NEVER GIVE UP ON THE GOOD TIMES: STUDENT ATTRITION IN THE UK*
Geraint Johnes1and Robert McNabb2
ABSTRACT
The determinants of students’ propensity to drop out of university is analysed using individual records of all students passing through the central applications process in 1993. The data set comprises about 100000 individuals and allows a much more thorough analysis of student wastage than has been possible heretofore. The main reasons for attrition, academic failure and ‘voluntary’ dropout, are modelled both separately and simultaneously. The results highlight, inter alia, the importance of matching and peer group effects.
JEL Classification: J24, I2
Keywords: Attrition, Higher Education
The authors are indebted to Richard Jones for excellent research assistance.
1 Department of Economics, The Management School, Lancaster University, Lancaster LA1 4YX UK, Voice: +44 1524 594215, FAX: +44 1524 594244, E-mail: .
2 Cardiff Business School, Cardiff University, Cardiff CF1 3EU UK, Voice:+44 1222 875210, FAX: +44 1222 874419, E-mail:
NEVER GIVE UP ON THE GOOD TIMES: STUDENT ATTRITION IN THE UK
- Introduction
The expansion of higher education in the UK over the last decade has been accompanied by an increase in the rate at which students drop out of university without completing the courses for which they were initially registered. Consistent statistical evidence is difficult to come by, but it would appear that the dropout rate has risen from about 13-16 per cent in the late 1970s and early 1980s to 19-25 per cent in the late 1990s.[1] In some respects this is unsurprising. If university places are filled by students in order of the (probability weighted) quality of the match between student and course, then an increase in attrition is likely to result from an expansion in the number of student places available, other things being equal. This might be mitigated in some measure by rising educational standards, both at higher education level and below.
The need to understand the determinants of student non-completion arises, first, because attrition rates are now an important component of the range of indicators being applied to measure university performance. In part, this reflects the fact that student attrition represents an inefficient use of resources if, having dropped out, students do not use whatever human capital is gained during their courses to advantage in the labour market. Understanding the determinants of why certain groups are more likely to drop out than others is also of some importance in underpinning government policy aimed at ensuring high completion rates.
Despite the significance of non-completion for both government policy and economic efficiency, very little work has been undertaken in the UK on student wastage. This is mainly because until now no national data have been available which allow empirical work to be conducted at the appropriate level of aggregation – that of the individual student. In this paper, we take advantage of a recently available data set that provides a considerable amount of information on all students entering the 'old' (ie pre-1992 universities) through the centralised admissions process. Hence, we are able to draw much firmer conclusions about the determinants of student attrition in the UK than have hitherto been possible, and in addition we are able to test a number of new and exciting hypotheses about matching and peer effects.
The remainder of the paper is structured as follows. Section 2 provides an overview of the current literature that brings to the fore the main explanatory variables that should be used in a rigorous study of student attrition. Some conceptual issues are discussed in section 3, where hypotheses of particular interest, suitable for testing with the data which are available, are developed. Section 4 concerns the assembly of the data set, while the results of the statistical analyses are reported and evaluated in section 5. Finally, section 6 draws together the main findings and makes suggestions for future research.
2. Previous Research
Although there is a lack of statistical analysis of student attrition in the UK, it is an issue that has attracted considerable attention elsewhere, most notably in the US where college dropout rates are very high. Tinto (1993), for example, observes that ‘of the nearly 2.4 million students who in 1993 entered higher education for the first time, over 1.5 million will leave their first institution without receiving a degree (and, of these) approximately 1.1 million will leave higher education altogether’.
The extensive US literature on college completion has, however, identified an extensive assortment of economic, sociological and psychological correlates of non-completion. Amongst the early studies to consider the determinants of attrition, Bayer (1968)finds that academic aptitude and family structure variables are key determinants of students’ propensities to drop out of US higher education institutions. Similar results have been obtained by Astin (1975). De Rome and Lewin (1984) also find that ability is important, with more able students less likely to drop out. In addition, De Rome and Lewin's results highlight a number of other correlates of the likelihood of non-completion. In particular, they suggest that full-time students have lower non-completion rates than part-time students, that student’s commitment to the course enhances the likelihood of graduation as does whether or not the university was the student’s preferred choice of institution. Completion rates also vary by subject studied and the student’s job prospects in the event of withdrawal. More recently, Okun et al. (1996) confirmed the importance of the same correlates. Psychological and sociological characteristics of students are given particular emphasis in the work of Eaton and Bean (1995).
