Calling it quits: a longitudinal study of factors associated with dropout among doctoral students

ABSTRACT

This article investigates whether experiences of the PhD trajectory during the first year of enrolment predict dropout during a later stage. More specifically, we studied how supervisor support, time pressure and passion for research relate to dropout among PhD students and assessed the role of discipline in this process. We used longitudinal data from four waves of the PhD Survey (2018 through 2021; N = 589), conducted at a university in Belgium, as well as administrative data on the enrolment status. Results show that supervisor support is negatively related to dropout, and that this is especially important for PhD students in the human sciences. Time pressure is positively related to dropout. When stratified by scientific discipline, this effect was only significant for PhD students in human sciences and in the life sciences and medicine. Passion for research showed a negative association with dropout. Stratification by discipline showed that this effect was only found among PhD students in natural sciences and engineering. Furthermore, teaching assistants showed higher dropout rates, and female PhD students in human sciences and life sciences and medicine were less likely to drop out.

KEYWORDS:

• Doctoral studies
• PhD students
• dropout
• supervisor support
• time pressure
• passion

Introduction

It is in universities’ interest to have high numbers of thriving and successful PhD candidates. PhD students are an essential part of the research system, as Larivière (2012) showed that one-third of the research output at universities was produced by PhD students. Moreover, the unsuccessful completion of a PhD trajectory goes hand in hand with some major costs. These costs include financial losses of departmental, institutional, and state resources (Golde 2005), costs for society – as dropout implies a loss of knowledge and talent –, as well as psychological costs and opportunity costs for the PhD students themselves (Allan and Dory 2001).

Informed by these considerations, we investigate the lack of effectiveness of a PhD trajectory in terms of dropout rate. This behavioural outcome variable is one of the most obvious indicators of an unsuccessful trajectory. Gaining insight into what elements are related to dropout and what contributes to it, will enable policymakers in universities to develop targeted measures. The latter is crucial. In general, dropout rates among doctoral researchers are expected (1) to be relatively low but also (2) non-systematic. Indeed, PhD students have had 20 years of education, which can be seen as a long horse race in which only the best students are selected. Although some dropout is probably unavoidable, one expects that dropout in such an over-selected group will be largely due to exceptional and/or personal circumstances and hence follow a random (i.e. non-systematic) pattern. Therefore, even weak systematic patterns will be highly informative for universities.

That said, the PhD track is very different from other phases of education and brings along its specific challenges. Where in previous stages of an academic career following courses and taking exams is the sole focus, most of the weight within a doctoral trajectory in Belgium, the region we study, lies on writing the dissertation. A substantial group of PhD students works alone on their project under the supervision of one or more supervisors. Working collaboratively with peers is not always part of the PhD trajectory, which can sometimes make it a lonely process and renders the role of the supervisor all the more important (Cantor 2020). Additionally, the academic environment in which PhD students work is characterised by ever-increasing job demands and competition, due to among other things a growing number of undergraduate students who increasingly fall under the responsibility of PhD students, an increasing pressure to get research funding and publish, and a growing demand to be involved in other activities next to research (Gill 2014). Both the high dependency on the supervisor and the demanding academic environment might incentivise PhD students to quit.

Indeed, research found that factors related to supervision, the project itself and psychosocial factors are associated with the intention to quit the PhD (van Rooij, Fokkens-Bruinsma, and Jansen 2021). Simultaneously, turnover intention does not always reliably predict actual turnover, nor are the variables explaining turnover intention necessarily the same as those explaining actual turnover (Cohen, Blake, and Dough Goodman 2016). Therefore, we add to this line of work by studying how the received support of the supervisor, the experienced time pressure during the project, and the amount of passion one has for research can predict actual dropout. Contrary to previous studies – that tend to focus solely on administrative data or survey data – we combine administrative data on actual dropout with survey data on the experiences of the doctoral trajectory. Gaining insight in how these predictors might influence dropout is interesting from a policy point of view, because it can lead to insights that can be translated relatively easily into implementable measures to reduce dropout rates.

In this study, we also pay attention to the heterogeneity within the group of PhD students. The contract under which they work (i.e. the tasks they take up, their salary …), personal attributes, the scientific discipline in which they work (i.e. the research culture) are all factors that may affect the perspective of the PhD trajectory. Although we argued earlier that systematic patterns in dropout among PhD candidates would not be expected, research does suggest that dropout rates between disciplines differ (Golde 1994; Wright and Cochrane 2000). However, deeper knowledge on the mechanisms behind this is lacking. The aim of this study, then, is to further examine this paradox by investigating whether certain characteristics of PhD students and certain experiences of the PhD trajectory are associated with dropout, and how the importance of these variables varies between scientific disciplines. More specifically, the two research questions for this article are: ‘to what extent do support, time pressure and passion for research predict dropout?’ and ‘does their potential predictive power vary across scientific disciplines?’.

