The land of the fee: the effect of Baden-Württemberg's tuition fees on international student outcomes

ABSTRACT Despite the increasing number of students learning abroad, little is known about the way international students migrate and how policies influence their decision. This article evaluates one German state's recent policy to charge international students for tuition, while tertiary education remains free elsewhere. For my difference-in-differences analysis, I collect and combine publicly available records for German higher education institutions since 1998. I find that the international enrollment rate decreases by 2 percentage points at treated institutions, driven by African and Asian students. In contrast to state government motivations, I find no evidence for a short-term decrease in exam failure rates.


Introduction
Recent decades have seen an unprecedented increase in the number of students abroad.In 2018, almost six million students enrolled outside of their home country, an increase of 68 percent since 2008.Despite this trend, economic research on the decision-making of students and host societies is only emerging (Beine, Noël, and Ragot 2014, Beine, Delogu, and Ragot 2020, Bound  et al. 2020, Dustmann and Glitz 2011, Arenas 2021).
This paper adds to our understanding by analyzing a recent policy change for international students in Germany, their main destination in Europe and the fourth-largest exporter of tertiary education worldwide.Germany is also the largest host country where English is not the primary language and stands out from other major destinations because tuition fees for public education are largely absent.This is an attractive feature of German education, especially for students from poorer regions worldwide. 1A policy change in one German state (out of sixteen), charging tuition from international students at public institutions only, allows a closer look at how tuition fees change education outcomes and migration flows.
To ease comparison throughout this paper, I refer to students outside of Europe as international students, whereas foreign students include all without German citizenship.Likewise, I use the term institution to refer to all different types of higher education in Germany such as universities, universities of applied sciences (Fachhochschulen), academies, or Hochschulen, and between separate campuses in different locations.Wherever necessary, I emphasize such a specific type of institution.
In 2017, the German state of Baden-Württemberg introduced a fee of 3000 euros per year, explicitly targeting students from outside the European Economic Area (EEA).Similar policies have been adopted in various European countries since the 1980s, especially after 2006 (Cai and Kivistö 2013,  European Education and Culture Executive Agency and Eurydice 2019). 2 The specific targeting of international students raises important questions regarding education and immigration that this article aims to answer: what is the policy's impact on migration flows?Who exactly is affected if the price of higher education increases?Are educational outcomes affected?
The differential treatment, deviating from free education at public institutions in only one state, provides well-defined treatment and control groups for a difference-in-difference analysis. 3The data used herein are collected from publicly available sources, the Federal Statistical Office, and the majority of states' statistical offices.Although both data sets stem from the same sources, the institutions themselves, they vary in depth and their units of observation.For example, one state reports data for each campus of an institution separately, whereas another state may lump them together by location or institution.Institutional mergers, campus closures, and institutions operating in multiple states further complicate the comparison of the two data sets, if they are treated in different ways by the offices.To assess the impact of these reporting differences, I analyze each estimation sample separately, obtaining similar results.
I examine the short-run impact of tuition fees on international enrollment and final exam failure rates at the institutional level.My results indicate that the introduction of tuition fees in the state of Baden-Württemberg leads to a drop of about 2 percentage points in the international enrollment rate there.The policy disproportionally affects students from Africa and Asia in the sense that their shares fall significantly on Baden-Württemberg's campuses.Estimates for the Americas are ambiguous and depend on the specified model.
Moreover, upon its introduction, state officials endorsed the tuition policy on the basis that fees might improve the academic performance of international students, who have historically performed worse than their native peers.Looking at foreign students' failure rate of final examinations, I do not find any evidence for an improvement in the short term.While more common student outcomes such as dropout rates or time to completion are not available, the measure of passing the final exam is rather novel and somewhat more likely among foreign students.
The results in this paper are robust to placebo tests on the policy's location and time, but also alternative specifications and identifications.These tests add further credence to my findings on the impact of the policy on international students' enrollment decisions and performance.
The main contributions of this paper are an empirical analysis of a recent state policy on the enrollment decisions and educational outcomes of a narrowly targeted group.In particular, the geographical distribution of changes in student migration flows adds to our understanding of the interconnectedness of regional educational policies and international students' decision-making.In contrast to previous studies, the policy that I analyze can be considered exogenous and takes place in a major destination for international students.Additionally, this paper highlights the growing importance of education export in non-English speaking countries, as well as the sensitivity of outcomes to the imposition of tuition.Thereby, I provide an estimate for similar policies and whose decisions may be affected.
While my analysis of the impact of the policy on the geographical distribution of students at German institutions is novel, the potential change in enrollment rates due to similar policies has produced a lot of research with contradicting results.To understand the source of the theoretical ambiguity, consider the following thought experiment: before the policy change, institutions fill all vacancies with approved applicants from Germany and abroad.The introduction of tuition fees might change a poor international student's mind such that she applies to another German institution.Her decision leads to a vacant spot, which is allocated to a domestic student, and so the international enrollment rate decreases.
Alternatively, the vacancy could be filled by a more affluent international student whose credentials surpass domestic applicants.In case such a student is accepted, the institution is presumably able to finance his spot each semester upon payment.If his tuition covers the spot fully, no other student is affected.In this scenario, no native student is involved as all vacancies transfer from one international student to another.Such transactions may even be without consequences for continental enrollment shares if both affected students hail from the same continent.For a modest amount of tuition charged, it is also possible that the institution can increase revenue without tradeoffs.Hence, there may be no effect of tuition on enrollment rates.
For well-renowned universities, the tuition charged could result in additional funds without stark changes in student numbers.This circumstance provides an incentive to broaden the university's international outreach, possibly at the expense of natives, especially if its pool of applicants is large.Less popular institutions, in contrast, may see a decrease in enrollment.In the extreme, international students apply elsewhere, thus causing less revenue, while being replaced by e.g.native students with lower skills.The institution in this scenario ends up with lower revenue and a decline in quality.Thus, the policy change might lead to a redistribution of students to institutions, possibly displacing higher-skilled natives.
There are other channels through which fees may affect enrollment.Expecting adverse consequences as a perceived lower probability of admission, some native students may be discouraged from enrolling in the treated state at all, similar to findings by Hübner (2012) and Dwenger,  Storck, and Wrohlich (2012). 4When looking at the impact of abolishing a time-consuming entry exam, Arenas (2021) finds an increase in enrollment (price effect) but also improved performances of the average student (quality effect).In contrast, if tuition fees signal quality, international enrollment rates may increase.On a broader level, following Kato and Sparber (2013), restrictive migration policies, in general, may reduce foreign applicants' quality.Likewise, lifting such restrictions on specific fields affects native groups in their choice of major, occupation, or labor force participation (Ransom and Winters 2021).Garibaldi et al. (2012), who focus on the link between tuition and degree completion at an Italian university, find a negative association.
To summarize, the adoption of tuition fees puts a markup onto a previously free, but restricted, good.Following the previous thought experiment, a reduction in international enrollment seems plausible, but not necessary, if institutions continue to restrict admission as before and applications exceed available spots.Thus, in theory, the potential effects of the policy change are ambiguous.
Empirical results from existing studies reflect this ambiguity.Previous research on different fees has focused primarily on the US and the UK, with mixed evidence. 5Apart from Beine, Delogu, and Ragot  (2020), who examine international enrollment in Italy, most studies focus on comparison across countries, thus neglecting regional variation within a single country and potentially raising endogeneity concerns.Beine, Delogu, and Ragot (2020) find a strong negative effect on international enrollment when universities increase their tuition.This paper adds to this strand of the literature by focusing on state-level policy changes on outcomes at the institutional level.My findings are consistent with Beine, Delogu, and Ragot (2020) and also show similar effects for state-mandated tuition policies.
My work is also closely related to research on the introduction of state-wide universal tuition fees in seven German states in the early 2000s, which generated a considerable amount of research, however contradictory in their findings.Hübner (2012), Bruckmeier and Wigger (2014), Dietrich  and Gerner (2012), and Dwenger, Storck, and Wrohlich (2012) focus on the implications of these policies for the enrollment and application decisions of high school graduates. 6Hübner (2012) finds a strong negative and significant impact (−2.7 percentage points) on high school graduates' enrollment rates, even −4.7 percentage points considering spillover effects (as tuition may affect high school graduates in non-fee states).For the same data, applying additional controls and bearing variation in states and over time in mind, research by both Bruckmeier and Wigger (2014) and Alecke,  Burgard, and Mitze (2013) finds no negative effect of tuition on the enrollment of high school graduates.My study differs from these papers in several ways.I analyze a state policy at the institutional level and emphasize the impact on foreign and international enrollment rates as well as possible implications for both international and native students.
For the same policy, Dwenger, Storck, and Wrohlich (2012) investigate the impact of tuition for medicine and dentistry applicants, both programs being centralized in Germany; they find that the probability of applying in one's home state decreases if tuition is charged.Moreover, their results indicate that the composition of students across states changes because better high school graduates are more likely to apply to universities in their home state.Bietenbeck, Marcus, and Weinhardt (2020) also evaluate the enrollment of current and prospective students in Germany on previous fees, when switching from free education to charging tuition fees.They find that tuition accelerates graduation, and generates financial gains for universities, but does not affect educational attainment.

