Is there gender discrimination in named professorships? An econometric analysis of economics departments in the US South

Abstract

This study examines the correlates of the probability that an individual academician holds a named professorship. Named professorships, like other positions within an organization, are determined by a mixture of market and non-market forces. Thus, both merit (both past and expected future productivity) and discrimination may play a role. Regression results and Blinder–Oaxaca decomposition tests presented here support a conclusion of gender discrimination in the named professorship process at American institutions of higher education. Specifically, it is found that gender discrimination results in a 7.6 percentage point disadvantage for females (relative to males) regarding the likelihood of holding a named professorship in economics.

Acknowledgements

The authors thank anonymous referees and Robert Fairlie for helpful assistance and Tom Lindley for providing useful comments on an earlier version of the paper. We also thank Chena Crocker and Zhang Wei for assistance with data collection. Any remaining errors are our own.

Notes

Metwalli and Tang (2002, pp. 133–4) report survey evidence indicating endowments to fund named professorship salaries often range from $0.5 million to $5 million, and they are typically funded by private sources. A majority of universities indicated that research or research excellence was the most important criterion for the awards. To enhance or improve chairholders’ job satisfaction, many schools offered preferential treatments. . . [such as] salary enhancement, supplementary budget[s] for research and related activities, teaching and research assistants, full-time or part-time secretarial assistance, and reduced teaching load[s] (Metwalli and Tang, 2002, p. 135).

The data collection process for this study was quite time consuming/costly. First, each university's website was visited and a search for professors’ names and titles performed. Where available, alma mater information was also recorded at this point; in other cases, it was obtained from the American Economic Association's database. Most difficulties in coding the FEMALE variable were alleviated using digital photographs from the various. edu websites. Lastly, names were matched with research records using the AEA's EconLitdatabase. The sample restriction (i.e. the ACC, Big XII and SEC schools) seemed reasonable given the resource costs involved. This restriction still resulted in a relatively large sample (i.e. n ∼ 500).

All of the empirical tests, from this point, employ the Version 1 results. A likelihood ratio test for deleting PRIVATE results in a X ² _1dfe of only 0.037. This result supports the omission of PRIVATE.

FEMALE was assigned as female=1, whereas other studies (e.g. Jackson and Lindley, etc.) assign GENDER as male=1. Therefore, the sign of the total effect here (−) is the opposite of that in other studies (+).

The significance level of the residual effect is found by performing an F-test on the joint significance of the female dummy variable (FEMALE) and its interactions with the other regressors in the model (i.e. PUBS and ALMA) using the pooled data, as in Jackson and Lindley (1989). The relevant F-statistic compares the SSE (or R²) from this unrestricted equation (k = 5) to that from a restricted version including only PUBS and ALMA on the right-hand side (k = 2).

One might also examine, as Jackson and Lindley (1989) argue, one of the subcomponents of the residual effect, the ‘constant effect’. The constant effect is the parameter estimate on FEMALE in the unrestricted regression (using the pooled data) that includes the interactions terms. In the present study, the difference between the intercept terms of the named professorship function between males and females (i.e. the constant effect) is −0.296. Although the negative sign supports gender discrimination, this effect is not significant (p-value is 0.239). More recent econometric advancements indicate that it is inappropriate to emphasize the decomposition of the residual effect. It has been noted that decomposition for subgroups of variables and the intercept term are not invariant to the scale of variables (Jones, 1983; Altonji and Blank, 1999, pp. 3158; Oaxaca and Ransom, 1999). Therefore, the Blinder decomposition of the residual term cannot be uniquely determined (Jones, 1983, p. 130).

Fairlie's (2003) study uses a pooled data set with about 40 000 observations.

See Caudill (1987, 1988), Anderson (1987) and Oskanen (1986) for a discussion of the advantages of LPM over probit or logit.