Evaluating Effects of Tuition Fees: Lasso for the Case of Germany

by   Konstantin Görgen, et al.

We study the effect of the introduction of university tuition fees on the enrollment behavior of students in Germany. For this, an appropriate Lasso-technique is crucial in order to identify the magnitude and significance of the effect due to potentially many relevant controlling factors and only a short time frame where fees existed. We show that a post-double selection strategy combined with stability selection determines a significant negative impact of fees on student enrollment and identifies relevant variables. This is in contrast to previous empirical studies and a plain linear panel regression which cannot detect any effect of tuition fees in this case. In our study, we explicitly deal with data challenges in the response variable in a transparent way and provide respective robust results. Moreover, we control for spatial cross-effects capturing the heterogeneity in the introduction scheme of fees across federal states ("Bundesländer"), which can set their own educational policy. We also confirm the validity of our Lasso approach in a comprehensive simulation study.


The fiscal response to revenue shocks

We study the impact of fiscal revenue shocks on local fiscal policy. We ...

Sélection de variables par le GLM-Lasso pour la prédiction du risque palustre

In this study, we propose an automatic learning method for variables sel...

Large-scale Feature Selection of Risk Genetic Factors for Alzheimer's Disease via Distributed Group Lasso Regression

Genome-wide association studies (GWAS) have achieved great success in th...

The effect of twitter-mediated activities on learning outcome and student engagement: A case study

The study reported in this paper, examines the effects of a rigorously d...

Fast and robust model selection based on ranks

We consider the problem of identifying important predictors in large dat...

The Price of Competition: Effect Size Heterogeneity Matters in High Dimensions

In high-dimensional linear regression, would increasing effect sizes alw...

When Does the First Spurious Variable Get Selected by Sequential Regression Procedures?

Applied statisticians use sequential regression procedures to produce a ...

1 Introduction

We study the effect of the introduction of tuition fees on university student enrollment behavior using publicly available data. In particular, we illustrate for the case of Germany that even low fees in only parts of the country already have a significant impact. In Germany, in contrast to other countries, the maximum fee amount was generally limited to € 1000 per year and fees were only present in parts of the country from 2006-2013111We always observe year at the beginning of the winter term in October of .. Moreover, the implementation and timing of the fees were no exogenous shock but strategic policy decisions on the federal state level (“Bundesländer”, denoted shortly as states in the following). Thus, the timing of the introduction and abolishment of fees varied considerably among states. Moreover, during the considered time period, major policy changes in different federal states significantly impacted the cohort size of prospective university students.222This comprises a decrease for the required compulsory years to high school graduation from nine to eight years of which the introduction varied on the state level, and the general German-wide abolishment of the 9 month compulsory military service for men in the age of 17-23. Specifically, the spatial time delay in the implementation of both tuition fees and different federal reforms induced migration effects that we show to have a direct substantial impact on student enrollment numbers.

All these issues make it impossible to identify tuition effects with standard methods that only compare sample means before and after policy reforms333As e.g. a simple difference-in-differences (DID) approach, see e.g.Card and Krueger (1994) or Ashenfelter and Card (1985).

. Capturing that the political decision for or against a tuition fee treatment was in fact strategic, we explicitly model this binary choice as a function of potentially many observable factors. Moreover, for the enrollments, we include spatial cross-effects allowing for migration in a fixed effects panel model besides the large set of all covariates marked as potentially relevant in different strains of literature on tuition fees. For the small number of publicly available observations on the state level and the large number of controls, however, we show the necessity of an appropriate data-driven selection method. We identify relevant factors from both the main regression and the treatment choice equation in order to determine the correct magnitude and significance of the marginal effect of tuition fees. In particular, we propose using a tailored Lasso-type procedure with stability selection that detects key factors and still correctly estimates confidence intervals of the desired effect.

Overall, we find a significant negative effect of tuition fees inducing an up to percentage point reduction in enrollment rates. Since the exact enrollment rate is hard to measure, we show the stability of our results over a large grid of values. While spatial cross-effects have been ignored in the previous literature on German tuition fees, they are identified as important drivers for enrollment rates by the Lasso, besides state specific factors such as the student-to-researcher ratio. We explicitly show that without Lasso-preselection of variables, the signal to noise ratio of the problem is too low for detecting the correct magnitude of the effect. Generally, these insights and our methodological solution are highly relevant for all cases of policy evaluation, where implementation occurs in a spatially time-delayed manner, as for example environmental policies that target global warming or financial regulations in different countries. In addition, we believe that our empirical findings cannot only contribute to the active ongoing discussions on reintroducing tuition fees in Germany, but might also be of independent interest for other countries such as the United Kingdom, where fees are on the rise.

For the analysis covering the years 2005-2014 and all 16 federal states in Germany, we include a comprehensive set of 18 covariates, which have appeared as potentially influential for tuition fees in the literature both inside and outside of Germany (e.g. Dynarski (2003); Kane (1994) and Bruckmeier and Wigger (2014); Mitze et al. (2015)). The variables are collected from different sources, but public data on student enrollment behavior is only available on the state level and not on a university level, which is due to strict German data protection laws.444Note that across states and universities, individual or household panel data from common sources such as e.g. the German SOEP is insufficient, incomplete and very unbalanced and cannot be employed for a general analysis. Please see Appendix A.2.1 for details.