Background

Dropout rates

Starting a PhD is preceded by a lengthy selection process, making it safe to assume that in principle those who start the trajectory have a good chance to be capable to finish it. This renders it remarkable that in a European institution about 34% of all PhD students do not obtain their PhD degree within six years (Hasgall, Saenen, and Borrell-Damian 2019). In Australian, British, Canadian, and American universities, average dropout rates range between 30% and 50%, depending on the discipline (Bowen and Rudenstine 2014; Lovitts 2002; Golde 2005; Council of Graduate Schools 2008). Indeed, dropout rates vary considerably between disciplines. Golde (1994), for example, found lower dropout rates among PhD students in natural sciences, and higher rates in social sciences and the humanities. Wright and Cochrane (2000) found higher success rates among PhD students in beta and gamma sciences (i.e. science, engineering, and medicine). However, research on the underlying tendencies that could explain these differences between disciplines is lacking.

According to various authors, the reasons for dropping out should not only be sought within PhD students themselves, but also in the underlying structures of the academic environment (Smith et al. 2006; Lovitts 2002). Issues such as an excessive workload, insufficient decision-making power, a low status in the academic hierarchy, and inadequate funding (Winefield et al. 2008), as well as low support and isolation (Van der Doef and Maes 1999; van Rooij, Fokkens-Bruinsma, and Jansen 2021), and an imbalance between work and private life (Bell, Rajendran, and Theiler 2012) all add up to the occupational stress of PhD students and eventually result in dropout.

One limitation of the existing literature on dropout among PhD students, however, is that it usually focuses either on PhD students who dropped out already (e.g. Bayley et al. 2012; Golde 2000; 2005; Willis and Carmichael 2011) or who are still doing their PhD (e.g. Martinsuo and Turkulainen 2011; Gardner 2010; Mazzola et al. 2011; Pyhältö and Keskinen 2012; Glorieux et al. forthcoming). This research has to rely on past experiences of the PhD students and their recollections of the situation. Moreover, it cannot be ruled out that the motives for dropping out among those students were not also perceived by the students who completed the programme. A limitation of studying the intention to quit is that it investigates what could impact the decision to drop out, and what elements play a role in the hypothetical intention to drop out.

Ideally, research on dropout is based on longitudinal data, where PhD students can be followed from the start of their trajectory until their completion or dropout. The limited existing research that uses longitudinal data usually relies on administrative data (e.g. Groenvynck, Vandevelde, and Van Rossem 2013; Visser, Luwel, and Moed 2007; Wright and Cochrane 2000). Such data, however, offer very limited insight into the process that leads to dropout. Combining administrative data with survey data offers interesting opportunities in this context.

Supervisor support

The PhD track is often characterised as a solitary process in the sense that much of research is performed by PhD students alone. When studying dropout, Tinto (1993) and Lovitts (2002) both focus on the importance of the integration of PhD students in their work environment. Tinto (1993) argues that the integration within the discipline and the department is one of the most important factors for doctoral success. Lovitts (2002) builds on this finding and argues that a lack of academic integration (i.e. integration within the academic system, intellectual development, receiving academic help etc.) is the most important factor in doctoral dropout – as opposed to social integration (i.e. having a sense of community among peers, extracurricular activities with fellow students etc.), which is important to undergraduate students, but less to PhD students.

The supervisors play a key role in the academic integration of PhD students. They are the first to advise students about the research project, to review the research plan, and to explain the programme requirements. Indeed, academic support from the supervisor is negatively associated with the intention to quit (van Rooij, Fokkens-Bruinsma, and Jansen 2021; Kis et al. 2022). Therefore, we expect that a stronger feeling of supervisor support is associated with a lower likelihood of dropout (H1a).

Research into the different supervisory needs between disciplines is lacking. In disciplines where research is done in labs, faculty members and PhD candidates tend to work more closely together as parts of a team, whereas in the humanities, research is mostly carried out in solitude (Cantor 2020). Golde (2005) argues that supervisors in hard sciences have more central roles in the research project and are a key-actors in the research team and thus their support plays a much bigger role in degree completion. However, it could also be argued that the quality of supervisor support is more important for successful completion for PhD students in humanities and social sciences, who are more prone to isolation and feeling unsupported (Chiang 2003). Because PhD candidates in the hard sciences tend to be more surrounded by peers and other actors that might function as a support system, we expect that in disciplines where PhD students mainly work individually (i.e. social sciences), the (lack of) received support of the supervisor plays a bigger role in the decision to drop out (H1b).

