One Average, Several Worker Profiles

After the clock and the map are made explicit, the H-1B wage question can be tested against other baselines and divided into worker profiles. This chapter asks whether the negative association survives changes in survey, job-tenure comparison, and immigrant baseline—and whether an average gap conceals premiums for some workers and deficits for others. The point is not to replace one headline with another, but to learn what a distribution of estimates can responsibly say.

Robustness is a range, not a yes-or-no test

The revised NBER analysis begins with the ACS baseline and then changes the survey used to represent native workers. The adjusted gap is −16.1 percent with the ACS. Using the CPS Annual Social and Economic Supplement produces −15.6 percent, a very similar result. A smaller sample that merges the ASEC with monthly CPS information produces a gap close to −20 percent, and restricting that sample to natives who reported only one job produces −19.4 percent. The direction is therefore not unique to one ACS calculation, although the magnitude moves with the sample and the information available.

That movement is informative in two ways. First, it makes a simple “ACS artifact” explanation less persuasive within the paper’s framework: a different survey with annual earnings produces a nearby negative estimate. Second, the alternative surveys are not perfect replacements. The ASEC reports metropolitan area of residence rather than place of employment, and the Survey of Income and Program Participation does not provide the same detailed metropolitan information. Changing the survey can solve one measurement problem while introducing another. Robustness means understanding what changed, not counting how many tables point in one direction.

SIPP shows why a result can be numerically different without automatically refuting the others. The SIPP estimate is close to zero, at +0.020, with a relatively large standard error. The paper cites research finding that SIPP wage and salary income is understated by roughly 11–15 percent. If the native baseline is measured too low, a zero observed gap need not mean that H-1B and native wages are actually equal. It may reflect an income-measurement difference. This is not evidence that the SIPP result is useless; it is evidence that the survey’s reporting properties must be part of its interpretation.

The broader lesson is that a survey can be a better instrument for one question and a worse instrument for another. The ASEC’s annual earnings window is attractive because it refers to the calendar year before the survey, while the ACS’s rolling twelve-month reporting creates the timing issue discussed earlier. But the ASEC baseline in this analysis uses residence rather than employment geography, and the smaller merged sample sacrifices observations when it adds information about multiple jobholding. A nearby coefficient across surveys is reassuring only about the dimension that the surveys share; it does not certify every underlying choice.

A second robustness issue concerns job stage. The I-129 wage is a starting offer for a new H-1B hire. Many native workers in the baseline have been with their employers for years. If wages rise with tenure, comparing those incumbents with new H-1B entrants could exaggerate the H-1B disadvantage even when the workers have similar age, education, occupation, industry, and location.

The NBER paper estimates the possible size of that bias using CPS job-tenure data.