Tuition fees for international students
In contrast to other developed countries, tuition fees are not common for public education in Germany.Apart from Hörergeld of 150 Deutschmarks per semester until 1970, tertiary education was free for all students at universities until the 2000s.Education is the domain of individual states, and so, beginning in 2006, seven states implemented universal fees.A particular reason was to motivate long-term university students (those exceeding the standard period of study by more than four semesters) to graduate.The typical fee was 500 euros per semester. 7Persistent public dislike led to their overall abolishment by 2014.
In March 2017, the state of Baden-Württemberg passed a new law (Gesetz zur Änderung des Landeshochschulgebührengesetzes und anderer Gesetze) to charge 1500 euros per semester from students outside of the EEA at public institutions.Exemptions are made for those with a German high school degree, also obtained abroad (Bildungsinländer), those married to or a child of an EEA citizen, as well as refugees and residents of Germany for five years or more.Further, institutions can negotiate bilateral exchange agreements and support talented students from the least developed countries.Likewise, exchange students are not subject to payments.The law also recommends against charging disabled students or those facing emergencies.Lastly, the fee does not apply to institutions for Public Service (Hochschulen für den öffentlichen Dienst) such as administration and policing.According to newspaper reports, about half of international students meet the criteria for exemptions. 8eports of the first announcement regarding this policy change circulated in October 2016 and insinuated cost-saving effects.In the following spring, the state parliament passed the law, which became effective in the 2017/18 winter term.The delay in legal adoption gave international students ample time, if aware of the prospective policy change, to reevaluate their potential applications or continue their studies at another institution within Germany.Although there may exist strong regional preferences for attending a university in Baden-Württemberg (such as higher reputation, access to local networks, or amenities), international students who desire to graduate in Germany have plenty of alternatives.Early reports expected the fee to apply to 7000 currently enrolled students and generate revenue of up to 45 million euros annually by 2022.Based on the data used below, total revenue of 42 million euros between 2017-2019 seems more realistic.
The responsible state ministry claims that the tuition guarantees improved conditions, in particular by reducing the excessive dropout rates of international students and by providing better support for the continuous growth of international student numbers.According to law, 300 euros of a student's semester fee remain at the institution to cover and further improve the general conditions for international students.The remaining amount of 1200 euros flows into the state budget.
At the time of implementation, the state ministry referred to similar fees in other countries, which ranged from 1500 euros in Austria up to 20,000 euros more in the Netherlands and the United Kingdom.Across the EU, fees are usually higher for international than for national students.For Bachelor students, other European countries charge fees of comparable magnitude but up to 10,000 euros for students in England and Wales, where institutions can determine fees themselves.For graduate students, Ireland charges a continental maximum of 30,000 euros (European Education and Culture Executive Agency and Eurydice 2019). 9Thus, in comparison, the tuition fee discussed herein is relatively modest.

