I.  Introduction

Which university entrance criteria matter for the academic performance of the students admitted? Most higher education institutions (HEIs) in developed countries rely on standardized tests to select their students (Zwick 2007; Sternberg 2010; Edwards, Coates, and Friedman 2012). Additionally, they can activate a second source of information. In some cases, the institution considers other elements in the admission process, such as student essays, interviews, as well as information on students’ extracurricular activities and experience. Other HEIs design their own entrance exams as a complement to standardized tests. The purpose is to obtain a more comprehensive assessment of higher education (HE) applicants (for reviews, see, e.g., Hoxby 2009; Zwick 2017). However, how broad the admission criteria should be is a question that has been given little attention in the literature.

I study the role of specialization of the admission criteria as a choice variable used by HEIs to maximize their objective function of admitting the best pool of students possible. These are expected to have better academic performance, better placement after degree completion, and better future labor market achievement. As a result, HEIs benefit from improvements in their own reputation.

The central finding of this paper is that universities with less specialized admission policies admit a pool of students who perform better at university by the end of the degree (namely, they obtain a higher final GPA on average). This is a surprising and relevant result. Many universities operate under the assumption that candidates who perform better at a subject-specific test will go on to perform better in that subject at university. For instance, in the artificial intelligence and data science course in Portugal, mathematics is often perceived as a relevant field-specific skill, while languages convey information on broader cognitive skills.¹ I define general skills as those that are not directly rewarded by the field of study. They might be informative about a student’s ability, but they are not specific to the field. Ability refers to the potential or innate ability that an individual is born with, while skills are learned and acquired with time. Strong emphasis on field-specific skills rewards students who specialize early in high school rather than cultivate a more versatile portfolio of skills.

In my theoretical framework, I consider a single university and a continuum of students. The university’s objective is to admit the best pool of students possible. In turn, students aim to maximize a utility function that depends on their own future performance and the level of effort exerted on the university admission tests. I assume that the university can select students on the basis of a wider and broader set of tests. Of the tests available, some tests measure skills that will be considered specialized, while others are tests of skills that serve solely as a signal of ability (the general skills). These skills are still informative about future professional life (Kautz et al. 2014), although in the model I assume a scenario where they might not be as strongly related to student performance as the specific skill. Thus, the university must decide how to set its admission criteria, either (i) by using only a field-specific test or (ii) by combining it with a test of a more general nature. In the latter case, the university gains additional information about a student’s ability. However, in return, it must contend with less effort being put by students into the specialized test, because students optimally reallocate the effort. In the case of a single field-specific test, the effort of students would be directed solely toward acquiring knowledge and preparing for the exam that contributes to future academic performance and is directly valued in the labor market (see, e.g., Card et al. 2018).

From a theoretical point of view, I find that the university chooses to use a general skills test as a criterion for selection alongside a test to evaluate subject-specific skills. Although, in equilibrium, students deviate effort from the field-specific test (the productive test) to the general skills test, a different set of students (more able on average) is admitted and ultimately performs better than when only a field-specific requirement is in place.

In my empirical analysis, I use Portugal as a case study. Although the Portuguese system has specificities, the lessons one learns from it provide generalizable insights to any HE system that considers at least one of two different metrics in the admission process: one that matches closely to what a student actually learns in the subject of study (a field-specific test for instance) or another one that conveys information of a more generic nature, not directly rewarded by the field of study.

I rely on an extremely rich administrative dataset of the population of HE applicants from 2008 to 2018. Over that period, I observe approximately 800,000 student applications to the first year of an HE program. For each applicant, there is information regarding personal characteristics, socioeconomic background, previous academic achievement, the application process (including all program preferences stated with the corresponding overall application score and each exam score), and the HEI placement. Moreover, for 5 years, I have information on the performance of students at HE for each year of their degree and subsequent graduation information (namely, the graduation date and the final GPA).

I consider the Portuguese exam as a general skills test. The Portuguese exam is compulsory and the only common exam in all academic tracks in high school. The performance in this exam is available for all applicants to HEIs. With some exceptions, most programs have multiple tracks of admission. They require either a subject-specific set of exams as admission criteria or a broader set of exams. The student will be ranked according to whichever set of exams yields the highest overall application score. For instance, suppose a student applies for an economics degree at a particular university. The score that determines the ranking in the application process is either the mathematics exam score or the average of mathematics and Portuguese exam scores, depending on which generates the highest application score and hence the highest feasible position in the ranking. The mathematics exam is considered to be the field-specific requirement, while the Portuguese exam is the general skills test. However, not all HE programs consider a nonspecific set of exams as an admission criterion. Thus, I perform my empirical analysis on HE programs that include the general skills test (the Portuguese exam) in the broader set of exams.² Therefore, my results do not apply to highly specialized subjects, such as medicine.

