Inequality of opportunity in selection procedures limits diversity in higher education: An intersectional study of Dutch selective higher education programs

RESEARCH ARTICLE Inequality of opportunity in selection procedures limits diversity in higher education: An intersectional study of Dutch selective higher education programs Lianne MulderID1,2*, Eddymurphy U. Akwiwu3, Jos W. R. Twisk3, Andries S. Koster4, Jan Hindrik Ravesloot5, Gerda Croiset6, Rashmi A. KusurkarID1,2,7, Anouk Wouters1,2 1 Amsterdam UMC Location Vrije Universiteit Amsterdam, Research in Education, Amsterdam, The Netherlands, 2 LEARN! Research Institute for Learning and Education, Faculty of Psychology and Education, Vrije Universiteit Amsterdam, Amsterdam, The Netherlands, 3 Amsterdam UMC, Vrije Universiteit Amsterdam, Epidemiology and Data Science, Amsterdam Public Health, Amsterdam, The Netherlands, 4 Department of Pharmaceutical Sciences, Utrecht University, Utrecht, The Netherlands, 5 Department of Medical Biology, Amsterdam UMC, University of Amsterdam, Amsterdam, Netherlands, 6 Wenckebach Institute for Education and Training, University Medical Center Groningen, Groningen, The Netherlands, 7 Amsterdam Public Health, Quality of Care, Amsterdam, The Netherlands * l.m.a.mulder@amsterdamumc.nl Abstract Selection for higher education (HE) programs may hinder equal opportunities for applicants and thereby reduce student diversity and representativeness. However, variables which could play a role in inequality of opportunity are often studied separately from each other. Therefore, this retrospective cohort study conducts an innovative intersectional analysis of the inequality of opportunity in admissions to selective HE programs. Using a combination of

multivariable logistic regression analyses and descriptive statistics, we aimed to investigate 1) the representativeness of student populations of selective HE programs, as compared to both the applicant pool and the demographics of the age cohort; 2) the demographic back- ground variables which are associated with an applicant’s odds of admission; and 3) the intersectional acceptance rates of applicants with all, some or none of the background char- acteristics positively associated with odds of admission. The study focused on all selective HE programs (n = 96) in The Netherlands in 2019 and 2020, using Studielink applicant data (N = 85,839) and Statistics Netherlands microdata of ten background characteristics. The results show that student diversity in selective HE programs is limited, partly due to the wide- spread inequality of opportunity in the selection procedures, and partly due to self-selection. Out of all ten variables, migration background was most often (negatively) associated with the odds of receiving an offer of admission. The intersectional analyses provide detailed insight into how (dis)advantage has different effects for different groups. We therefore rec- ommend the implementation of equitable admissions procedures which take intersectional- ity into account. PLOS ONE PLOS ONE | https://doi.org/10.1371/journal.pone.0292805 October 13, 2023 1 / 24 a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Mulder L, Akwiwu EU, Twisk JWR, Koster AS, Ravesloot J, Croiset G, et al. (2023) Inequality

of opportunity in selection procedures limits diversity in higher education: An intersectional study of Dutch selective higher education programs. PLoS ONE 18(10): e0292805. https:// doi.org/10.1371/journal.pone.0292805 Editor: Stefan Grosek, University Medical Centre Ljubljana (UMCL) / Faculty of Medicine, University Ljubljana (FM,UL), SLOVENIA Received: March 7, 2023 Accepted: September 28, 2023 Published: October 13, 2023 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0292805 Copyright: © 2023 Mulder et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: The OSF link (https:// osf.io/rwhg3/?view_only=

Introduction Globally, the odds of entering higher education (HE) are neither divided equally nor equitably amongst all adolescents [1, 2]. There are several sociodemographic characteristics which, around the world, can result in barriers to accessing HE. These relate to socio-economic status (SES), ethnic/migration background, parental education and occupation, age, sex and having an urban versus rural background. However, these factors are often studied separately from each other, instead of combining elements of applicants’ identities in one analysis (e.g., analyz- ing outcomes by combining sex, ethnic background, and age). Therefore, this retrospective cohort study conducts an innovative intersectional analysis of the inequality of opportunity in admissions to selective HE programs. By doing so, we aim to improve intersectional insights into educational (dis)advantage in selective admissions, which affects the diversity of the matriculating student population. As a diverse and representative student population is essen- tial to achieving a better quality of education [3–5], stimulating improved representation is crucial. Furthermore, a diverse classroom prepares students for their role as professional in a labor market where they will work with patients, customers, clients, and colleagues from a wide range of backgrounds [6]. Additionally, for certain labor sectors, such as healthcare, it is essential to educate a diverse workforce to be able to provide the best care to a diverse patient population [7].

