The attractiveness of programmes in higher education: an empirical approach

ABSTRACT

Higher education institutions have experienced an increase in student enrolment over the past decades. At the same time, universities increasingly attempt to attract students by offering a variety of study programmes. Using a Dutch panel data set of 1300 programmes in 50 institutions, this study investigates what explains the attractiveness of study programmes. We hypothesize that the distance of study programmes plays a major role in student decisions to attend. Based on an instrumental variables identification strategy, we demonstrate that the closest distance between similar programmes offered and competition between programmes have significant effects on the enrolment of students in higher education. The results indicate that a one-kilometer increase in the closest distance between similar programmes decreases the number of students to enrol in a programme by – seven students after controlling for programme type and other characteristics.

KEYWORDS:

• Higher education
• attractiveness
• distance
• competition

1. Introduction

Higher education institutions in the Netherlands have experienced a number of significant changes in the decades prior to this study. The number of students entering universities has increased from 660,891 to 700,226 between 2012/2013 and 2014/2015 (Nuffic 2013, 2015). Moreover, universities in the Netherlands have taken measures to increase their efficiency by dropping duplicate programmes. Research universities and universities of applied sciences – which together constitute the higher education sector  –  compete in a competitive market (Soutar and Turner 2002), as funding for higher education is highly dependent on the number of students enrolled. Most of this competition is related to higher educational programmes. This raises the question of what factors affect the competitiveness of programmes. Apart from competition, assessments of quality and programme diversity have become more important in order ‘to understand students’ needs and how they choose their institution/programs' (Tavares et al. 2008, 107). Universities are required to meet student demands by for increased programme quality and diversity. One of the arguments from previous higher education literature is that higher education institutions with a more diversified system are supposed to be able to offer access to higher education to more students (Van Vught 2007). Each student is offered an equal opportunity to choose the best study programme for their success and to compete with other students of similar background. Diversity of programmes can be defined as ‘diversity among programs which emphases in subject or field, in level of academic degrees, in orientation, in quality and in forms of program delivery’ (Dill and Teixeira 2000, 100).

The Netherlands governments’ aim is to pressure their higher education institutions to become more accountable and more efficient in terms of all factors involved in determining a programme’s attractiveness. In this process, some factors are more important than others. Some of the factors likely to be considered highly important are the type of programme, the distance of the institution, and the competition between programmes. Improving these three factors should provide a good foundation that would appeal to the high school leavers’ market. Presently, the Netherlands’ higher education institutions are facing the challenge of determining their programmes’ attractiveness. In many countries the majority of university students attend a nearby institution. Proximity might sufficiently influence potential students to choose a certain institution/programme. Despite the significance of distance in determining a programme’s attractiveness, competition between programmes was expected to make higher education institutions more efficient and more effective in terms of programmes offered (Teixeira et al. 2012). No previous study exists on the relationship between competition and attractiveness of programmes, as the policy changes in recent years have focused instead on stimulating diversity in higher education. The present study therefore focused on the role of distance and competition between universities in determining programme attractiveness.

This paper examines what makes some study programmes more attractive than others by observing the determinants that influence student programme preferences (Soutar and Turner 2002). While the effects of distance variables on attractiveness of higher education institutions have been widely discussed, relatively little has been identified about whether and how distance affects the choice of a programme. A previous study analysing the relationship between distance and university choice showed that there is a strong negative relationship between the two (Frenette 2004). Apart from this, some economic studies used distance-to-university as an exogenous variable in their estimations. For instance, and in disagreement with the above indications, Spiess and Wrohlich (2010) found that distance has no significant effect on enrolment in higher education. Thus, whether distance affects student enrolment remains an open question. This study tests this question. The study also hypothesizes that competition between programmes has a positive effect on programme attractiveness to students. Researchers have pointed out that competition among higher education institution programmes has extensively influenced a wider range of student preferences. For example, higher education institutions that offer more of programmes that are similar will attract more students. In either case, however, it is not immediately clear how distance or competition would affect the enrolment of students. This paper is the first to test the following research hypothesis:

A small distance between institutions offering similar programmes and offering programmes that are competing with other similar programmes offered in other higher education institutions influences the attractiveness of programmes.

To account for the endogeneity in the relationship between distance and enrolment, an instrumental variable technique is used. Population density is used as an instrument, as it is a strong predictor of distance to the closest between higher education institutes. The hypothesis is tested in the following two-step empirical framework: (1) we estimate the influences of an increase in population density on the closest distance between similar programmes offered, and (2) we exploit the rich panel structure of the data and identify the influence of a close distance between similar programmes offered on programmes’ attractiveness to students by using the two-stage least squares (2SLS) estimator.

Compared to the existing literature, this paper contributes to current knowledge on the attractiveness of programmes in higher education in at least three ways. First, a review of the literature shows that attractiveness is commonly estimated at the institutional level. Conversely, this study focuses on the study programme level. By using this level of analysis, this study avoids inaccuracies that may arise, such as measurement error. Second, this study is able to estimate the influence of location (i.e. distance), competition and other characteristics on the attractiveness of a programme. While previous studies have mostly been conceptual or used descriptive statistics, this paper also looks at the associations between characteristics of programmes and their attractiveness. Third, using the richness of the data collected, this paper presents a more complete picture of programmes attractiveness over the period from 2008 to 2013. To the best of our knowledge, this is the first paper that studies the attractiveness of programmes using data in the Netherlands empirically.

