Date: Wed, 11 Feb 2009 00:27:06 -0800 From: Norm Matloff To: Norm Matloff Subject: new working paper on H-1B by Kerr and Lincoln To: H-1B/L-1/offshoring e-newsletter Profs. Wm. Kerr and Wm. Lincoln of the Harvard Business School just released a study in which they claim that Chinese and Indian H-1Bs, especially the Chinese, are highly innovative, and that "shocking" (their term) our economy with a large inflow of H-1B would be greatly beneficial. The paper is already making the rounds--Kerr will speak at the Center for Immigration Studies on Feb. 13, Vivek Wadhwa lauded it in his BusinessWeek column and Thomas Friedman mentioned it in his NYT column--and is sure to be highly cited by the industry lobbyists. Sadly, people in Congress will give major credibility to this study even though it is very badly flawed. The authors' finding that the Chinese are the most innovative might strike some readers as odd. Aren't East Asians known for lack of innovation? Actually, the governments of China, Japan, South Korea and Taiwan themselves believe their people lack the ability to innovate, and have publicly wondered how to solve the problem. Nobel physicist C.N. Yang has said, "...[those] trained in the Orient tend to be too [intellectually] timid...This attitude prevents them from jumping over hurdles to make important contributions...This too timid attitude is a handicap later in life when they want to be more creative or more imaginative." I must assure you that there are indeed many East Asians that are fantastic innovators (Yang himself of course being a prime example)--but at the same time, I share the views of the above-mentioned governments and Dr. Yang; the cultures generally do not foster innovation, and in fact unwittingly inhibit it. As I will explain, the study is indeed fundamentally flawed. It is impressive looking, 50 pages in length, extensive data analysis, lots of references, and math that goes beyond the capabilities of people on Capitol Hill (i.e. logarithms). But, sadly, it makes a number of major errors. Back in December, Prof. Kerr kindly e-mailed me a draft of the study, asking for comments, and I responded with a detailed list of suggestions. Some of them involved important aspects of H-1B that he seemed to be unaware of, while others were methodological. (I am a former statistics professor, and continue to do research and consulting in the field.) But as far as I can tell, he did not heed any of my comments in the final version of his paper, though of course he does cite my work, primarily "On the Need for Reform of the H-1B Nonimmigrant Work Visa in Computer-Related Occupations," N.S. Matloff, University of Michigan Journal of Law Reform, Fall 2003, Vol. 36, Issue 4, pages 815-914. Where, then, did the authors of this study go wrong? The most glaring error is their failure to properly account for the relation of H-1B hiring to the boom/bust nature of the tech field. They do note that boom/bust nature, but don't understand the implications. Clearly, there is more patenting during boom times and less during busts. During booms, there are more jobs, in particular more R&D jobs, since R&D is a luxury for almost all firms, and thus more patents. Moreover, there is much more venture capital available during booms, and since startup firms tend to be more innovative, this makes the boom/bust variable even more important. This ties directly to the number of H-1Bs. Congress has been willing to expand the H-1B program during times of boom in the tech sector, and has declined to expand it during times of bust. Thus the authors' analysis boils down to showing that more patents are produced during booms, rather than showing a direct H-1B effect on patenting. Another flaw I pointed out to Prof. Kerr was a methodological problem called multicollinearity, in which the variables are so intercorrelated that it can change regression coefficients from positive to negative and vice versa. No need to go into the details here (the Wikipedia entry is actually a pretty good exposition of the issue if you want a basic summary), but the point is that since the author's entire analysis rests on the signs of these coefficients--positive meaning that H-1Bs have a positive effect on U.S. innovation--the authors have real statistical problems in addition to the "common related variable" flaw due to the boom/bust pattern discussed above. One aspect the authors were especially interested in was whether Americans were "crowded out" of the field--i.e. displaced--or "crowded in," meaning that the presence of the H-1Bs actually enhanced the innovation efforts of the Americans. They find that there appears to be a crowding-in effect, though they mention that it is not statistically significant, i.e. their