In addition to standard economic, social and educational factors from the literature on student enrollment rates, we also include specific effects for Germany which play a major role in the considered period. Particularly, policy changes such as the abolishment of mandatory military service or the heterogeneous introduction of a one-year reduced secondary education (”G8”) in different states are key policies. Moreover, we use spatial variables that capture state cross-effects in the policy decisions for or against fees as the proportion of students migrating to each state from states with and without tuition fees based on their proximity. These are crucial to control for migration effects due to heterogeneous implementation and time delay of policies across states that could otherwise bias the estimated effect of tuition fees. We work with relative enrollment rates instead of absolute numbers as the dependent variable to ensure compatibility of effects across federal states of different population sizes. For correct ratios, however, we require the population size of all high school graduates affected by the introduction of tuition fees in a specific state. This quantity is hard to measure and thus prone to measurement errors as it consists not only of recent and less recent high school graduates from this specific state, but also of a parts of cohorts from other states and abroad from where students migrate to study. We transparently treat this measurement ambiguity and thus provide results that are robust in this respect. Overall, the limitation to only state-level data results in a relatively small number of available observations where single observations could gain substantial influence on the overall result. Thus in total, we face a situation of many potentially influential but correlated covariates and relatively few observations with possible outliers due to data quality problems.

We tackle these challenges with tailored variable selection techniques combined with subsampling in a fixed effects panel regression setting. Tuition fees are modeled as a binary variable and are used to explain enrollment rates in federal states conditional on a set of control variables. These controls can linearly influence the enrollment rate as well as the treatment effect of tuition fees (auxiliary step). We assume that only a small number of all controls in fact have a true non-zero influence on these two variables that we detect by appropriate Lasso double selection

(Belloni et al., 2014b). This double selection is key to our situation as it helps to mitigate the potential bias caused by variables which are highly correlated with the tuition fee treatment that would be deselected by a simple Lasso only on the main equation. As finite samples estimates from the Lasso can be biased due to the penalization, we conduct a post-Lasso step using pooled OLS for only those control variables with non-zero coefficients in both Lasso-steps. For valid results in our situation of highly correlated covariates and where single data points might be imprecisely measured but still appear very influential given only a few total observations, we propose an adaptation of the double selection method with a stability selection mechanism (see Meinshausen and Bühlmann (2010) in the two Lasso steps. In such cases, pure Lasso might have difficulties in correctly predicting the influence of each variable, which can lead to the choice of too many variables. The stability selection employs subsampling to make the selection more robust in such settings. We conduct a thorough simulation study which shows that applying stability double selection always improves results, while maintaining good statistical finite sample performance when influential observations are present.

Up to our knowledge, the literature on student enrollment behavior generally works with only small sets of covariates on which there is no consensus and often only ad-hoc justifications. Therefore, we propose a data-driven statistical procedure in order to empirically identify relevant factors. Nevertheless, there are analyses on effects of tuition fees in various countries that mostly find significant effects only for certain subgroups of the population. Kane (1994), Noorbakhsh and Culp (2002) and McPherson and Schapiro (1991) find negative effects of tuition fees555In the study of McPherson and Schapiro (1991), the authors find that the net costs (tuition fees minus student aid) have a negative impact, which is an even stronger argument. for low-income groups or groups with African-American ethnicity for the US. More generally, Neill (2009) finds that an increase in tuition fees reduces enrollments significantly for the Canadian system. With the availability of individual data in the presence of much higher fees, but also an established scholarship system, US and Canadian studies can identify effects of tuition fees on enrollment that range between pp and pp. For countries where the situation is more comparable to the German system, and the particular case of Germany, previous studies generally cannot to detect significant effects of tuition fees on enrollment rates (see e.g. for Germany Baier and Helbig (2011); Hübner (2012); Dwenger et al. (2012); Bruckmeier and Wigger (2014); Mitze et al. (2015), but also Huijsman et al. (1986) for the Netherlands and Denny (2014) for Ireland). This seems to be caused by the small number of included covariates, while missing out on the key ones according to our statistical selection technique. Variables possibly correlated with the tuition fee decision are mostly ignored, as well as state cross effects through differences in timing, which we show both to be relevant. Moreover, we cover the comprehensive list of all German tuition fee periods and states, which helps to increase precision of estimated effects in contrast to previous studied, who focused only on subperiods, specific states or subgroups. With mostly insignificant effects between pp and pp, the previous German studies seem to systematically underestimate the true impact of fees.

The remainder of the paper is structured as follows. A description of the data set and variables is presented in Section 2. It also contains the transparent construction of (a set of) response variables from the limited available information. Section 3 introduces the linear panel model and the Lasso-type selection methods featuring the stability double selection. In Section 4, a Monte Carlo simulation shows the advantages of these methods with different distortions in a controlled environment. After discussing the main results of our empirical study in Section 5, we conclude in Section 6.