The remainder of the paper is organized as follows. The next section presents a literature review on study attractiveness in higher education, distance, competition and programme characteristics. Section 3 presents the Dutch higher education setting. Section 4 presents the data and variables. Section 5 details the empirical framework, and Section 6 details the empirical results. The final section presents the conclusion of the study.

2. Literature review

This section highlights studies that focus on the determinants of programme attractiveness to students. We combine this literature review with studies on the influence of distance and competition. This literature section aims to explore the relationship between these variables (i.e. distance and competition) and the attractiveness of programmes.

2.1. A review of programme attractiveness

There are many ways to assess what makes a programme attractive to students. Shanka, Quintal, and Taylor (2006) studied factors that influence international students’ choice of education destination. Shanka, Quintal, and Taylor (2006) analysed international student’s choice of study destination by conducting a survey among international students in a major Australian higher education institution. They observed a significant influence from factors such as distance from student’s home to the city, safety, education quality, cost of living and tuition fees. They identified that ‘academic reputation, the variety of courses, the quality of education, campus safety, costs/fees and campus location include the number and variety of programs’ that represent major influences to student choice (p. 34). They designed a questionnaire in their ‘correspondences’ analysis based on ‘present literature’ (p. 36). Shanka, Quintal, and Taylor (2006) found that choice criteria have a significant effect on international student enrolment. Moreover, they concluded that higher education institutions should consider a ‘unique selling point’ for prospective students (p. 43). Soutar and Turner (2002) examined student university preferences using conjoint analysis to identify several factors that might influence student partiality. Their study indicated that course suitability, academic reputation, job prospects and teaching quality are the most highly ranked attributes. Soutar and Turner (2002) described several factors that might be considered by students when they choose a particular university, indicating that distance may be among them. Soutar and Turner (2002) found that ‘course suitability, academic reputation, job prospects and teaching quality’ are the most important determinants in university preference (p. 44).

There are similar studies that have described factors that influenced student educational programme choice. Pang and Appleton (2004) analysed the factors that influenced Chinese students to study in US higher education institutions. They found that the reputation of US academic programmes influences this choice. Pang and Appleton (2004) described a number of factors that influence the decision of students to study in the United States. The researchers found that one of the main factors is the academic reputation of US higher education institutions. Some students were not satisfied with the current local (Chinese) education system and decided to migrate to the United States. The findings of this study provide a first step to understanding the allure of specific higher education institutions. Beekhoven, De Jong, and Van Hout (2003) studied why some courses are more attractive than others and examining the composition of student populations, differences in curricula and individual student characteristics. They analysed the study progress of first-year students in various higher education courses in the Netherlands. A comparison was presented before and after controlling for several characteristics. Using multilevel analysis, the researchers showed that individual student characteristics and course characteristics only partially explain student choice. It was found that some courses were more attractive than others after controlling for individual characteristics. Beekhoven, De Jong, and Van Hout (2003) concluded that course characteristics only partially influence study progress for first-year students. Conard and Conard (2000) described that a college’s academic reputation plays a significant role in student choice to attend (p. 69). They found three factors that described the effects of academic reputation, including curricular concerns, exclusivity and career preparation. Moreover, Conard and Conard (2000) showed that in order to attract students (i.e. college-bound high school seniors), higher education institutions must engage in strategic planning. These institutions must encourage students to participate in career development and urge external organizations to participate in student recruitment. Conard and Conard (2000) also found that a targeted marketing strategy might enhance perceived academic reputation and student attractiveness.

In conclusion, some authors describe factors that influence the tendency of students to choose their programmes, while others arrive at the opposing conclusion that programme characteristics only partially affect student partiality.

2.2. Distance

Recent studies have found varying results on the importance of distance to student decision to apply to specific universities. In a study focusing on the distance between individual residents and the nearest universities, Kjellström and Regnér (1999) examined the influence of the geographical distance between resident homes and the nearest university to enrolment decisions. Their findings indicate that students are not concerned about how their residential locations compare to university sites. In another study conducted more recently, Denzler and Wolter (2011) examined how distance to a nearby university influences the probability of student enrolment. They found that distance to a university does not influence study choice among students from the highest socioeconomic group. These findings indicate that distance to a university is an element that influences differences in the cost of a university education.