Because of this admission criteria design, I observe students being admitted to the same program on the basis of performance in different exam sets. In my empirical analysis, the treatment is the inclusion of the Portuguese exam as an admission criterion at university. Nevertheless, students might self-select into programs on the basis of unobserved attributes. I tackle the problem of selection into treatment by first determining an alternative application score for each student: the application score in the case that the student was admitted with a different exam set. For each student, I compute (i) her application score in the case the Portuguese exam was considered (my measure of general skills) and (ii) her application score in the case the Portuguese exam score was not considered (my field-specific skills measure).

I define exposure to treatment in the following way. I focus my analysis around the program admission threshold (the admission score of the last admitted student). Within each program, I define three categories of admitted students: the specialists (those admitted with a field-specific test and whose general skills would not have been enough to gain entry to the HEI), the generalists (those admitted purely on the basis of their performance on the general skills test who would not have been admitted on the basis of only the field-specific test result), and the all-rounders (those who could have been admitted with each one of the two types of tests). I observe differences in academic student performance across the different groups at distinct landmarks of the academic program. In particular, I find that (i) the generalists perform no worse than their specialist peers by the end of the first academic year, and (ii) by the end of the program, the specialists are outperformed by the all-rounders and the generalists.

My results have substantial implications for university admission practices. Universities that include a general skills test in their set of entrance exams benefit at the margin of admission, because this has a positive effect on the average student performance. Although this effect is not observable by the end of the first year, students admitted on the basis of a broader set of skills and knowledge perform better on average toward the end of the program.

An additional policy implication surrounds the role of the field-specific exam, because it may or may not be fit for purpose. Differences in student performance across groups suggest that the field-specific exam on its own might not be an effective test of the specialist skills associated with high achievement at university, which is the main policy message of my paper. Although I find evidence that the field-specific exam is a predictor of student performance by the end of the first academic year, I observe that the generalists and all-rounders have a better performance by the end of the degree when compared with specialists. Thus, considering only the field-specific exam does not guarantee that those who perform well on it will perform better at university.

Alternatively, another possible interpretation is that the general and field-specific exams measure the same characteristics, but the general exam is just a more precise signal. In that case, the result would be driven by the exams’ precision rather than the differences between general and specific skills. Thus, the generalization of the previous policy implication will depend on how exams are designed in other settings.

Finally, in my empirical analysis, I consider only programs that allow for the Portuguese exam as an alternative requirement, that is, programs where there is a variation of admission requirements on the standardized tests. Thus, my empirical results should not be interpreted as meaning that universities should all ask students to take the Portuguese exam, while my empirical analysis does not allow me to conclude that. In the mechanism section (sec. X), I discuss and acknowledge that my empirical results may not be generalizable across all fields of study.

The remainder of the paper proceeds as follows. Section II summarizes the literature review, and section III sets up the theoretical framework and its assumptions. Section IV defines the equilibrium in the two stages: the effort of students and the university’s choice of admission criteria. In that section, I also consider an extension of the model where the social planner plays a role. Section V describes the institutional setting, and section VI presents the datasets. Section VII presents the empirical setup, and section VIII presents the estimation strategy. Section IX presents the results, and section X discusses possible mechanisms. Finally, section XI concludes.

II.  Literature Review

I contribute to different strands of literature. First, my paper contributes to the body of work that seeks to evaluate the role of standardized tests in the process of admission to HEIs (e.g., Rothstein 2004; Zwick and Green 2007). Examples of standardized tests include A-levels in the United Kingdom and SAT/ACT in the United States. The widespread use of and reliance on these tests has been criticized in the literature for favoring ethnicity (e.g., Bridgeman, McCamley-Jenkins, and Ervin 2000; Freedle 2003), gender (e.g., Leonard and Jiang 1999), or socioeconomic background (e.g., Zwick 2019; Campbell et al. 2022). Despite a lack of consensus in the literature on whether such tests predict performance at university (e.g., Burton and Ramist 2001; Kuncel and Hezlett 2007; Radunzel and Noble 2012), they are the most common instrument used to for admission. In this paper, I conclude that admission tests indeed help to predict student performance at university. Namely, the nonspecific tests are informative of overall performance, while the specific tests are good predictors of student performance in the first year of the HE program.

Second, my paper contributes to the literature on the effects of broadening admission criteria in HE (e.g., Sternberg 2010; Sternberg et al. 2012; Schmitt 2012; Stemler 2012; Niessen and Meijer 2017) and the effect of “noncognitive” skills, including for instance social skills and leadership skills (e.g., Deming 2017). Some HEIs rely on a second source of information. For instance, in the United States, some Ivy League universities rely on interviews. In the United Kingdom, Oxford and Cambridge establish their own entrance exams for some degrees. Often, this extra step in admission broadens the type of information gathered about the student by the institution. My work suggests that broadening the nature of cognitive admission tests leads to increased performance at university because abler students are admitted. Nevertheless, the general admission test should be considered as a complement to field-specific requirements.

