennial 5-year rates in 2000. Then, beginning in 2005, net migration data is available from the ACS as well, and I add that source into the average. Since
2011, my harmonized estimate is essentially an average of IRS SOI, PEP, and ACS. Chart 2 simply divides these numbers by the annual population according to the U.S. Census Bureau. To assess migration before 1900, it becomes necessary to make imputations from decennial Census population data. This method suggests it is likely that Illinois experienced slightly negative domestic migration in the 1890s and 1880s, and more severe negative migration in the 1870s. Net domestic migration was almost certainly strongly positive for all decades before the 1860s. Chart 3 shows each decade by whether its total domestic net migration is estimated to be positive or negative. Chart 3: Long Run Historic Migration 1800s 1810s 1820s 1830s 1840s 1850s 1860s 1870s 1880s 1890s 1900s 1910s 1920s 1930s 1940s 1950s 1960s 1970s 1980s 1990s 2000s 2010s Estimated Balance of Migration As can be seen, Illinois has not had any period of sustained net in-migration since the 1910s, and the most recent out-migration rates are among the most severe in Illinois history. This article makes no
suggestion as to the cause or consequences of these migrations, but hopefully will serve to provide a common base of understanding about what the facts really are.