Another strand of literature described how distance influences student choice on programmes in college or university. Sa, Florax, and Rietveld (2004) examine the determinants of university entrance for Dutch high school graduates in the year 2000. They investigate how a university, in terms of programme accessibility and quality, can attract students. They also use a production-constrained gravity model to investigate the impact of distance between regions and university region on attractiveness. They find that the choices of prospective university students are influenced by a distance deterrence effect and a downward rent effect, but poor educational quality of a university programme does have a negative impact. Frenette (2006) described the role of geographic distance in student decisions to attend university after high school graduation. He estimated the distance between the homes of high school students and the nearest university, and his findings indicated that students living ‘out-of-commuting distance’ are less likely to apply to the university than students living ‘within commuting distance’. In a very recent paper, Spiess and Wrohlich (2010) examined whether students who live near a university enrol in higher education. The researchers reported that a lower enrolment in college is caused by a greater distance from the nearest university. They showed that living nearby the university when students completed secondary school affects the decision to enrol in higher education. Griffith and Rothstein (2009) studied factors that affect low-income student enrolment in higher education institutes, focusing on decisions to apply for selective four-year colleges. They hypothesized that distance from students’ homes has an impact on enrolment applications, and their findings confirmed that distance influences student choice to apply to universities. Finally, Drewes and Michael (2006) indicated that students prefer to enrol in universities that are closer to their residence.

The studies reviewed above show a similar point of view regarding the influence of distance from individual residence to a college or university. Although distance to universities has often been mentioned as a factor in student enrolment choice, the closest distance between similar programmes offered has not been sufficiently specified. Differences between programmes offered have not been examined. This study takes the approach of analysing factors such as distance at study programme level and combines distance with other factors such as competition in order to obtain a more complete picture of what attracts students to universities (see Figure 1).

Figure 1. Conceptual model of the relationship between the independent (distance and competition) and dependent (attractiveness of programmes) with two empirical approaches.

2.3. Competition and programmes

Abramo, Cicero, and D’Angelo (2012) conducted a study assessing competition between universities. They measure and investigate the dispersion of research performance within and between Italian universities. They applied a Gini coefficient to measure the degree of concentration of performance by the researchers in the various disciplines. Indicators of productivity, such as the number of publications and the number of fields’ standardized citations, were used in their observation. Their overall main result indicated that the levels of concentration of performance within universities are high compared to the levels between universities.

Rossi (2009) offers another theoretical approach to the issue of university competition. He argued that strengthening competition is a strategic method of reducing diversity in the higher education system. To test this, Rossi (2009) used Italian data on specific disciplines for the period 1999/2000–2005/2006 and employed the Herfindhal index to measure competition and diversity. With respect to the difference between competition and diversity, he argued that higher levels of competition have an effect on university diversification strategies. The study concluded that there is a positive relationship between competition and diversity. However, this relationship is not robust.

The topic of programme diversity has not been discussed often in the literature. Goedegebuure et al. (1996) defined diversity in universities as the variety in the organization or products of higher education, differences among the programme or services provided by academic institutions, and differences among the types of institutions themselves. Birnbaum (1983) conducted the first study on the topic of programme diversity. According to Birnbaum (1983), ‘programmatic diversity relates to the degree level, degree area, comprehensiveness, mission and emphasis of programmes and services provided by institutions’ (as cited in Van Vught 2008, 152). Dill and Teixeira (2000) studied programme diversity from the economic perspective, defining and measuring academic diversity in terms of programme innovation. Marginson (1999) defined programme diversity as the variety in programmes or services, whether between institutions or within institutions.

3. Higher education in the Netherlands

Higher education in the Netherlands is organized as a binary system (De Boer, Enders, and Leisyte 2007). This paper focusses only on a group of 37 universities of applied sciences (‘Hogescholen’) and 13 research universities (‘Universiteiten’). While universities of applied sciences prepare students for the practice of a profession, research universities prepare students for independent scientific work in an academic or professional setting. This paper focuses on study programmes with a total of 1356 in universities of applied sciences and research universities are observed.

As mentioned briefly above, there are two higher education types in the Netherlands: higher professional education (‘Hoger beroepsonderwijs; HBO’) and scientific education (‘Wetenchappelijk onderwijs; WO’). Figure 2 provides an overview of the higher education system in the Netherlands. HBO offers four-year higher professional study programmes in a wide range of disciplines at the bachelor level and in the form of one- to two-year master programmes. WO offer three-year bachelor programmes, one- to two-year master programmes, and four year Ph.D./doctoral programmes (Canton and Jongbloed 2001).

Figure 2. The Netherlands higher education system.

Higher education in the Netherlands has become more international over the past decade, as Dutch institutes for higher education have designed more programmes for international students. All higher education institutions in the Netherlands receive funding for teaching and basic research from the government (Goedegebuure and Westerheijden 1991).

4. Data and descriptive statistics

4.1. Data

Data for this analysis was taken from the Dutch Ministry of Education (‘Dienst Uitvoering Onderwijs’ or DUO). They include information at the study programme level (e.g. business management and sociology) concerning the number of enrolled students in the period 2008–2013. However, data limitations influence the scope of the analysis. This study was unable to access information on individual characteristics, for example, the distance between students’ homes and universities, the quality of teaching, and the academic reputation of the higher education institutions. This study also has a limitation in that we could not make a distinction between programmes at different locations of the institutions. In this study, the study programme is simply defined as the set of courses that a student follows in order to obtain a degree (e.g. a Master of Science degree). The output variable is defined as follows. The number of enrolled students is the number of students in certain academic years enrolled in particular higher education institutions. The advantage of using the number of enrolled students at the study programme level in the period 2008–2013 is that this number can be used as a longitudinal data. This can be particularly important in assessing the role of distance between study programmes since students are most likely to select a study programme that is near to their home. The first step consisted of identifying a group of enrolled students based on study programme, higher education institutions and cities. The final sample consists of all students enrolled in the similar study programme at different higher education institutions and cities.