Third, this paper contributes to the literature on the predictive performance of test scores and its implications for admission policies. For instance, although high-stakes assessments influence the decisions of students to apply to HEIs (Papay, Murnane, and Willett 2016), Bettinger, Evans, and Pope (2013) propose to reduce the number of ACT components to improve college admissions. At the same time, other papers show the predictive power of past performance, such as high school transcripts (e.g., Belfield and Crosta 2012; Cyrenne and Chan 2012; Dooley, Payne, and Robb 2012). Moreover, prior work shows that combining teacher scores and high-stakes assessments is a better practice for selection of students (Westrick et al. 2015; Zwick 2019; Silva et al. 2020). My results corroborate that high-stakes assessments are a predictive tool of students’ future academic performance.

My work relates to literature in the field of the economics of HE. While previous work has focused on different aspects of admission to HEIs, such as the role of student portfolios, admission tuition, or college choice and selectivity (e.g., Epple, Romano, and Sieg 2006; Avery and Levin 2010; Avery et al. 2012; Hoxby and Turner 2013), my paper focuses on one novel aspect of the process of admission to HEIs: the choice of the examination structure, namely, how specific the admission exams should be. Even under the assumption that some skills have no direct influence on student performance, and thus evaluating them may distract from learning more productive skills, I show that the best way for the university to resolve the resulting trade-off is to have a nonzero weight on the general skills test. My empirical evidence also shows that universities with broader admission exams select a better pool of incumbents.

Finally, the results of this paper are also linked to previous literature that shows it is useful to employ different types of instruments to select into HEI a pool of high-ability students. Although in this paper I focus on broadening the nature of standardized tests, countries often use a wide system of selection mechanisms to identify best-matching options that are not exclusively based on standardized tests. Most European countries rely on a combination of admission criteria of standardized tests and final high school GPA (an average of all courses taken at high school). The literature finds that high school GPA tends to be a better predictor of student performance in HE compared with standardized tests and that both measures should be considered as complements in the process of admission to HEIs (Cerdeira et al. 2018; Zwick 2019; Silva et al. 2020). I find that in Portugal, where high school GPA is considered together with standardized tests as admission requirements to HEIs, by broadening the nature of standardized tests, universities can select more high-ability students who have a broader set of skills. In other countries where admission is based solely on admission tests, considering high school GPA or teacher grades might also help to broaden the type of skills required to gain admission to HEIs.

III.  Theoretical Framework

The design of admission policies aims at overcoming problems of imperfect information (Stiglitz 2000), namely, information asymmetries (Teixeira et al. 2006) that characterize HE markets, given that it is not possible to directly observe a candidate’s ability.³ College admission assessments provide incentives for signaling via both productive and nonproductive activities. In this section, I build a simple model where I compare two types of admission requirements (field-specific and general) that can potentially be used to overcome the problems of imperfect information.

In my theoretical framework, I focus on the choice of the examination structure in an HEI during the process of admission.⁴ I study a single university (or college) offering a single degree. A continuum of individuals of measure 1 want to apply to the university. However, there are more applicants than university places. I assume that only half of the applicants are able to be admitted to the university. The time of the sequential game that I model below is as follows: the university sets the admission exams required to obtain admission, and then students decide on the amount of effort to exert on each exam. The text below describes the two types of admission exams considered, the goals of each one of the agents, and the key assumptions of the model.

IV.  Equilibrium

In this section, I find the equilibrium by first determining the student’s optimal level of effort exerted on the two admission tests and then incorporating this into the university’s admission problem.

A.  Effort of Students

In equilibrium, each student knows the effort distribution of the entire cohort, so each student takes as given the effort choice of all her fellow students ($e_1^{*}$, $e_2^{*}$), thus τ* is a function of ($e_1^{*}$, $e_2^{*}$).

The assumptions that all students have the same expectation about their ability and ability is revealed in the job market implies that the student expected wage is as follows:¹³

$$\begin{array}{}\text{(3)}& w=(\gamma \overline{\alpha }+e_1)\mathrm{\Pi }+(\beta \overline{\alpha }+e_1)(1-\mathrm{\Pi })=v\overline{\alpha }\mathrm{\Pi }+e_1+\beta \overline{\alpha },\end{array}$$

where $\mathrm{\Pi }=\mathrm{Prob}(T_1+\lambda T_2\ge {\tau }^{*})$. The student’s objective function is the maximization of the difference between the expected wage and the cost of effort.

The maximization problem of a student is given by

$$\begin{array}{}\text{(4)}& \underset{e_1,e_2}{\mathrm{Max}}v\overline{\alpha }\times \mathrm{P}\mathrm{r}\mathrm{o}\mathrm{b}(T_1+\lambda T_2\ge {\tau }^{*})+e_1+\beta \overline{\alpha }-C(e_1,e_2).\end{array}$$

The solution of the student’s problem is stated in the following result.¹⁴

Proposition 1. 