Intersectionality is a theory developed by Kimberle´ Crenshaw [8] which holds that identi- ties are multi-layered. On each layer, a person can occupy a position which is privileged in their particular society and seen as ‘the Norm’, or a position which is disadvantaged, and seen as the non-normative ‘Other’. This places an individual on multiple hierarchical axes of privi- lege/oppression relating to social structures [8–10]. Each axis, therefore, may relate to struc- tural advantages or disadvantages of students in the educational sphere. From an intersectional point of view, a person’s experience cannot be understood sufficiently by consid- ering these layers or axes independently [11]. Instead, the combination of their position on these axes influences a person’s experience and opportunities in a given context. Using inter- sectionality theory to analyze the inequality of opportunity in HE selection is important, because applicants to selective HE programs can belong to multiple groups which are disad- vantaged in the education system. For example, they may have a lower SES and an underrepre- sented migration background and be ‘social newcomers’ [12] in their chosen field of study (meaning: no social ties such as parents in that field). These applicants could thereby face mul- tiple barriers in the route to entering a selective HE program, such as in the case of health pro- fessions education [13]. Similarly, applicants without a migration background, from high SES families, who are ‘insiders’ in the field (e.g. they have parents who practice the same profession

as the program is intended to train the student in), may experience accumulated advantages which are not earned on the basis of their own merit [14]. However, the size and significance of these combined (dis)advantages are unknown. This study aims to fill that gap in the literature. Background characteristics and educational inequality On the level of individual background characteristics, there are several general patterns of edu- cational inequities. For example, students from socio-economically disadvantaged backgrounds are known to face numerous obstacles on the route to HE [15], including disadvantages in com- petitive selection procedures [3, 16, 17]. Applicants from a low SES background are often already underrepresented in the applicant pools of competitive HE programs [18]. There, they find themselves contending against applicants from high SES backgrounds, who had more resources to adequately prepare them [1]. Oftentimes low SES applicants lose this competition, PLOS ONE Inequality of opportunity in Dutch selection procedures limits diversity in higher education PLOS ONE | https://doi.org/10.1371/journal.pone.0292805 October 13, 2023 2 / 24 7114b77c9f8b4062ab31f885ce331a65) is now public will remain public forever. The rest of the data availability statement is accurate. Funding: This work was supported by a Microdata Access Discount award from ODISSEI, the Open Data Infrastructure for Social Science and Economic Innovations (https://ror.org/03m8v6t10) (awarded to LM). The funders had no role in study design, data collection and analysis, decision to

publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist.

not because they are less qualified, but because the selection procedures are (unintentionally) biased against them [19–21]. This is a major problem, as socio-economic conditions are likely the most important underlying factor influencing educational inequalities [1, 22]. Parental occupations may also influence odds of HE participation [1] or odds of admission in a selection process. For example, students with at least one parent in a Science, Technology, Engineering, and Mathematics (STEM) related job show higher levels of STEM interest, moti- vation and achievement, and choice to major in a STEM field [23]. Their higher achievement levels [23] are likely to result in higher odds of admission in selection procedures which give weight to previous scholastic achievement. Another example are the higher odds of admission to health professions education programs for applicants with at least one parent who is a regis- tered healthcare professional [13]. Interviews with high school students who were interested in health professions education programs also showed how a social network connection in the medical field helped students to develop their interest for such programs, and why a network connection was the most important facilitator in preparing for the selection procedure [24]. Furthermore, having an underrepresented ethnic or migration background is associated with educational disadvantage and barriers to optimal achievement due to e.g. microaggres- sions and racism [25], lower expectations by the education system [26], the effects of testing

language [27], and differential access to high-quality and highly-resourced schools [3, 26, 28]. As a consequence, in many countries, applicants with such backgrounds have lower odds of admission than their white peers without a migration background [13, 29–31]. Finally, rural applicants may be at a disadvantage in entering HE. For example, Canadian research has reported lower performance in a multiple mini interview of applicants graduating from a rural high school [32]. In China, educational stratification is mainly driven by a rural- urban divide in which rural schools are disadvantaged in numerous ways. For example, they receive lower levels of funding and have poorer teacher qualifications, negatively affecting school enrollment and educational attainment [33]. Elite universities have witnessed an increasing rural-urban enrollment disparity: In the 1970s, one in two first-year students were from a rural background; whereas in 2011, this was one in seven [34]. Altogether, the aforementioned examples of unequal access to HE represent a major prob- lem for educational quality, as a diverse, representative student population has been shown to be associated with better educational experiences for students [3–5]. Furthermore, disadvan- taging applicants who are already underrepresented in HE reproduces societal and educational inequities [14, 21]. Admissions committees should use valid selection methods [35], but research by the Dutch Inspectorate of Education suggests that this is not always the case [36].

However, it is unknown what the effects are of belonging to multiple (dis)advantaged groups at the same time. This study aims to address the gap in the literature of the intersectional (dis) advantages which applicants face in accessing selective programs in HE, based on nationwide data from The Netherlands. The context: Higher education in The Netherlands In The Netherlands, the typical route to HE consists of several steps. In the transition from ele- mentary to high school, around ages 11–12, pupils are placed in either the pre-vocational (vmbo, 4 years), higher general (havo, 5 years), or pre-university (vwo, 6 years) track. Research has shown that there is inequality of opportunity in the placement into these tracks, and that the capabilities and potential of certain groups of children (e.g. lower SES, with certain migra- tion backgrounds, from rural areas) are underestimated [37–39]. Although transitioning between tracks is possible, this is an exception [26], and the rates at which this happens differ greatly between provinces [40]. This means that for many children, their educational route is PLOS ONE Inequality of opportunity in Dutch selection procedures limits diversity in higher education PLOS ONE | https://doi.org/10.1371/journal.pone.0292805 October 13, 2023 3 / 24

