Date: Sun, 18 Dec 2011 23:01:19 -0800 From: Norm Matloff Subject: BRAIN Act, Zavodny study To: H-1B/L-1/offshoring e-newsletter As some of you know, yet another "staple a green card to their diplomas" bill is being introduced in the House, to be called the BRAIN Act. Many such bills have been introduced in the last several years, but it's my understanding that this one (or possibly a variant) will actually make it through committee, and may have enough bipartisan support for passage. As readers of this e-newsletter know (but unfortunately readers of the San Jose Mercury News/Contra Costa Times don't know, as I was misquoted in a joint article that ran today), I regard these "staple" bills as disasters, and consider employer-sponsored green cards just as harmful as the H-1B work visa.. Pat Thibodeau of Computerworld once sent a kind compliment my way by giving me credit for predicting in 1998 that university computer science enrollment would plummet a few years afterward, in part because of the H-1B work visa. It must have seemed like an odd prediction at the time, since CS enrollment was skyrocketing then, but yes, it did come true (http://blogs.computerworld.com/lets_end_the_h_1b_best_and_brightest_nonsense). Well, that encourages me to make a similar prediction now: If the BRAIN Act or its cousin does become law, I believe that future economic historians will point to this legislation as a cause of a permanent decline in American tech superiority. The tech field, already suffering from an oversupply of labor, will become so overcrowded that the Yogi Berra "That restaurant is so crowded that nobody goes there anymore" phenomenon will take hold. Research has shown that when a profession is severely overscribed, the first workers to bail out are the most talented (because they have so many other good options). We'll be left with a very mediocre tech workforce, whether domestic or foreign-origin, because wages will be suppressed to such low levels. It's already true now to a large extent. Georgetown University researcher Tony Carnevale found that engineering has the slowest growth rate of any major occupation. Carnevale told the WSJ, "If you're good at math, then you'd have to be crazy [financially] to pursue a STEM career." It ought to be clear to anyone that the foreign worker programs are a primary cause (though Carnevale has not made such a connection). As I've often mentioned, in 1989 an NSF internal report recommended bringing in lots of foreign STEM students in order to hold down PhD salaries--and noted that the resulting stagnant wage levels would drive American students away from doctoral studies. Well, that forecast proved quite accurate. And guess what!--the NSF also recommended a "staple a green card to their diplomas" policy in order to attract the large numbers of foreign students in order to make that all happen. At least we haven't had such a policy until now, but if my sources are correct, that soon will change. I believe the BRAIN Act would be one of the worst cases in history of the U.S. shooting itself in the foot. Even accounting for the corruption in Congress and the huge sums of money the tech industry spends on the Hill, putting all that aside, how could anyone in Congress even think of supporting this kind of legislation, one that would drive young Americans away from STEM? I may have more to say on BRAIN here the next couple of days, but for now I'll focus on the latest report by Madeline Zavodny, which is being presented as justification for the BRAIN Act. Zavodny's 2003 study, written while she was at the Fed, became a favorite citation for the American Immigration Lawyers Association, but it was very badly flawed (see http://heather.cs.ucdavis.edu/Archive/Fed03.txt). Her new study suffers from many of the same problems. Here are two major issues: 1. Zavodny is confusing correlation with causation. Zavodny claims to show that when the number of immigrants (she includes technically nonimmigrant workers such as H-1B in this) in the labor force rises, the number of employed U.S. natives rises by a large amount. She concludes that immigration produces a net job gain for natives. This claim, of course, is similar to that of a study by NFAP, whose fallacy was cogently exposed by Carl Bialik, the excellent Wall Street Journal columnist ("The Numbers Guy"). I reviewed Bialik's criticism of the NFAP claim in http://heather.cs.ucdavis.edu/Archive/WSJOnNFAPClaim.txt where I pointed out an important aspect that even the astute Bialik missed. I'll explain this shortly. (I wish to mention first that I usually don't use the term "native," preferring to say "U.S. residents and permanent residents." This includes nonnatives who are now naturalized citizens, and people who immigrated here under family immigration policies etc. I use the term "American" for short, though the greencard holders are only potential Americans at that point. Unfortunately, those who write in favor of H-1B and employer-spoonsored green cards love the term "native," with the implication that those of us who disagree must be "nativist." But since Zavodny's study separates out natives, I'm forced to work in those terms here.) Did you know that I invented a wonderful new kind of chocolate cake? Well, actually I haven't, but let's indulge in a bit of fiction for a moment, and then relate it to fact. So, I invent this new kind of cake, super tasty, low in fat and even safe for diabetics. Now I've got to mass produce it. I'll need to hire some food science graduates to make the production line work properly, so as to not lose the essence of that great cake. But I'm cheap, and don't want to pay high salaries. And on top of that, the HR people