Data
For my analysis, I collect and combine publicly available data, covering institutions of higher education, from two sourcesthe Federal Statistical Office of Germany (Statistisches Bundesamt (Destatis) 2020b) and the states' statistical offices.Institutions report to the state's entity, which then forwards these numbers to the federal level, so both sources are indirectly linked.
State and federal data are not aggregated at the same level, differ in their depth of information, and thus they are not readily comparable. 10Most importantly, the data from the Federal Statistical Office lump foreign students into one category, while the states provide additional information on citizenship by continent.Hence, the federal data exclusively refers to foreign students, and the states' data to international students.
The federal data include observations on students' gender (male/female) and nationality (German/foreign) at campus levels between 1998 and 2019.I merge the Federal student statistics with an additional data set from the same source covering final examinations before graduation to evaluate student performance (Statistisches Bundesamt (Destatis) 2020a).Descriptive statistics in Table 1 show that failure in the final examination is a rare event, but somewhat more likely for foreigners.Unfortunately, other relevant measures such as dropout rates or time to completion were not available.While failed final exams only apply to students nearing graduation, and thus also come with a delay for the policy discussed, they give an idea of the policy's impact on students' quality.One additional caveat is that such exams often require an oral defense and thus there is a lower degree of anonymity than for other performance measures.
To deal with the imprecision in students' origins, arising from the binary value, I have also collected data from the states' statistical offices. 11Those data identify international students' continents of origin by citizenship, which allows me to observe changes in migration patterns.
Given mergers and openings of new campuses and institutions, resulting in different units of observation, I refrain from assembling all information into one data set and run two separate analyzes on the policy's effect on both foreign and international students.Doing so allows me to draw more insightful conclusions on enrollment rates by continent of origin and student success, which is not included in the states' data.
Because most students from European countries outside of Germany will be covered by bilateral agreements (as in exchange programs like Erasmus), I exclude those from international students. 12oth data sets include years from 1998 to 2019 to cover the number of foreign and international students.For any year, student data, such as enrollment numbers and the share of males, refer to the beginning of the winter semester, starting in October.Many institutions only accept new enrollment for the winter semester.I also identify public institutions of tertiary education separately in case students value these differently than private institutions.To account for other factors relevant to migration decisions, I obtained state and regional GDP per capita from the statistical portal of the Federal and states' offices (Arbeitskreis Volkswirtschaftliche Gesamtrechnungen der Länder 2020), and unemployment rates from Eurostat (2020).Additionally, I control for the share of foreigners within each state, taken from the Federal Statistical Office, and an institution's Shanghai (ARWU) alumni rank, which is a measure of institutional quality. 13Lastly, highschool reforms (G-8 Abitur), which temporarily influence the number of graduates within a state, enter as a binary variable per state and year.All of these variables are included in both samples.
In total, my panel for the federal data includes 10,603 observations for 698 institutions or locations.The second sample, collected from the states themselves, contains 6921 observations for 466 institutions.Between 1998 and 2019, several institutions opened, closed, or merged campuses, leaving the panels unbalanced.Data provision by the states' statistical offices explains large parts of the gap.Table 1 lists the descriptive statistics for relevant variables in the analysis.
For all key variables, both samples are comparable as means are within one standard deviation.If the states' sample was complete, it would overlap entirely with the federal one.The different means in Table 1 for unemployment rates and GDP indicate that economically stronger years or states are slightly overrepresented in the states' sample, as are campuses with more students.The slightly higher number on ARWU rank in column (2) signals that prestigious campuses are more likely to be included in the state sample.Despite the slight differences between the two samples, student characteristics are similar in both.
International student numbers in Germany have been increasing in recent years.Figure 1 illustrates that both the number and share of foreign students go up in comparison to 1998.After an initial decline in the share, both the number and the share have increased monotonically since 2012.The graph also illustrates that the increase in international student numbers is driven by foreign students, but not at the expense of Germans.Even though foreign enrollment has been more than doubling, the overall fraction remains relatively low. 14One reason for the increase is the Bologna Process, which streamlined the educational system within Europe, thus facilitating the recognition of foreign degrees.