A number of studies have shown that students with non-traditional academic backgrounds are found to have higher attrition propensities than those with more traditional attributes and that attrition rates for the two groups are determined by different sets of factors. For example, Bean and Metzner (1985) find that mature and part-time students have higher attrition rates and these are typically linked to problems associated with finance, family or work commitments. Attrition amongst students with more conventional academic backgrounds are more likely to reflect problems of social integration. Recent work by Robst et al. (1998) distinguish students by gender and focused in particular on the gender composition of faculty as a determinant of students’ attrition rates. This work suggests that retention rates are higher for female students on courses taught by faculty a high proportion of whom are themselves female, even after controlling for such variables as subject mix. The same study also examines the impact of ethnicity though this is not found to have a significant effect on the likelihood of non-completion. One exception is that black women are significantly less likely than other students to drop out.
Johnson (1994) draws a distinction between voluntary and involuntary dropouts. While both groups tend to be of lower ability than those who persist, those who withdraw voluntarily tend to be older than the norm, and those who are forced to withdraw are younger than the average persisting student. Later in the present paper we shall disaggregate students who do not complete their degree courses in order to investigate this issue further in the British context.
One of the most wide-ranging studies of student attrition is that by Tinto (1993). This study identifies eight causes of student wastage: intention (occupational and other aspirations), commitment (to the goal of completion and to the institution), adjustment (to student life), difficulty (an aspect studied also by Robst et al., 1998), congruence (erroneous expectations resulting in mistakes being made in choosing where or what to study), isolation (failure to adapt to college life), obligations (to family, for instance), and finances.[2] While this is an impressive sociological contribution, empirical verification of Tinto’s arguments is hindered by the difficulty of obtaining quantifiable measures of some of the personal characteristics which are key to his model. Nevertheless, this work highlights the importance of the quality of match between student and institution in determining whether a student drops out or persists and that the quality of the match is determined by the qualities of both the institution and the students themselves. The concept of match quality is one that we examinein the present paper.
Most recently, Light and Stayer (2000) have examined the match between student ability and university quality as a determinant of non-completion. In contrast to earlier work, they model both the initial decision to attend college (defined in terms of quality) as well as the determinants of completion. These are estimated simultaneously to allow for the fact that the choice of college is determined by observables and unobservable factors that also affect the subsequent decision to withdraw or graduate.[3] Light and Stayer's findings confirm earlier the results of earlier work in that academic aptitude has a positive impact on the likelihood of graduation. However, they also find important matching effects: low ability students are more likely to drop out from higher quality institutions than they are from low quality institutions. Non-completion rates for more able students, on the other hand, at first increase but then fall as the quality of the institution increases.
Whilst the above studies are instructive, there are differences between the American higher education system and those of many other countries which may render it hazardous to make inferences about, say, British experience, based on these studies alone. For instance, the participation rate has been especially high in American higher education, and education at this level has been provided by a greater heterogeneity of institutions. The paucity of British work on this subject has been remarkable, however. A clutch of studies by Johnes (1990) and Johnes and Taylor (1989, 1990) was followed by a study, specific to the Scottish universities, by Woodley et al. (1992). The thrust of these exercises revealed that, in Britain, attrition rates depend on aptitude, the subject studied, commitment to the university, and a measure of the distance of the university from a student’s parental home. Scottish universities tend to have higher dropout rates than those elsewhere in the UK, but this is largely explained by the higher age participation rate in Scotland than in other countries of the Union and by idiosyncrasies of the education system in Scotland. In particular, the broader curriculum pursued between the ages of 16 and 18 as students prepare for the Higher (rather than, as in the rest of the UK, the General Certificate of Education Advanced Level) examination.
3. Some Conceptual Issues
Two interrelated themes that have run through much work in labour and education economics in recent years are matching and the influence of peer effects. Matching models have enhanced our understanding of unemployment by developing bilateral job search theories so that the quality of a match between worker and firm determines either the probability that a worker is hired and/or the tenure of the worker with the firm. An immediate extension of this idea to the context of higher education would provide a model of student retention. Indeed, such ideas underpin the work of Booth and Satchell (1995) who have empirically investigated the hazard rate for doctoral students. In the context of education (especially at levels up to first degree), the quality of the education provider is partly determined by the characteristics of students within the peer group. Hence, peer group effects have been widely studied, for example, as determinants of pupil performance in secondary schools (Feinstein and Symons, 1999).