2 Data

We construct a panel from publicly available data on enrollment numbers and socio-economic and university-related covariates for the 16 German states () in the years 2005 to 2014 (). We use a widespread set of potential controls for determining the effect of tuition fees, which only existed in the years 2006-2013 in at least one state (see Figure  2 for an overview of the timing of fees in different states). The years 2005 and 2014 serve as a base for comparison before and after the introduction and complete abolishment of tuition fees. Note that we are limited to state level aggregated data, since available individual or household type data from common sources such as e.g. the German Socio-Economic Panel (SOEP) is highly incomplete and very unevenly distributed across states and universities and thus cannot be employed for a general analysis on the effects of tuition fees. Please see Appendix A.2.1 for details.

As the response variable we use the enrollment rate of high school graduates into university in state at the winter term (WT) of year to (denoted as ).666The academic year starts with the winter semester usually beginning in September or October of year and ending in February of year . We use data from public institutions, which account for the majority (more than 90%) of higher educational institutions in Germany. As higher educational institutions, we denote general university type institutions comprising universities, specialized technical, arts and music universities but also universities of applied sciences (Fachhochschule) and cooperative state universities (Duale Hochschule). As the population size among German states varies substantially, relative enrollment rates ensure comparability of results across states, in contrast to the absolute number of new enrollments (from anywhere) in state at WT of year . The percentage is obtained as the quotient of the number of enrollments in state and the so-called eligible set of high school graduates for year coming to or staying in state , which can generally differ substantially from the own-state high school graduates in of this specific year. We set

Figure 1: Illustration of the composition of the eligible set .

where we model to consist of three main different groups, namely own -specific high school graduates , “affected” graduates from other German states and the number of new international enrollments in , (see Figure 1):


While respective enrollment numbers from in , from to and of international students in are publicly available for any state in WT , there is, however, no available direct data for the respective eligible quantities in(2). For the from to component, this can be well approximated by its upper bound of the number of all high school graduates in as in the German federal system, the “home state” of the high-school diploma is often part of the immediate choice set of university entrants. Since the share of international students remains stable at around 15% over the years due to effects such as language barriers in German undergraduate programs, we assume that the low amount of tuition fees in the international context has no effect and we therefore only use the lower bound in the eligible set. Though for the eligible part of potential movers from to within Germany, extreme approximations by its lower bound of the number of enrollments or the upper bound of all graduates in are too coarse. In particular in view of tuition fee interventions, it is clear that is affected, but unclear how. We therefore model it explicitly as a convex combination between the potential extremes.


with . Of course, choosing too low, i.e. giving too much influence, will yield values that are unrealistically low. An absolute lower boundary would be a mean enrollment of , which is achieved at . Looking at the aggregated number of all new enrollments (not just first-time students) in all of Germany from German high schools over 2003-2014 divided by all high school graduations in Germany at that time in our data, we have a mean enrollment rate of around , which can serve as a very rough proxy for where to expect realistic values. If we only look at first-time enrollments, the rates have monotonically increased from 40% in 2009 over the years.777Data source: federal ministry of education (BMBF) data webspace http://www.datenportal.bmbf.de/portal/de/K253.html Table 1.9.3 We therefore take as a reasonable lower -boundary, which yields . We then conduct our analysis transparently over a grid of -values in between 0.98 and 1 which we denote as admissible s and which yield mean enrollment rates . Figure 10 in Appendix B shows the mean enrollment rates over indicating the sensitivity of with respect to in the considered range.

With additional information using the number of new enrollments with graduation in state enrolling anywhere in Germany at combined with and we can augment the approximation of . Moreover, in order to additionally control for effects from postponers in , we employ extra non-public information888Provided by the Federal Statistics Office on request for a fee. on the number of new enrollments in state in WT with high school diploma obtained in year . With this, we can obtain an alternative approximation of the number of high school graduates in potentially moving to at


with share of enrollments from to within the cohort of relative to all enrollments from in year , approximating the potentially moving share of the graduates (See Table 4 and Figure 9 in Appendix B for a (graphical) overview of involved sets and their role) .999As it can happen that , we ensure that is at least . We focus on numbers up to a time lag of in , which cover generally more than 75% of enrollments (on the German level), and use this graduation time specific information also for state to get a refined approximation of by


Note that for a choice of , the empirical mean squared and mean absolute deviation of and over all and are minimized and both almost coincide. As a robustness check to our pure public data analysis, we also report results for a response .

Figure 2: Overview of the timing of tuition fees in German states (presence in gray). The winter term (starting October) and summer term (starting April) are indicated with small ticks. States not listed had no tuition fees at all.

In the covariates, we model the treatment effect of a tuition fee as a dummy, with indicating an existing tuition fee in state in the winter term starting in year and otherwise.101010

In Germany, there were no fees for students studying for their first degree in public institutions from WT of 2014 and onwards. Before that, the maximum amount for first degree studies was limited to €1000 per year. Almost all universities made use of the maximum amount, thus suggesting a dummy variable design.

