A Paper on H-1B of Great Importance But Likely to Be Ignored

A couple of people have brought to my attention a new research paper on H-1B by Kirk Doran (Notre Dame), Alexander Gelber (UCB) and Adam Isen (U.S. Treasury). Its findings are certainly provocative, as they directly contradict most (or all) of the recent major work that is positive on H-1B.

Though some might expect me to automatically endorse every paper with findings on H-1B similar to mine, on the contrary, I tend to be rather critical of such work. For example, Georgetown’s Tony Carnevale found that a sizable number of Computer Science graduates are not finding work in their field. This would seem to starkly contradict the industry PR line that there is a dire shortage of CS grads, but I’ve pointed out that (a) a large number of CS grads are simply not sharp enough to do software development and (b) many aren’t suitable for the CS “talking jobs” (say, customer support) either, as they are not the outgoing articulate types that are needed.

In the case of this paper, though, it is one of the most careful, insightful works I’ve seen on either side of the H-1B field. No, not perfect by any means, as I will discuss below, but a fine job at the end of the day.

Contrary to the results — real or misinterpreted — of previous work on whether H-1B increases patenting, the authors here do not find such an effect. Similarly, they find that H-1B does NOT increase employment, and in fact the arrival of new H-1Bs causes some displacement of existing workers, both American and foreign — quite contrary to the obligatory statement in every pro-industry press release, editorial and research paper that “each H-1B creates 2.62 new jobs.”

But mark my words — you won’t be hearing much about this paper. I doubt that it will be highlighted in the press, or even mentioned as counterpoint in articles that bring up the claimed 2.62 figure. The problem is that these authors don’t have the massive PR machine that the pro-H-1B researchers have, in which journalists, editorial boards and above all, offices on Capitol Hill are constantly bombarded with “educational” material featuring H-1B poster boys (rarely girls), tales of woe that “Johnny Can’t Do Math,” and of course the ubiquitous 2.62 figure. I make this point here specifically for the activist programmers and engineers hoping for reduction of H-1B, who often don’t realize that, sadly, facts are not enough in politics.

Well, what does the paper do specifically, and what are its weaknesses as I see them?  First, what are its main findings?

  • “Winning additional H-1B visas has an insignificant effect on patenting within eight years…”
  • “H-1Bs substantially crowd out employment of other workers.”
  • “We find some evidence that additional H-1Bs lead to lower average employee wages while raising firm profits….and rules out the scenario in which H-1Bs replace natives one-for-one.”

One disappointment I have with the paper is that it doesn’t sufficiently treat the question of the variability of the quality of the H-1Bs, and of corresponding variability of the firms that hire them. The authors do perform some separate analyses in which they omit the large IT services firms, but they do so only because such firms don’t do much patenting. That is true, but it ignores the point that the quality of the H-1Bs hired by those firms is low. Why didn’t the authors do separate analyses in which they kept only the mainstream firms that are most vocal in pushing Congress to expand H-1B, such as Google, Facebook, Intel, IBM, Oracle, Cisco and Texas Instruments? Conceivably they might find a salutatory effect of H-1B on patenting in those firms. For that matter, assuming the authors’ data separated visa counts for the main 65,000 cap and the 20,000 supplementary cap for foreign MS/PhD grads of U.S. universities, they should have done separate analyses here too. This is the industry lobbyists’ favorite group, so it would have been useful to see if H-1B had a postive effect on patenting in those firms.

There is also no recognition in the paper about the role of the OPT extension to the F-1 student visa, which serves as a buffer to a “shortage” of H-1B visas. If there were a positive effect of H-1B/foreign workers on patenting, it might be missed if the OPTs are being granted a lot of patents. The time period for OPTs was extended during the George W. Bush administration, and now the Obama people want to lengthen it even further. So, this is a very important factor policywise.

Mind you, my own position is that the H-1Bs who were formerly foreign students are actually WEAKER than their American peers, thus causing a net LOSS in patenting for the nation, due to the displacement effects (the latter of which, I hope you recall, even Giovanni’s research has shown). This was the subject of my EPI paper, and of work by John Bound, who ironically is prominently acknowledged by the authors as being helpful to them. So, I believe H-1B should be curtailed, not expanded, and I feel the same about OPT. But though confirming this with the data of the present paper may have been difficult, if they had at least found that H-1Bs don’t increase patenting even at firms like Google, this would have been very significant.

I was also disappointed at the authors’ weak understanding of H-1B wage law and the manner in which H-1B displacement of Americans works. The authors remark that the wage laws “may not be effective,” but they seem to have no inkling of the central role of age in the use of H-1Bs as cheap labor, in which young H-1Bs are hired in lieu of older (35+) Americans, in full compliance with the law. And while the authors seem surprised that new H-1Bs often displace earlier-hired ones, it is probably due to the same problem; those earlier H-1Bs may have gotten enough raises in the last couple of years to make them likely targets for dismissal too.