The second step is to create a database of the higher education institutions that graduating high school students could select. This database was constructed from the ‘Studiekeuze123’ website, which is a collaboration of the Dutch Ministry of Education, students, and higher education institutions. The website lists the addresses of higher education institutions as well as their postal codes in the Netherlands. The selection criteria necessitated that students be enrolled at higher education institutions that are in the DUO database. The database includes research universities and universities of applied sciences offering degree programmes from a wide of range of study programmes. In total, 50 of the original 233 higher education institutions were selected in the final sample. The 50 selected institutions were described according to the most recent data available on the DUO (2012–2013 academic years for any institution). To obtain our final sample, we combine the DUO and Studiekeuze123 data. We take a few restrictions into account. First, we focus on data in the period 2008–2013. We also focus on higher education institutions that offer the similar study programmes. Next, the sample is restricted to study programme. Thus, the final sample includes all similar study programmes offered in research universities and universities of applied sciences in the years 2008–2013.

To explore empirically the dynamics of distance in the attractiveness of programmes in higher education institutions, we develop a set of operational definitions of the concepts of distance. The distance definition builds upon many previous literatures that have argued that distance would affect the enrolment decisions of students. Various factors have influenced the decision to enrol in programmes in new higher education institutions (for comprehensive literature reviews, see Kjellström and Regnér 1999; Frenette 2006; Spiess and Wrohlich 2010; Denzler and Wolter 2011).

The next step is to calculate the closest distance between the similar study programmes offered. The geographic co-ordinates (latitude and longitude) of higher education institutions were derived and calculated from the postal codes of each study programme offered by each institution. We calculated the distance between similar study programmes (in km) formula by determining the shortest route between similar study programmes offered at the higher education institutions’ location.¹ To qualify as the closest distance between similar study programmes offered, the different higher education institutions have to offer similar study programmes. For example, a study programme such as psychology could be offered by three or four higher education institutions.

This study also addresses the concepts of competition between the similar programmes offered at higher education institutions. We intend to explore these concepts via an empirical framework that seeks to describe how competition takes place in higher education institutions. Higher education institutions have been increasingly competitive in an already competitive market to recruit new eligible students. Competition has become a well-known concept in the higher education fields, and many higher education scholars have contributed to the further theoretical conceptualization of competition processes (see, e.g. Rossi 2009; Teixeira et al. 2012). In this study, we obtained competition data by using two different indices – ̶ the Herfindhal index approach and the Gini coefficient. First, competition was measured using the Herfindahl index, which can be used to measure the market concentration ‘in a variety of contexts’ (Rhoades 1993, 188). It is useful in analysing the competitive effects of between study programmes in higher education institutions. In this study, we used the competition between similar study programmes offered, and we defined competition as the concentration index of the similar study programme offered at different higher education institutions. For the statistical measurement, the study used the enrolment of students in a similar study programme offered according to particular higher education institutions. The Herfindahl index can be constructed as(1) where is the number of enrolment students in a similar study programme offered in higher education institution . The expression denotes the total number of enrolment students in a similar study programme offered in higher education institution . The Herfindhal index varies from 0 to 1. If there is no competition, this is indicated by 0, and complete competition is indicated by 1. A low value for the Herfindhal index indicates that a similar study programmes offered in the higher education institutions experience less competition, while a high value index indicates that similar study programmes are offered in a highly competitive environment.

The second approach to measure competition by using the Gini coefficient. The Gini coefficient is a measure of inequality of distribution, and it can be used to compare similar programmes offered across different higher education institutions. The ratio has values between 0 and 1. We applied the Gini coefficient by calculating the total number of similar programmes in each institution and the number enrolment of students in each institution. The Gini coefficient is computed as(2) where is the cumulative proportion of the number enrolment of students in each institution, and denotes the cumulative proportion of the number of similar programmes offered across different higher education institutions. The Gini coefficient varies from o to 1, if corresponds to perfect equality (less competition between similar programmes offered), and corresponds to perfect inequality (high competition between similar programmes offered).

This data set contains information about the enrolment students, the closest distance between similar programmes offered, programme characteristics and competition between similar programmes offered. The final sample includes 1356 study programme within 50 different higher education institutions in 26 cities in the Netherlands.

4.2. Descriptive statistics

Table 1 presents descriptive statistics for the variables distance between institutions offering the similar programmes, competition between similar programmes offered at different higher education institutions, programme and other characteristics. The average distance between institutions offering the similar programmes is 71.03 kilometers. Competition between similar programmes offered (using the Herfindahl index has a lower average (4.66%) compared to other different indices, for example, the Gini coefficient, which yields 55.06%.

Table 1. Descriptive statistics.