Because of German laws, each state could strategically decide on the introduction and timing of fees.

We generate spatial controls that capture migration behavior to each state from other state groups, which are formed depending on proximity and fees. This is necessary because of the heterogeneity of introduction and abolishment of tuition fees over states that can be seen in Figure 2. Additionally, there are many cases where fee-states border non-fee states, which is highlighted in Figure 3. We therefore construct the spatial controls to measure the share of new enrollments in state that obtained their high school diploma in another state group. For each state , we measure the proportion of new enrollments from a specific group relative to all enrollments in . The groups consist of fee states that have a shared border with , fee-states without a shared border with , non-fee states, and enrollments from outside Germany (Migration.international). A detailed description can be found in Table 6 in Appendix B.

Figure 3: Overview of the presence of tuition fees (left) and the G8-reform (right) in the 16 German states until 2015. Darker colors represent longer presence of the respective variable.

Furthermore, to control for non-constant state specific effects, we employ 18 control variables using data from the socio-economic panel (SOEP)111111We use the SOEP-long version 31. More information at https://www.diw.de/en/diw_01.c.519381.en/1984_2014_v31.html; for the usage, see Wagner et al. (2007) and Destatis121212More information at https://www.destatis.de/EN. Some variables were generated using data from Genesis-online database of Destatis accessible at https://www-genesis.destatis.de., the Federal Statistical Office in Germany. A detailed description can be found in Table 5 and Table 7 in Appendix B. Together with the spatial variables, we have a set of potentially relevant covariates plus the binary variable of tuition fees. Among others, we capture socio-economic variables comprised of urbanization level, income, rent, life satisfaction, unemployment rate and university and student related controls on staff and graduation statistics, the student-to-researcher ratio and data on the funding of universities. In particular, this set of variables contains all types of relevant controls from similar, previous studies (e.g. Bruckmeier and Wigger (2014); Mitze et al. (2015)). Moreover, we include two variables on the G8-reform that reduced the time of secondary education from nine to eight years. The implementation of this major educational policy change was also heterogeneous across states and is illustrated in blue in Figure 3. This reform almost immediately substantially impacted the timing and the overall likelihood of much younger high school graduates to enroll to a university. We control for this effect with a dummy , where positive values indicate that the G8-reform was implemented in this state , and additionally mark transition period years of double cohorts of G8 and G9 cohorts graduating by .

Figure 4: Left: DFFITS for and all controls: influential observations in red. Right: Boxplot of for the 18 covariates and the treatment with threshold: . Figures for each covariate available from authors upon request.

Inspecting the data, we find that single observations are highly influential. The left-hand side of Figure 4 shows that the fitted enrollment rates heavily change when specific single observations are dropped from the regression estimation. More importantly, when looking at the leverage of covariates, we can inspect how coefficients change when these specific single observations are left out of the regression. If many influential observations affect one covariate, its selection by Lasso would depend strongly on these observations. The diagnostic tools used here are the DFFITS for changes in and the DFBETAS for changes in coefficients of covariates. Thresholds to decide whether or not observations are influential are calculated as for DFFITS and for DFBETAS, with as the stacked number of observations. More specifically, with , and as the DFBETAS measure of the th observation of covariate , let . therefore measures how many influential observations exist for each covariate . The boxplot of in Figure 4 shows that all covariates suffer from this phenomenon, indicating that the selection is unstable. In addition to high expected correlations between regressors, it further encourages the use of stability selection instead of using all data points just once.

3 Model and Method

The key goal of our study is to derive a finite sample precise estimate of the effect of tuition fees on enrollment rates . For this, we use a linear panel model with time-fixed effects , where covariates consist of socio-economic variables and spatial factors . For each admissible in (3), the influence on is linear in the tuition fee dummy and can be nonlinear in the other components, described by an unknown function .131313For ease of exposition, we omit in the following in . In an auxiliary equation, we explicitly model the introduction of tuition fees as a function of the observed covariates which will help to fine-tune the data-driven selection of relevant covariates in the main equation. Using states, years , with and , we obtain the following form:


with and . Given the few available observations, we assume that the functions and are linear, i.e. and with , . Note that from plugging in the linear form of (7) into (6), we obtain a reduced form of the main equation


with and . We model the state-specific fixed effect as time-constant, and are the error terms for which we assume strict exogeneity, i.e. Then standard fixed effects estimates for (7) and a linear version of (6) can be obtained by time-demeaning the entire equation using with and similarly , , , , which removes the unobserved effects and then allows for standard pooled OLS estimation. As some time-constant effects

comprising e.g. unobserved regional aspects such as climate conditions, culture, or the topography of a state might be correlated with at least some of the covariates such as e.g. rent or the urbanization level, a fixed effects model is in fact necessary. Thus for statistical testing, the usual degrees of freedom (

) correction for fixed effects panel models applies.141414The of the residuals reduce from to , which is due to the demeaning process. For each observation , one degree of freedom is lost because of the error term . The latter is now comparable to a parameter that needs to be estimated (see Wooldridge (2002, p. 271-272) for more information).