Event study
As outlined above, the data do not report exemptions at the individual level, hence estimations below focus on average effects. 15To provide preliminary evidence on the effects and validity of the identification strategy, I first conduct an event study of the dynamic effects of the policy on foreign and international students, at the institutional level, controlling for the usual unit and time-fixed effects, as well as a vector of time-varying observable characteristics with the following model: Equation ( 1) captures the effects of tuition fees on the international enrollment rate, the dependent variable international it , at treated institutions.The coefficients on the lead represent tests for pretrends, and those on the lag capture the dynamic effects of the policy.The vector X it consists of all the covariates that are listed in Table 1.These include the share of male students to account for gender disparities in the student body, and whether the institution had implemented a tuition fee previously in a particular year.Student numbers and shares are taken or derived from the original samples. 16Standard errors are clustered at the institutional level i.I focus on enrollment rates for two reasons: first, the total of enrolled students differs widely between institutions, even more so for international students.Most institutions are rather small but may attract a large fraction of international students, which would be obscured when looking at the mere total.Second, using a logarithmic scale eliminates observations with zero students, which distorts the analysis.In contrast, as shown in Figure 1, the share of foreign students is fairly stable and reflective of changes within the student body.The shares also include observations for year-institution pairs with no international students enrolled.Nevertheless, I discuss alternative approaches, in particular on the use of zero value, in Section 5.
Corresponding to the event study specification in Equation (1), Figure 2 illustrates the impact of the policy change in foreign enrollment rates, using the federal data, with and without controls (Jann  2014).To allow for more variation at the local level, I use GDP per capita at the NUTS 2 level. 17The year before the implementation (2016) serves as a benchmark and is thus omitted from the regression.Whenever data are available, I include all post-periods, i.e. q 1 ≤ 3 in Equation (1).The first leading variable includes all years from 1998 to 2009, and so p 1 = 7.
The foreign enrollment rate in Baden-Württemberg is larger than in control states for years before the introduction of tuition fees but the difference diminishes over time.Without adjusting for timevariant covariates, there appear to be pre-trends differing across states.Yet, the difference in foreign enrollment is statistically indistinguishable from zero for all years save 2010, thus being consistent with the parallel trend assumption when controls are considered.
According to Figure 2, the tuition fee reduces foreign enrollment by 1.0-2.3percentage points.The deviation from the previous trend is highly statistically significant for all lags.After international students are subject to tuition, the share of foreign enrollment decreases with a change in trend after implementation.The gap also becomes larger over the years, which may hint at a better knowledge of the policy or a network effect that affects students' decision-making.The lack of individual data on migration decisions prevents a clearer picture of these channels, however, the discussion below highlights the importance of geographical variation that may play a role.
For international enrollment, using the states' data, the dynamics shown in Figure 3 mirror the previous findings.The inclusion of controls results in pre-trend estimates that are statistically indistinguishable from zero.After tuition is charged, international enrollment rates decrease significantly by 0.9-2.8percentage points with similar trends as above for the foreign share.
Both graphs support the hypothesis that a deviation from zero-price education for international students reduces their enrollment rates.Despite the smaller and incomplete sample but higher precision, both results are reassuringly similar.
Further validity for these preliminary findings may be provided by replacing enrollment shares with student numbers.Such figures (not included herein) depict similar trends for the total numbers of foreign and international students.Therefore, I focus on enrollment rates below, which also facilitate comparison across institutions of different types and sizes.

Difference-in-differences analysis
For my main analysis, I apply a multivariate panel regression to estimate the effects of the policy with the following model: where international it refers to the variable of interest, e.g.international enrollment rate, while Post is a binary variable on the post-treatment period starting in 2017.Fee it indicates public institutions in Baden-Württemberg, i.e. the treated institutions, state i and year t are state and time-fixed effects and X it is a vector of all other covariates as outlined above.The main coefficient of interest is β, which describes the effect of the policy on international enrollment into higher education in Baden-Württemberg.
In further analyses, I apply Equation (2) to evaluate geographical variation for international students' regions of origin as the outcome variable.Likewise, I estimate the impact of tuition on student performance in terms of failed final exams of foreign students.

Enrollment rates
Under parallel trends, estimates of β in Equation ( 2) represent the effect of fees on international enrollment rates.As no other state introduced such a policy, their institutions and private ones in Baden-Württemberg form a natural control group.
Table 2 summarizes regression results in separate panels for foreign and international students, listing the results for the federal and states' data respectively.The estimates indicate a highly significant drop in foreign enrollment rates in Baden-Württemberg after tuition fees were introduced for international students.Despite the use of a different, and much smaller, sample, estimates are similar for both foreign and international students.
This finding also holds when only considering public institutions in the last two columns of each panel.Charging tuition from international students is associated with a decrease in their share of the student body by about 2 percentage points in the short run.For all specifications, estimates are similar and highly statistically significant.Note: The columns show the estimates for foreign enrollment rate (columns 1-4) and share of international students (columns 5-8) on a set of independent variables.Students with citizenship other than German are considered as foreign, while international students include those from outside of Europe.Tuition fee refers to public institutions in Baden-Württemberg, which introduced tuition fees for international students in 2017.In line with Figures 2 and 3, the enrollment rates of treated students are declining over the sample period, whereas the descriptive statistics and Figure 1 show that they are increasing on average in Germany throughout the period.The insignificant and sometimes negative estimates for rank are contrary to previous findings in the literature but may be absorbed by other covariates or reflect the relatively high international enrollment rates across Germany. 18Reforms, which reduced the years of schooling to twelve years (G-8 Abitur), temporarily increased the enrollment of native students in a state the same year or prior.These reforms do not affect the share of foreign or international students at statistically significant levels.An increasing number of foreign high school graduates may explain why these coefficients remain positive.Lastly, Table 2 highlights preferences for stronger economic and more culturally diverse regions for students' enrollment decisions.

