What the Individual-Level Evidence Actually Shows
The individual-level evidence asks a narrower question than “Do licensed workers move?” It asks whether, among people who make a substantial move, workers in occupations with more state-specific licensing are less likely to cross a state line than comparable workers in occupations with more portable exams. The value of this design is that it brings the suspected credential-transfer cost closer to the decision being studied. Its limitation is that the answer remains conditional on how the comparison is constructed and on what the data cannot observe.
The result is often summarized as a roughly 7 percent reduction in interstate migration for state-specific licensed occupations. That summary is useful only if the denominator, sample, comparison group, and outcome are kept in view. The same study reports a much larger raw difference before those choices are made. Reading the estimates in sequence shows why the numbers change and what the preferred estimate does—and does not—establish.
How the comparison is constructed
Johnson and Kleiner use the American Community Survey from 2005 through 2017, focusing on people ages 18 to 64 who live in the 50 states or the District of Columbia and have usable information on migration, occupation, and the main demographic controls. The ACS identifies a person's current residence and, for movers, the prior year's residence. Beginning in 2005, it also provides sub-state geography through Public Use Microdata Areas of Migration, or MIGPUMAs. These areas are roughly county-sized in dense places and larger in rural places.
That geography lets the researchers estimate the distance of a move rather than treating every change of residence as equivalent. They calculate the straight-line distance between the centroids of the current and prior MIGPUMAs. A move between states is necessarily a move between MIGPUMAs because the migration areas are contained within states, but a move within a state can be local or long-distance. The design can therefore ask whether a substantial move crossed a state border.
The sequence of restrictions changes the meaning of the outcome:
In the full sample, the dependent variable is an indicator for having moved between states in the prior year. People who did not cross a state border include nonmovers, people who moved within a MIGPUMA, and people who moved between MIGPUMAs while remaining in the same state. The comparison therefore mixes the decision to move at all with the decision to move across a state line.
The next restriction keeps people who moved at least 50 miles. Now everyone in the sample has demonstrated a willingness to make a meaningful move, even if that move stayed within a state. The outcome asks whether the move crossed a state border. This is closer to the cost of transferring a license, because a long-distance move is more likely to change the relevant labor market and to destroy local career capital for occupations that depend on it.
The preferred specification goes one step further. It keeps long-distance movers who lived outside their state of birth in the previous year. State of birth is an imperfect measure of home-state attachment, but it helps address a plausible selection story: people who strongly prefer to remain near home may choose state-specific occupations and also be less likely to cross a state border. Restricting the sample to people already living away from their birthplace removes some of that pattern from the comparison.
The regressions then adjust for observable differences between workers and places. Controls include education, age, income, race, marital status, employment status, citizenship, and number of children. Prior-state-by-year fixed effects account for conditions associated with where the person lived before the move and when the move occurred. State-of-birth fixed effects account for persistent differences associated with birthplace. These controls do not make the comparison experimental, but they narrow the set of differences that the licensing indicator is asked to capture.
The comparison group is also important. The main estimate compares state-specific licensed occupations with quasi-national licensed occupations. Both groups are licensed in every state, which makes them more similar than a licensed-versus-unlicensed comparison on some institutional dimensions. The quasi-national group is expected to face lower transfer costs because it does not have to repeat a state-specific licensing examination, although it may still face other state requirements. A second analysis compares state-specific occupations with all other occupations, but that broader group includes some occupations with licensing requirements that are not cleanly observed. The authors therefore treat it as a less precise comparison and, in some specifications, a lower-bound contrast.
This design is aimed at a particular causal question: conditional on making a long move, and after accounting for observed characteristics and place effects, is a worker in a state-specific occupation less likely to cross a state border than a worker in a quasi-national occupation? It is not estimating the probability that any randomly selected licensed worker will move next year. It is also not measuring how long a license transfer takes or how much income the worker loses while waiting.
What remains after the filters
The descriptive pattern explains why the sample design matters. In the full ACS sample, about 2.0 percent of people in state-specific licensed occupations moved between states in the prior year, compared with about 2.9 percent of people in quasi-national licensed occupations. Among those who moved at least 50 miles, 61.3 percent of state-specific licensed movers crossed a state line, compared with 70.5 percent of quasi-national licensed movers.
Those figures are meaningful descriptions, but they are not yet the causal effect of re-licensure. The first comparison includes people who never moved and occupations with different levels of local capital and migration preference. The second conditions on a long move but still includes differences in birthplace and other characteristics. A reader who reports the larger gap as “the effect of licensing” is combining several stages of the migration decision.
The regression makes the difference explicit. In the full sample, the estimated coefficient for state-specific licensing relative to quasi-national licensing is −0.014. The dependent-variable mean is about 0.024, so the reported percentage effect is −58.33 percent. That is the large headline gap that appears when the analysis asks about interstate migration across the whole population and controls for observable characteristics without first conditioning on distance.
Among all movers of at least 50 miles, the coefficient is −0.073 and the dependent-variable mean is 0.652. The corresponding relative effect is −11.20 percent. The gap has fallen because the sample no longer compares people who do not move with people who do. It compares the border-crossing choice among people who have already accepted a substantial relocation.
In the preferred sample—50-mile movers who lived outside their state of birth in the prior year—the coefficient is −0.051, against a dependent-variable mean of 0.774. The reported percentage effect is −6.59 percent, which the authors summarize as about 7 percent. In percentage-point terms, the estimated difference is about 5.1 points in the probability of crossing a state line among these long-distance movers; in relative terms, it is about 6.6 percent below the comparison mean. Those are different descriptions of the same estimate, and keeping them separate prevents the result from sounding larger or smaller than it is.