I hire don't want to spend much effort in finding people to hire. So we hire foreign students who just graduated from the food science program at the local university. Suppose things work out splendidly. My business grows, with bakeries in every major American city. I hire marketing people, mechanical engineers to automate part of the production and packaging process, operations research people for optimizing product distribution, accountants, and of course lawyers. Based on the positive result I got from hiring the foreign student food science grads--competent work at an inexpensive wage--I'll probably hire those engineers and OR specialists from the foreign student pool too. And of course some of those lawyers will handle immigration papers for the foreign students I hire. The marketers, accountants and lawyers will probably be U.S. natives or U.S. permanent residents, and so will a number of others who work for me. Zavodny, Stuart Anderson (NFAP) and so on consider the above scenario to be one in which immigrants create jobs. I hired immigrants, and voila!, a year or two later, I was hiring natives too. Ergo, the immigrants supposedly "created" jobs. But the flaw is clear: Those foreign workers whom I hired didn't CAUSE my business to expand and hire some natives; I could have hired U.S. natives with food science degrees, and gotten the same results. And that is exactly what happens in the tech industry. They hire foreign students from U.S. campuses because (a) they are cheaper (they are young, thus cheap, and are willing to work for less, in exchange for sponsorship for a green card), and (b) it's convenient--just make a recruiting trip to a couple of local universities, which will have many foreign students to choose from. Of course, if in my cake story there were a shortage of qualified American food science professionals, then the "immigrants create jobs" argument would have some validity. But we don't have a shortage of qualified Americans in the tech industry, as I've shown before (wages rising only 3% per year, etc.). I could have hired Americans for my cake business in the story above. So, NO, the immigrant workers in the tech area are NOT creating jobs. So, even if Zavodny's regression models were strong (which I'll show below they are not), her conclusion that the immigrants "caused" more jobs to be created for natives would still be invalid. 2. Zavodny's regression models are fragile at best. In its basic form, a regression model estimates a presumed linear relationship between the mean of a response variable Y and predictor variables X1, X2 and so on. Here the word "relationship" means that the mean of Y for fixed values of X1, X2 and so on is a linear function: mean Y = c0 + c1 X1 + c2 X2 + ... for some constants c0, c1, c2 and so on to be estimated by the sample data. For instance, we might analyze human mean weight as a linear function of height and age. (Note the word "mean," often overlooked.) One might also include interaction terms in the equation, say X1 times X2 above. In the weight/height/age example, for instance, it may be that the increase in mean weight associated with a one-inch increase in height is different for older people than for younger people, thus an "interaction" between height and age. In Zavodny's regression models, Y is (the logarithm of) the native employment rate; X1 is the log of the percentage of jobs held by immigrants; and X2, X3 etc. are variables for the 50 U.S. states, and for Time, to account for different years and thus different levels of economic activity. Now, let's look at typical examples from Zavodny's paper (http://www.renewoureconomy.org/sites/all/themes/pnae/img/NAE_Im-AmerJobs.pdf). On the one hand, she writes the attention-grabbing statments like An additional 100 immigrants with advanced degrees in STEM fields from either U.S. or foreign universities is associated with an additional eighty-six jobs among US natives. On the other hand, her actual regression coefficients are all tiny, e.g. ...[the] results for immigrants with a bachelor's degree or higher indicated that a 10 percent increase in their share of the total workforce is associated with a 0.03 percent increase in the overall native employment rate during 2000-2007..." Note the figure 0.03 percent--meaning 0.0003. This and all her other regression coefficients are really minuscule. So, how can she get such large job creation numbers from such minuscule regression coefficients? The answer is that they in effect multiply the total overall number of jobs. Take a very small number and multiply by a number in the many millions, and you are able to get numbers like 86 above. The problem with this is that it makes the analyses exceedingly fragile to errors in her model. Needless to say, for example, no relation in economics is exactly linear. Linear models can be good, useful approximations, but they are far from exact. The errors in those models could easily be much larger than Zavodny's tiny regression coefficients, for instance making what actually is a negative relationship look to be positive. Zavodny's lack of interaction terms brings similar potential errors and potential changes of algebraic sign (positive to negative and vice versa). Though the statistical methodology can be a bit sophisticated, the problems I'm describing here are just common sense: Any time one multiplies a huge number by a tiny one, the result is of questionable value. And again, this is especially true if one is trying to determine whether the relationship of two variables is positive or negative. Points 1 and 2 above are the main issues, especially Point 1. There are various other issues, but I'll leave it at thse two. Norm