The geographical distribution of changes in enrollment rates
As one may expect, international students react to financial incentives.The states' data provide an opportunity to consider continents of origin and assess whose behavior the policy affects, and how.Unfortunately, more detailed data on countries were not available for all states considered.Although continents are a crude measure, as they may ignore differences across countries on the same continent such as trade links and colonial past, they also provide clearly and exogenously defined groups for comparison by difference-in-differences. This allows me to examine heterogeneity in the effects of international fees on confined groups by place of origin.
Table 3 lists the changes in continental shares upon the policy change in tuition.Controls are not reported here.According to the underlying hypothesis, shares of the student population follow similar trends over time.Shares for international students do not differ at treated institutions in Baden-Württemberg, as indicated by Tuition fee.None of the continents experienced a statistically significant divergence in the Post period in control states.
The claim for lower enrollment rates due to the policy cannot be rejected for Africa, the Americas, and Asia.Coefficient estimates indicate a drop of 0.4 percentage points for African students and a (less statistically relevant) decline of 0.2 percentage points for Americans.The impact is strongest for Asians, whose post-treatment enrollment share fell by about 1.1 percentage points.The estimates are approximately equivalent to a decrease of 37.7 percent (total international share), 59.2 percent (African share), 20.3 percent (Americas), and 29.9 percent (Asia). 19uropean students are by and large not subject to the policy and thus unresponsive to the policy introduction, thereby providing further credibility to the policy's impact.The absence of effects for Oceania and Other (i.e.unknown origin and stateless students) likely reflects the small sample sizes or greater variance in enrollment rates, which is supported by the panels in Figure 4.These illustrate the per-year changes in reference to the year of 2016, i.e. before the introduction of tuition fees.The graphs align with the significant changes found in Table 3, except for the Americas.None of the per- year estimates is statistically significant after tuition was charged, whereas pre-trends diverge significantly from 2016 on, thus casting doubt on a particular impact of the policy on American students.
The larger impact of tuition on international students could potentially be explained by a gravity model, wherein students migrating at higher social costs prefer less financial costs (Bessey 2012,  Beine, Noël, and Ragot 2014).As education in a foreign country comes with higher social costs, students may aim to study at lower financial costs.Thus, tuition fees may provide an important disincentive if students have weak location preferences. 20For the state of Baden-Württemberg, non-accessible data by country (except for China and selected European countries) prevent a more detailed analysis of compositional effects.Over the post-period, the total number of Chinese students increased by exactly 17, which contrasts with increasing numbers elsewhere.This may be of particular importance.As noted above, the enrollment of young Syrian refugees should not be relevant as they are exempt from paying tuition.

Failure rates of final exams
As pointed out previously, Baden-Württemberg's government claimed to improve the performance of international students by making them pay for education.To test this claim, I apply Equation ( 2) with the failure rate of foreign students in final exams before graduation as the dependent variable.Typically, these exams include bachelor's and master's theses.
Facing a switch from no to any fee may provide students with an incentive to change behavior such as dropping out, switching to a non-fee institution, or accelerating graduation. 21Unfortunately, there are no publicly available data on dropout rates by institutions, which means that only students approaching graduation are present in the data. 22Also, keep in mind that the federal sample used includes only information on foreign students.All institution-year pairs with at least one observation are considered, whereas those with no specific value on final exams are treated as missing.If there is any value reported for an institution in a specific year, I assume that the unreported numbers of passed and failed exams are zero.
Recall that the descriptive statistics in Table 1 show that foreigners are slightly more at risk of failing final exams.Figure 5 indicates an initial increase and subsequent decline in foreign students' failure rate since the implementation of tuition fees, though this is not statistically different from the reference or prior years.The deviation from zero for foreign students' failure rates at Baden-Württemberg's public institutions in 2009 (or earlier) signifies that not all variation is picked up by the model.
The difference-in-differences specification, which pools over all periods, provides inconsistent results on average during the post-implementation period.Estimates are listed in the first two columns of Table 4.If controls are included, the interaction term becomes statistically significant, albeit marginally at the ten percent significance level.In this sense, foreign students are less likely to fail their final exams at treated institutions in the post-period, in line with the state government's claims.
However, the event study in Figure 5 points toward a more relevant change after 2012, not being causally related to the discussed imposition of a tuition fee.The event study estimates suggest that after the introduction of tuition fees, the share of failed final examinations of foreign students increases at first, but then decreases again.A possible explanation for this pattern could be that already enrolled students, attempting to avoid payments by graduating early, increase the failure rate temporarily.
Apart from such short-run consequences, this finding suggests that a sizable improvement in international students' performance, as anticipated by the state government at the time of implementation, has yet to realize.So far, there is no strong empirical support for the state's claim that the raised funding leads to a quality effect.
One potential explanation could be that the fees discourage foreign students despite their quality without an impact on student performance.In addition, the reallocated money per student (300 euros) may be inadequate for quality improvement.Given the decrease in international student rates altogether, fewer funds are raised and spent than anticipated, and thus do not advance institutional support.Alternatively, international students need to spend more hours working to cover the tuition than prior cohorts.In that sense, learning conditions differ for pre-and post-periods, which could obscure a potentially positive quality effect as well.For a better understanding, individual-level data analysis is necessary to evaluate student performance in more detail.