The preferred estimate is therefore not evidence that licensing prevents most interstate moves. It says that within a selected group of people already willing to move a long distance, and relative to a group facing lower expected exam-transfer costs, the state-specific group crosses state lines at a lower rate. The interpretation depends on the assumption that conditioning on the long move and the included controls removes other differences in the returns or costs of crossing a state line.
The estimate is also not a direct observation of a worker taking a second examination. The ACS records occupation and migration, not the board's exact decision, the worker's application history, or the counterfactual job they would have taken in another state. The study uses the examination classification to create a plausibly relevant contrast. That makes the result informative about portability, but it does not turn the coefficient into an invoice for a specific license.
Robustness is evidence about stability, not proof of causation
The authors test whether the preferred pattern changes substantially when the model is specified differently. They add controls in stages, including education, age and sex, other personal characteristics, state-year fixed effects, state-of-birth fixed effects, and interactions between age categories and state-year effects. The estimated difference changes only modestly while the model's explanatory power rises. That pattern makes it less likely that the result is simply a consequence of omitting the measured demographic and place variables.
They also use sensitivity calculations developed by Altonji, Elder, Taber, and Oster. In the preferred outside-state-of-birth sample, the reported calculation requires unobservable selection roughly 15.77 times as important as observable selection to move the effect to zero under the suggested maximum R-squared. Even when the hypothetical maximum R-squared is set to 1, the required ratio is 1.98. In plain language, the result would need remaining selection on unobserved characteristics to be substantially stronger than the selection captured by the observed controls under those assumptions.
That is useful evidence, but it is not a random-assignment guarantee. The sensitivity method assumes that unobserved selection operates in a direction related to observed selection and depends on a hypothetical maximum R-squared. The study still lacks broad, historical information about changes in re-licensure requirements that could provide a more direct source of exogenous variation.
The result is similar when a 100-mile rather than a 50-mile threshold defines a long move. It is also nearly identical across the earlier 2005–2011 period and the later 2012–2017 period, despite a change in the ACS geography definitions. Alternative ways of clustering standard errors leave the estimated effect statistically distinguishable from zero, and randomization-inference procedures produce similarly strong evidence in the reported specifications. These checks reduce concern that one threshold, one period, or one standard-error convention is doing all the work.
The authors repeat related analyses with the Current Population Survey Annual Social and Economic Supplement. The CPS has a much smaller sample and cannot measure sub-state move distance or state of birth, so it cannot reproduce the preferred ACS design. It does, however, record the prior year's occupation and provides an independent check on whether the pattern depends entirely on observing only the current occupation. The results remain similar, with the important caveat that the alternative data solve some problems while creating others.
The average hides occupational differences
A single average is not a good description of every licensed career. In the study's occupation-specific analysis, 12 of the 16 state-specific occupations have significantly lower interstate migration than quasi-national occupations when move distance is not restricted. After conditioning on a 50-mile move, six remain statistically significant. In the preferred sample of long-distance movers outside their state of birth, five show a significant reduction: teachers, pest-control workers, electricians, pharmacists, and lawyers.
The reported effects among those occupations vary. Pharmacists and pest-control workers show the largest reductions, about 16 percent and 14 percent respectively. The other affected occupations are roughly 7 percent lower. Several state-specific occupations do not show a statistically significant difference in the preferred comparison. That does not prove that their transfer costs are zero; it means the study does not estimate a reliably different migration rate for them under that specification.
The variation fits the mechanism. A pharmacist may work for a large firm and have less local clientele than a self-employed professional, but a state-specific jurisprudence examination can still create a distinct transfer step. A teacher may carry subject knowledge across a border but lose district-specific seniority and benefits. A lawyer may face both a state exam and a practice built around local law. These are not interchangeable cases, so an average estimate should be used to identify a pattern, not to predict the experience of every worker.
The data can miss the most serious career interruption
The title's idea of a career stalling raises an outcome that the main ACS design cannot observe directly. The survey records a person's occupation after the move, not necessarily the occupation before it. A worker who moved because of a household decision and then could not transfer a license might switch occupations, leave the labor force, or accept work outside the measured occupation. If the analysis retains only people observed in the occupation after the move, some of the most consequential effects disappear from the sample.
The CPS comparison helps because it records the prior occupation. Among interstate movers, continuing members of state-specific licensed occupations are more common than continuing members of other occupations, and the difference between state-specific and quasi-national groups remains similar when the analysis is limited to people reporting the same occupation in both years. This makes it less likely that the ACS limitation overturns the reported migration pattern. It does not measure the workers who left the occupation altogether, the earnings they lost, or the effect on a partner's career.
The authors specifically note the possibility of trailing spouses and other household-level decisions. Migration is often chosen for the household, not by each worker independently. A licensed worker may move because a partner receives a better opportunity and then bear the cost through a career change. The worker is observable as a mover, but the licensing-related loss may be visible only in an outcome the data do not follow.
The individual-level evidence thus supports a careful conclusion. After narrowing the question to long-distance movers and reducing several sources of selection, state-specific licensed occupations show about a 7 percent lower interstate-crossing rate than quasi-national licensed occupations in the preferred sample. Robustness checks make that pattern harder to dismiss as a simple artifact, and occupation-specific results show that it is not uniform. But the estimate remains conditional, the exam category is a proxy, and the data do not capture every stalled career.
The remaining question is one of scale. If a portability penalty matters for selected workers, does it also explain the total volume of migration between states? A second study approaches that question with a different unit of analysis, a state-level licensing measure, and a spatial flow model. The contrast is not a reason to discard the individual result; it is the next test of how far that result can be generalized.