Time pressure

It is expected that the academic environment in which PhD students work also strongly influences their perception of the PhD trajectory and their drive to continue their studies. Some PhD students are required to take up multiple roles (i.e. student, employee, teacher) and juggle multiple responsibilities (i.e. doing research, teaching, attend meetings, being involved in other projects or service contracts) (Winstone and Moore 2017; Siltanen et al. 2019). This high workload and role diversity, in combination with ‘the gift of autonomy’ that characterises the academic work environment (i.e. the flexibility of choosing one’s working hours), can easily result in a form of self-exploitation in which researchers find themselves working late at night and during the weekend (Ross 2008). Working long hours and working on irregular moments is not necessarily a bad thing. In fact, van Tienoven et al. (2023) found a positive association between working long and non-standard hours, and the engagement of PhD students with their research. However, when these long, irregular hours are accompanied by a high feeling of time pressure and a lack of control over one’s own time, this positive association with research engagement of PhD students is strongly weakened.

Indeed, 70% to 80% of the PhD students indicate experiencing time pressure (Kolmos, Kofoed, and Du 2008; El-Ghoroury et al. 2012). Additionally, it appears to be challenging for PhD students to keep control over working hours so that they can be adjusted to their private situation. Both Herman (2011) and Castelló et al. (2017) identified the imbalance between private and academic life, and the difficulty to combine a PhD with family commitments or another job as one of the most prevalent obstacles PhD students face.

High job demands and a multitude of roles result in a strong feeling of being pressed for time, and the prevailing academic culture perpetuates that feeling. Gill (2014) argues that universities became businesses, by adopting corporate management techniques and shifting towards neoliberal funding regimes. Universities have been focussing more and more on ‘massification, marketisation and internationalisation’ (Barcan 2013, 31), and these growing expectations of productivity go hand in hand with a decrease in available time (Tham and Holland 2018). Such a high workload ultimately leads PhD students to consider stopping their doctoral studies (van Rooij, Fokkens-Bruinsma, and Jansen 2021). Against that background, we expect that a stronger feeling of time pressure is associated with a higher likelihood of dropout (H2a).

Regarding differences between disciplines, Lindholm and Szelényi (2008) found that academics in the hard, life sciences (e.g. medicine, biology, agriculture …) experienced the most time pressure of all disciplinary groups. Aittola (2001) mentions that the work of doctoral students conducting research in labs is highly demanding, as the operation of the labs depends on them. Overall, research on differences between disciplines regarding time pressure is scarce, but we expect that in disciplines where PhD students tend to combine their research with other work (i.e. medical sciences), the role of time pressure plays a bigger role in the decision to drop out (H2b).

Passion for research

In addition to these structural elements that can affect dropout, personal characteristics of PhD students cannot be overlooked. Having the right skills and knowledge, as well as certain psychological attributes also determine the success or failure of a PhD (Lindsay 2015; Herman 2011). Moreover, such characteristics are also relevant from a policy perspective as they can be measured during the selection process.

Dropout is often urged by a discrepancy between the goals and expectations of PhD students, and the norms and practices of the discipline and department (Golde 2005). PhD students dropped out because they realised that the practice of doing research in a certain discipline was not for them – or was not as meaningful to them as they thought. Similarly, Busso and Rivetti (2014) report that finding meaning in work and being passionate about it plays an important role in the work of researchers. They compare researchers with artists, and refer to passion as necessary to incite curiosity, motivation, and performance. Especially in the academic work environment with flexible and often long working hours, passion is seen as an important element to remain disciplined. Indeed, van Tienoven et al. (2023) found a positive association between engagement with PhD research and working long and non-standard hours.

Yet passion does not only play a role in productivity; it is also a necessary factor for the well-being of researchers. Especially for young researchers, who face a lot of insecurity, periods of doubt, and competition, liking what they do can be a great element of support and plays a crucial role in persevering. Fetherston et al. (2021) refer to this as the ‘academic calling’; meaning that for some academics, their work is part of their identity, and they enjoy it to the extent it does not feel like hard work anymore. Their research confirms that job satisfaction has a positive effect on the well-being of researchers.

In turn, these personal attributes may affect more external elements of the job. For example, the support and involvement of the supervisor might be related to the intrinsic motivation and capabilities of the PhD student and vice versa (Young et al. 2019; Lindsay 2015).

Based on these findings, we expect that a lower sense of passion for research is associated with a higher likelihood to drop out (H3a). We did not come across any research on disciplinary differences in the amount of passion PhD students have for their research. Therefore, we do not expect that the effect of passion for research on dropout is different between the disciplines (H3b).

Data and method

Data

To answer our research questions, we rely on longitudinal data from the VUB PhD Survey as well as administrative data of PhD students of the Vrije Universiteit Brussel (VUB). VUB is a medium sized university located in Brussels, the capital of Belgium. The VUB PhD Survey contains information on the subjective experiences of PhD students. The survey is organised on an annual basis by the Researcher Training & Development Office (RTDO) and executed by the Research Group TOR (Tempus Omnia Relevat), both affiliated with the VUB. The PhD Survey is used to monitor the job satisfaction of PhD students at VUB and has been conducted annually between April and May in all faculties of the VUB since 2018. The PhD Survey consists of one questionnaire, hosted on the data collection platform MOTUS (https://www.motusresearch.io/en). The design of the study was approved by the ethics committee of the VUB (file number ECHW_318). The response rates for the waves vary between 42% and 49%.