Robustness
I use several strategies to judge the robustness of my estimates to alternative interpretations.To help assess whether the difference-in-differences estimates are causal, I generate placebo treatments in states where no fees were charged.If my results are consistent with the underlying theory, I would not expect any visible effect on these placebo treatments.Students may enroll in other states' institutions after Baden-Württemberg implemented fees.Such a substitution may lead to positive estimates in other states.On the other hand, negative estimation coefficients may challenge the robustness of my findings.No other state introduced international tuition fees during the observed period.Figure A1 summarizes the analysis of foreign enrollment shares for all other states. 23However, none of these nonfee states shows the same pattern as the fee state in the pre-and post-period.Bavaria, e.g. is arguably the most similar in terms of size, population, economy, and political preferences to Baden-Württemberg.If anything, the post-treatment estimates imply an increase, e.g. a substitution to non-fee institutions in Bavaria.
Post-period trends are decreasing in the states of Hesse, Lower Saxony, and Schleswig-Holstein.Yet, there are pre-trends for at least one specification in each state and outcomes are nowhere near as pronounced as for Baden-Württemberg, especially for both foreign and international student shares.For average rates, only Schleswig-Holstein observes negative estimates for both foreign and international enrollment shares.These are, however, not reflected in the event studies, in contrast to Baden-Württemberg's. 24 Thus, I consider these checks as evidence that the change in tuition policy led to a drop in Baden-Württemberg's foreign and international enrollment rates.Following the reasoning above, Figure 6 shows the total international student number for Baden-Württemberg (i.e.treated and untreated institutions) and Bavaria.The figure provides further support for the parallel trend assumption, which also holds for the foreign student number (not pictured).In both states, the pre-trend is fairly stable, apart from a slight lag in 2012/3, which coincides with abolished tuition fees and high school reforms discussed above.In the relevant post-treatment period, the total number of international students declines in Baden-Württemberg, whereas it continues to increase in Bavaria.
In an additional check, I follow Silva and Tenreyro (2006) by applying a Poisson Probability Maximum Likelihood Model to consider the large amount of zeros in the total numbers, especially among international students and failed exams.The results continue to be significantly negative for Post × Baden-Württemberg across all specifications for foreign and international students  (Table 5).For continental groups, the previous results hold, although the relative magnitudes differ a bit: the impact is now largest for African students, whereas the low share of non-EEA students could explain the negative effect for Europe (Table 6).The impact on failed exams remains statistically irrelevant (Table 7).For further evidence, I implement an inverse hyperbolic sine transformation, which approximates the natural logarithm of my outcome variables while keeping observations with student shares of zero (Bellemare and Wichman 2020, Amuedo-Dorantes, Shih, and Xu 2020).Since shares are nonnegative, my estimates follow the arcsinh-linear case transformation by Bellemare and Wichman (2020).The results in Tables 8-10, though not readily interpretable as percent changes, resemble the previous ones closely.I consider these results to support my findings above.
I have also conducted a synthetic control approach, where the other institutions replicate the foreign share at treated ones.Figure A2 shows the changes in foreign share over the years.As one may expect, the synthetic control continues to increase, whereas the foreign share at fee institutions levels once the policy is in place.I consider this divergence of trends, which coincides with the policy change, as further evidence for my hypotheses.
To allow for non-linearities in the share and other determinants, I replace the share with the log odds of being a foreign student.The dependent variable in each case is constructed by P = ln (share) 1−ln(share) .Table 4 indicates that there are no systematic deviations in enrollment rates.Students' failure rates are statistically insignificant, strengthening the case for no impact of the tuition fee on student performance.

Conclusion
In this paper, I link the policy of tuition fees, affecting only international students in higher education in the German state of Baden-Württemberg, to a decrease in international enrollment rates, i.e. a price effect.My estimates are consistent with those in both Hübner (2012) and Dwenger, Storck,  and Wrohlich (2012), although theirs target a more narrow group.Qualitatively, my estimates are comparable to Beine, Delogu, and Ragot (2020) for Italy in the sense that there exists a negative relationship between tuition fees and enrollment shares of international students.
There are sizable differences across continents of origin in the effect of charging tuition fees on international students.The enrollment of African and Asian students exhibits the most sensitivity to the implementation of fees, which suggests that their geographical preferences are weak.The imposition of tuition fees causes African and Asian students to substitute from Baden-Württemberg to institutions elsewhere.The measure of student performance in passing a final exam before graduation is a novel feature in the literature.In contrast to the official purpose, quality improvement has not yet translated into a measure of student performance, which is consistent with experiences in Denmark (Cai and Kivistö 2013).However, it is perhaps too premature to draw general conclusions about students' overall performance.All results are robust across a battery of placebo estimations.
The findings in this paper have implications for education policy.Estimates suggest that charging only international students significantly changes the composition of the student body.Even the modest amount of 3000 euros per year disincentivizes international students.Potentially, institutions can offset the found price effects by visible quality improvement.Such a quality effect would be necessary to attract the immigration of highly qualified students and facilitate their integration.From another perspective, the policy may be successful to do so by selecting students, who are more likely to stay in the vicinity.Further research is needed to assess the policy's long-term outcomes.