determined at an age which is lower than in many other countries [39]. This contributes to inequality of opportunity [39]. In the final year of high school, pupils can apply to post-secondary education programs. Most higher general pupils enter HBO, but they could also enter vocational education. Most pre-university pupils enter university, but they too have the option to enter HBO or vocational education. Of these post-secondary education programs, only higher education institutions (HEIs) use selection to admit students. These include higher professional education (in Dutch: HBO–often referred to as university of applied sciences), and university. For sake of readabil- ity, we have used the Dutch abbreviation HBO in this manuscript. Within HEIs, most study programs are not selective, but programs may use specific eligibil- ity criteria. Applicants with other (non-traditional) educational backgrounds than specified in the eligibility criteria must prove that they meet equivalent educational levels [41]. For exam- ple, for many HBO programs, it is possible to apply with either a higher general diploma or a vocational education diploma. For many university programs, it is possible to apply with either a pre-university diploma, or with a completion certificate of all first-year courses of a (related) HBO study program called propedeuse. Some study programs have to use selection procedures, due to capacity limitations called numerus fixus (fixed number). When there are more eligible applicants than seats, institutions

must use at least two types of qualitative selection criteria (e.g. intelligence, motivation, study skills) to determine who will be offered a place in the program [42]. In the selection procedure, diversity may be lost, which could result in underrepresentation of certain groups. However, apart from four university health professions programs for which inequalities based on sex, migration background, socio-economic status and having healthcare professionals as parents were found [13], it is unknown to what extent this is the case across the entire HE domain. Therefore, we wanted to investigate the possible inequality of opportunity in all selective HE programs. We hypothesized that there could be study programs with less or no inequality of opportunity, and programs in which underrepresented applicants have higher odds of admis- sion. Programs with proven inequality of opportunity in the selection procedure, could then learn from the programs who offer equal or equitable opportunities to applicants. Additionally, based on the importance of a representative workforce in certain labor markets (e.g. health- care), we were interested in whether there are study programs where potential best practices could be found in admitting underrepresented applicant groups. Thus, we aimed to answer the following research questions: 1. How representative are student populations of selective HE programs, compared to: a) the applicant pool; b) the higher general population and pre-university population; and c) their age cohort?

2. Which demographic background variables (e.g. sex, migration background, year of birth, socio-economic status, parental work sector) are associated with an applicant’s odds of admission into selective HE programs? 3. What are, from an intersectional perspective, the acceptance rates of applicants with all, some or none of the background characteristics positively associated with odds of admission? PLOS ONE Inequality of opportunity in Dutch selection procedures limits diversity in higher education PLOS ONE | https://doi.org/10.1371/journal.pone.0292805 October 13, 2023 4 / 24

Methods For this study, we used the STROBE cohort reporting guidelines [43] and the SAMPL guide- lines [44]. These are statistical and methodological guidelines suited to the type of study we executed. Study design, setting and eligibility criteria, study size We conducted a retrospective multi-cohort study using anonymized non-public microdata from Statistics Netherlands, focusing on applicants and students in academic years 2019–2020 and 2020–2021. These years were chosen, as they were the two most recent years for which essential variables were available in Statistics Netherlands microdata at the start of the research. Statistics Netherlands does not allow for analysis on single-institution data. Therefore, we cre- ated several clusters of studies which each consisted of at least three institutions, to make data non-traceable to any particular institution. In cases where different studies had to be combined to achieve this, clusters were made on the basis of similarity of programs (e.g., related to per- sonal healthcare or not). Table 1 contains information on the study programs in all HBO clus- ters (starting with ‘SH’) and all university clusters (starting with ‘SU’). The six HBO clusters included 19 HBO study programs, the nine university clusters included 28 study programs. We only included study programs taught at HEIs which are pub- licly funded and which are members of either the Netherlands Association of Universities of Applied Sciences (Vereniging Hogescholen) or Universities of the Netherlands (Universiteiten

van Nederland), as these are the organizations which granted us permission to perform data analysis on their anonymized applicant pool and student population. It is important to note that many selective programs (e.g. Nursing) are only selective at one or a few HEIs, but that students can enroll in these programs at other HEIs without having to go through a selection procedure. Therefore, S1 Table also contains information on the number of HEIs that did not use selection for the particular study programs. Since this research focuses on inequality of opportunity in selection, the student populations of non-selective programs are not included in our analyses. We only used the term ‘selective program’ for programs which have a capacity limit and use selection to decide which applicants to admit for the limited places available. We created the following groups: Applicants for selective programs (N = 85,839). This group consists of all domestic applicants who applied through Studielink (the national online application portal for post-sec- ondary education in The Netherlands) for at least one selective program, and who received a ranking number. The group is divided into HBO applicants and university applicants. Appli- cants who never received a ranking number (e.g., because they decided to withdraw from the selection procedure, or were not eligible) were not included, since the selection committee was not able to offer them a place (see Fig 1). Age cohort (N = 204,075). To compare the demographics of study clusters with the

demographic characteristics of the young population, we used Statistics Netherlands micro- data of all 16-year-olds who were registered in The Netherlands on October 1st in 2015. Within the age cohort, we also distinguished whether 16-year-olds were higher general or pre-univer- sity students, since these are the most common educational routes to HBO and university. Data sources Studielink provided all applicant data of all selective HE programs from 2019 and 2020 to Sta- tistics Netherlands, which pseudonymized the Citizen Service Numbers (CSN) of domestic applicants. The researchers used pseudonymized CSN to merge other Statistics Netherlands microdata to the dataset. International applicants were excluded from this research. PLOS ONE Inequality of opportunity in Dutch selection procedures limits diversity in higher education PLOS ONE | https://doi.org/10.1371/journal.pone.0292805 October 13, 2023 5 / 24