The data of the VUB PhD survey were matched with administrative information on the current administrative enrolment status of PhD students: (1) successfully completed the PhD programme, (2) still active in the programme, or (3) dropped out of the programme.

For this paper, we limit the data to the VUB PhD Survey waves from 2018 to 2021 and restrict the sample to PhD students who were in their first year of enrolment when completing the survey (n = 589).

The combination of administrative data on the enrolment status with survey data on the subjective experiences of PhD students during their first year of enrolment enable us to investigate the effects of subjective indicators at moment t on moment t + 1, and see whether they can predict dropout. Moreover, the university-wide data enable us to study differences within the heterogenous group of PhD students, by focusing on a group of PhD students (1) from various disciplines who (2) work under different contracts.

Dependent variable

The dependent variable is a dummy-coded variable that indicates whether a PhD student dropped out (see Table 1). In our sample, 10.5% dropped out. Note that this is not the actual dropout rate of the PhD programme at the VUB, as our sample only consists of PhD students that completed the PhD survey during the first year of their enrolment and were enrolled in the academic years 2017–2018 to 2020–2021. Since dropout rates are typically calculated over a 6-year or 8-year period, the dropout rate in our sample is lower than average.

Table 1. Frequency table of dropout.

	N	%
Dropped out	62	10.5
Did not drop out	527	89.5
Total	589	100

Independent variables

In this study we investigate how supervisor support, time pressure and passion for the research play a role in the decision to drop out from the doctoral programme. Below we describe how we operationalised these variables.¹

Experienced time pressure in the first year of enrolment was measured by the items ‘Too much is expected of me’, ‘I never catch up with my work’, ‘I never have time for myself’, ‘There are not enough hours in the day for me’, ‘I frequently have to cancel arrangements I have made’, ‘I have to do more than I want to do’, ‘I have no time to do the things I have to do’, and ‘More is expected from me than I can handle’ using a 5-point Likert scale ranging from 1 = strongly disagree to 5 = totally agree (van Tienoven, Minnen, and Glorieux 2017). One component was identified using a principal component analysis, with Eigenvalues between 4.30 and 4.72 (depending on the wave), and 53.7% to 56.5% of the variance explained.² The Cronbach’s alpha of the scale ranged between 0.88 and 0.90. Based on this, a summative scale was computed ranging from 0 to 10. The descriptive information of this variable can be found in Table A1 in the Appendix.

Support of the supervisor in the first year of enrolment is measured by the items ‘Satisfaction with the quality of the meetings’, ‘Satisfaction with the frequency of the meetings’, ‘Satisfaction with the support you receive in writing articles’, ‘Satisfaction with the stimulation/inspiration to solve research problems/issues’, ‘Satisfaction with the expertise the supervisor has on the research subject’, ‘Satisfaction with the introduction to other prominent researchers’ and ‘Is you supervisor involved in your research?’ using a 5-point Likert scale ranging from 1 = not at all satisfied to 5 = totally satisfied. One component was identified using a principal component analysis, with Eigenvalues between 4.88 and 5.99 (depending on the wave), and 48.8% to 54.5% of the variance explained.³ The Cronbach’s alpha of the scale ranged between 0.87 and 0.91. Based on this, a summative scale was computed ranging from 0 to 10.

Passion for research in the first year of enrolment was measured by a scale ranging from 0 to 10 on which PhD students indicated how passionate they are about their PhD research (0 = not passionate at all, 10 = totally passionate). The descriptive information of this variable can be found in Table A1 in the Appendix. The correlations between the supervisor support, time pressure and passion for research can be found in Table A2 in the Appendix.

To assess differences across disciplines, we used PhD students’ enrolment in the doctoral schools as a proxy. At VUB, PhD students are enrolled in one of three doctoral schools depending on the faculty they belong to. The doctoral schools support PhD students throughout their trajectory with workshops, courses and activities. The first doctoral school is the doctoral school of Life Science and Medicine (LSM; ‘Medicine & Pharmacy’ and ‘Physical Education & Physiotherapy’), the second one is the doctoral school of Natural Science and (bioscience) Engineering (NSE; ‘Sciences & Bio-science Engineering’ and ‘Engineering Sciences’) and the third one is the doctoral school of Human Sciences (DSh; ‘Arts & Philosophy’, ‘Economics & Social Sciences’, ‘Law & Criminology’ and ‘Psychology & Educational Sciences’). The doctoral schools overarch faculties with a common nature and generate a common culture and policy within them. To increase the statistical power in our analyses, we used the doctoral schools as a proxy for discipline, and not the individual faculties.