As noted earlier, Light and Stayer (2000) confirmed the significance of matching effects in their study of the determinants of completion. That study focused on one aspect of matching, that between the quality of the institution and student ability. In this paper, a second matching hypothesis isa also examined and concerns the integration of students within the university. In particular, we examine whether students who reside at home are more likely to drop out because they do not integrate with their peer groups in the same way as other students.
Secondly, the impact of peer group effects on the likelihood that an individual will persist or to drop out of a university course is also considered. A priori, one would expect students who are more able, relative to the average for their peer group, will also be more likely than others to drop out of university. Those of lower than average ability, on the other hand, may be more likely to fail, other things equal. Peer group effects may work also along dimensions other than ability. As noted above, Robst et al. (1998) emphasised the role that the gender composition of staff can have on the propensity to persist. In terms of peer group effects, we extend this analysis to investigate the role played by the gender composition of the peer group of students..
4. Data
The empirical work that follows is based upon a data set that we have constructed from four sources. The main source is the set of Universities Central Council on Admissions (UCCA) records of all individual students in traditional universities.[4] In addition to information collected at time of application to university, the data set includes follow up information about the progress of each student. The data set therefore contains information about persistence through the university course in addition to degree obtained and first destination. These data have been used in recent studies of degree results (McNabb et al., 1998; Naylor and Smith, 1999a) and labour market attainment (Naylor and Smith, 1999b).
The background information provided for each student is extremely rich. It includes, inter alia, information about the type of secondary school attended by the student and the occupation of the main income-earning parent. These data are available for a long run of years, and in each year there is a large number of observations. So we concentrate our analysis only on the most recent year for which data are available, namely 1993.[5] Moreover, we include only English and Welsh universities in our analysis owing to the difficulties associated with the interpretation of results obtained at Scottish higher.[6]
A number of institution-specific variables are included to measure institution quality. Data on income and expenditure patters of the universities and on staff:student ratios are available from the Higher Education Statistical Agency; this is the second source of data. The third source of information concerns teaching quality. This is obtained from the results of subject-specific Teaching Quality Assessments (TQA) carried out by the Higher Education Funding Councils and available from Likewise, the final source of data - on the quality of research - is available from the Funding Councils’ Research Assessment Exercise (RAE) for 1992. These data may be obtained from
The number of variables used in our analysis is quite large. Definitions are therefore provided in Appendix Table 1.
5. Empirical Analysis
Tables 1 through 3 report the results of a multinomial logit analysis of student attrition. Three regimes are identified: completion of the degree, academic failure ('involuntary' attrition), and dropout (which we describe as ‘voluntary’, although it may include, inter alia, attrition due to poor health, indigence, imminent failure, or even death). The model is estimated using all observations; in addition separate sets of results are reported for males and females. The choice of explanatory variables is informed by the received literature and the conceptual discussion provided earlier.
The results indicate, first, that both voluntary and involuntary non-completion are negatively related to performance in national examinations prior to university entry (A-Levels or Scottish Highers). This lends some support to previous studies that show that academic aptitude is an important factor affecting the likelihood of non-completion. This is also reflected in the fact that those students who enter university on the basis of A-Levels or Scottish Highers are less likely to drop out of university than those with other types of academic qualifications. Indeed, students who enter university with no formal educational qualifications are also more likely to graduate than those with less traditional academic qualifications. The latter students are typically mature, entering higher education post-experience.
The findings reported here also indicate that older students are, ceteris paribus, less likely to drop out or to fail their degrees than are their younger counterparts. For many mature students, though, these effects may be offset by the fact that they are married which increases the likelihood of dropping out.
The results for the pooled sample indicate that men are both more likely to drop out and to fail their degrees, other things equal, than women. This result echoes the finding of several earlier studies, and has commonly been attributed to gender differences in the determinants of attrition – for example, Tinto (1993) argues that social factors dominate women's decisions to quit, while academic considerations are at the forefront in the case of men. We explore the gender issue further by estimating separate equations for men and women. These will be discussed later.