The optimal effort ${\mathbf{e}}^{*}(\lambda )=(e_1^{*},e_2^{*})$ is

$$\begin{array}{}\text{(5)}& \begin{array}{rl}{e}_1^{*}(\lambda )& =\{\begin{array}{cl}\frac{v\overline{\alpha }}{a(1-{\delta }^2)}(1-\lambda \delta )+\frac{1}{1-{\delta }^2},& \mathrm{if }\delta \widehat{\lambda }\le \lambda <\frac{\widehat{\lambda }}{\delta },\\ \frac{v}{a}\overline{\alpha }+1,& \mathrm{if }\lambda <\delta \widehat{\lambda },\end{array}\\ e_2^{*}(\lambda )& =\{\begin{array}{cl}\frac{v\overline{\alpha }}{a(1-{\delta }^2)}(\lambda -\delta )-\frac{\delta }{1-{\delta }^2},& \mathrm{if }\delta \widehat{\lambda }\le \lambda <\frac{\widehat{\lambda }}{\delta },\\ 0,& \mathrm{if }\lambda <\delta \widehat{\lambda },\end{array}\end{array}\end{array}$$

where $\widehat{\lambda }=1+(a/v\overline{\alpha })$.

Proof. Available in the online appendix.

According to proposition 1, for small values of λ I obtain a partial corner solution (for $\lambda <\delta \widehat{\lambda }$), and thereafter I have an interior solution (for $\lambda >\delta \widehat{\lambda }$). In other words, when the weight on T₂ is too small, the student would not exert effort for that test, and e₁ is constant on λ.¹⁵

As a result, the optimal amount of effort varies with λ as shown in figure 1.¹⁶

Figure 1 represents the student’s optimal strategy in a Nash equilibrium. For low values of λ, the student would not change her decision when compared with the case where $\lambda =0$. The same occurs if the cost of switching from one test to another is too high (which means a high value for δ). Alternatively, when the university increases the weight on T₂, for values of $\lambda >\delta \widehat{\lambda }$ the student reallocates effort from T₁ to T₂. Thus, under some conditions, allowing for a second exam deviates effort from the productive test (T₁). From a student’s perspective, this would be the optimal strategy to maximize the difference between expected wage and costs.

C.  Interpreting the Results

I model the university admission problem with the focus on two aspects: the noise and the nature of the admission tests. Figure 2 presents the two test score distributions for students of both abilities. On the horizontal axis, I have the test score distribution of T₁, and the test score distribution of T₂ is represented on the vertical axis. Ability is fixed for each student type, and (e₁, e₂) is also fixed in the equilibrium. Given that the two idiosyncratic terms of both tests follow a uniform distribution, the shape of the distribution is rectangular. Besides the gap on ability, the noise of the two tests is a key factor in both distributions.²⁰

According to figure 2, when λ increases, the university can select more high-ability students.²¹ The increment in λ is illustrative of the following: if a university is faced with two students who have the same score on T₁, the university should select the one with the highest score on T₂. Indeed, as argued by Holmström (1979), additional information would allow a more accurate judgment of a student’s performance. For students who have the same score on T₁, the university would like to pick the ones who are also good on T₂ (even if T₂ conveys less relevant skills).²²

The other important feature of my model is the nature of the second test. I assume that there are no administrative costs of introducing a second test. However, there is an inefficiency associated with that test: it reduces the productive effort, e₁ (see fig. 1).²³ The cost of running a less relevant test (the general admission exam) is a decrease in the student’s future productivity, which reduces the university payoff. In the model, I assume the extreme scenario where the less productive effort, e₂, is less informative about future wages than e₁. Hence, when the university is deciding about introducing a second admission test, it faces a trade-off between gaining new information about the student’s ability and losing productive effort. Given this, there are two effects on equilibrium occurring at the same time. On the one hand, geometric intuition shows that by increasing λ, the university is able to increase the number of high-ability students. On the other hand, from figure 1, I conclude that an increase of λ has a detrimental effect on e₁, which has a negative effect on the student’s future wage. The two effects move in opposite directions.

Nevertheless, in theorem 1, I have shown that the overall effect of introducing a second admission test is positive. Even if students deviate effort from the relevant test, the university still benefits from including a general admission exam in its admission criteria.²⁴ A separated question is whether that is socially desirable.

Alternatively, I also considered the case of the social planner. I took a utilitarian approach to government intervention. The government’s goal is to maximize the unweighted sum of wages of all individuals, internalizing the cost of effort exerted in the two admission tests. This approach is common in the literature since Arrow (1971). I find that the university and the government do not always agree about λ. Nevertheless, both the university and the government agree that the weight on T₂ should be positive.

V.  Admission Policies in Portugal

The theoretical framework can be applied to any HE system where different admission conditions are combined, such as field-specific tests and additional requirements. This combination is common across countries, and Portugal is one such case. In Portugal, universities use national central exams and the high school GPA as mandatory requirements.