Variable	Obs	Mean	Std. dev.	Min	Max
Location
Closest distance at closest between similar programme offered (km)	18,066	71.03	50.34	0	181.16
Programme Characteristics Type of Education (%)
Full time	18,066	22.74	41.92	0	100
Dual time	18,066	4.74	21.26	0	100
Part time	18,066	72.50	44.65	0	100
Type of Programmes (%)
Bachelor	18,066	30.38	45.99	0	100
Master	18,066	42.54	49.44	0	100
Other Characteristics Gender (%)
Male	18,066	50.54	49.99	0	100
Female	18,066	49.45	49.99	0	100
Competition between similar programmes offered (%)
Using Herfindahl index	18,066	4.66	6.89	0	100
Using Gini coefficient	18,066	55.06	23.09	0	100
Number of institutions	50
Number of study programmes	1356
Number of cities	26

Furthermore, this study controlled for programme and other characteristics such as the type of education, type of programmes, and gender composition. One aspect of this line of research observes the influence of programme characteristics (e.g. type of programmes and type of education) on the enrolment of students. Besides programme characteristics, we also accounted for other characteristics such as gender, which possibly influences our output variable. We use the percentage of male and female students as the percentages for gender. The assumption behind the use of gender is that study programmes with a majority of male or female students have a significantly higher number of enrolled students. Enrolled students are, on average, about 51% male; however, significant heterogeneity exists, as there are study programmes with only males and only females.

5. Empirical framework

5.1. The ordinary least square regression analysis

To measure the attractiveness of a programme, we specified an OLS model with the closest distance as an explanatory variable. Programme attractiveness to students is possibly influenced by the closest distance between institutions that offer similar programmes together with unobserved characteristics. Therefore, there is an association between the closest distance between institutions offering similar programmes and the number of student enrolments. In order to deal with unobserved characteristics, we used a fixed effect estimator.

In addition to the distance variable, we included competition between similar programmes offered by using differences indices such as the Herfindhal index and the Gini coefficient. This was done in order to understand the level of competition that occurs between similar programmes. Programme attractiveness to students is possibly influenced by competition between similar programmes offered. In addition, we included a dummy variable indicating whether type of programmes, type of education and gender influence the attractiveness of programmes. The regression model that is estimated was formulated as follows(3) where the outcome variable () is equal to the number of students in programme at institution in year . Variable is equal to competition between similar programmes. Variables constitute a vector of programmes and other characteristics such as type of programmes, type of education and gender. Variable depends on distance to closest similar programmes and competition between similar programmes . Equation (3) was estimated through OLS regression without control variables (1a) and subsequently with control variables (1b). Equation (3) is referred to as model (1) in Section 6. Each model clusters the standard error at the programme level. We controlled for the type of programme, expressed as bachelor’s degree and master’s degree. Variables were also included for type of education and gender. These variables are represented as part time, dual time, full time, male and female. All models included dummy variables for the city and the year in order to control for city and year fixed effects. As this study’s data are cross-sectional, these fixed effects were used to avoid potential omitted variable bias. By including these fixed effects, the study controlled for average differences across cities and year in any observable or unobservable predictor. This reduces the risk of omitted variable bias.

OLS has the possibility of yielding biased estimates, as the distance and competition between similar programmes between institutions could be correlated with other unobserved variables that also correlate with the outcome variable (the number of students enrolled). We therefore also applied a 2SLS.

5.2. 2SLS regression analysis

This study is generally interested in whether the model represents a relationship between the closest distance and programme attractiveness. It estimated a regression where the closest distance is the dependent variable and population density serves as the instrument. Population density is largely dependent on density in a 50-kilometer radius and density in provinces. Therefore it is strongly related to the closest distance between similar programmes. A two-stage estimation approach was applied. In the first regression, we used the closest distance between similar programmes offered as the dependent variable and the population density functions as our instrument. Our hypothesis is that more programmes are offered in more dense populations, and distance between similar programmes is shorter in these areas. There is a strong relationship between population density and the closest distance between similar programmes offered. The first stage regression model was estimated as follows(4) where, corresponds to the closest distance in institutions i in similar programme j in year t. represents the population density in year within city , which serves as the instrument. The parameter of interest is . All models were estimated without control variables (2a) and were consecutively estimated with the inclusion of several control variables (2b). The study controlled for programme and other characteristics such as type of programmes, type of education and gender. Each model contains dummy variables for city and year in order to control for city and year fixed effect. The equation is also estimated by replacing by the average population density in a 50-kilometer radius and the population by province, respectively. The study used population density in a 50-kilometer radius, which allowed us ‘to control the different categories of population density between urban and rural’ (Spiess and Wrohlich 2010, 6). The closest distance between similar programmes offered is dependent on the population density in a 50-kilometer radius and by province. Therefore, there is a significant relationship between the closest distance and density (in terms of population average). Equation (4) then becomes Equation (5), which is(5)

In Equation (5), denotes population density within 50 kilometers, and denotes population density within province. The model was estimated without control variables (3a) and subsequently with control variables (3b). Therefore, Equation (4) represents the model with one instrument, whereas Equation (5) contains two instruments.

The second stage regression estimates the influence of the closest distance, which offers similar programmes on attractiveness to students in the programmes coming from the variation in . This is captured by the instrument in model (2). This part was used to estimate parameter in the following equation.(6)

Likewise, we estimated model (5) with the two instruments and . In order to do so, the following model was employed, where is the parameter of interest.(7)

6. Empirical results

6.1. The OLS results

Table 2 presents the results of the OLS regression as defined in Equation (3). There are four estimated specifications, including model 1(a) without control variables and model 1(b) with control variables. All models have clustered standard errors at programme level. A dummy is added for type of education, type of programmes and gender with fixed effects for city and year.

