Date: Tue, 17 Apr 2007 00:15:15 -0700 From: Norm Matloff To: Norm Matloff To: H-1B/L-1/offshoring e-newsletter I have long warned that analyzing H-1B wage data is a treacherous business. Indeed, I devote an entire section in my university law journal article on the methodological issues which arise in studying H-1B wages. (Section V.B, Methodological Issues, "On the Need for Reform of the H-1B Nonimmigrant Work Visa in Computer-Related Occupations", University of Michigan Journal of Law Reform, Fall 2003, Vol. 36, Issue 4, 815-914.) For this reason, I often find myself being sharply critical of a study even if its findings support my point of view. Such is the case for a new study by longtime immigration analyst B. Lindsay Lowell and Johanna Schneider (Avato), currently with the Institute for the Study of International Migration, Georgetown University. The paper, "The Wages of Skilled Temporary Migrants: Effects of Job Portability and Student Status," was presented last month at the annual meeting of the Population Association of America. The paper's findings are roughly in the same direction as mine, i.e. they conclude in certain senses that H-1B workers make less than U.S. citizens and permanent residents. Here is a brief summary from the April 13 issue of Science magazine: # ...science, technology, engineering, and mathematics (STEM) workers # holding an H-1B--a temporary visa granted to skilled foreign # workers--earn 5% less than natives employed in similar positions with # similar skills and experience earn. It also shows that H-1B visa # holders who don't job-hop make 11% less than natives and that those # who enter the workforce after graduating from a U.S. university earn # 16% less. # There is one group of foreigners who do not seem handicapped by their # H-1B visa status, however: Those hired directly from overseas--45% # of the total--make 14% more than native workers. Sadly, none of these conclusions can be taken as valid. The paper: (a) has severe methodological flaws, (b) is factually incorrect on H-1B regulations, and (c) fails to properly use the existing literature, notably the very references cited by the paper. Bottom line: I regret to say that this paper is of very little value and in fact worsens the confusion on the H-1B topic. Following is my critique of the paper. I should mention that I am basing it on the March 2007 version, not the September 2006 extended abstract available on the PAA Web site, http://paa2007.princeton.edu/sessionViewer.aspx?sessionId=706 The paper relies solely on the use of regression analysis. This is of course a standard statistical tool. I use it in some of my own research on H-1B, and for that matter, have done research on regression analysis itself. (I am a former statistics professor.) But regression analysis can be very, very dangerous. An excellent example (please bear with me on this, as it is of great importance) is given in Introduction to Linear Models and Statistical Inference, by S. Janke and F. Tinsley, Wiley, 2005, p.396. Suppose one is doing a study of factors underlying crime rates. Specifically, Janke and Tinsley analyze real data from 1986 to 2000 in El Paso County, CO, with the following variables: Robberies: number of robberies (per 100,000 people) Income: average income in the county Year: calendar (to guage a time trend separate from income) Sheep: number of big horn sheep shot (per hunter) That last variable, Sheep, is obviously irrelevant, but Janke and Tinsley threw it in to prove their point about how misleading regression analysis can be if it is not tempered by domain knowledge. Well, they ran the numbers, and the regression model "showed" that the Sheep variable (and the other predictors) was "highly significant"! In other words, it would appear that Sheep is a big factor driving Robberies. Why did the statistical analysis produce such a nonsensical conclusion? The problem is that Sheep was highly correlated with Year. (Apparently the hunters used more lethal equipment over time.) So the "significant" characterization of the Sheep variable was actually a reflection of the fact that both Sheep and Robberies were correlated with Year, i.e. there was a time trend in both. There are several key lessons here: 1. One should not simply run a regression analysis and present the results as if they are "truth." 2. Interactions between predictor variables can be of the utmost importance, and if ignored, can lead to highly misleading findings. 3. Omissions of important predictors can complete distort the results. 4. Quantitative analyses without qualitative understanding of the phenomena being studied can lead to highly misleading findings. Similar problems can result from the *omission* of relevant predictors. If for instance Year had not been included in the above example, then the Sheep variable would have been EVEN MORE significant. The H-1B paper by Lowell and Avato has very serious problems in this regard. For example, they do not include two very important predictor variables, Industry Sector and Region. Foreign-student PhDs in science and engineering (S&E) tend to disproportionately take jobs in industry rather than in academia (David S. North, Soothing the Establishment: The Impact of Foreign-Born Scientists and Engineers on America, University Press of America, 1995, p.64). Without the Industry Sector predictor in the regression analysis, the higher salary effect due to Industry Sector would misleadingly