Nontraditional student withdrawal from undergraduate accounting programmes: a holistic perspective

ABSTRACT

A collaborative project of several Quebec universities, this study investigates nontraditional student withdrawal from undergraduate accounting programmes. A nontraditional student is older than 24, or is a commuter or a part-time student, or combines some of these characteristics. Univariate and multivariate analyses of student dropout factors were performed. A logistic regression for full-time students indicates several significant determinants of student withdrawal: returning to school after working for some time, enrolment in a non-first choice programme, dissatisfaction with programme choice and courses, and low grade point average (GPA). For part-time students, low GPA is the main explanatory factor for student withdrawal. Other factors appear to be instrumental in withdrawal decisions, such as management of external resources (time and family responsibilities) for women. The results suggest that students would benefit from university support services to acquire learning strategies that improve perseverance. Lastly, in-class learning activities that help bolster grades could decrease student withdrawal rates.

KEYWORDS:

• Accounting studies
• dropping out
• interactive multimedia system
• university students
• withdrawal

Acknowledgements

The authors thank the following collaborators of the participating institutions: Sylvain Beaudry, Diane Bigras, Pierrette Doré, Bruce Lagrange, Isabelle Lemay. The authors are grateful for the helpful comments of the two anonymous reviewers.

Disclosure statement

No potential conflict of interest was reported by the authors.

Funding

This study was supported by Fonds de développement académique du réseau (FODAR) de l'Université du Québec.

Notes

1. Respondents answered all questionnaires by identifying items that applied to them. All variables that contained several items were computed in the same manner as the financial strain variable.

2. Six questions on moral support from parents and friends were also included. However, no results are reported since the scale’s reliability was poor (Cronbach’s alpha = 0.451).

3. Seven universities originally participated in the study. However, the students surveyed in three universities were in a second-year accounting course while all students surveyed in one university were in a distance education university. Since the reasons for dropping out may be different for students in their first year than for those in later years (Willcoxson et al., 2011), we kept only students who were in their first year of study. The distance education university was dropped from the sample since the impact from the institution could not be differentiated from that of distance education. Further, there were too few distance education students to analyse their withdrawal characteristics separately (only 35 students). This provided us with a more homogeneous sample.

4. There is no significant difference at p ≤ .05 in the withdrawal rate between the largest university and the other two institutions (chi-squared, not tabulated).

5. In a receiver operating characteristic (ROC) curve, sensitivity (vertical axis) is plotted in function of 1-specificity (horizontal axis). Sensitivity is the true positive rate, that is, number of true positives/total number of positives (Hosmer & Lemeshow, 2000). Specificity is the true negative rate, that is, number of true negatives/total number of negatives (Hosmer & Lemeshow, 2000). The area below the curve indicates the quality of the model. For the logistic regression in Table 4, the area under the ROC curve is 0.851, considered an excellent result (Hosmer & Lemeshow, 2000). It can therefore be concluded that the logistic model discriminates between subjects well. Specificity is very high (98.2%), although sensitivity is not (45.9%) but accuracy is good (89.7%). The regression was generally effective in classifying those who did not leave their programme of study but relatively ineffective in classifying dropouts using a cutoff point of 50%. We use a cutoff point of 16.3%, which corresponds to the actual dropout rate for full-time students considered in the regression, to present model 3’s rate of classification in Table 4. Adding a control for the largest university in model 3 does not change the results for the other variables, and the coefficient for the added variable is not significant at p ≤ .05 (not tabulated). This is also the case for the analysis for part-time students presented in Table 5. Adding an interaction in model 3 between the current programme of study not being students’ first choice and dissatisfaction with programme choice and courses does not change the results for the other variables. The interaction is not significant even at p ≤ .10.

6. To present the model’s rate of classification in Table 5, we use a cutoff point of 22.3%, which is the actual dropout rate for part-time students considered in the regression. The variance inflation factors are low, under 1.63, for all variables considered in regressions presented in Tables 4 and 5. For condition indices over 15, the regression coefficient decomposition matrix shows that none of the condition indices reflect variance proportions over 90% for two or more coefficients, indicating no problem with multicollinearity (Hair, Anderson, Tatham, & Black, 1998).