Table 1. Information on clusters of study programs. Cluster HBO program HBOs using selection Average acceptance rates 2019 Average acceptance rate 2020 SH1 Allied Medical Care 3 52.9% (1028/1942) 52.2% (994/1904) Nursing 1 Midwifery 3 SH2 Dental Hygiene 4 36.5% (539/1477) 38.6% (529/1372) Denturism 1 (2020) Optometry 1 SH3 Biology and Medical Laboratory Research 2 (2019), 1 (2020) 72.6% (864/1190) 64.6% (674/1044) Forensic Science 2 Medical Imaging and Radiation Therapy 1 SH4 Physiotherapy 6 78.5% (2263/2881) 74.1% (2214/2987) Psychomotoric Therapy/Psychomotricity 1 Sport Studies 1 SH5 Creative Media and Game Technologies 1 68.9% (1081/1570) 70.0% (996/1422) Fashion & Textile Technologies 1 Industrial Design Engineering 1 Art and Economics 1 (2020) SH6 Applied Psychology 4 63.8% (2041/3199) 54.5% (1719/3157) Applied Biology 1 Skin Therapy 2 Cluster University program Universities using selection SU1 Medicine 8 45.4% (2986/6571) 43.0% (2969/6912) SU2 Dentistry 4 41.6% (717/1725) 38.1% (643/1689) Pharmacy 1 SU3 Psychobiology 1 85.1% (4631/5439) 75.8% (6845/9030) Psychology 6 (2019), 9 (2020) SU4 Biomedical Sciences 4 57.4% (1391/2425) 48.1% (1068/2219) Biomedical Engineering 1 (2019) Clinical Technology 2 SU5 Biology 1 59.3% (902/1521) 46.9% (555/1184) Biotechnology 1 Nutrition and Health 1 Veterinary Medicine 1 Nanobiology 1 SU6 Artificial Intelligence 4 81.1% (1421/1753) 86.2% (1469/1704) Industrial Design 2 SU7 Architecture, Urbanism & Building Sciences 2 74.5% (2838/3810) 68.9% (2411/3498)

Ethics and replicability. The research project was based on anonymized data from Statis- tics Netherlands and Studielink. Therefore, participant consent was not required. Statistics Netherlands microdata and Studielink data are non-public. The statistical results comply to all Statistics Netherlands privacy regulations and the Dutch law regarding use of their non-public microdata (Wet op het Centraal Bureau voor de Statistiek, 2004). Selected data from Studielink of the applicant pool was anonymized by Statistics Netherlands. The researchers had no access to identifiable information. This study could be replicated if access to the datasets is gained. Our research protocol, including all procedures, sources of variables, and software syntax for statistical analysis, can be found in S2 File and on https://osf.io/rwhg3/? view_only = 7114b77c9f8b4062ab31f885ce331a65. Two government documents with data on HEIs and study programs were used to cross-check the accuracy of cluster sizes (number of institutions per cluster) in S1 Table. These can be found in the same database. Data of small groups. Statistics Netherlands regulations prescribe that descriptive statis- tics and analyses performed on groups smaller than 10 persons cannot be published in detail. As a consequence, we had to replace frequencies between 0–4 by ‘<5’ and frequencies between 5–9 by ‘<10’, to avoid traceability to an individual or small groups of individuals. If one Fig 1. Study size and exclusion criteria. https://doi.org/10.1371/journal.pone.0292805.g001 PLOS ONE

Inequality of opportunity in Dutch selection procedures limits diversity in higher education PLOS ONE | https://doi.org/10.1371/journal.pone.0292805 October 13, 2023 7 / 24

category within a variable had <5 or <10 persons, the next smallest category also had to be rounded to the nearest pentad (e.g. <15, <20). Variables. The study includes the following applicant characteristics: sex, year of birth, migration background, country of birth, and degree of urbanity of postal code of address in 2015. It also includes the following applicants’ parental data: Income percentile of parent with highest income; household assets percentile; number of parents who receive social welfare; number of parents who receive a social services income (excl. social welfare); number of parents who are registered healthcare professionals (HP) according to the official Dutch BIG- register; and the number of parents who work in the primary, secondary or vocational educa- tion sector. Details on these variables are described in Table 2. The Studielink dataset also con- tained information on whether an applicant had received an offer of admission, whether this offer was accepted, and whether an applicant had enrolled in the program by October 1st in the year of application. Bias. We followed Sˇimundić’s [54] classifications of bias in research to study potential sources of bias in advance of the analyses. By including all selective programs in HE, and an age cohort of all 16-year-olds on October 1st, 2015 as reference group for the demographics of the young population, we aimed to eliminate potential bias in sampling. Statistical analysis. To determine the representativeness of each cluster’s student popula-

tion compared to the applicant pools, age cohort, and all first-year students in HBO and uni- versity, we used frequency tables to compare the distribution of students on all variables. Not all applicants who received a placement offer became students in that program. There can be several reasons for this. For example, they may have received an offer of admission in two dif- ferent study clusters, and only accepted one. They could also decide to enter a non-selective study, or refrain from entering HE. It is also possible that after a student accepted an offer of placement, they failed their final exams in high school. This excludes them from entering HE. After that, we examined data for evidence of multicollinearity amongst the independent variables using both the Pearson correlation coefficients between variables, and the variance inflation factor (VIF) of each variable. We then performed univariable logistic regression anal- yses to examine which of the independent variables were associated with being offered admis- sion (the dependent variable). The results are shown with odds ratios (ORs) and the corresponding 95% confidence intervals (95% CIs). Statistical level of significance was set at 0.05. Multivariable logistic regression was used to create a model for each cluster in each aca- demic year. This was done to investigate the effect of background variables of applicants on the odds of receiving an offer of placement, while adjusting for the other variables in the model. The results are shown with adjusted ORs and 95% CIs. We did not apply forward or