show up in the predictor Foreign Student, falsely making it look like the H-1Bs get paid MORE than Americans. Or, if the regression still says that H-1Bs get paid less, the reported gap would be smaller than what it actually is. The same is true for Region. H-1Bs, as with other foreign-born people, tend to cluster in the large urban areas, where wages are higher. By omitting Region, it could falsely make it look like H-1Bs are paid more than Americans, or again underestimate the amount that they are paid less than Americans. I mentioned the importance of the Industry Sector and Region predictors, and the distortionary effects of omitting them, in my law journal article. Lowell and Avato cite my article, but apparently missed this information. They apparently did not look closely at some of the other sources they cited. A notable example is their citing of the 2003 GAO report. Lowell and Avato say that the GAO did an employer survey in which they found that younger H-1Bs made more than younger Americans while older H-1Bs made less than older Americans. Here Lowell and Avato mixed two different aspects of the GAO study. The GAO did do an employer survey, but the age-related findings did NOT come from the survey. Instead, those findings came from statistical analysis of H-1B approvals and the Current Population Survey. Not only did Lowell and Avato incorrectly cite the GAO report in this regard, but also they failed to make the important point that what the GAO did find in its employer survey was that many of the employers admitted to hiring H-1Bs because the employers could pay them less than Americans. (As to the GAO's statistical analysis, it also had huge methodological flaws; see http://heather.cs.ucdavis.edu/Archive/GAO03.txt) It's regrettable that Lowell and Avato did not see the GAO's statement that many of the employers they surveyed admitted to using the H-1Bs as cheap labor. (The statement is in the executive summary, thus hard to miss.) The exact statement is the one I've often quoted: "Some employers said that they hired H-1B workers in part because these workers would often accept lower salaries than similarly qualified U.S. workers; however, these employers said they never paid H-1B workers less than the required wage." Not only is important that some employers admitted to this (imagine how many do it but didn't admit it), but what is crucial is the last part of GAO's statement--"however, these employers said they never paid H-1B workers less than the required wage." This is key, as it shows that it is fully legal to underpay H-1Bs, due to the gaping loopholes. As I said, I regard this as a key point, and thus I consider it a major flaw in the Lowell and Avato paper that they don't understand this. They assume that the underpayment of H-1Bs comes from violations on the law. This in turn causes them to misunderstand the significance of the Kirkegaard paper that they reference. Lowell and Avato included two indicator variables for nationality, one for Asian origin and the other for European origin. Let's call these predictors Asian and European. The authors found a larger positive regression coefficient for the Asian variable than for the European one. In other words, they found that Asian H-1Bs make more than European ones. This is in contrast to the data I obtained from the INS (for the computer-related H-1Bs only), which showed exactly the opposite trend. For example, the median wages for H-1Bs from the UK, France and Germany were $74,000, $64,000 and $70,000, while for China, India and the Philippines they were $55,000, $52,000 and $52,000. So, why did they and I have different results? Note carefully that their analysis is for all S&E H-1Bs, while mine is only for the computer-related ones. If, as is plausible, the European H-1Bs tend to be less focused on the computer occupations (which are higher paying than, say biology) than their Asian counterparts, this would lead to a larger positive regression coefficient for Asian than European. But that would not mean that Asians make more than Europeans IN THE SAME JOBS. Once one holds occupation constant, as I said, one would then see that it is the Europeans who make more, not the Asians. But Lowell and Avato did not do that. They did not have variables indicating occupation, which clearly are of high importance, as seen above. I found cite numerous instances in the paper which suffer from the same problems: Omission of important variables, lack of interaction terms for the variables they do have, and so on. To keep this brief, I will not go into those, but just touch on a couple of other points. Lowell and Avato note that well-known point that H-1Bs from Third World countries have a lower "reservation wage," i.e. they are willing to work for less, and explain this by saying that even lower wages in the U.S. as far above what the H-1Bs would make back home. True, but it's only part of the story. If the H-1Bs are being sponsored for a green card, that is a nonmonetary form of compensation which most H-1Bs value very highly, thus contributing to a lower reservation wage. The authors make various errors on H-1B regulations, and this too probably caused problems with their analyses. For example, they state that the Optional Practical Training portion of the F-1 student visa can be renewed automatically until the student gets an H-1B visa. I'm fairly sure this is not correct. All in all, a very disappointing paper. Norm