As explained before, I aim to test whether the inclusion of a general skills test as an admission criterion generates a better pool of students. However, it is crucial to clarify the admission procedure of HEIs in Portugal before presenting my estimation strategy.

VI.  Datasets

In my study I link three primary datasets: (i) applications to public HEIs (DGES 2019b), (ii) students’ performance, and (iii) graduation from HEIs (DGEEC 2019).²⁹

The application dataset provides information on the population of applicants to the HE system. For each student, I have demographic characteristics, socioeconomic background, previous academic achievement, the application process including all program preferences stated, and the placement. I have microdata for 11 years, from 2008–9 to 2018–19.

The students’ performance and graduation from HEI datasets are comprehensive data sources that cover all the HEIs in Portugal and have information on all the students enrolled. They report a student’s performance each year she has been enrolled and her graduation, whenever applicable. The performance dataset also reports information regarding the mobility status, such as the placement at an exchange program.

Over the period 2008–18, I observe approximately 800,000 student applications to the first year of an HE program under the General Access Regime (GAR; the centralized process of application to public HEIs). Additionally, I track enrolled students over the 5 years from 2013–14 to 2017–18. This panel comprises 1.5 million observations on individual-year for 760,000 individuals.³⁰ I also observe approximately 330,000 individuals graduating from 2012 to 2017, irrespective of their application year and years of enrollment.³¹ I link the three datasets (application, performance, and graduation) restricting the link to those individuals who applied to public HEIs through the centralized national system, who were offered a place, and who enrolled.³² From the individuals observed in the performance dataset, I was able to match 96% to the application dataset, namely, 205,297 individuals. Table 2 presents descriptive statistics for the linked individuals. The majority of the individuals studied at a public high school, and 57% of the individuals are female. One-third of the individuals are nonlocal students, which means these students live in geographic areas different from those of their households.³³ Moreover, only 30% of the individuals have a mother or father who has an HE degree.

Once I define the outcomes of interest, the variables of interest, and the methodology to be used, I will impose further constraints that should be applied to the linked dataset in order to determine the analysis population.

VII.  Empirical Setup: Tests and Performance

VIII.  Estimation Strategy

The treatment is the inclusion of the Portuguese exam as an admission criterion at the individual level. I define y(1) as the observed outcome of student performance in case of treatment and y(0) otherwise. The problem I must tackle is that selection into treatment might not be random. However, I can accommodate selection in my model by understanding how individuals are assigned to treatment.

In my analysis, I define three different groups of students: (i) the specialists, (ii) the generalists, and (iii) the all-rounders (see fig. 3). First, I observe students admitted without the Portuguese exam, and so their field-specific skills are better than their general skills ($T_1>T_2$ and $T_1>{\tau }^{*}$). Among these students, some of them could not have been admitted with a general admission exam (specialists). Some could have been admitted with such an exam (all-rounders). Second, I also observe students admitted with Portuguese but could have been admitted even without it (another type of all-rounders). These are students who are relatively better at general skills but whose field-specific skills would have been enough to get them admitted ($T_2>T_1>{\tau }^{*}$). Finally, I observe students who were admitted only because their Portuguese exam score was sufficiently high (generalists). These are students whose field-specific skills are relatively weak compared with their peers and would not have been admitted without the Portuguese exam ($(T_1+T_2)/2>{\tau }^{*}>T_1$).

There is a potential trade-off in allowing generalists to compete with specialists. The university can increase the average quality of the pool of students by admitting all-rounders. At the same time, the university might get pure generalists in its pool. In the example, all students gained entry to the program. A consequence of including Portuguese as an alternative requirement (T₂) is that generalists may take places that would otherwise have been allocated to specialists. Thus, I frame my research question in the following way: Do the generalists perform better at university relative to their peers?

To answer the question, I look at performance in T₁ and T₂. For individuals that were admitted with Portuguese, I dichotomize them according to whether their performance in T₁ was sufficient by itself to meet the entry requirement (τ*).³⁸ For individuals that were admitted without Portuguese, I look at performance in T₂. In my estimation strategy, the specialists will be the comparison group. I hypothesize that the effect of allowing the inclusion of the Portuguese exam on individual performance will be different across groups. Namely, I expect generalists to perform worst at university compared with their peers, given that the generalists are weaker in terms of field-specific skills.

In the data, I attempt to determine the application score for each student in the case that she was admitted with a different exam combination. For each student, I compute (i) her application score in the case the Portuguese (PT) exam was considered (τ_PT) and (ii) her application score in the case the Portuguese exam was not considered (${\tau }_{\sim \mathrm{P}\mathrm{T}}$). On the basis of four blocks of information (admitted with Portuguese, τ*, τ_PT, and ${\tau }_{\sim \mathrm{P}\mathrm{T}}$), I can distinguish between the different comparison groups. Table 6 reports the number of students for whom I can identify each one of the two types of application scores.³⁹

In the case that ${\tau }_{\sim \mathrm{P}\mathrm{T}}^i<{\tau }^{*}$, student i is classified as being a generalist. Otherwise, she is classified as an all-rounder. As a result, according to table 7, I observe that 8% of the overall sample of enrolled students are all-rounders that were admitted with Portuguese, and 4% of the enrolled students are generalists. Additionally, for 39% of the overall sample, I cannot distinguish between the two groups. This occurs because I cannot observe their scores on the exam combination required in the case that Portuguese was not considered.

