Joel Waldfogel's “Faculty Preferences Over Unionization": A Skeptical View

By: Rosemary BattJack FioritoJeffrey Keefe; Marlene Kim; Elaine McCrate; Jake Rosenfeld; Chris Tilly; Paula Voos
November 2016


 Download PDF version here

Joel Waldfogel's “Faculty Preferences Over Unionization": A Skeptical View

By: Rosemary Batt, CornellJack Fiorito, Florida State UniversityJeffrey Keefe, Rutgers; Marlene Kim, University of Massachusetts Boston; Elaine McCrate, University of Vermont; Jake Rosenfeld, Washington University; Chris Tilly, UCLA (communicating author, [email protected]); Paula Voos, Rutgers

Joel Waldfogel’s recent National Bureau of Economic Research working paper “Faculty preferences over unionization” (NBER #22149, April 2016) has been brought into debates about the costs and benefits of unionization for higher education faculty.  His paper purports to show a statistical link between higher scholarly productivity and greater likelihood of opposing a union, based on the analysis of the characteristics of signatories of faculty open letters for and against unionization at the Universities of Washington and Minnesota.  Some have used this paper to bolster arguments against faculty unionization.  We write as labor and employment relations scholars to explain why we think it is inappropriate to use this paper to justify a vote against unionization. (We expect that he will be further updating the paper, but here we direct our comments at the April 2016 working paper, the only one publicly available.) 

In this paper, Waldfogel uses open letters of support or opposition to unionization from over 1,000 faculty at the University of Washington and support from over 200 at Minnesota and combines these with publicly available data on salary, job titles, department affiliation, research productivity, teaching success, and political contributions from over 5,000 faculty. Based on econometric analyses, he finds that “… faculty with higher pay and greater research productivity are less supportive of unionization, after controlling for job title and department.”  Based on these correlations, he concludes that the findings raise “the possibility that universities that unionize will face difficulty attracting and retaining the most productive scholars.”  However, our review of the paper yields several reasons to question this claim. 

Waldfogel’s conclusion substantially overstates the actual empirical findings. First, the findings only hold for one of the two campuses he studied – the University of Washington but not Minnesota.  The magnitude of the effects is small (in absolute terms, and relative to other correlates of his productivity measures); and the statistical results linking productivity and unionization only measure correlation, not causation — as Waldfogel is careful to state elsewhere in the paper.  Thus, reverse causality is plausible: that is, when faculty are unhappy – perhaps about poor working conditions or unequal allocation of teaching loads that undermine research productivity -- then they are more likely both to have lower productivity and to support a union. To draw inferences about recruitment and retention of new faculty based on a few correlations is a huge leap.  Faculty decisions about accepting a job offer are affected by many other factors that the paper does not address – such as availability of research support, teaching loads, department reputation, quality of colleagues, labor market opportunities for a spouse or partner, and locational preferences.

Waldfogel’s inference also flies in the face of empirical reality.  Campuses with unionized full-time faculty include the entire CUNY and SUNY, Rutgers, and Florida state university systems, the public university systems in all the New England states, and flagship state institutions in Oregon and half a dozen other states.  Unions also represent tenure-track faculty at numerous state schools in other states, including the University of Illinois Chicago, Wayne State, and parts of the University of Wisconsin system.  Though between the National Labor Relations Board’s 1980 Yeshiva decision and the subsequent 2014 Pacific Lutheran relaxation of Yeshiva, private university professors were virtually barred from the right to collective bargaining, faculty unions also have contracts at numerous private colleges, including Adelphi, Bard, and Hofstra.  Certainly union representation has not prevented all these institutions from recruiting and retaining productive researchers.

This article also omits important parts of the broader context – what do faculty unions really do? First, no faculty contract we know of prevents universities from giving a variety of rewards to more productive faculty. Second, no faculty contract we know of compels a university to tenure incompetent or unproductive junior faculty.  Rather, union contracts require due process for evaluation, retention, and promotion decisions – reducing the likelihood that rewards are based on grudges, favoritism, or arbitrary criteria. Third, although higher average salaries are often the most visible benefit of unionization, other less visible benefits are particularly noteworthy and long lasting: transparent due process rights and collective voice mechanisms.  At a time when university administrations are under intense financial or other external pressures, unions allow faculty to have greater voice in university decisions over resource allocation -- for example, regarding their health and retirement plans, teaching loads, investments in library or classroom technology, or use of adjunct faculty.