We controlled for the following characteristics of PhD students. Gender, which was measured by a dummy for being female. Nationality, which was measured by using a dummy for being Belgian. Type of contract which measures to the type of funding PhD students have. PhD students can (1) have a personal mandate, meaning their funding is directly related to themselves (reference category), (2) have project funding, meaning the funding is granted to the supervisor of a project, who in turn hires the PhD student to execute it, (3) be teaching assistants, meaning they divide their time between research and teaching, or (4) be self-financed or have another type of contract. Note that in the Belgian context, there is only a relatively small number of PhD students that do a PhD on a voluntarily (i.e. non-paid) basis (10.6% in our sample). The majority of PhD students receives a salary and is considered a fulltime professional. Table A3 in the Appendix shows the distribution of the background characteristics.

Research strategy

To answer our research questions, we use a two-step analysis. Firstly, we performed a logistic regression analysis predicting dropout. Model 1 included background characteristics only (gender, nationality, doctoral school, and type of contract). In separate models, we successively combined the background characteristics with the following predictor variables: the experienced support of the supervisor during the first year (model 2), the experienced time pressure during the first year (model 3), and passion of PhD students for their research in the first year (model 4). The fifth and final model included all variables.

Secondly, we stratified the final model by doctoral schools. We tested whether the effect parameters varied significantly between disciplines using calculations suggested by Paternoster and colleagues (1998).

Results

Model 1 (Table 2) shows that teaching assistants had a higher likelihood to drop out compared to PhD students with a personal mandate, and so did those with project funding (OR = 3.17 and OR = 2.11, respectively). PhD students in the doctoral school of DSh and those in the LSM had a higher likelihood to drop out compared to those in NSE (OR = 1.87 and OR = 2.14, respectively).

Table 2. Logistic regression models of predictors of dropout during PhD track (n = 589).

	Base model (1)			Model 2: supervisor support			Model 3: time pressure			Model 4: passion for research			Final model (5)
	B	OR	Sig.	B	OR	Sig.	B	OR	Sig.	B	OR	Sig.	B	OR	Sig.
Gender (ref. Male)	−.37	.69		−.41	.66		−.46	.63		−.45	.64		−.52	.59	(*)
Nationality (ref. Non-Belgian)	.12	1.13		−.02	.98		.13	1.14		−.02	.98		−.05	.95
Doctoral school (ref. NSE)
DSh	.62	1.87	(*)	.61	1.83	(*)	.47	1.59		.66	1.94	*	.49	1.63
LSM	.76	2.14	(*)	.84	2.31	*	.51	1.67		.92	2.51	*	.69	1.99
Type of contract (ref. personal mandate)
Teaching assistant	1.16	3.17	*	1.07	2.91	*	1.14	3.12	*	1.22	3.38	*	1.13	3.09	*
Project funding	.75	2.11	(*)	.67	1.96		.79	2.20	(*)	.72	2.06	(*)	.70	2.01
Other or self-financed	.68	1.97		.60	1.82		.62	1.86		.64	1.91		.53	1.69
Support supervisor				−.28	.76	***							−.17	.84	*
Time pressure							.31	1.37	***				.27	1.30	***
Passion for research										−.30	.74	***	−.18	.83	*
Constant	−3.1	.05	***	−.89	.41	n.s.	−4.25	.01	***	−.63	.53	n.s.	−1.19	.30	n.s.
Adjusted R^2	4.7			9.9			11.3			9.0			15.7

Notes: Levels of significance: ***p ≤ 0.001, **p ≤ 0.01, *p ≤ 0.05, (*)p ≤ 0.1.

The effect parameter of type of funding decreased and became non-significant after including supervisor support in Model 2. The effect parameter of the experienced supervisor support was statistically significant: the more support one experienced during the first year of one’s PhD, the less likely one was to drop out (OR = 0.76), confirming H1a.

Model 3 includes time pressure and shows that this also had a significant effect: the more time pressure one experienced during the first year of the trajectory, the more likely one was to drop out (OR = 1.37), confirming H2a. After the introduction of this variable, the effect of doctoral schools disappeared.

As shown in Model 4, passion for research also played a role in dropout. The more passion for one’s research, the less likely one was to drop out (OR = 0.74), confirming H3a. The effects of doctoral schools, of being a teaching assistant, and of having project funding remained significant in this model.

The final model includes all three independent variables as well as the control variables. All indicators remained statistically significant: experienced time pressure was positively associated with dropout (OR = 1.30), while experienced support and passion for research were negatively associated with dropout (OR = 0.84 and OR = 0.83, respectively). For the control variables, we found that being a teaching assistant led to a higher likelihood of dropout (OR = 3.09) and being female was related to a lower likelihood of dropout (OR = 0.59).

It is remarkable that the effects of these control variables did not substantially decrease after controlling for the indicators related to doctoral experiences. Some even became larger. This indicates that, next to the experienced support of the supervisor, experienced time pressure, and passion for research, there are other processes at play that influence the decision to drop out. We reflect on this more in the discussion.

Finally, it is important to note that the odd’s ratios represented the relative difference to the 10.5% likelihood of dropout (see Table 1). Some effects might seem large, but they still resulted in a relatively low absolute likelihood of dropout.