To distinguish between specialists and all-rounders (who were admitted without Portuguese), I need to look at performance in T₂. Individual i is classified as being a specialist when I observe that ${\tau }_{\mathrm{P}\mathrm{T}}^i<{\tau }^{*}$. Otherwise, she is classified as an all-rounder. In fact, I observe 21% of the enrolled students are specialists, and 15% of the enrolled students are all-rounders, while for 13% I am not able to distinguish between the two groups.

In both distinctions, I assume that the pool of allocated students and the threshold (τ*) would remain the same in the case of changing the admission criteria.⁴⁰ Because of data constraints, the division between specialists and all-rounders is more accurate than the division between generalists and all-rounders. In the dataset, I observe the Portuguese exam score for the majority of students. As a result, when looking at those admitted without Portuguese, I cannot identify the group for 13% of the students, while for those admitted with Portuguese the percentage increases to 39%. The unbalance on assigning the sample into groups can potentially bias the results.

Therefore, I run my analysis around the program admission threshold. I am interested in comparing student performance among those whose Portuguese exam was, at the margin, the only reason why they were admitted (generalists). In this setup, I consider the reduced-form regression:

where yⁱ is the outcome, Xⁱ is a vector of controls (gender, high school GPA, and nonlocal student), Θ represents the fixed effects (I consider year, preferences, and program fixed effects), and generalists is an indicator equal to 1 if the student’s application score without Portuguese is smaller than or equal to the threshold. The treatment toggles on whether the student needed her result on the Portuguese exam to gain entry into the HE program. The term α₁ is the coefficient of interest, and it tells the reader whether the generalists performed better at university in comparison with the specialists (the omitted category) at the margin of gaining entry into the HE program, and α₂ and α₃ measure differences in student performance when comparing all-rounders with specialists.

The advantage of comparing students close to the threshold in the distribution of the admission score is that students were admitted to the same program, and I can observe their outcome. I assume that individuals around the assignment rule are comparable (comparing students who at the margin were admitted only because of the Portuguese exam with the students for whom it was not necessary). I run my analysis within 10 and 5 points distance from the threshold, and table 8 provides descriptive statistics for all variables considered in the estimations (according to all constraints imposed in the analysis population).

IX.  Results

I perform my analysis at the margin of gaining entry in HE. All results in this section should be interpreted in that context. Table 9 presents the different estimations proposed in equation (9). In column 1, I look at differences in student performance across the four different groups, including year and preference fixed effects only. The omitted and comparison group is the specialists. In column 2, I additionally control for individual characteristics, and in column 3, I include program fixed effects as well. For that reason, the complete estimation is reported in column 3. Coefficients in panels A and C should be interpreted as standard deviation changes.

Starting by the final GPA, in panel C of table 9, I find that both generalists and all-rounders perform better than specialists (the omitted category). This result also holds within program (col. 3). An all-rounder admitted without Portuguese would have, on average, an increment of 0.089 standard deviations on her final GPA when compared with a specialist student of the same ability (within the same program). Moreover, a generalist has, on average, an increment of 0.126 standard deviations on her final GPA when compared with a specialist within the same program. Therefore, all groups outperform specialists in all specifications considered concerning final GPA.

Panel A of table 9 similarly reports that generalists and all-rounders (with and without Portuguese) seem to perform better than their specialists peers. However, these differences in the number of ECTS credits accumulated at the end of the first year do not hold once I introduce the program fixed effects. Within program, generalists and all-rounders perform no worse (and no better) than specialists.

Finally, when I consider completion on time as an outcome (panel B), the effect associated with the all-rounders who were admitted without Portuguese is positive and statistically significant in columns 1 and 3. The effect of all-rounders who were admitted with Portuguese is also positive but not statistically significant.

At the margin of the cutoff, I expect students to have a similar ability, and I would not expect many differences across groups. My results show that differences in performance across groups exist, depending on the outcome measure considered. First, I find no evidence that students admitted purely on the basis of their performance on the Portuguese exam perform worse at university than the others. I do not observe generalists performing any worse than the specialists, which is a surprising result. In fact, when considering the final GPA, generalists on average have a better performance when compared with specialists, and the effect is very similar across the different estimations.

In turn, I observe consistent differences in performance regarding all-rounders relative to specialists. Overall, I still observe that for both types of all-rounders, they seem to obtain slightly better final GPA when compared with their specialists peers, and all-rounders (without Portuguese) have a higher probability of completion on time. These results suggest that general skills matter at the margin of gaining entry in HE.

