Professor Blasts Research on H-1B Visa Workers Earning Higher Wages

A University of California professor refutes claims made in a recent University of Maryland study that H-1B visa holders make more money than U.S. technology workers. Professor Norman Matloff dissects the wage numbers used, finds what he claims are critical omissions and discusses his own studies and others which found the majority of H-1B visa holders earn 15 to 20 percent less than U.S. workers.

In response to a recently published University of Maryland study, eWEEK interviewed Norman Matloff of the University of California, Davis, to get his perspective on the findings. The Maryland findings claim H-1B visa holders typically earn more in wages in the U.S. when you control for education level and several other factors. eWEEK interviewed the Maryland study's authors, Assistant Professor Sunil Mithas and Professor Henry Lucas, last week. Why are H-1B visa holders earning more than U.S. workers? According to the Maryland study, "because of their intangible human capital, rigorous screening and selection processes and willingness to work across borders, [foreign visa workers] are likely to earn higher wages than U.S. citizen IT professionals."
Matloff believes after evaluating the Maryland research that there is a substantial conflict in the findings with other academic research. Matloff sat down with eWEEK to discuss the plethora of problems he calls "disturbing" in the Maryland study. The following is the question-and-answer interview with Matloff that occurred on May 26 over e-mail.
eWEEK: What are your main problems with the University of Maryland study that found H-1B visa holders are being paid more than U.S. citizens?
The study is fatally flawed, because its data source is grossly unrepresentative of H-1Bs in the computer field. The readership of InformationWeek [a good portion of the University of Maryland's research is based on a 2006 study done by Information Week and the staffing firm Hewlitt Associates] tends to be much more managerial and less technical, so the survey is not picking up the typical H-1B computer professionals.
The Maryland study's own data shows this. In years of experience, for instance, the H-1B respondents in the study are much more experienced than the government H-1B data show. Not surprisingly, then, the H-1B wages in the Maryland study don't match those of the government H-1B data either, with the InformationWeek respondents earning much higher wages. The median for H-1Bs in computer-related occupations was about $60,000 (in 2003, the middle of the authors' data period), according to the U.S. Citizen and Immigration Services (the former INS). This is far short of the Maryland authors' mean of $79,087. Even the 75th percentile in USCIS was only $72,815. The data do vary from year to year, but all of the years of USCIS show that the authors' claimed wage figures simply do not jibe with those of the USCIS. Again, this shows that the InformationWeek sample involves the wrong population.
In addition, as an academic I'm very disturbed at the appalling lack of even-handedness on the part of the Maryland authors. Their statements to the press do not jibe with their own findings, and their coverage of previous research is severely one-sided. I'll cite three (of many) examples:

First, the Maryland authors have failed to tell the press that they correctly recognize the basic economic principle that the H-1Bs, due to mobility limitations, will be paid less than comparable Americans. A U.S. citizen or permanent resident can move to another employer if offered more money, whereas an H-1B faces major obstacles. Mithas and Lucas do say this in their paper, noting that this is the reason why they found green card holder wages to be higher than H-1Bs. In other words, the authors agree with those of us whose research has found that H-1Bs are paid less than comparable (key word) Americans, Yet the authors' statements to the press don't mention this, nor do they even mention it in their paper's Abstract and Conclusions section.

A second example concerns controlling for relevant variables. The authors correctly state that this must be done, and indeed in one small section of their paper they do control for job location and job title, two crucial variables. They find that the claimed wage premiums for H-1Bs are greatly reduced once these variables are accounted for: Their calculated premium is 6.8 percent without factoring in these variables, but it shrinks down to 2.3 percent with these controls included. The former figure is almost three times as large as the latter, yet they've cited only the larger figure in the press interviews I've seen. Again, given the emphasis they place, correctly, on controlling for relevant variables, it is irresponsible to cite the larger number and not the smaller--but more accurate--one.

Third, a central component of academic research is proper reporting of prior research. The Maryland authors' paper is sorely lacking in this regard. They give detailed criticism of previous work that finds problems with the H-1B program, while never offering any criticism at all of the work they cite that portrays the program in a positive light. For instance, they fault some research showing wage problems for H-1Bs for not reporting the results of statistical significance tests, while not mentioning that many of the papers they highlight as supporting H-1B don't present those test results either.

Moreover, the authors do not mention at all the fact that two Congressional-commissioned employer surveys have found that H-1Bs are indeed often paid less than comparable Americans. These surveys are far more relevant than regression analysis (the tool used by the authors, myself and others), as regression can only attempt an indirect, approximate answer to the questions at hand. Employer surveys address this question directly, and thus are very important. It was unconscionable for the Maryland authors to omit discussion of these official employer surveys.

In short, this paper was not the evenhanded treatment that is expected in academia.