Table 2. Summary results of the OLS estimates (Equation (3)).

Attractiveness to students	Model 1(a)	Model 1(b)
Distance to closest programmes ()	−12.93***	−3.40**
	(2.74)	(2.74)
Competition () using Herfindahl	−10.79***	−10.23***
Index	(0.35)	(0.33)
Competition () using Gini	−5.72***	−4.48***
coefficient	(0.58)	(0.55)
Control variables	No	Yes
Dummies for city and year	Yes	Yes
Number of observations	18066	18066
R^2	0.13	0.23

Note: The control variables include type of education, type of programmes, and gender.

All models estimate city and year fixed effects. Standard errors clustered at the programme unit level are in parentheses.

Asterisks show significance levels as *p < .05, **p < .01, and ***p < .001.

The estimates for are significantly negative in model (a) and (b). Model (a) suggests that an increase by one kilometer in the distance between similar programmes offered decreases the enrolment of students in the programme by 13 students. Controlling for various programmes and other characteristics in model (b), the OLS estimate for the influence of distance to the closest similar programme remains negative and statistically significant. The estimate for using the Herfindahl index is significantly negative in model 1(a) and (b). Model (a) (without control variables) suggests that an increase of 1% in competition between similar programmes within cities decreases the enrolment of students in the programme by 11 students. After controlling for various programmes and other characteristics in model (b), the OLS estimate for the influence of competition between similar programmes offered to attractiveness to students remains negative and significant. Competition () is measured using the Gini coefficient. The estimate of is statistically significant and negative in model (a) and (b). This result suggests that an increase of 1% in competition between similar programmes within cities decreases the enrolment of students in the programme by six students. Controlling for various programmes and other characteristics in model (b), the OLS estimate for the influence of competition on attractiveness to students also remains negative and significant.

6.2. IV-results

6.2.1. First stage estimates

Table 3 shows the first-stage IV regression results as specified in Equation (4) of Section 5. Each model clusters the standard error at the programme level, and each model contains dummies for the city and years 2008–2013 in our data set. Population density is used as a variation of closest distance in order to estimate the effect on the outcome variable and because higher population density is probably associated with a shorter distance between institutions. Population density has a positive relationship with the closest distance between similar programmes offered. A one-person increase in population density (per square mile) is associated with a 0.37 kilometer increase in the closest distance. This indicates that population density is positively correlated with the distance between institutions that offer a similar programme. This result holds when controlling for several programme characteristics. The sign of the population density coefficients remain significantly positive. The significance level for this variable is at a level of 1%.

Table 3. Summary results of the first-stage estimates using as an instrument (Equation (4)).

Distance to closest programmes ()	Model 2(a)	Model 2(b)
Population	0.37***	0.37***
Density	(0.023)	(0.023)
Control variables	No	Yes
Dummies for city and year	Yes	Yes
Number of obs	18066	18066
R^2	0.10	0.10

Note: The control variables include type of education, type of programmes, and gender.

All models estimate city and year fixed effects. Standard errors clustered at the programme unit level are in parentheses.

Asterisks show significance levels as *p < .05, **p < .01, ***p < .001.

Table 4 presents the first-stage regression results as specified in Equation (5) of Section 5 using the 2SLS estimator. Recall that Equation (5) uses population density in a 50- kilometer radius and the population by province as instruments. The coefficients in all four models – population density in a 50-kilometer radius – are positively and highly significant. Consequently, the population density by province are negatively significant. Thus, the coefficients in Table 4 are similar to those in Table 3, but Table 4 is divided by the 50-kilometer radius and by province. The Netherlands is divided into 12 regions or provinces. The population density in a 50-kilometer radius seems a better predictor than population size in a province for the increase in closest distance between similar programmes offered.

Table 4. Summary results of the first-stage estimates using and as instruments (Equation (5)).

Distance to closest programmes ()	Model 3(a)	Mode 3(b)
Population	0.40***	0.40***
density in a 50 km radius PCct	(0.021)	(0.026)
Population	−0.41***	−0.40***
Density by province PPct	(0.013)	(0.014)
Control variables	No	Yes
Dummies for city and year	Yes	Yes
Number of obs	18066	18066
R^2	0.10	0.10

Note: The control variables include type of education, type of programmes, and gender.

All models estimate city and year fixed effects. Standard errors clustered at programme level unit are in parentheses.

Asterisks show significance levels as *p < .05, **p < .01, ***p < .001.