The next section further discusses these results and potential associated mechanisms.

X.  Discussion and Mechanisms

Universities that allow for Portuguese as an alternative requirement benefit at the margin of admission. Students that perform better in general skills exams perform no worse at university than their specialist peers. Alternatively, there is some evidence suggesting that students with a sufficiently high score on the Portuguese exam perform better at university, even when they did not need the general skills exam to gain entry in HE. As a result, the inclusion of a general skills test as an admission criterion can have a positive effect on student performance at the margin of admission.

The evidence suggests that all groups performed better relative to specialists in terms of final GPA. However, that effect is less pronounced and not statistically significant when I look at completion on time and at the number of ECTS credits by the end of the first year. Differences across groups vary according to the type of measure I consider; namely, at the beginning of the program, differences are less pronounced than toward the end. I interpret that result from two perspectives.

Final GPA includes all course scores, while ECTS credits measure only the number of credits accumulated independently of the scores. Specialists might have a head start, but they do not capitalize on that advantage. Students with better general skills do not substantially outperform specialists by the end of the first year. On average, both groups have the same quantity of ECTS credits accumulated within the same program. Even so, generalists outperform specialists when I consider the final GPA as outcome. To a certain extent, this signals that generalists are able to adapt to changes. Students with different skills need time to adapt.

Additionally, I can also infer that the specialist exam may not be well designed. This result is particularly relevant in a centralized admission system that emphasizes the importance of field-specific skills. Throughout the paper, I made the assumption that the field-specific exam provides a good measure of specialist skills. However, in the light of my results, I can hypothesize that the specialist exam is not very accurate at checking the field-specific skills required. In fact, by the end of the program, specialists obtain a worse final GPA on average. The field-specific exam seems not to match the skills needed to obtain good performance in the degree.

Moreover, policy implications should be drawn carefully. The outcome measures considered (ECTS credits, completion on time, and final GPA) are different in their nature. They all measure student performance but at different levels. ECTS credits (number of credits completed by the end of the first year) evaluates initial performance and conveys limited quantitative information, because it reports whether the student completed a course but not its evaluation; completion on time is a qualitative measure of students’ performance, while the final GPA is a quantitative one. In other words, the first two measures inform the policy maker about how fast the student was on completing her studies, while the final GPA aims to quantify student performance. Hence, the measures considered in the empirical analysis are different regarding their nature, and the number of observations in the estimations is not constant across the three outcome measures, given that they evaluate performance at different moments of the student’s degree. Thus, each outcome measure is estimated on a different sample subject to attrition that might vary by both observable and unobservable dimensions between different groups of interest (generalists, specialists, and all-rounders). To address this issue, I adopt two different approaches. First, I reestimate the models on common sets of students (either those who completed on time or those who finished the degree and have a final GPA). Second, I estimate a selection model.

First, in table A3, I consider the final GPA only for those students who completed the degree on time. There, although the effect for all-rounders is not statistically significant, the effect for generalists remains statistically significant and of the same magnitude.

I also ran additional regressions on credits at the end of the first year on a constant sample (students who finished the degree and have a final GPA) to impose the same level of attrition. These results are presented in table A6. By construction, panel C is the same as panel C in table 9, given that we are considering only students with final GPA. Outcome completion on time (panel B of table 9) was not reestimated on this sample of students, given that the students who have a final GPA invariantly completed on time (94%). In panel A of table A6, which refers to the number of ECTS credits completed by the end of the first year, results on all-rounders (with and without Portuguese) are robust when compared with those in panel A of table 9. The results on generalists are now stronger. This suggests the set of generalists who finish their degree and have a final GPA is positively selected. It might be that only the better generalists survived and finished the degree.

Second, I undertake an analysis of selection in table A4, where I estimate a Heckman selection model. In the selection equation, I estimate the probability of students not dropping out by the end of the first year. In the outcome equation, I model students’ final GPA. Estimates in table A4 are very similar to those in table 9. Generalists and all-rounders obtain a better final GPA on average than specialists.

Nevertheless, other selection issues my play a role in my results. My empirical analysis considers only programs that already allow for Portuguese as an alternative requirement. A skeptical reader might argue that such a policy has different effects according to the goal and type of institution. For instance, in less well-regarded programs, often the goal is to admit as many students as possible. They have an incentive to allow for as many exam combinations as possible in order to fill the number of vacancies. Opening the application process to all types of students might be perceived in different ways. Allowing for different alternative entry requirements can be perceived as a sign of being a less rewarded program. Students that perceive themselves as high-ability students might not apply to those programs.

I acknowledge that the type of institution can potentially drive the results.⁴¹ My results might not be verified for elite institutions. Nevertheless, the reader should keep in mind that those institutions are able to select the top distribution of students, even with only one admission criterion. Regardless of the nature of the admission exams, there is always a high demand for programs in elite institutions. In any case, my results are relevant for the average institution, which represents the majority of the HE system.