Table 3 shows the results of the final model stratified by doctoral schools. Note that in this study we use doctoral schools as proxies for scientific discipline. In the human sciences (DSh) both the supervisor support and experienced time pressure showed a significant effect parameter. Experiencing substantial support from a supervisor related to a lower likelihood of dropping out (OR = 0.75). Besides, experiencing a lot of time pressure related to a higher likelihood of dropping out (OR = 1.50), partly rejecting H2b. Similar to the overall model (Model 5 in Table 2) the effects of being female and type of contract were significant. Female PhD students in human sciences had a lower likelihood of dropping out (OR = 0.46). Contrarily, teaching assistants as well as those with project funding and those who financed their own research or had another type of contract had a higher likelihood of dropping out when compared to those with a personal mandate (OR = 10.97, OR = 6.18 and OR = 9.57, respectively).

Table 3. Logistic regression models of predictors of dropout during PhD track stratified by doctoral school (n_DSh= 236, n_NSE= 252, n_LSM= 101).

	Human Sciences (DSh)			Natural Sciences & Engineering (NSE)			Life Sciences & Medicine (LSM)
	B	OR	Sig.	B	OR	Sig.	B	OR	Sig.
Gender (ref. Male)	−.78	.46	(*)	−.26	.77		−1.26	.28	(*)
Nationality (ref. Non-Belgian)	.03	1.03		−.32	.73		.44	1.55
Type of contract (ref. Personal mandate)
Teaching assistant	2.40	10.97^a	*	.36	1.43^a		1.22	3.38
Funded	1.82	6.18^a	(*)	.01	1.01^a		1.29	3.65
Self-financed or other	2.26	9.57^ab	*	−1.31	.27^a		−.77	.46^b
Support supervisor	−.29	.75^a	*	−.08	.93		.06	1.07^a
Time pressure	.40	1.50^a	***	.14	1.16^a		.35	1.42	(*)
Passion for research	−.04	.96^a		−.44	.65^ab	**	−.04	.96^b
Constant	−2.91	.06	(*)	1.22	3.40	n.s.	−3.78	.02	n.s.
Adjusted R^2	27.5			12.8			18.6

Notes: Levels of significance: ***p ≤ 0.001, **p ≤ 0.01, *p ≤ 0.05, (*)p ≤ 0.1. Odd’s ratios sharing a letter in their subscript are significantly different from each other at α = 0.1.

In natural sciences and engineering (doctoral school of NSE), only passion for research was statistically significant. The more passion for research in the first year of enrolment, the lower the likelihood of dropout (OR = 0.65). This effect was significantly stronger than in the other two doctoral schools, rejecting H3b.

In life sciences and medicine (doctoral school of LSM), only the experienced time pressure was statistically significant, confirming H2b. The more experienced time pressure, the higher the likelihood to drop out (OR = 1.42). Additionally, like female PhD students in human sciences, female PhD students in life sciences and medicine too had a lower likelihood of dropping out than their male peers (OR = 0.28).

Discussion

This article assesses which factors predict dropout for a PhD programme and whether these factors varied by scientific discipline. We argued that finding even small differences would be meaningful given that the group of PhD students is an over selected group. Previous work suggested that supervisor support, experienced time pressure that arises from increasing job demands, and being passionate about doing academic research are crucial elements in the (successful) completion of the PhD programme (e.g. van Rooij, Fokkens-Bruinsma, and Jansen 2021; Kis et al. 2022; Herman 2011; Castelló et al. 2017; Busso and Rivetti 2014). This study tested their systematic effect on dropout from a PhD programme in general and by scientific discipline. The necessity of the latter is confirmed by the general finding that PhD students in the natural sciences and engineering have a lower likelihood to drop out than PhD students from the human sciences and life sciences, which is in line with previous findings (Golde 1994; Wright and Cochrane 2000).

We found that more satisfaction with supervisor support is related to a lower likelihood of dropout (confirming H1a). This turns out to be especially important for PhD students in human sciences (confirming H1b) and might relate to the important role supervisors play in the academic integration of PhD students. PhD students in the human sciences are more likely to work in solitude on their own project when compared to, for example, PhD students in the natural sciences and life sciences, whose work is more likely to be centralised around teamwork tied to a specific location or a larger research project (Golde 2005; Cantor 2020; Chiang 2003). Consequently, the latter PhD students are more likely to have closer contacts with peers and faculty staff. Integration within the discipline and the department is one of the most important factors for doctoral success (Tinto 1993; Golde 2005).