sample size is not large enough to be very sure. What they don't mention, though, is that the confidence intervals they have for those regression coefficients suggest almost as strongly that the coefficients are negative. Also on the crowding-in/out issue, the authors correctly recognized that patents typically arise from the work of people with graduate degrees, and thus raised the question as to whether the presence of the foreign students has displaced Americans from university graduate programs. The authors cite the work of fellow Harvard professor George Borjas that "natives are crowded-out from graduate school enrollments by foreign students, especially in the most elite institutions, and suffer lower wages after graduation due to the increased labor supply." The authors say that Borjas' findings don't jibe with other work on this topic, but in fact the papers they cite don't address the Borjas issue at all. This is a disturbing logical error that is so blatant as to cast serious doubt on every other claim in the paper. Moreover, the National Science Foundation (one of the funders of the Kerr/Lincoln study) itself stated at the time Congress was considering instituting the H-1B program that the program would indeed crowd OUT the Americans. In the paper I've quoted often here, the NSF said that "A growing influx of foreign PhDs into U.S. labor markets will hold down the level of PhD salaries...[If] doctoral studies are failing to appeal to a large (or growing) percentage of the best citizen baccalaureates, then a key issue is pay...A number of [the Americans] will select alternative career paths." That of course is exactly what happened in the subsequent years; enrollment by Americans in tech grad programs has gone way down, and tech PhD salaries have not kept pace with those of comparable professions, exactly as Borjas found. So the crowding-out is quite clear and, as mentioned, actually forecast by the NSF. That invalidates Kerr's and Lincoln's regression analyses right at the outset, because they are based on data that doesn't account for the crowding-out at the graduate school level, which the authors agree is the source of most of the later patent activity. In this regard, the authors expressed surprise that the Chinese seemed to generate more patents per capita than the Indians. I explained to Prof. Kerr in December that this is partly due to the fact that proportionally more of the Chinese H-1Bs hold graduate degrees, compared to the Indians. Another key issue is that the authors do not address the counterfactual. What if natives were to hold the positions taken by H-1Bs? Would the number of patent applications filed be the same as, greater than or less than the number we now see? Again, failure to address those questions renders the authors' findings invalid. The Kerr/Lincoln study counted everyone with a Chinese or Indian surname as foreign-born, and by implication, a current or former H-1B. This ignores the huge numbers of Chinese- and Indo-Americans who were born here or immigrated as minors with their families. On the other hand, anyone with a British surname is counted as a U.S. native. This type of accounting obviously creates quite a distortion in the statistical analysis. Worse, the authors repeatedly describe U.S. natives in this manner, using terms such as "English inventors," "English patenting," "English ethnicity," and so on. I pointed out that it might be offensive to some that one is considered to be American if and only if one has British ancestry (especially given the "English" surnames of the two authors). But the authors made no change. I'm enclosing my December message to Prof. Kerr below. In which I point out the flaws I've described above, and show him how they might be remedied. But as I said earlier, the authors apparently did not use any of my suggestions. While I do appreciate their having asked for my comments, I am now baffled as to why they did so. Note carefully that Kerr's and Lincoln's work here has not been peer-reviewed. This is a major lack, and no study is worth much until it undergoes peer review, especially with regard to the changes demanded by the reviewers. These are literally "suggestions that can't be refused," unlike mine. You can download the paper at http://www.hbs.edu/research/pdf/09-005.pdf Norm From matloff Tue Dec 16 20:23:22 2008 Date: Tue, 16 Dec 2008 20:23:22 -0800 From: Norm Matloff To: "Kerr, William" <*************> Subject: comments on your H-1B paper On Tue, Dec 16, 2008 at 07:58:46PM -0500, Kerr, William wrote: > I hope that this note finds you well. Attached is a recent working > paper of mine on the H-1B visa program and US invention. In the > paper, we cite two of your articles on the abuses of the H-1B program. > Any comments that you have on our paper would be greatly appreciated. > I hope that you have a very nice holiday season! Again, thanks for sending it. As I anticipated, though, I do find that it has various serious methodological problems. I hope that my comments below will prove useful to you. (I apologize in advance if my tone below comes across as impolite, which is not my intention.) 