Additionally, the reader might ask herself whether my results would depend on the field of study. Even controlling for differences across fields, my main results remain the same. The access to some fields might be more restricted than others and may affect students’ choices (see Hastings, Neilson, and Zimmerman 2013; Kirkeboen, Leuven, and Mogstad 2016). My results might be driven by the type of field that allows more often for Portuguese as an alternative admission requirement. Therefore, the distribution of students across different fields of study is relevant to understand my results.⁴² In the subpopulation of analysis, I observe very few natural sciences programs, and these results might not be representative for that field of study.⁴³

Finally, in this work, I focus only on the public HE system, which is the one for which I observe the entrance and admission exam grades. The private sector enrolls a small share of students, approximately 20% of the system (Biscaia, Sá, and Teixeira 2021). Moreover, over the past decades, private institutions have been facing a declining student demand (Teixeira and Amaral 2007). Given that admissions to the private system are not centralized, we do not observe the entrance grade or admission exam grades for the private sector in our datasets. Nevertheless, the major difference between the private and public sectors is related to the fields of study they offer. The public sector offers all fields, while the private sector offers mainly courses in the fields of social sciences, business and law, and health and welfare (Teixeira et al. 2013). My analysis of the field of study for the public sector (table A5) shows that the performance of the different groups of students (generalists, specialists, and all-rounders) does not vary significantly across fields. Therefore, I believe the bias of not considering the private sector is, most likely, small.

XI.  Conclusion

Intended and unintended consequences of university admission practices have been a focus of the academic, political, and judicial agendas.⁴⁴ I contribute to that debate by analyzing the nature of the admission requirements. I test whether the inclusion of a general admission exam generates a better pool of admitted students.

In this paper, my first contribution is theoretical. I set up a simple model in which a university faces a particular trade-off when deciding to combine a general admission requirement with a field-specific exam. The university gains information about a student’s ability at the cost of reducing the weight allocated to the field-specific requirement. As a result, in equilibrium, the student would decrease the effort level in the productive test (the field-specific exam). Nevertheless, I find that the university benefits from including a general admission test in its requirements.

My second contribution is the provision of an empirical application of the model by using Portugal as a case study. I find evidence that including a general exam as an entry requirement increases the average student performance at university at the margin of admission to an HEI. In particular, I conclude that generalists outperform specialists in terms of final GPA. This is a surprising result that has not been a consistent finding in the literature and has two possible interpretations.

First, universities that allow for general admission requirements should not put too much weight on first-year assessment. Albeit the fact that performance in the first year is often considered as a good proxy to evaluate who the good students are, generalists might need time to adapt in order to obtain a better performance. This is a common practice in the United Kingdom, for instance, where first-year grades do not count for final GPA.

Second, my results show that the field-specific exam may not be well designed in some fields. Students admitted purely on the basis of their field-specific skills perform worse at university. That means the field-specific exam is not accurate at checking the specialist’s skills needed to complete the degree.

Additionally, I also find that all-rounders have better performance at university when compared with specialists. Although this result is expected, it reinforces the idea that having knowledge in multiple subjects can provide all-rounders with a more holistic understanding of the HE system and its interconnections, which can be an advantage in certain academic contexts. Additionally, a broader range of skills and knowledge across different subjects that all-rounders have might be beneficial in multidisciplinary university programs or in scenarios where the curriculum covers a wide range of topics. Nevertheless, it is also true that a student’s performance depends on various factors, including the specific university program, the student’s strengths and interests, and the learning environment, and this should be taken into consideration when interpreting the results.

My findings have substantial implications for the determination of university selection policies. Universities should rethink their admission practices by either designing admission tests that evaluate field-specific skills more accurately or introducing general admission requirements to effectively distinguish their candidates. In short, diversity in admission criteria has a positive effect on student performance at university. However, the empirical results of the paper might not apply to all fields of study, particularly in engineering and physical sciences. In my analysis, I considered only HE programs where there was variation in the type of admission exams required by HEIs, and so my empirical results are not necessarily verified in all HE programs. Nevertheless, the theoretical results are valid for all fields of study.

In fact, my results offer an optimistic view of the possibility to affect the pool of admitted students at university. Even though I focus on the Portuguese system, this paper provides generalizable insights about the importance of the choice of admission criteria.

Finally, this paper poses some interesting questions that are worthwhile to explore in future research. I have shown that the general admission exam has an important role to play in university admission practices. However, this paper does not provide an answer to what should be the optimal weight to allocate to that exam. Additionally, there might be some unintended consequences of increasing the weight associated with the general exam. For instance, by introducing a general admission exam, a university may change the gender composition of the admitted pool of students and/or it can also change the way students rank their preferences when applying to HEIs. In sum, these possibilities open the scope for future research.