Waldfogel’s methodology is also problematic, raising questions about his actual empirical findings.  First, it assumes that the probability of signing an open letter in support of or opposition to a union, during an active union campaign, is only determined by one’s level of support for a union — not any confounding factors that may be independently correlated with productivity — and correspondingly that those who did not sign a “pro” or “con” letter (80% or more of the faculty on both campuses) fall between those two groups in their attitudes toward unionization.  It is unlikely, however, that a public expression or lack of it is the same as a true preference, which is why secret ballot procedures are used in contested settings such as union and political elections. Publicly supporting a union may be risky for assistant professors whose tenure and promotion depends on the votes of senior faculty.  Normative pressure may lead people to voice a public position but vote differently, as exit polls in political elections have shown.  Waldfogel acknowledges the problems that his assumptions may pose, but his only check is to try omitting those who did not sign a letter on either side of the union issue. He does not, for instance, try controlling for other factors that might confound the outcomes (age, years at the university).

Second, Waldfogel’s chosen productivity variables, while often used as quick indicators of performance, are problematic as they measure quantity but not quality; and the variables have a large proportion of missing values. The criteria for promotion to tenured or full professor rarely rely on publication numbers alone, and many top research universities emphasize quality over quantity.  Waldfogel’s first measure, the number of publications reported by ISI (converted to a publication rate by dividing by the number of years at the university), does not take into account the number of authors per article nor the quality of the journal outlet. It is well known that scholars who regularly co-author articles with 2, 3, or more colleagues have a higher number of publications, with each author receiving full credit for the same co-authored article.  The issue of quality is more serious, as the highest quality and most original research often requires the longest lead times for original data collection and analysis.

Waldfogel’s second productivity measure is the number of citations reported by Google Scholar (as a cumulative total, though on a logarithmic scale that weights each additional citation a bit less).  Google Scholar citations must be viewed with caution as a reliable indicator of scholarly productivity for several reasons.  Google Scholar has an idiosyncratic approach to counting citations, for instance including citations in theses, unpublished working papers, and technical reports, but omitting citations in books that are not part of the Google Books collection. Google Scholar makes no adjustment for the quality of the output – whether it is an unpublished technical report or journal publication, or the quality of a journal outlet. Moreover, articles may receive a high number of citations not because the quality of the research is high, but because the subject matter of the article is popular or “hot” – certainly not an indicator of “productivity” or “quality”, but taste.  The Google Scholar variable also appears to simply be a raw count of number of citations, with no adjustment for the number of years publishing—a potentially misleading indicator that may just be standing in for age, experience, or point in one’s career.

A high proportion of missing observations for the productivity measures is also a concern because we cannot be certain how the results would change if all the data were present.  In this paper, actual productivity data exist for only 58 percent of the faculty along one measure (ISI citations) and only 17 percent of faculty for the second measure (Google Scholar citations). Thus, Waldfogel’s findings are based on imputing (estimating) the real value of the remaining data.  While he attempts various corrections to compensate for the missing data, the accuracy of such imputations is highly questionable when such a high proportion of the data is missing. 

Third, Waldfogel’s sample is constructed to include lecturers and instructors, as well as regular, tenure-track faculty, which casts added doubt on the findings.  Because non-tenure-track faculty are typically evaluated on teaching and not research, their research productivity will be lower, and because of their greater vulnerability they may be more likely to support the union—but interpreting this correlation as implying that lower productivity leads to greater support for the union would be misleading.  Waldfogel does control for job title, but that just assumes instructors and lecturers each differ from regular faculty by a fixed shift in probability of supporting or opposing a union, when the relationship between this variable, other independent variables (such as productivity measures), and the dependent variable is almost surely more complex. A sensible check on this would be to run regressions with and without instructors and lecturers, but Waldfogel has not done so.

In sum, there are many reasons why the findings from this article should not be inserted into a debate over how faculty should vote in a union election.

Questions/Comments contact Chris Tilly at [email protected] 


Rosemary Batt

Jack Fiorito
Florida State University

Jeffrey Keefe

Marlene Kim
University of Massachusetts Boston


Elaine McCrate
University of Vermont

Jake Rosenfeld
Washington University
Chris Tilly
Paula Voos