6.2.2. Second-stage estimates

The second-stage IV regression results of our analyses are presented in Tables 5 and 6. The coefficients in Table 5 are estimated using only one instrument, whereas Table 6 presents the results using two instruments (i.e. population density in a 50-kilometer radius and population by province). The results show a negative and significant effect of closest distance between institutions that offer similar programmes on attractiveness to students (model 4a). This result holds when controlling for programme and other characteristics (model 4b). Similarly, the competition variable using two indices (e.g. Herfindhal index, Gini coefficient) are first estimated without control variables (model 4a) and subsequently with control variables. The control variables include dummies for programme (type of education and type of programmes) and gender (male and female to control for city and year fixed effect (model 4b). Model 4a shows a negative and significant effect of competition on attractiveness to students. The results also remain unchanged when control variables are added to the model. The effect of competition on attractiveness to students remains negative and significant at the 1% level.

Table 5. Summary results of the second-stage estimates using as an instrument (Equation (6)).

Attractiveness to students	Model 4(a)	Model 4(b)
Distance to closest programmes ()	−20.80***	−17.41***
	(3.13)	(3.04)
Competition () using Herfindahl	−10.73***	−9.69***
Index	(0.35)	(0.34)
Competition () using Gini	−5.71***	−5.38***
coefficient	(0.58)	(0.56)
Control variables	No	Yes
Dummies for city and year	Yes	Yes
Number of obs	18066	18066
R^2	0.13	0.18

Note: The control variables include type of education, type of programmes, and gender.

All models estimate city and year fixed effects. Standard errors clustered at the programme unit level are in parentheses.

Asterisks show significance levels as * p < .5, ** p < .01, *** p < .001.

Table 6. Summary results of the second-stage estimates using and as instruments (Equation (7)).

Attractiveness to students	Model 5(a)	Model 5(b)
Distance to closest programmes ()	−17.50***	−6.68**
	(2.98)	(2.88)
Competition () using Herfindahl	−10.80***	−10.03***
index	(0.44)	(0.42)
Competition () using Gini	−5.11***	−4.27***
coefficient	(0.80)	(0.76)
Control variables	No	Yes
Dummies for city and year	Yes	Yes
Number of obs	18066	18066
R^2	0.15	0.15

Note: The control variables include type of education, type of programmes, and gender.

All models estimate city and year fixed effects. Standard errors clustered at the programme unit level are in parentheses.

Asterisks show significance levels as * p < .05, ** p < .01, *** p < .001.

Table 6 also presents model 5 (a and b), which contains the closest distance and competition variables using two indices (i.e. Herfindhal index, Gini coefficient). Model 5 is first estimated without control variables (5a) and subsequently with control variables (5b). The results show a negative and significant effect on attractiveness to students. This result holds when controlling for programme and other characteristics (5b). Likewise, the second-stage model (5a and 5b) shows a negative and significant effect on attractiveness to students. The effect of distance and competition on student attractiveness remains negative and significant at 1% level.

6.2.3. Identification tests

To assess the validity of our instrumental variables estimation, we applied statistical tests such as the under-identification test, the weak identification test and the over-identification test. These tests were performed for our model with population density in a 50-kilometer radius and population density by province as two instruments and used for the closest distance between similar programmes offered, as defined in Equation (7).

First, the under-identification test was performed to identify whether the equation with exogenous instruments are correlated with the endogenous regressors for the model that includes one instrument in Equation (6). The Anderson canonical correlations LM statistic is 1.1e+04. It is significant at the 1% level in the initial model (a) and holds at 1.1e+04 in the extended model with control variables (model b). Therefore, we reject the null hypothesis of under-identification.

Furthermore, in the statistic expresses whether the instrument (the population density in 50-kilometer radius and by province) is sufficiently correlated with the endogenous regressor (the closest distance which offers similar programme). Concerning the equation with two instruments, the Cragg–Donald Wald F statistic gives a statistic of 4.9e+05 in the initial model specification (model a). This is higher than the Stock and Yogo threshold of 16.38 for size level of 10%. Also, in the extended model including control variables (model b), the F statistic is higher than the threshold of 10 (F statistic > 16.38). These statistics indicate that both instruments in Equation (5) are sufficiently strong to continue with our analysis.

Finally, this study reports on the over-identification test, which tests whether the correlation with the error term is zero. The Sargan statistics display whether the instrument is truly exogenous. With respect to the equation with two instruments (Equation (7)) without and with control variables, the over-identification test yields a Sargan statistic equal to 0.00. Therefore, we argue that the population density instruments used are valid instruments for the distance to closest similar programmes.

7. Discussion and conclusion

The number of students enrolled in higher education institutions has increased tremendously in recent years. As students become more selective in choosing study programmes, higher education institutes must compete more intensely to attract new prospective students. Although a few studies have investigated student behaviours by using distance between higher education institutions as an input variable (e.g. Frenette 2004; Denzler and Wolter 2011), little is known about the influence distance between institutes with identical programmes on attractiveness of programmes. Using a data set from the Dutch Ministry of Education (‘Dienst Uitvoering Onderwijs’) consisting of students from 1356 study programmes spread across 50 institutions within the period of 2008–2013, this study has investigated the relationship between higher education institutes’ study programme and distance included competition, programme variety and other characteristics such as gender among the dependent variables. This paper is the first to explore how variation in proximity of institutes to other institutes with similar programmes contributes to programme attractiveness. We defined programme attractiveness to students as the number of students who enrol in programmes within institutions and cities, and we focused our attention on the programme of study context.