The phrase “hired in lieu of” above is quite different from the “replace Americans by H-1Bs one-for-one” model the authors seem to have in mind. This model of course likely stems from the huge publicity given to cases like SCE, in which that phenomenon does indeed occur, but the much more common use of the H-1B program is at the hiring stage. This is something the authors should have pursued, though again it’s not clear that their data would have been sufficient for this kind of analysis.

The authors use the word causal, something I’ve taken my colleague Giovanni Peri to task for. The authors seem to believe that this is justified by the “random treatment assignment” nature of the data, in which visas were awarded by USCIS by random lottery, but I still consider use of the word to be reckless.

Consider for instance a point I’ve made before on Giovanni’s “lottery” analysis. (By the way, the authors of the present paper don’t consider his paper to be truly based on randomization.) He found that the cities that “won” the H-1B lottery saw the wages of Americans in IT rise. But consider a scenario in which a firm fires most of its American IT workers except for the managers, and hires a number of H-1Bs to replace them. (Again, it does occur sometimes, as we’ve seen.) Then, since the managers tend to make more money than the programmers, the wages of Americans at that firm — those Americans who REMAIN — does rise, even though none of the retained Americans got a nickel in raises. In such a scenario, the claim that the H-1Bs “caused” the American salaries to rise might be technically correct, but certainly not in the sense people assume upon hearing the claim.

Bottom line, though, this paper is generally done extremely carefully and its results are highly remarkable. It flatly contradicts the work by Peri and Zavodny (“2.62”!), and for that matter, contradicts the inaccurate interpretation by the industry of Bill Kerr’s paper on patenting. Too bad no one will notice.

12 thoughts on “A Paper on H-1B of Great Importance But Likely to Be Ignored

  1. I am mystified by their purported methodology. So, Microsoft gets an “unexpected” extra H-1B, so they look at Microsoft’s annual profits to determine what “causal” effect that extra hire had? A company with 128,000 employees of whom several thousand (?) are also already H-1Bs? I am not cheered when their results even in part agree with my opinions.


    • Josh, the paper’s analysis does not literally look at the impact of adding one new H-1B. What they measure is a rate, a derivative. And actually, they did a separate analysis of small companies, with similar results.


      • A rate of what against what? I saw that they at least charted against an axis of the size of the companies and the lines seemed smooth BUT even the small companies may have been too large for this to work. At least they have not assuaged my concerns. Do you know otherwise?

        At best they have a sub-critical phenomenon they hope to squeeze out of a second order analysis – or whatever statistical terminology you would prefer.

        You can throw the challenge back at me – what methodology would work better, at the external, statistical level? That may be especially tough because there may be effectively no training set, no cases where H-1Bs are absent to compare against. So as I keep ranting about in different ways, a micro- rather than macro- study would be much more credible, go in and study individual cases in depth. As you properly do when you remind people of Microsoft’s low acceptance rates, etc.

        The causal relationships are never evident from just the numbers and have to be deduced in other manners. What offends everyone the most is when causality is claimed like the 2.66 number, when there is nothing at all in the methodology to establish it and COMMON SENSE argues strongly for other interpretations.


        • In any regression analysis, there is always the question of the range in which the model (typically linear) is valid. Your concern essentially is that one.

          A related concern, which they also bring up, is whether one can extrapolate from the firms’ behavior at the last minute of the H-1B season. They try to address this in various ways; see p.22.

          So there are many possible issues that might be brought up in addition to the ones I did in my posting. There are various others than I had but did not want mention, as my post was already rather long. This is the nature of economics. This is why, for instance, there was such a lively discussion at the econ seminar I attended on Tuesday and reported on here.

          I’ve mentioned many times that any statistical analysis should be accompanied by a qualitative analysis, bringing what we know and seeing whether it jibes with the numbers. That has been my argument against the Peri and Zavodny studies — since we don’t have a labor shortage (the flat wages pretty much rule that out) and since the average quality of the H-1Bs is lower than that of the Americans (by multiple direct measures, not just regression), the H-1Bs should have no magical employment-increasing powers. The burden of proof is on Peri and Zavodny to refute that. The only possible refutation I can see is that since the workers are cheaper, employers can run more projects.

          Given the sample size, a positive relation between H-1Bs and employment and patenting should have been discovered. The fact that it wasn’t says a lot, so this already makes the paper very valuable. Second, the fact that the paper did find displacement effects is important, even if there is a range issue.


          • > since the workers are cheaper, employers can run more projects.

            Even that would not be probative, since many projects fail entirely, and even a successful project staffed entirely with H-1Bs at low wages might have been done by fewer Americans at higher wages for a much lower total expenditure.

            This is certainly the feeling of many of us out in the world, that companies waste money on H-1Bs, take far more time, more iterations, incur huge opportunity costs not to mention complete failures of 8, 9, and 10 digit budget projects, because the ONLY staff they consider is that of the lowest cost, irrespective of quality.


          • The paper under discussion found no increase in employment, so this scenario doesn’t apply here.

            Your point is valid, but what you are really saying is that though the paper found no positive effects from H-1B, it should have gone further and picked up negative ones. I agree, of course, and made related statements in my original post.


  2. > The problem is that these authors don’t have the massive PR machine that the pro-H-1B researchers have, in which journalists, editorial boards and above all, offices on Capitol Hill are constantly bombarded with “educational” material featuring H-1B poster boys (rarely girls), tales of woe that “Johnny Can’t Do Math,” and of course the ubiquitous 2.62 figure.

    Agreed. Regarding the 2.62 figure, that number comes from a regression of data between 2000 and 2007. If you do the exact same regression but shift the data forward two years, between 2002 and 2009, you get a LOSS of 1.21 jobs. I believe that the reason for the difference is that the latter range avoids the steep job losses suffered by both native and H-1B workers after the tech crash. During this period, the change in jobs was heavily correlated but it was a correlation in the loss of jobs caused by the tech crash, not the job-creation story pushed by pro-H-1B researchers. I’ve posted all of the numbers at http://econdataus.com/amerjobs.htm .

    There has been no response from Zavodny regarding this fact and I don’t expect one. As Norm says, they have the massive PR machine and feel no need to defend their numbers. There also seem to be a large number of people who are more than willing to parrot any number that is given to them. I’ve posted references to the 2.62 number at http://econdataus.com/claim262.htm . As can be seen, recent “parrots” include Senators Orrin Hatch and Jeff Flake. A White House paper and posting referenced the 2.62 number in mid-2013 and mid-2014. That is the genius (some might say evil genius) of the Zavodny paper. It gives several simple numbers that anyone can mindlessly repeat.


    • I’m not a statistics type as I prefer simple arguments that simple people will be able to comprehend and act on, but I do want to point out something.

      They do have a massive PR campaign with millions of bucks spent.

      But, I have been targeting every one of their campaigns that I run across and I’m noticing more and more from Washington, DC looking at those articles.

      Granted, it may be lobbyists looking for ways to use my arguments against me, but it is important that we do not give up.

      There are 3.8 million computer and math people working in the occupational code 15-0000 (Google BLS OES) as of 2013.

      Figure 50% are now temporary workers on temporary visas, but that still leaves nearly 2 million Americans.

      If we would each chip in 20 bucks, we could out advertise them.
      If we would each right a letter to the editors of our papers (not the bought and paid for shills, but the back home, small town papers) showing them the data from BLS CES showing where we have only created 375,000 jobs since 2007, I believe it would have a ripple effect across America.

      But we have to do it.
      We can’t just talk about it


      • > I’m not a statistics type as I prefer simple arguments that simple people will be able to comprehend and act on, but I do want to point out something.

        In fact, that is why I posted the simplest argument against claims of 2.62 and 1.83 jobs created by each foreign-born worker at http://econdataus.com/amerjobs.htm . For the 2.62 number, that simple argument is that the same formula that suggests that 2.62 jobs were created for 2000-2007, suggests that -1.21 jobs were LOST from 2002-2009. For the 1.83 number, that simple argument is that the same formula that suggests that 1.83 jobs were created for 2000-2010, suggests that 0.69 jobs were LOST for 2003-2008. In short, the number depends heavily on the dates selected (or some would say, cherry-picked).

        I’ve been involved in political debates for many years and I believe that you need to approach the debate on all levels. It is important that we have experts like Hira, Miano, Salzman, Borjas, Norm, and many others who can do serious studies on this issue. As you say, however, we also need to debate this issue on all of the levels down to the casual listener. For example, even your figure of 375,000 jobs having been created since 2007 is far too complex for most people to conclude anything from. They don’t know how many STEM students are turned out from U.S. universities. Even if they do know, they may be misled by arguments from the other side that some of these STEM degrees are for medieval studies and so on. Of course, those people who are hitting your web site, especially those who are hitting more than one page, may be picking up this information, especially if you are sourcing your numbers well as I believe you are. Still, this is just addressing one more level of the debate.

        Regarding the 2.62 and 1.83 numbers, I fear that we have basically surrendered this part of the debate to the other side. Unless a statistician of some reknown can check the calculations (I’ve uploaded the code at https://github.com/econdataus/amjobs ) and is able to get one or more serious publications or politicians to take note, I’m sure that the other side will keep using those numbers. When the casual listener hears the claim that each foreign worker creates 2.62 jobs, they typically think that there must be something positive there. Even if they understand that correlation does not equal causation and that these estimations may have errors, they tend to think that it represents so much smoke, there must be some fire there. The fact that moving the time span forward two years shows a LOSS in jobs shows that it is all smoke. So we do need to make all of the positive arguments that we can but we also need to debunk those arguments that can easily be shown to be wrong and misleading.

        I do agree that it would help if all STEM workers, at least those who are aware and/or been affected of these issues, contribute to their solution, either in contributions or in letters to the editor. It does seem that there are many responses to articles online. We need to engage the debate on all levels.


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s