The impact of institutional student support on graduation rates in US Ph.D. programmes

Abstract

Using National Research Council data, we investigate the determinants of graduation rates in US Ph.D. programmes. We emphasise the impact that support and facilities offered to doctoral students have on completion rates. Significant, strong and positive effects are found for the provision of on-site graduate conferences and dedicated workspace, though there are differences across disciplines in the impact that these have on completion. Other facilities have more limited impact, though results from a quantile regression analysis suggest that some support measures – including student appraisal – may have a stronger positive impact at the bottom end of the distribution.

Keywords:

• research
• doctoral programmes
• skills
• education production function

JEL Classification:

• I23

Acknowledgements

While retaining responsibility for any errors, the authors acknowledge helpful comments from two referees.

Notes

1 Ehrenberg and Mavros (1995) link this literature by modelling the decision in each period between dropping out, remaining in the programme or completing the Ph.D. Jenkins (2008) provides a useful review of survival and competing risks models in this context.

2 Agriculture (N = 292), Biology and Health (N = 1066), Engineering (N = 701), Humanities (N = 756), Physics and Math (N = 835), and Social and Behavioural Sciences (N = 829).

3 The National Research Council considers this distinction between 6 and 8 year cut-offs for different subject areas as appropriate, given the nature of the doctoral qualification in these fields. It means that the fields are not strictly comparable – though we try to accommodate this in the sequel by using fixed effects for fields. In addition, we estimate the effects separately for the two broad types of fields by interacting our main variables with a dummy indicating these two areas. Unfortunately, this is a characteristic of the data-set about which we can do nothing beyond this.

4 Recognising the censored nature of the dependent variable, we have repeated the preferred specification of the model using Tobit methods. As is typical in applications of this sort, the results differ little from those obtained using standard least squares methods, but are less straightforward to interpret, and so we do not report the results of the Tobit analysis here.

5 The fixed effects, particularly those attached to institutions, are likely to be of interest to prospective doctoral students in that they may be regarded as a kind of performance indicator. However, due to the incidental parameter problem, we cannot estimate them consistently (see, e.g. Neyman and Scott, 1948; Lancaster, 2000).

6 Indeed, a referee has suggested that the negative sign on this coefficient may reflect the fact that the existence of annual review gives programmes the opportunity to terminate the registrations of students who are not making sufficient progress.

7 These results are available upon request from the authors.

8 The results that are obtained in broad field specific equations without these fixed effects are qualitatively similar.