The study employed a two-step empirical framework to address the endogeneity of population density, expressed as population density in a 50-kilometer radius and population density by province over the period of 2008–2013. This framework was then used as an instrument to identify the closest distance between similar programmes offered. We assumed that population density is exogenous to the closest distance. Population density is positively associated with distance (McDonald 2008). The first-stage estimates indicate that a one-person increase in population density (per square mile) increases the level of the closest distance between similar programmes offered by 0.4 kilometers. For example, some cities, such as Amsterdam and Rotterdam, have very high-population densities; they are quite crowded, and that population density may lead to more distance between similar programmes at their higher education institutes. This finding, again, is based on population density in a 50-kilometer radius and shows significance. There is a positive association between the closest distance between institutions that offer similar programmes and population density within a 50-kilometer radius. Population density by province is not associated with the closest distance between similar programmes offered. The second-stage results indicate that a one-kilometer increase in the closest distance between similar programmes decreases the number of student enrolments by – seven students. This result can be explained by the relationship between distance and enrolment of students, as distance has negative influence on study choice among students (Kjellström and Regnér 1999; Denzler and Wolter 2011). In this context, the number of enrolled students is lower due to the increase in the closest distance between similar programmes offered.

Additional findings indicate that competition does affect higher education institutes’ attractiveness to students negatively. The competition might be explained by programme characteristics or other characteristics as well. To disentangle these influences, we control for type of programmes, type of education, and gender in our empirical models. However, our estimation results give no empirical evidence for such an explanation. Particularly, these results suggest that competition between similar programmes offered, when programme and other characteristics are considered, has a significantly lower percentage points in comparison to competition when programme and other characteristics are not considered.

In this paper we conclude that the closest distance between similar programmes and the competition between programmes offered have significantly negative influences on the enrolment of students in the respective higher education institutions. Breakdown of enrolment of students by type of programmes and other characteristics further shows that the closest distance between similar programmes and the competition between programmes offered do affect the enrolment of students.

This paper yields some insightful implications. The two major reasons (distance and competition) that students choose certain study programmes in higher education institutions are no longer related to students’ motives of interest. Implications arising from this finding could include the need for higher education institutions to re-examine their diversity system in study programmes to encompass student expectations. Higher education institutions may also need to consider other ways to recruit prospective students. Nevertheless, the negative influence of the two major reasons (distance and competition) on attractiveness of programmes should be improved by measures to increase the enrolment of students.

In conclusion, we emphasize some of the limitations of our study and provide some suggestions for further research. First, we limited our analyses to the distance to the closest institutions between similar programmes offered and competition between similar programmes. The interpretation of the results captures some changes related to attractiveness of programmes, although these results differ from this study’s hypothesis. Further research could investigate this result more intensively by refining the analysis using a more rigorous methodology. For instance, this could be achieved by using a regression discontinuity approach to exploit discontinuities in the attractiveness of programmes.

Acknowledgements

The authors would like to thank participants at the 2016 Workshop on Education Economics at Maastricht University for their useful comments on previous versions of this paper.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes on contributors

Ferdi Widiputera was born in Hamburg, Germany, on April 9, 1978. He completed a Bachelor Degree in Economics at Trisakti University between 1996 and 2000. He also holds a Master Degree in Economics from University of Indonesia, graduating in 2004. He has several years of working experience in multinational company from 2001 to 2009. Since 2010, he worked in the government sector at the Ministry of Education and Culture of The Republic of Indonesia. Ferdi has received a scholarship from his institution and he is currently a Ph.D. candidate at the Top Institute of Evidence Based Education Research, Maastricht University.

Kristof De Witte is a tenured associate professor at the Faculty of Economics and Business at KU Leuven, Belgium, and he holds the chair in ‘Effectiveness and Efficiency of Educational Innovations’ at Top Institute for Evidence Based Education Research (TIER) at Maastricht University, the Netherlands. Kristof De Witte is further an affiliated member of the CESifo Network (Ludwig-Maximilians-University and Ifo Institute).

Wim Groot is Professor of Health Economics since 1998 and Professor of Evidence Based Education since 2008, both at Maastricht University. In 1986 he received a M.A. degree in Economics from the University of Amsterdam and in 1992 a Ph.D degree in Economics from the same university. He is the co-founder and scientific-director of the Teachers Academy of Maastricht University and the Top Institute for Evidence Based Education Research. Since 2015 he is also Professor of Evidence Based Education and Labor Market Policy at the University of Amsterdam.

Henriëtte Maassen van den Brink is professor of Economics (Education, Labor Market and Economic Development) at the Faculty of Economics and Business at the Department of General Economics at the University of Amsterdam and professor of Evidence Based Education at Maastricht University. She obtained her Ph.D. in Economics (cum laude) at the University of Amsterdam in 1994. Since 2008 she is the Scientific Program-Director of TIER (Interuniversity Top Institute for Evidence Based Education Research). Since 2015 she is Chairman of the Education Council of the Netherlands. Her research interests are in the areas of education, labor and health economics. Her publications can be found on www.tierweb.nl.

Notes

1. Calculations based on the stata module, known as ‘nearstat’, generated the closest distance between neighbour programmes based on geographic co-ordinates (latitude and longitude).