Notes
1.All numbers above are taken from Heublein and Hutzsch (2021, Fig. 2).Note that there are no official records for China.As discussed in more detail below, several German states charged tuition for all students between 2006 and 2014.2. As of 2023, the EEA includes all current EU member states as well as Iceland, Liechtenstein, and Norway; Croatia provisionally applies the EEA agreement since 2014.3. Additionally, the Free State of Saxony allows a similar measure for its institutions.To the best of my knowledge, only Hochschule für Musik und Theater Leipzig has taken advantage of the possibility by charging non-EU citizens for most courses since 2013.I exclude this institution from my analysis.Moreover, the Free State of Bavaria has introduced a similar policy to the one discussed herein in 2022, which is not included in the sample.4. Both of these papers focus on student mobility of German high school graduates under a universal fee.For the same policy, Alecke, Burgard, and Mitze (2013) and Bruckmeier and Wigger (2014) find no such effect.5. See Beine, Delogu, and Ragot (2020) for more details and Bound et al. (2021) for a review of current trends in US Higher Education.6.As in the papers mentioned, I do not consider second-round migration decisions as in Lange (2013).7. Exceptions were possible at the university level in the states of Bavaria and North Rhine-Westphalia but were only used in two cases.As I only focus on the presence of fees below, I ignore the lower fees at the universities in Bielefeld and Münster.To the best of my knowledge only the University of Münster deferred tuition until 2007, which is considered in the analysis below.It should also be noted that German higher education is relatively selective as only about half of the native population is entitled to enroll.8.The large increase in Syrian, and thus Asian, refugees in 2015 should not qualitatively change the results below as there is no plausible impact on the decision of where to study given the exemption.There is also a considerable delay until they start with tertiary education.For more information on exemptions granted see https:// www.stuttgarter-nachrichten.de/inhalt.semestergebuehren-in-baden-wuerttemberg-nur-jeder-zweiteinternationale-student-zahlt.19c5e4d6-65bf-482c-93c5-c2906032af48.html.9.I want to thank an anonymous referee who brought this source to my attention.10.To match reports with summarized data only, I combine campuses affected by mergers, renaming, or recoding in the same region to one observation.A full list of such instances can be found in Appendix 1. 11.All major destinations are present in the states' data.Existent data protection regulations prevent usage from educational institutions in Rhineland-Palatinate. Additionally, Lower Saxony, Mecklenburg-Western Pomerania, and Saxony did not reply to data requests.For Bremen, only years since 2000 are included, while data for Berlin, Brandenburg, and Saxony-Anhalt cover only years from 2007 onward.Hamburg and Schleswig-Holstein round the number of enrolled students up or down to threes.12. Doing so also mitigates the problem of having foreign citizenship but having completed an education in Germany, which mostly applies to Turks, but also Bosnians and Kosovars.One potential confounder remains as Tunisians and Moroccans with a German high school degree, i.e.Bildungsinländer, account for around onefourth of international students.In 2019, the 21,400 Erasmus students in Germany made up 6.7 percent of all international and less than 20 percent of European students.Baden-Württemberg had the highest ratio of international students with 10.6 percent (Deutsches Zentrum für Hochschul-und Wissenschaftsforschung and Deutscher Akademischer Austauschdienst 2021).13.There are no data available for 2003, the earliest year of the ranking, so I use the Nobel rank for that year.The values for rank follow Beine, Delogu, and Ragot (2020); I cap ranks at 500 to allow for consistent comparison over all years and calculate the value for ARWU by (500 + 2) − ranking.In additional checks on institutional quality, I control for the inclusion of a university, a graduate school, or a cluster (all at the institutional level) in a federal elite program (Exzellenzcluster); results are unchanged.14.According to Statistisches Bundesamt; DZHW-Berechnungen, the fraction of Bildungsinländer is remarkably stable, fluctuating between 2.87 and 3.5 percent between 1998 and 2019.15.Income differences across continents potentially discourage students, who would have applied in the absence of a fee.I argue that visa regulations to permit work after immigration as an international student ought to assuage such concerns.These assumptions are supported by the most recent figures from 2016 as employment rates for low and lower-middle, high, and upper-middle-income countries all fall between 46 and 52 percent (Apolinarski  and Brandt 2018, p. 45).16.The inclusion of an index for the cost of living does not substantially affect the results presented herein but reduces the sample size to 10 percent of the used sample (data source: numbeo.com).Its estimates are statistically equivalent to zero and do not affect my conclusions.In consequence, I do not consider the cost of living further.17.As data for regional GDP per capita were unavailable for 2019 at the time of the study, there exists a trade-off regarding its inclusion as a control.Figures 2 and 3 show that exclusion does not sufficiently alter estimates when other controls enter the regression.In consequence, state GDP serves as a control in further estimations.18.A potential confounder is the non-consideration of the prestigious Humboldt-Universität zu Berlin in the ranking.Enserink (2007) offers insights into the underlying quarrel.19.Anecdotal evidence shows a drastic decline in the African share in Heilbronn, Karlsruhe, and Reutlingen.See https://www.stuttgarter-zeitung.de/inhalt.studiengebuehren-fuer-auslaendische-studierende-studenten-ausafrika-bleiben-weg.73e213fb-4a51-4806-90e8-b6e100974e31.html and https://www.forschung-und-lehre.de/ lehre/einige-hochschulen-leiden-unter-studiengebuehren-1240/.20.A gravity model allows to evaluate previously inaccessible factors of international migration flows such as the network effect, poverty constraints, and linguistic and cultural links (cf.Beine, Bertoli, and Fernández-Huertas  Moraga 2016).With all institutions in place, such a model would include 698 destinations for dyadic pairs with the continents of origin, which would be beyond the scope of this paper.Moreover, the lack of more fine-grained data will also misspecify inward multilateral resistance because I am unable to identify countryspecific changes within the dyads.21.Bietenbeck, Marcus, and Weinhardt (2020) find that mid-2000s tuition fees accelerated the time of graduation for enrolled students, but decreased the overall rate of enrollment.22. Dropout rates are particularly high for international bachelor students.Most recent numbers for 2014 reveal that almost half end their studies without a degree, whereas only up to 30 percent of Germans do so.For master programs, dropout rates for 2016 are 26 percent for internationals and 17 percent for Germans.The rates are similar across continents, except for remarkably high dropout rates of up to 46 percent of African master students (Heublein and Hutzsch 2021, p. 55).
23.The trends follow similar paths for international enrollment rates, whenever data are available (not pictured).Furthermore, a similar placebo test for non-treated private institutions in Baden-Württemberg gives pre-and post-treatment estimates that are statistically equal to zero, possibly reflecting small sample sizes.Using alternative years as another placebo treatment does not challenge the findings above.24.As Lower Saxony is not included in the states' data, I cannot draw further conclusions here.The event study for the foreign student share does not indicate a clear negative trend.

Figure 1 .
Figure 1.Number and mean share of foreign students at German Institutions of Higher Education (1998-2019).Both total number and share of foreign students have been increasing relatively to 1998.Data source: Statistisches Bundesamt (Destatis) (2020b).

Figure 2 .
Figure 2. Dynamic effects of foreign enrollment rates in German Institutions of Higher Education.Foreign students include all students without German citizenship.The data cover the period between 1998-2019, years until 2009 being included in the first estimate.The year before the policy change (2016) serves as benchmark.Treated institutions are public and located in the state of Baden-Württemberg, all others serve as control group.The graph pictures the estimates with a 95 percent confidence interval.Estimations include the full set of covariates: number of students, share of males, ARWU rank, public institution, previous tuition fee, high school reforms (G-8) in state and year, share of foreigners in state, GDP per capita, and unemployment rate.Specifications differ by the inclusion of log GDP per capita as controls.Regional controls refer to NUTS 2. Data source: Statistisches Bundesamt (Destatis) (2020b).