The current academic culture is characterised by high job demands and juggling responsibilities from multiple academic roles, which result in a strong feeling of being pressed for time (Tham and Holland 2018). We showed that time pressure positively associated with dropout (confirming H2a). However, when stratified by scientific discipline, it turned out that only PhD students in the humanities and in life sciences and medicine systematically faced a higher likelihood on dropout if they experienced more time pressure (partly confirming H2b). A possible explanation for this finding might relate to the structure of the PhD programme. In general, PhD students in the human sciences are less tied to (the availability of) physical infrastructure and thus have more flexibility regarding their working hours. Exactly this characteristic, however, makes time management an important personal responsibility and along that way may increase experienced time pressure. In contrast, the programme of PhD students in natural sciences and engineering is often very structured, and their research direction is usually much more straightforward (Lovitts 2002; Golde 2005). Moreover, their research is typically structured in time and place (i.e. pre-booked time slots in a lab), which results in a more predictable and delineated work schedule. The PhD programme of PhD students in the life sciences and medicine often implies combining research with clinical work, which might result in hectic work schedules and increased feeling of time pressure.

Apart from supervisor and time pressure, we also studied how the passion for research was associated with dropout. Overall, reduced passion for research increases the likelihood of dropout (confirming H3a). However, stratification of predictors by scientific discipline revealed that this effect can be attributed to PhD students in natural sciences and engineering (rejecting H3b). The absence of a significant effect for the other scientific disciplines could be explained by the fact that it is tempered by a higher feeling of time pressure.

To summarise, supervisor support, time pressure and passion for research systematically predict dropout of PhD students from the PhD programme, although their predictive power varies across scientific disciplines. From our results, it would thus be advisable to put the emphasis in support policies for PhD students accordingly. Supervisor support and academic integration should be a key focus in human sciences. Reducing the experienced time pressure by guiding PhD students in creating healthy working schedules – by teaching them to deal with a high degree of time flexibility and watch over potentially conflicting time schedules (e.g. through the combination of research and teaching or research and clinical tasks) – should have priority in the human sciences and in life sciences and medicine. Perhaps the most difficult focus lies with students of the natural sciences and engineering, where passion for their research needs to be stimulated.

Next to predictive effects from variables related to the experience of the PhD trajectory several control variables turned out to be systematically predictive of dropout. Firstly, the type of contract predicted dropout. Teaching assistants had a higher likelihood to drop out, which results from their need to juggle more tasks than regular PhD students (Glorieux et al. forthcoming) and this relationship persisted after controlling for time pressure. This indicates that other dynamics must be at play. Indeed, earlier research showed that teaching assistants often do not have a well-developed research plan, are less involved with their colleagues, and often spend more time on teaching than is stipulated in their contract, which leads to the tendency to estimate their likelihood of successful completion lower than regular PhD students (Glorieux et al. forthcoming).

Also, gender plays a role in dropout. Female PhD students in human sciences and in life sciences and medicine were less likely to drop out. It is not an uncommon finding that women outperform men in education (DiPrete and Buchmann 2013; Carbonaro, Ellison, and Covay 2011; Goldin, Katz, and Kuziemko 2006), but it is interesting that even at this late stage of education, having already gone through a very long selection process, this gender difference persists. Previous research on gender differences in doctoral completion is inconclusive: some authors found that male PhD students have a higher likelihood to complete their PhD (Groenvynck, Vandevelde, and Van Rossem 2013; Van Ours and Ridder 2003; Visser, Luwel, and Moed 2007), whereas others found no effect of gender (Wright and Cochrane 2000; Spronken-Smith, Cameron, and Quigg 2017), or found that women are more likely to complete their PhD (Wao 2010). More nuanced findings are reported in studies that distinguish between disciplines. In the humanities and social sciences, female PhD students are shown to have higher completion rates, but they are less successful in science, engineering, and mathematics (Bell 2008), concurring with our findings. Mastekaasa (2005) found no overall difference between genders but did see that in applied sciences men were more likely to drop out than women. In fundamental sciences, no difference was found. One explanation for this could be that in applied sciences, a PhD is often non-essential for a further career, and men are more likely than women to leave their PhD when other career opportunities come up. Women in academia receive fewer external recruitment offers than men (Martinez, O’Brien, and Hebl 2017), and promotions for women often come with less attractive benefits than for men (Booth, Francesconi, and Frank 2003). Indeed, Van de Velde, Levecque, and Mortier (2019) found that male PhD students more often than their female peers indicated that interesting job offers by another employer would be a reason for them to quit their doctoral research. All these arguments underscore the importance of considering the discipline when studying the work experiences and success rates of PhD students.

In none of the analyses, nationality played a significant role in dropout. This is surprising to some extent because the perception often holds that PhD students that move to another country to complete their PhD are often highly motivated and competent, and their stakes to successfully complete the trajectory are typically higher than for domestic students. Reports of the PhD survey show that international PhD students, especially non-European students, tend to be more satisfied and self-confident about the PhD trajectory (Glorieux et al. 2023). On the other hand, it is possible to argue that international students need to familiarise with the Belgian regulations and culture. Additionally, personal problems outside of the PhD trajectory might have a stronger influence on them because the lack of a social network. It can be assumed that these two dynamics cancel each other out.