1. Clearly, there is more patenting during boom times and less during busts; during booms, there are more jobs, in particular more R&D jobs, since R&D is a luxury for almost all firms, and thus more patents. Moreover, there is much more venture capital available during booms, and since startup firms tend to be more innovative, this makes the boom/bust variable even more important. This ties directly to the number of H-1Bs. Congress has been willing to expand the H-1B program during times of boom in the tech sector, and has declined to expand it (and even allow the cap to revert to a lower level) during times of bust. That in turn suggests that your analysis boils down to showing that more patents are produced during booms, rather than showing a direct H-1B effect. You do have your "year fixed effects" variables, which you say "absorb the main effects" from the H-1B variable, but that worsens the problem, as now you have a serious multicollinearity issue, which renders questionable not only the magnitudes of your regression coefficients but more importantly their signs. A remedy would be to use the numbers of jobs as a covariate. This would not be perfect (and you'd have to be very careful as to which jobs to count), and there are other problems (your analysis allows for a one-year lag time to get the patent filed, but the gestation of the ideas could be several years), but it would be an improvement. 2. In the tech field, most patents are filed by people with graduate degrees (Master's or PhD). Your analysis fails to disaggregate this factor, and it is quite important, as a substantial proportion of H-1Bs hold graduate degrees. (By the way, it varies quite a bit by field, much more in the life sciences than in engineering, though still substantial in the latter.) You cite George Borjas' work finding that natives are crowded out of graduate programs, but it's actually much worse than what George found. See Eric Weinstein's work which found that the NSF pushed for more foreign students in order to hold PhD salaries down, especially the point that NSF explicitly recognized that the low salaries would drive natives away from grad school. Without accounting for such effects, your finding that natives are not crowded out from patenting is invalid. One possible remedy for this flaw would be to look at numbers of foreign students earning U.S. graduate degrees in each year. Of course you'd still have the multicollinearity problems as well as others (the natives, with better English, tend more to be hired for less-R&D type positions, even with a grad degree), but this would be more accurate. By the way, this is relevant to your surmising that the Chinese may be more patent-active than the Indians. The fact is that proportionally more of the Chinese H-1Bs hold graduate degrees, compared to the Indians. The "H-1B bodyshops," which seldom employ MS or PhD workers, are primarily and Indian, not Chinese, phenomenon. 3. You do not address the counterfactual. What if natives were to hold the positions taken by H-1Bs? Would the number of patents filed be the same as, greater than or less than the number we now see? One way to assess this would to look the number of native students earning grad degrees in SE, and comparing their marginal patent activity (per capita) to that of the foreign students. This is really central, as your paper will be taken to mean that "H-1Bs are more innovative than Americans"; it's not what you said, but it will be used that way by the industry lobbyists. 4. You miss the point on the prevailing wage issue. So does Kirkegaard, whom you cite. The key point is that the legally defined prevailing wage is not the MARKET wage. The legal definition of prevailing wage is riddled with huge loopholes, and these enable employers to pay H-1Bs much less than Americans while still being fully compliant with the law. (The 95% figure Kirkegaard cites is minor, not the main wage savings.) 5. Lots of people named Gupta and Chen are natives, not immigrants, or immigrated with their families when they are children--as opposed to coming to the U.S. as a foreign student or a guest worker. This is a major problem in many papers in this field. Also, your terms "English" and "ethnic" for the natives and immigrant worker populations are inaccurate and will be, I believe, offensive to some. Norm