Figure 3 .
Figure 3. Dynamic effects of international enrollment rates in German Institutions of Higher Education.International students include all students from outside of Europe.The data cover the period between 1998-2019, years until 2009 being included in the first estimate.The year before the policy change (2016) serves as benchmark.Treated institutions are public and located in the state of Baden-Württemberg, all others serve as control group.The graph pictures the estimates with a 95 percent confidence interval.Estimations include the full set of covariates: number of students, share of males, ARWU rank, public institution, previous tuition fee, high school reforms (G-8) in state and year, share of foreigners in state, GDP per capita, and unemployment rate.Specifications differ by the inclusion of log GDP per capita as controls.Regional controls refer to NUTS 2.Data source: States' statistical offices, 2020.

Figure 4 .
Figure 4. Event studies for international enrollment rates in Baden-Württemberg by continents.The horizontal reference line at zero represents values for 2016.All estimations include the full set of covariates: number of students, share of males, ARWU rank, public institution, previous tuition fee, high school reforms (G-8) in state and year, share of foreigners in state, GDP per capita (at regional level), and unemployment rate.(a) Africa.(b) Americas.(c) Asia.(d) Europe.(e) Oceania.(f) Other.Data source: States' statistical offices, 2020.

Figure 5 .
Figure 5. Dynamic effects of foreign students' failure rate in final examinations.The data cover the period between 1999-2019 with the excluded year of 2016 as the reference line.Treated institutions are located in the state of Baden-Württemberg, all others serve as control group.The graph pictures the estimates with a 95 percent confidence interval.All estimations include the full set of covariates: number of students, share of males, ARWU rank, public institution, previous tuition fee, high school reforms (G-8) in state and year, share of foreigners in state, GDP per capita, and unemployment rate.Data sources: Statistisches Bundesamt (Destatis) (2020b 2020a).

Figure 6 .
Figure 6.Total number of international students in Baden-Württemberg and Bavaria.The figure depicts the total number of international students in Baden-Württemberg (treated and untreated institutions) and in Bavaria.The vertical line represents the introduction of tuition fees for international students in 2017.Data source: States' statistical office, 2020.

Table 1 .
Descriptive statistics for both samples on variables used.Column (1) includes data from the Federal Statistical Office, whereas column (2) considers data from the states' statistical offices on campuses of German institutions of higher education.

Table 2 .
Panel regression results for foreign and international enrollment rates.

Table 3 .
Changes in continental shares of total student populations of institutions.Tuition fee refers to public institutions in Baden-Württemberg, which introduced tuition fees for international students in 2017.All estimations include the full set of covariates: number of students, share of males, ARWU rank, public institution, previous tuition fee, high school reforms (G-8) in state and year, share of foreigners in state, GDP per capita, and unemployment rate.Standard errors in parentheses.*p<0.1, ** p<0.05, *** p<0.01.

Table 4 .
Changes in foreign students' final exam failure rates.−ln(share)).Tuition fee refers to public institutions in Baden-Württemberg, which introduced tuition fees for international students in 2017.The estimations in columns (2) and (4) include the full set of covariates: number of students, share of males, ARWU rank, public institution, previous tuition fee, high school reforms (G-8) in state and year, share of foreigners in state, GDP per capita, and unemployment rate.Standard errors in parentheses.* p<0.1, ** p<0.05, *** p<0.01.

Table 5 .
Poisson PML for total numbers.

Table 6 .
Poisson PML for total numbers by continents.Tuition fee refers to public institutions in Baden-Württemberg, which introduced tuition fees for international students in 2017.All estimations include the full set of covariates: number of students, share of males, ARWU rank, public institution, previous tuition fee, high school reforms (G-8) in state and year, share of foreigners in state, GDP per capita, and unemployment rate.Standard errors in parentheses.* p<0.1, ** p<0.05, *** p<0.01.

Table 7 .
Poisson PML for total number of failed final exams.Tuition fee refers to public institutions in Baden-Württemberg, which introduced tuition fees for international students in 2017.All estimations include the full set of covariates: number of students, share of males, ARWU rank, public institution, previous tuition fee, high school reforms (G-8) in state and year, share of foreigners in state, GDP per capita, and unemployment rate.Standard errors in parentheses.* p<0.1, ** p<0.05, *** p<0.01.

Table 8 .
Inverse hyperbolic sine transformation.The estimates follow the arcsinh-linear case transformation as described byBellemare and Wichman (2020).Students with citizenship other than German are considered as foreign, while international students include those from outside of Europe.Tuition fee refers to public institutions in Baden-Württemberg, which introduced tuition fees for international students in 2017.Observations cover the years 1998-2019.Standard errors in parentheses.* p<0.1, ** p<0.05, *** p<0.01.Tuition fee refers to public institutions in Baden-Württemberg, which introduced tuition fees for international students in 2017.All estimations include the full set of covariates: number of students, share of males, ARWU rank, public institution, previous tuition fee, high school reforms (G-8) in state and year, share of foreigners in state, GDP per capita, and unemployment rate.Standard errors in parentheses.* p<0.1, ** p<0.05, *** p<0.01.

Table 10 .
Inverse hyperbolic sine transformation: foreign students' final exam failure rates.Tuition fee refers to public institutions in Baden-Württemberg, which introduced tuition fees for international students in 2017.All estimations include the full set of covariates: number of students, share of males, ARWU rank, public institution, previous tuition fee, high school reforms (G-8) in state and year, share of foreigners in state, GDP per capita, and unemployment rate.Standard errors in parentheses.*p<0.1, ** p<0.05, *** p<0.01.