Though, in our situation of , observations are so scarce relative to the dimensionality of the problem that plain OLS-type estimates are extremely imprecise. Thus for proper estimation of our main coefficient of interest we assume that in fact only a few () of the other controls and are relevant for each state in the equation of (). We use the Lasso (Tibshirani, 1996) as a data-driven tool to select the respective relevant covariates from an penalized minimization problem. We obtain the Lasso estimates as


with regularization parameters that are estimated by cross-validation and 151515In practice, there exist several techniques for solving this problem, while we use coordinate-descent algorithms (Friedman et al., 2007; Friedman et al., 2010) provided in the glmnet package in R.. Note that we use the reduced form of the main equation (8) and therefore implicitly penalize the treatment also in (9). Selecting all variables with non-zero and from Equation (9) and (10), we obtain corresponding index sets and of sizes smaller than for the prevailing components in and . Hence we get the following model:


where , and , as the corresponding union of respective selected variables from (9) and (10). We obtain final estimates for and of the coefficients via OLS in (11) for each response , with the respective relevant active set from the Lasso steps (9) and (10) for each admissible in (3).

Note that consists of variables either influencing the treatment or the response . Hence the selection choice from the auxiliary equation (10) corrects wrong de-selection choices in the main enrollment equation (9) due to highly correlated control variables and thus significantly helps to robustify the selection against biased estimates from underspecification, therefore implicitly reducing also post-selection concerns of Leeb and Pötscher (2008). The price of a potential overspecification in this respect is small for our data and mitigated by the post-selection estimation in (11). This procedure is known as post-double selection and yields a consistent -estimator as well as correct coverage of confidence intervals (see Belloni et al. (2014a)). For comparison, we also calculate simple post-Lasso estimates, which employ OLS only on variables selected from (6). With this, however, we would miss out on factors with small effects on , but a high correlation with which can bias estimation of . We illustrate this in detail in the simulation study in Section 4. Omitting the post-OLS step entirely also leads to a biased measurement of , since the penalization shrinks lasso estimates towards zero in finite samples.

Though, for the very strong correlation of control variables in combination with single influential observations in our data, standard Lasso runs into problems generally selecting too many variables, especially in small samples. We therefore propose a subsampling modification based on stability selection (Meinshausen and Bühlmann, 2010) in the Lasso selection steps (9) and (10). For this, we generate subsamples of size of the data points and thus obtain estimates for each coefficient in (9) and (10). Hence, from the estimates and , only those variables will be included in with a relative inclusion frequency above a threshold and respectively, where both are in . The relative inclusion frequencies over all are computed as and for each variable . Thus, we only include variable in the model if or . Typically, the stability double selected index set is a subset of the standard double selected set and depends on the choice of sufficiently large and and the number of repetitions . For the empirical results and the simulation, we use and .161616For the robustness checks using only the control year 2008 and 2014, we increase the subsample to to deal with the small data set. For a data-driven threshold choice, we set minimum thresholds as lower bounds to make sure to screen out irrelevant variables. Since the response values change with in (9), the corresponding minimum thresholds also depend on . The selection of effective thresholds is then performed over a grid of threshold values starting from the minima increasing the threshold level to the first points where small changes in the thresholds do no longer change the model. The algorithm for the threshold choice can be found in Appendix A.1. In the simulation, we also report estimates with and for comparison.

4 Simulation

We conduct a Monte-Carlo Simulation to show the importance of stability selection when it is hard to disentangle effects of different covariates. This can be further adapted to our data by including influential observations and by inducing strong correlation among covariates. Using , , and with , , , and , we simulate a linear panel model of the following form:

with coefficients depending on : , , and for , zero otherwise. The coefficients of covariates are up to 10 times higher than the coefficient of the treatment, since such large differences are also likely to arrive in our empirical application, where the expected treatment effect is relatively small. We generate the fixed effects as and 171717: for , represents a covariate that is standard normal with a correlation of to and , ., with , representing the rows and the columns of , . For , , and for , . The errors are independently distributed as and

with a heteroscedastic structure given by

Given this structure, we distort the last

of observations by a vector

, and we generate each , where and depending on the scenario. In each scenario (i.e. different inf-values), we distort covariates either from the active set (), the inactive set () or the response . For distortion of covariates, we modify them to . This means that for either (inactive set) or (active set). When is distorted, we have and . We report mean values over 1000 replications for the absolute bias of estimators from , the root mean squared error for with , the number of selected covariates, the true positive rate TPR=, and the false positive rate FPR=

. We also report the rejection rate, which is based on conventional t-tests on the estimated

against the true

with heteroscedasticity consistent standard errors

(MacKinnon and White, 1985). We report results from post-Lasso and post-double selection as the described in Section 3, using no subsampling at all and using the subsampling similiar to stability selection with . Additionally, we report the two extreme cases using all covariates without selection (Fixed Effects all) and using only the true influencing variables (Oracle).