The inconclusive findings concerning gender and the latter assumption about nationality imply that elements outside of the PhD trajectory also contribute to dropout, and that these influences can also be systematic. Other processes seem to be at play that are likely to be not related to experiences of the PhD trajectory but have an external origin. For example, other job opportunities, adaptation challenges for foreign students, parenthood, or other personal reasons are elements we did not account for in our analyses, but which could provide an additional angle for further research, to get a deeper understanding of structural differences in doctoral dropout.

A limitation of the study is that it only includes data from a single university. A reproduction of the study on data of other (international) universities would be interesting, especially because of the universality of the topic being studied. For further research, it would also be interesting to investigate the underlying trends of our findings with qualitative data. Indeed, our analyses show clear links between dropout, characteristics of the doctoral trajectory and disciplines. We also have raised possible explanations for our observations, but to get a really clear view of why certain elements appear to be more important for certain disciplines, a more in-depth conversation with PhD students would be valuable. Finally, our focus on the experiences of PhD candidates in their first year of enrolment could be expanded in further studies to examine how these experiences evolve throughout further phases of the trajectory and eventually affect dropout.

Conclusion

In contrast to general expectations that follow from the long educational selection process, dropout from the PhD programme is systematic to a substantial extent. Supervisor support and academic integration, time pressure that arises from multiple academic roles and the corporatisation of academia, and passion for research, predict the likelihood of dropout. These predictors vary across scientific disciplines, which highlights the need for universities to be aware of the diversity of PhD students when formulating support policies for PhD students. These policies could include facilitating supervisors to support academic integration of first-year PhD students and create better job resources; monitoring the implementation of research plans and the balance between research and teaching or clinical tasks to reduce experience time pressure; or facilitating state-of-the art research infrastructure to keep PhD students passionate about their research. Finally, special attention should be paid to the needs of teaching assistants, specifically to those in the human sciences, because even after taking supervisor support, time pressure and passion for research into account, they are still more likely to drop out.

Acknowledgements

R&D Department – Doctoral Schools Vrije Universiteit Brussel for financial support. – Dr. Hannelore De Grande & Prof. Dr. Gerd Vandersteen for initial research design. We thank the members of the steering committee for their constructive feedback

Disclosure statement

The authors declare no conflict of interest.

Additional information

Funding

This work was supported by the Research Council of the Vrije Universiteit Brussel.

Notes

1 Variables that were tested but did not show a significant effect were: satisfaction with received support of the university, involvement of colleagues, working on irregular hours, performing other tasks next to research, self-efficacy, experiencing personal doubts and expecting an academic career.

2 To verify the reliability of the PCA, a factor analysis was also performed. The resulting factors were very similar to the components of the PCA and showed very high correlations (between 0.996 and 0.998, depending on the wave).

3 To verify the reliability of the PCA, a factor analysis was also performed. The resulting factors were very similar to the components of the PCA and showed very high correlations (between 0.996 and 0.998, depending on the wave).

Appendix A

Table A1. Average scores and standard deviations on supervisor support, time pressure and passion for research, by doctoral school (score between 0 and 10).

	DSh		NSE		LSM		Total
Support	7.64 (1.90)		7.78 (1.72)		7.84 (1.44)		7.73 (1.75)
Time pressure	3.87 (2.02)	^a	3.29 (1.87)	^ab	4.09 (2.10)	^b	3.66 (1.99)
Passion for research	8.20 (1.54)		8.26 (1.45)		8.40 (1.23)		8.26 (1.45)

Note: Values sharing a letter in their subscript are significantly different from each other at α = 0.01.

Table A2. Pearson correlations between supervisor support, time pressure and passion, by doctoral school.

	DSh			NSE			LSM			Total
	Support	Time pressure	Passion	Support	Time pressure	Passion	Support	Time pressure	Passion	Support	Time pressure	Passion
Support	1	−−0.26**	0.43**	1	−−0.20**	0.44**	1	−−0.17	0.27**	1	−−0.22**	0.42**
Time pressure		1	−−0.14*		1	−−0.11		1	−−0.24*		1	−−0.14**
Passion for research			1			1			1			1

Notes: Levels of significance: ***p ≤ 0.001, **p ≤ 0.01, *p ≤ 0.05, (*)p ≤ 0.1.

Table A3. Distribution of background characteristics (column percentages).

	DSh	NSE	LSM	Total
Gender ***
Male	41.5	63.1	25.7	48.0
Female	58.5	36.9	74.3	52.0
Nationality ***
Belgian	46.2	44.4	69.3	50.6
Non-Belgian	53.8	55.6	30.7	49.4
Type of contract ***
Teaching assistant	16.5	14.7	13.9	15.3
Personal mandate	18.2	27.4	32.7	24.6
Project funding	30.1	48.0	37.6	39.0
Other or self-financed	35.2	9.9	15.8	21.1
Doctoral school
DSh	–	–	–	40.1
NSE	–	–	–	42.8
LSM	–	–	–	17.1

Note: Significance level: ***p < 0.001.