backward selection. The regression analyses were performed on the applicants who received a ranking number. Next, we used each cluster’s multivariable model to create an intersectional analysis of the admission rates of two groups of applicants within that cluster: 1) applicants who possessed all demographic attributes which were significantly positively associated with the odds of being offered admission in the final model of that cluster (referred to as group ALL); and 2) appli- cants who possessed none of these demographic attributes (group NONE). If a variable had 3 categories (e.g. A, B, C) and only category B was shown to have significantly different odds of admission compared to reference group A, additional regression analyses were performed with reversed reference categories, to investigate if B and C also significantly differed from each other. The double analyses with different reference categories provided the information which of the three categories had significantly higher odds of admission. For group ALL and NONE, we calculated the percentage of applicants which were offered admission. The comparison between these groups (and the ratio of the acceptance rate of PLOS ONE Inequality of opportunity in Dutch selection procedures limits diversity in higher education PLOS ONE | https://doi.org/10.1371/journal.pone.0292805 October 13, 2023 8 / 24

Table 2. Demographic data recorded for each applicant and their parents. Variable Values Rationale Appplicant data Sex 0 = Male The known male:female ratio differences in some fields of study [45] 1 = Femalea Year of birth Recoded in 3 categories to create groups of sufficient size for logistic regression analysis: The known higher performance of older students (>21 years), compared to younger students, in Dutch medical education [46]. 0 = 1999 or earlier 1 = 2000–2001 2 = 2002 or later Migration background Statistics Netherlands original values are based on country of birth of the person, or the country of birth of parents. Each country has its own value, which we recoded in 5 categories for the descriptive statistics: Earlier research showed the inequality of opportunity in selection to Dutch university health professions education programs for applicants with a Turkish, Moroccan, Surinamese and Dutch Caribbean migration background [13]. In this research, applicants with an Indonesian migration background constituted a group of similar size, with a similar (post-)colonial migration history as Surinamese and Dutch Caribbean migrants. 0 = No migration background 1 = Turkish/Moroccan migration background 2 = Surinamese/Dutch Caribbean/Indonesian migration background 3 = EU/EEA/Swiss (European) migration background 4 = Other migration background. These were recoded in 3 categories for the regression analyses: 0 = No migration background 1 = Turkish/Moroccan/Surinamese/Dutch Caribbean/ Indonesian (TMSDI) migration background 2 = Other migration background (OMB) migration

background Country of birth 0 = Born in The Netherlands 1 = Born abroad Degree of urbanity of postal code, of address in 2015 Based on average number of addresses per km2 of the area: Urban areas may offer more opportunities to build one’s CV and engage in extracurricular activities, which may be (an element of) a selection instrument used in The Netherlands [47]. A Canadian study showed lower performance in a multiple mini interview of applicants who graduated from a rural high school [32]. 1. Very strong (2500 or more) 2. Strong (1500–2499); 3. Average (1000–1499); 4. Weak (500–999); 5. Not at all (less than 500) Recoded in 3 categories to create groups of sufficient size for logistic regression analysis: 0 = (Very) strongly urban 1 = Averagely urban 2 = Weakly/not urban Parental data Income percentile of parent with highest income Scale of 1–100, recoded in 3 categories to create groups of sufficient size for logistic regression analysis: The known barriers of low SES in the education field in general or in access to higher education, and the disproportionate share of students from high-income families among health professions education students [48–50]. Income and assets percentiles, rather than their values in euros, were included because percentiles indicate the relative position one occupies compared to the rest of the population. 0 = Percentile 1–70 1 = Percentile 71–90 2 = Percentile 91–100 Household assets percentile b Scale of 1–100, recoded in 3 categories to create groups of sufficient size for logistic regression analysis: 0 = Percentiles 1–40

1 = Percentiles 41–80 2 = Percentiles 81–100 (Continued) PLOS ONE Inequality of opportunity in Dutch selection procedures limits diversity in higher education PLOS ONE | https://doi.org/10.1371/journal.pone.0292805 October 13, 2023 9 / 24

group ALL divided by that of group NONE) shows the contrast in opportunities for the groups that differed the most from each other. To provide insight into the acceptance rates of appli- cants who had some, but not all, of the characteristics positively associated with odds of admis- sion, we created figures which showcase these rates for every combination of characteristics. All analyses were performed using IBM SPSS software for Windows, Version 25.0 (IBM Corp, Armonk, NY). Results Representativeness S3 Table summarizes the demographic data of all applicants who received a ranking number, all applicants who were offered a seat in the program, and all those who became registered stu- dents, for each cluster. With these results, we can answer research question 1, on how repre- sentative student populations of selective HE programs are, compared to: a) the applicant pool; b) the higher general population and pre-university population; and c) their age cohort. Due to the extensive amount of descriptive statistics, we focus here on the variables with the largest differences between categories. Sex. The results show that when student populations are not representative of their age cohort or the higher general/pre-university population, this is partially due to self-selection in applications. For example, in SH1 (healthcare related), men constituted 13–14.5% of the stu- dent population in 2019, respectively 2020. This is primarily due to them applying in lower numbers (approximately 16% of the applicant pool). Most clusters in which men are underrep-

resented, concern health professions education programs. SU7 (mostly consisting of engineer- ing programs) was the only cluster in which women were significantly underrepresented: they made up 28–33% of the student population, but also only 28–31% of the applicant pool. Table 2. (Continued) Variable Values Rationale Number of parents who receive social welfare 0, 1, 2, recoded in 2 categories to create groups of sufficient size for logistic regression analysis: The low disposable income of families on social welfare, and the lack of a current network in the workforce, are hypothesized to have a negative effect on the child’s opportunity to enter a selective program. 0 = 0 parents 1 = 1 or 2 parents Number of parents who receive a social services income (excl. social welfare) 0, 1, 2, recoded in 2 categories to create groups of sufficient size for logistic regression analysis: There are different types of social services income, including types that are based on long-term illness or a labor disability. Combined with the lack of a current network in the workforce, these factors are hypothesized to have a negative effect on the child’s opportunity to enter a selective program. 0 = 0 parents 1 = 1 or 2 parents Number of parents who are registered healthcare professionals 0, 1, 2, recoded in 2 categories to create groups of sufficient size for logistic regression analysis: The known influence of having a network in the medical field as a facilitator in preparing for selection in health professions education programs [15, 16, 51–53] 0 = 0 parents 1 = 1 or 2 parents

Number of parents who work in the primary, secondary or vocational education sector 0, 1, 2, recoded in 2 categories to create groups of sufficient size for logistic regression analysis: We hypothesized that parents who work in the primary, secondary or vocational education sector can provide access to information about higher education and selection procedures. 0 = 0 parents 1 = 1 or 2 parents a It is acknowledged that not every individual is, or identifies as, ‘male’ or ‘female’, but Statistics Netherlands only has two possible sex categories. This means that e.g. intersex persons either have missing data on their sex registration, or are categorized as male or female. Transgender persons who have changed their sex registration in the national Personal Records Database are included in this study according to the sex which was registered in 2021. b When parents live in different households, they may each have a different household assets percentile. In that case, we selected the highest percentile. https://doi.org/10.1371/journal.pone.0292805.t002 PLOS ONE Inequality of opportunity in Dutch selection procedures limits diversity in higher education PLOS ONE | https://doi.org/10.1371/journal.pone.0292805 October 13, 2023 10 / 24

Socio-economic status indicators. When it comes to parental assets category, we see that compared to the age cohort, applicants and students from percentile category 1–40 (the 40% of households with the lowest assets levels) are underrepresented in every cluster except SH2. This is also the case when compared to the eligible pool (higher general for HBO programs, pre-university for university programs), except in SH2, SH6 and SU2. The same can be observed for parental income category: compared to the age cohort, per- centile category 1–70 is underrepresented in every cluster except SH2 and SU2. Compared to the eligible pool, students from this income category are underrepresented in every cluster except SH2, SH3_2019, SH6, SU1, and SU2. This means that in the majority of clusters, appli- cants and students who have parents belonging to the top-30% of the income distribution are disproportionately present. This is especially the case for cluster SU4 (Biomedical Sciences, Biomedical Engineering & Clinical Technology), where more than 85% of students have parents in the top-30% income category, and cluster SU1 (Medicine), where this is the case for at least 78% of students. Compared to the age cohort, students with parents who receive social welfare are underrep- resented in all clusters except SH2 and SU2. Relative to the eligible pool, they are underrepre- sented in all clusters except SH2, SH3, SH6, SU1, SU2, and SU3_2020. Migration background. Compared to the age cohort, students without a migration back-

ground are overrepresented in SH1, SH4, SU4, and SU5. Students with certain migration back- grounds, such as Turkish or Moroccan, Surinamese, Dutch Caribbean or Indonesian, are underrepresented in the majority of clusters. This pattern is less evident when compared to the eligible pools. Students with an EU/EEA/Swiss migration background are overrepresented in most university clusters, both compared to the age cohort and the pre-university population. Finally, students with an Other migration background are sometimes (slightly) underrepre- sented, but more often overrepresented compared to the age cohort and eligible pools. Variables associated with odds of admission Research question 2 aimed to find out which demographic background variables are associated with an applicant’s odds of admission into selective HE programs. We excluded the variable ‘born in The Netherlands or abroad’ from our logistic analyses, as the variable ‘migration back- ground’ is also based on (parental) country of birth. We used ‘migration background’ as it is able to give more detailed information on someone’s background. However, the variable ‘born in The Netherlands or abroad’ remained in the descriptive statistics (S3 Table), to provide a better insight into those applicants who, according to the Statistics Netherlands definition, have a ‘migration background’ despite being born in The Netherlands. Their ‘migration back- ground’, as defined by Statistics Netherlands, is then based on the country of birth of at least

one of their parents. No evidence was found for multicollinearity in the remaining variables (the highest Pearson correlation was 0.389). S2 Table contains the univariable logistic regres- sion results for each cluster, and S4 Table gives the multivariable regression models. The univariable logistic regression results show that only in study cluster SU6_2020 (4 Arti- ficial Intelligence programs; 2 Industrial Design programs), none of the individual variables were significantly associated with the odds of admission. All other study clusters showed sig- nificantly different odds based on at least one variable. The variables most often associated with the odds of admission, were migration background (23 out of 30 clusters), assets category (20 out of 30 clusters), and income category, sex and urbanity degree (all 17 out of 30 clusters). The variable associated the least often, was having 1 or 2 parents working in the primary, sec- ondary or vocational education sector (6 out of 30 clusters). PLOS ONE Inequality of opportunity in Dutch selection procedures limits diversity in higher education PLOS ONE | https://doi.org/10.1371/journal.pone.0292805 October 13, 2023 11 / 24

After adjusting for all other variables, the picture of inequality of opportunity changes only slightly. The multivariable regression models in S4 Table show that the inequality of opportu- nity was widespread. SU6_2020 remained the only cluster that did not show significant differ- ences in odds of admission between applicants. The variables which most often influenced the odds of receiving an offer of placement, were migration background, sex, and age. However, the groups within these variables which had higher odds varied between clusters. For example, in seven out of 15 clusters, male applicants had higher odds, whereas in eight out of 15 clusters, women had higher odds. These mixed results were also visible for the urbanity degree of the postal code where applicants lived in 2015, as in some cases, those from (very) strongly urban areas had higher odds, whereas in other clusters, this was the case for applicants from weakly/ non urban areas. In clusters where year of birth mattered, younger applicants (born in 2000 or later) were usually at an advantage. Where groups are underrepresented based on sex, the inequality of opportunity in selection sometimes (slightly) compensates for their underrepresentation, whereas in others, it contrib- utes to it further. In some clusters, however, there is a loss of diversity after offers of admission have been made as well. For example, in SH5_2019, women make up 55% of the applicant pool, 62.4% of the applicants who have been offered admission, but only 50.3% of the enrolled

students. We do not have an explanation for this large difference. In almost all clusters where migration background was significant, applicants without a migration background had higher odds (in 18 clusters). However, in one cluster (SH3_2020), applicants with a Turkish, Moroccan, Surinamese, Dutch Caribbean, or Indonesian (TMSDI) migration background had higher odds, and in one cluster (SU5_2020), applicants with an Other migration background had higher odds than applicants without a migration back- ground. The multivariable models show that inequality based on migration background can- not be explained by the other nine variables included in the model: in the large majority of clusters, migration background remained significant after adjusting for those other back- ground characteristics of applicants. Analyses of the descriptive statistics shows that in 12 clusters where applicants without a migration background had higher odds, they were actually underrepresented in the applicant pool as compared to the higher general or pre-university population. Four of those clusters concerned health professions education programs. In some cases, the inequality of opportunity in admissions is the main reason for underrepresentation amongst students. For example, in SH1_2019 (Allied Medical Care, Nursing & Midwifery), 6% of applicants (117/1942) had a Turkish or Moroccan migration background, which is almost equal to their proportion in the age cohort (6.7%). However, these applicants only made up 1.3% of applicants offered admis-

sion (13/1028). Out of the 491 students who finally enrolled, less than 10 had a Turkish or Moroccan migration background. If enrolment rates had been equal to application rates, then there would have been 30 students in SH1_2019 with such a background. In clusters where parental assets category was significant, those from higher categories con- sistently had higher odds. This was also the case for having healthcare professional parents, and parents who worked in the education sector. Having parents with either a social welfare or other social services income was negatively associated with odds of admission twice. Intersectional acceptance rates To answer the third research question, we calculated the acceptance rates of applicants with either all, some or none of the background characteristics positively associated with odds of admission, based on the final regression models in S4 Table. Table 3 shows the results of this intersectional analysis for each study cluster. For example, the row of cluster SH1_2020 shows PLOS ONE Inequality of opportunity in Dutch selection procedures limits diversity in higher education PLOS ONE | https://doi.org/10.1371/journal.pone.0292805 October 13, 2023 12 / 24

Table 3. Intersectional analysis of the inequality of opportunity in Dutch higher education selection procedures: Comparison between group ALL* and group NONE**. Background characteristics associated with significantly higher odds of admission in each multivariable logistic regression model Cluster Sex Age, based on Y. O.B. Migration background Urbanity degree of postal code in 2015 Income category Assets category Parents on social welfare Parents with SSI HP parents Parents working in PSVE sector Acceptance rate of group ALL* Acceptance rate of group NONE ** Ratio accept. rate ALL/ NONE p- value of the ratioa Total dataset M 2000 or later None 71–100 41–100 None 1 or 2 69.7% (975/ 1398) 34.9% (290/ 832) 2,00 < 0.001 SH1_ 2019 F 1999 or earlier None AWNU 1–90 1 or 2 1 or 2 >97% (data hidden) b 11.1% (data hidden) b 8.74– 9.01 c c SH1_ 2020 F 2001 or earlier None AWNU 71–90 1 or 2 80% (20/25) 15.4% (data hidden) b 5,19 < 0.001 SH2_ 2019 None Averagely 81–100 1 or 2 >97% (data hidden) b 25% (158/ 632) 3.88– 4.00 c c SH2_ 2020 None 41–100 1 or 2 64.8% (46/ 71) 25% (126/ 503) 2,59 < 0.001 SH3_ 2019 AWNU 78.3% (517/ 660) 64.8% (327/ 505) 1,21 < 0.001 SH3_ 2020 TMSDI AWNU 64.3% (9/14) 57.2% (218/ 381) 1,12 0.599 SH4_ 2019 F VSU 71–100 89.8% (317/ 353) 69.7% (83/ 119) 1,29 < 0.001 SH4_ 2020 F 1999 or earlier VSAU 1 or 2 88.2% (30/ 34) 72.1% (320/ 444) 1,22 0.040 SH5_ 2019 F 41–100 79.7% (441/ 553) 49.5% (45/ 91) 1,61 < 0.001 SH5_ 2020 F 2000– 2001 None VSU 82.9% (87/ 105) 50% (12/24) 1,66 < 0.001 SH6_ 2019 M 2002 or later

None 41–100 1 or 2 60% (3/5) 40.1% (139/ 347) 1,50 0.367 SH6_ 2020 M None 81–100 64.7% (88/ 136) 42.4% (300/ 708) 1,53 < 0.001 SU1_ 2019 F 2002 or later None or Other MB 60% (54/90) 31.6% (78/ 247) 1,90 < 0.001 SU1_ 2020 41–100 44.7% (2344/ 5248) 34.3% (390/ 1136) 1,30 < 0.001 SU2_ 2019 2000– 2001 45.2% (484/ 1070 35.6% (233/ 655) 1,27 0.010 SU2_ 2020 2002 or later VSU 42.9% (109/ 254) 31.3% (142/ 454) 1,37 0.002 SU3_ 2019 None 71–90 89.8% (727/ 810) 81.8% (361/ 445) 1,10 < 0.001 SU3_ 2020 None 1 or 2 78.8% (688/ 873) 74.4% (2886/ 3877) 1,06 0.007 SU4_ 2019 2002 or later None 1–70 41–100 85.7% (6/7) 38.9% (14/ 36) 2,20 0.023 (Continued) PLOS ONE Inequality of opportunity in Dutch selection procedures limits diversity in higher education PLOS ONE | https://doi.org/10.1371/journal.pone.0292805 October 13, 2023 13 / 24

Table 3. (Continued) Background characteristics associated with significantly higher odds of admission in each multivariable logistic regression model Cluster Sex Age, based on Y. O.B. Migration background Urbanity degree of postal code in 2015 Income category Assets category Parents on social welfare Parents with SSI HP parents Parents working in PSVE sector Acceptance rate of group ALL* Acceptance rate of group NONE ** Ratio accept. rate ALL/ NONE p- value of the ratioa SU4_ 2020 None 41–100 50.9% (799/ 1571) 35.6% (48/ 135) 1,43 < 0.001 SU5_ 2019 M 2000 or later VSU 71–90 77.1% (27/ 35) 32.4% (34/ 105) 2,38 < 0.001 SU5_ 2020 M 2000 or later Other MB 75.6% (34/ 45) 26.1% (30/ 115) 2,90 < 0.001 SU6_ 2019 71–100 81.3% (1077/ 1324) 73.9% (136/ 184) 1,10 0.017 SU6_ 2020 NA NA NA SU7_ 2019 2000 or later None None 80.3% (1357/ 1689) 41.4% (12/ 29) 1,94 < 0.001 SU7_ 2020 F None 81–100 80.4% (270/ 336) 54.1 (170/ 314) 1,49 < 0.001 SU8_ 2019 None WNU 1 or 2 95.3% (41/ 43) 68.1% (222/ 326) 1,40 < 0.001 SU8_ 2020 None WNU 1–90 87.1% (101/ 116) 66% (103/ 156) 1,32 < 0.001 SU9_ 2019 M 2000 or later None 78.4% (344/ 439) 41.7% (15/ 36) 1,88 < 0.001 SU9_ 2020 M 2000 or later None or Other MB 41–100 None 1 or 2 76.7% (33/ 43) <3% (data hidden) b c c *Acceptance rate group ALL: calculated as the percentage of applicants who were offered a place, out of the total number of applicants in group ALL (Group ALL: the group of applicants who possessed all demographic characteristics positively associated with odds of admission in this cluster’s final model)

** Acceptance rate group NONE: calculated as the percentage of applicants who were offered a place, out of the total number of applicants in group NONE (Group NONE: the group of applicants who possessed none of the demographic characteristics positively associated with odds of admission in this cluster’s final model) a based on chi-square test of association with α < 0.05 b data hidden due to CBS data regulations regarding frequencies <10 c cannot be calculated or shown precisely, due to CBS data regulations regarding frequencies <10 M = Male; F = Female; Y.O.B. = Year of birth; SSI = Social services income, excluding social welfare; PSVE = primary, secondary or vocational education; HP = registered healthcare professional TMSDI = Turkish, Moroccan, Surinamese, Dutch Caribbean or Indonesian migration background; MB = Migration background; VSU = (Very) strongly urban; VSAU = (Very) strongly to averagely urban; AWNU = Averagely to weakly/not urban; WNU = Weakly/not urban; NA = not applicable (no significant variables in multivariable logistic regression model) Note 1: acceptance rates of applicants who had some, but not all, of the characteristics positively associated with odds of admission, fall between the rates of group ALL and group NONE. Due to the high amount of possible combinations of variables, it was not possible to showcase acceptance rates of each type of combination in this table. Information on the acceptance rates of other groups than ALL and NONE can be retrieved from the figures in S1 File. Note 2: If a variable has 3 categories (e.g. A,

B, C) and only category B was shown to have significantly different odds of admission compared to reference group A, additional regression analyses were performed with reversed reference categories, to investigate if B and C also significantly differed from each other. The double analyses with different reference categories provided the information which of the three categories had significantly higher odds of admission. https://doi.org/10.1371/journal.pone.0292805.t003 PLOS ONE Inequality of opportunity in Dutch selection procedures limits diversity in higher education PLOS ONE | https://doi.org/10.1371/journal.pone.0292805 October 13, 2023 14 / 24