Table LABEL:Tab:Sim_results summarizes our simulation results. First of all, as expected, the proposed double selection procedure combined with stability selection performs best overall and is almost identical to the oracle procedure that knows the true active set. Using of or does not affect results much in most cases. When distorting the inactive set, using a higher minimum threshold reduces the FPR even more than in other cases, as the noise variables have more influence. When regarding post-Lasso, however, seems to perform better in general, which can be explained by the post-Lasso not detecting all relevant covariates in the simulated data, where a lower threshold leads to the inclusion of more relevant variables compared to noise variables and improves the method here. For the double selection, only more noise variables are added since all relevant variables are already (almost) always detected. When distorting the response, bias and RMSE values go up in general for all procedures, but their relative performance compared to the oracle does not get worse. Comparing stability procedures to their non-stable counterparts, we see that the latter include up to twice as many covariates without much improvement on the TPR, but high increases in the FPR. This confirms the hypothesis that without stability selection, many irrelevant covariates are included in the model, which increases the bias and RMSE. The rejection rate is especially high for all post-Lasso procedures, which is not surprising given their high bias and relatively low standard errors that are a result of including fewer variables in the model. Small standard errors also affect the RMSE values, and in scenarios with high distortions in the response y, the post-Lasso has a lower RMSE than its double selection counterpart (regarding the stability procedures).

Taking a closer look at the different forms of distortion, we do not observe much change for high -values when we distort variables from the inactive set. As expected, when influential observations are only present in the noise variables, they do not affect the selection procedures much. When distorting the active set only, however, procedures with the post-Lasso select fewer (relevant) variables due to the added noise, which leads to a higher bias (for the stability cases), and increases RMSE values. The double selection procedures seem to be very robust against such distortions, with all measures remaining relatively unchanged. This is not surprising, since the double selection procedure helps to reduce such a bias by taking the second equation into account. Finally, distorting the response is interesting, since both relevant and irrelevant covariates are affected at the same time. Even with extremely high distortions, the double selection procedures keep a lower bias compared to the other methods and double selection with stability selection has very low FPRs, while selecting almost all variables from the active set. All in all, the simulation shows that only when we use stability selection, we can select the right variables without including too many noise variables. In our simulated model, where it is hard to distinguish between covariates and the treatment effect is relatively small compared to the effects of other covariates, the non-stable methods perform worse over all distortion scenarios181818Results are similar using a lower correlation among covariates. Additional simulations are available upon request.. Furthermore, we see that when some covariates explain the treatment well, but only have a moderate effect on the response (which is the case in the application), double selection outperforms the post-Lasso in terms of bias and rejection rate.

5 Empirical Results

5.1 Main Findings

In this section, we present the results of our empirical study. Generally, with only publicly available data and the proposed post stability double selection methodology, we find that tuition fees in Germany significantly reduced the enrollment rate by 3.8 percentage points (pp) to up to 4.5pp on average over all possible cases of response variables. For all admissible values of , the procedure consistently identifies the same one university specific and one educational policy change control variable in and the four spatial variables as important drivers highlighting the importance of fee induced migration effects. Moreover, we find that during the considered period, other socio-economic factors only played a minor role. Given the transparency in and the data-driven stability double selection, we judge these findings are very robust.

Double Selection + Stability All Controls
Effects on
Double.Cohort /
Migration.neighbor.fees /
/ / /
Note: Response values are scaled to a percentage level. Standard errors are heteroscedasticity consistent (HC3, see White,Mckinnon 1985). P-values for Tuition fees are depicted in parentheses. Variables in blue appeared similarly in previous studies (not necessarily together).
Table 1: Estimates of the Effect of Tuition Fees for Different -Values
Figure 5: We report estimates for in (6) for each of the three methods in bold over the grid of admissible in from (3). Post-Lasso and post-double selection are both combined with the stability procedure (i.e. repeated subsampling). In thin versions of the respective line types, we also depict the corresponding upper bounds of the robust confidence intervals. For the confidence intervals, we use HC3 heteroscedasticity robust standard errors which adjust for high-leverage points..

Table 1 summarizes the post-selection estimation results. Most importantly, we find a significant negative effect over the whole grid of -values only when using post-double selection with repeated subsampling (Double Selection + Stability). The reference point from additional non-public information in (5) suggests in fact that values very close to the right boundary of are the most plausible, i.e. the number of effective enrollments of migrating students from to within Germany almost coincides with the number of potentially enrolling ones at . For such large -values in particular, using all controls in a plain panel OLS clearly underestimates the effect and thus leads to inflated p-values, which is illustrated in Figure 5. Post-double selection Lasso without the stabilizing subsampling does not work as it leads to the same results as a pooled OLS with all controls. In those cases, the magnitude of the effect from tuition fees is roughly four times smaller than for the post stability double selection and the impact becomes insignificant. Across all admissible , only about a third of the controls are selected with our proposed procedure, which indicates that many plausible controlling factors from the literature are in fact not relevant and dominated in this period of heterogeneous changes in educational policies across states.

Looking more closely at Figure 5, we see that over the entire grid of admissible -values, only the double selection procedure with subsampling guarantees good performance, whereas with all controls the estimated effect for vanishes with approaching the upper bound 1. With an effect of tuition fees close to zero for the upper -boundary, and only half the size of the one by the stable double selection at the lower -boundary, the pooled OLS appears biased in detecting individual influences in this situation, where observations are scarce relative to the dimension of the model. This behavior is not surprising, as many irrelevant controlling factors that might be spuriously correlated with the response and the treatment are present without selection. This is more critical at the upper -boundary, where the variability of the response is higher. Furthermore, using the post-Lasso, even with stability selection, gives less stable and often insignificant results. The insignificance can be traced back to the lack of additional controls that are only added in the second step of the double selection procedure, whereas the rather unstable results can furthermore be accounted for by the difference in the selection procedure in the first step that includes the treatment in the equation. All this emphasizes the importance of using a post stability double selection as proposed.

Figure 6:

Controls with high inclusion probabilities in the first step depending on

(x-axis). Green indicates that the respective variable is selected in the final model, which depends on the threshold that is depicted as a red line. The y-axis shows the selection probability from the Lasso step.

Figure 6 shows all controls that were selected in the main equation (9) (i.e. with as the dependent variable). We find the spatial variables to be highly relevant, which implies that mobility and migration effects played a major role for enrollments in the presence of heterogeneous timing and implementation of tuition fees and major educational policy decisions across states. In size, they largely contribute in explaining the variability of the enrollment rates. At the lower boundary of , only one of the four spatial variables Migration.neighbor.fees is included less often over different subsamples and is thus deselected by the stability selection for low -values. As there is only a small limited number of overall neighbours of each state, their impact on enrollments in state is generally much smaller as from the aggregated rest of the country and thus more sensitive to a variation in the response variable.

Furthermore, the variable Double.cohort that indicates if there were two cohorts of high school students graduating in the same year, caused by the G8 reform reducing time to graduation, is identified as an important controlling factor. Double.Cohort has a negative sign, which at first might appear counter-intuitive, as with a double cohort, one would expect enrollment numbers of students to rise. For relative enrollment rates, however, a negative sign of double cohort seems justified, since universities did not double their admission numbers when there was a double cohort. Moreover, when the competition for universities is extremely high in a double cohort situation, fewer people might decide to actually compete and rather consider outside options or postpone university entrance with a gap year. Note that for the extreme boundary case (), however, the variable is deselected, which can be attributed to the pre-dominance of the migration factors with large size effects at the extreme upper -boundary. Repeating the analysis with Double.Cohort in the extreme case for , however, does not change results and only alters coefficient values in an minor insignificant way. This behavior can be expected when taking into account that the effect of Double.Cohort is relatively small compared to the other variables close to the upper boundary of .

In line with theory, the variables Student.to.researcher.ratio and the share of international enrollments Migration.international that are additionally selected in the auxiliary equation of the double selection procedure only have a minor direct influence on enrollment rates, while having a large impact on tuition fees. Thus, this socio-economic factor and the financial situation of universities drives the political decision for the introduction of fees. Overall, the double selection step is key yielding additional necessary variables for accurate estimation of (see Figure  5).

Generally, these findings show that spatial factors and the double cohort variable are crucial for identifying the effect of tuition fees on enrollments. In the existing empirical literature, however, they have been largely ignored yielding downward biased insignificant estimates. Moreover, the auxiliary equation and the stability double selection are key for detecting the magnitude .

5.2 Robustness Checks

Apart from using all available data, we also analyze two subsets that either contain only periods with tuition fees (2006-2013) or that consist of the peak year 2008 of the presence of tuition fees and the year 2014 after their abolishment. Furthermore, we work with the alternative response variable constructed from additional non-public information in the eligible set in (5). Estimates of for these adaptations are summarized in Table  2.

Data sets No. of Variables
Tuition Fees All Fees Small All/Fees/Small
min MSD with :
All Controls - 19/19/-
Post-Lasso Stability 4/4/3
Double Selection Stability 7/6/7
min MAD with :
All Controls - 19/19/-
Post-Lasso Stability 4/4/3
Double Selection Stability 7/6/7
with /:
All Controls - 19/19/-
Post-Lasso Stability 3/9/2
Double Selection Stability 6/10/6
Note: Response values are scaled to a percentage level. Standard errors in parentheses are heteroscedasticity consistent (HC3, see MacKinnon and White (1985)). p0.05; p0.01; p0.001 indicate p-values from a t-test on significance from zero. is chosen according to minimum mean squared deviation (MSD) and minimum mean absolute deviation (MAD).
Table 2: Estimates of the Effect of Tuition Fees for in Different Time Frames

First, when comparing the effect with -response values over different time frames, we find that the main results prevail over the variation in the data set. The double selection is still the only reliable method, while post-Lasso and pooled OLS with all controls cannot capture the strength of the effect nor its statistical significance. Post-Lasso generally de-selects too many relevant controls, yielding smaller effects in absolute values of tuition fees on enrollments. Omitting the first and last year from the data only causes mild changes in the amount of included controls, but the size of the estimate for from double selection decreases in absolute terms, probably due to fewer available observations. Though, in the extreme case of the smallest data set, where only two years with either “no fees at all” or “fees in seven states” are considered, the magnitude of the effect increases substantially. The results of the extra response confirm the above observations. The size of the estimates for for different time frames and the amount of included controls mostly coincide with results for the response . In this case, however, the pure post Lasso double selection estimate is much closer to the estimate of the stability double selection procedure in size and becomes even mildly significant.

In summary, we conclude that the effect is rather robust to changes of the time frame and double selection consistently identifies the effect, where the other methods mostly fail. While changes in the strength of the effect arise mostly in very high-dimensional situations (i.e. small data set), the effect is also identified using the additionally constructed . Comparing the strength of the effect to previous studies, which estimated (mostly insignificant) effects from pp to pp, we see that for almost all cases, our estimated effect lies rather between and pp using double selection, and is always highly significant. On the contrary, using fixed effects with all controls and without selection yields estimates that appear to be downwards biased and closer to the lower bound found in other studies, while in almost all cases, this cannot identify significant effects.

6 Conclusions

In this article, we propose a subsampling stabilized double selection technique in order to identify the effect of tuition fees on enrollment rates from public state-level data in Germany. We show that for extracting size and significance of the effect, such techniques are key in this particular setting, with few observations, correlated covariates, and influential observations, where the implementation and timing of tuition fees varied across states and time.

With our tailored post-Lasso approach, we are the first to find an overall significant negative effect of tuition fees in Germany. With the stability double selection we identify the relevant factors, which are crucial for political decision-making. In particular, previously neglected spatial migration effects and the major shift in educational policy by G8 appear as key control variables for enrollment rates in the considered period. The detected effect is robust over a large grid of different response values and different subsets of the full data set. These empirical findings therefore contribute to the existing literature on education economics. In the active ongoing discussion about the reintroduction of tuition fees in Germany, the results might also be of political interest.

Moreover, this study strongly advocates the use of data-driven variable selection to choose relevant controls from a broad set of possible influencing factors. We explicitly show that standard fixed effects panel regressions without selecting variables fails to detect correct and precise effects for such small sample sizes relative to the dimensionality of the problem. Furthermore, appropriate statistical selection techniques determine and justify the relevance of chosen controlling factors, yielding an easily interpretable post-selection model that outperforms all ad-hoc choices. For future research, it would be interesting to use the data-driven identification of relevant controls also for other countries, e.g. the United Kingdom or France, aiming for a comprehensive European study with increasingly relevant spatial cross-effects across country borders. This is particularly relevant given the reintroduction of fees for international students in parts of Germany, that could trigger such cross-effects.


  • Ashenfelter and Card (1985) Ashenfelter, O. and D. Card (1985): “Using the Longitudinal Structure of Earnings to Estimate the Effect of Training Programs,” Review of Economics and Statistics, 67, 648–660.
  • Baier and Helbig (2011) Baier, T. and M. Helbig (2011): “War all die Aufregung umsonst? Über die Auswirkung der Einführung von Studiengebühren auf die Studienbereitschaft in Deutschland,” WZB Dicussion Paper, 2011-001.
  • Belloni et al. (2014a) Belloni, A., V. Chernozhukov, and C. Hansen (2014a): “High-Dimensional Methods and Inference on Structural and Treatment Effects,” Journal of Economic Perspectives, 28, 29–50.
  • Belloni et al. (2014b) ——— (2014b): “Inference on treatment effects after selection among high-dimensional controls,” Review of Economic Studies, 81, 608–650.
  • Bruckmeier and Wigger (2014) Bruckmeier, K. and B. U. Wigger (2014): “The effects of tuition fees on transition from high school to university in Germany,” Economics of Education Review, 41, 14–23.
  • Card and Krueger (1994) Card, B. D. and A. B. Krueger (1994): “Minimum Wages and Employment: A Case Study of the Fast-Food Industry in New Jersey and Pennsylvania,” American Economic Review, 84, 772–793.
  • Denny (2014) Denny, K. (2014): “The effect of abolishing university tuition costs : Evidence from Ireland,” Labour Economics, 26, 26–33.
  • Dwenger et al. (2012) Dwenger, N., J. Storck, and K. Wrohlich (2012): “Do tuition fees affect the mobility of university applicants? Evidence from a natural experiment,” Economics of Education Review, 31, 155–167.
  • Dynarski (2003) Dynarski, S. (2003): “Does Aid Matter? Measuring the Effect of Student Aid on College Attendance and Completion,” American Economic Review, 93, 279–288.
  • Friedman et al. (2007) Friedman, J., T. Hastie, H. Höfling, and R. Tibshirani (2007): “Pathwise coordinate optimization,” Annals of Applied Statistics, 1, 302–332.
  • Friedman et al. (2010) Friedman, J., T. Hastie, and R. Tibshirani (2010): “Regularization Paths for Generalized Linear Models via Coordinate Descent,” Journal of Statistical Software, 33, 1–22.
  • Hübner (2012) