The impact of the H-1B cap exemption on Ph.D. labor markets

We refer here to an employment-based visa. It is also possible for foreign workers to enter the labor market by marrying a US citizen.

https://​www.​uscis.​gov/​.

Kerr (2013) provides a review of the contributions of immigrants to US innovation and entrepreneurship.

As one of the three surveys in the Scientists and Engineers Statistical Data System (SESTAT), NSCG concentrates its target population on science and engineering workers. https://​highered.​ipums.​org/​highered/​survey_​designs.​shtml#NSCG.

Ignoring individuals’ job-switching patterns is nontrivial, especially when the individual switching pattern is differential across fields of expertise and types of job. The final observed job choice could be driven by unknown factors developed over the years after graduation instead of the original immigration policy change.

The Survey of Doctorate Recipients (SDR) collects additional individual-level data on postgraduation plans, but only for a stratified subsample of science and engineering Ph.D.s.

We exclude the cohort of permanent residents from the analysis for two reasons. First, there is an endogeneity concern of visa status among foreign-born Ph.D. recipients. There may be some underlying factors affecting both their job preferences and visa status (e.g., unobserved family background). It is also possible that permanent visa status was obtained after earning their Ph.D. due to unobserved individual skills and research experience or due to a green card lottery procedure. Both of these factors can drive respondents to favor or reject an academic position. The second concern is the parallel trend assumption, and we test this empirically in Fig. 1. As seen, there is a preexisting jump (dip) around 1999 which disqualifies them as a comparison group in our case.

Working in government as a researcher could be inferred from the database; however, it is not equivalent to working in a government research institution. Note that self-employment is not permitted for foreign nationals. Hence, we concentrate on identifying the potential causal effects on academia and industry.

Due to a violation of the common trend assumption, we concentrate our analysis to US citizens and temporary residents.

We perform the equation-by-equation strategy over the course of the estimation procedure. Since there is a potential cross-equation correlation in the error terms, one may consider the alternative way of employing system of equations (SUR model) to obtain estimates with higher asymptotic efficiency. But, in our case this will give the identical results as the same regressors show up in each equation (Wooldridge 2010).

We follow the classification in the SED dataset to define broad doctoral academic fields. The list of fields is presented in Table 1.

Throughout the OPT duration, foreign Ph.D. holders can acquire postdoctoral training under the same student visa status. Notice that the postdoctoral training to which we refer here is not the short period binding with a subsequent determined employment prospect. That said, the sample of graduates we extract for the test are not individuals working on OPT just to wait for the processing of their H-1B petitions. Instead, postdoctoral participants are Ph.D. recipients spending a formal time on postgraduation training and will be on the job market again when the training ends. Though in both cases of foreign-born graduates working under the OPT program, the difference in job prospects is significant; and with the SED survey, we can differentiate one from the other. Specifically, the survey respondents will choose “Employment (other than postdoctoral fellowship, postdoctoral research associateship and traineeship)” to the question of “What best describes your (within the next year) postgraduate plans?” in the case of having a determined employment plan. Otherwise, they will choose having determined postdoctoral training plan.

Notice that the sample analyzed in the estimation of postdoctoral participation does not overlap with data used in the previous estimation (i.e., Tables 2, 3, and 4), where the outcome variables are the employment preferences.

In Appendix Figure A3, ESM, we show the trend of yearly Ph.D. graduates from Canada and Mexico and the total H-1B and TN working visa issuances for these two countries from 1997 to 2006 (data source: U.S. State Department). The total number of working visas assigned to Canadian citizen is way lower than the labor force. It reflects what Kato and Sparber (2013) suggested: “Canadians do not have to apply for TN or H-1B visas, but must instead simply meet the criteria to qualify as a TN or H-1B type of worker.” This makes Mexico possibly the more effective comparison group. Yet, the amount of Mexican Ph.D. graduates is way lower than the rest of foreign-born graduates (the treatment group). With approximately 12,000 foreign-born Ph.D. graduates per year, only 200 are from Mexico. The amount of Mexican Ph.D. graduates intending to participate in the US labor market is even lower.

Amuedo-Dorantes and Furtado (2017) documented a shift in foreign-born college graduates’ career choices due to the visa cap change in 2004 using this identification approach and found an increase in the percentage of working in academia for the US labor market participants graduating after 2004. From Fig. 1, the reduction in the visa cap in 2004 did not affect the career choices of Ph.D. students in our case because the preference of working in academia started to decrease for certain academic fields starting in 2004 (an issue we will return to in a later section). In this part, we just run robustness checks using this design by limiting the data from 1996 to 2003.

Since the employer type does not need to be answered when individuals respond not having certain employment plans, the observations of employer types for these participants are considered missing in the main estimates.

To make the magnitude of point estimates comparable with our main analysis, we list in the last row of each panel the mean outcome variable for the treatment group prior to the policy change.

Dividing the DID percentage point estimate by the likelihood of the treated group in the pre-policy period.

http://​news.​bbc.​co.​uk/​2/​hi/​business/​8558257.​stm.

Here, we only plot graphs for the academic fields that were significantly affected by AC21 (See Tables 6 and 7).

https://​www.​uscis.​gov/​sites/​default/​files/​files/​pressrelease/​H-1B_​12_​9_​04.​pdf.

Though the yearly cap became nonbinding for four consecutive years after 2001.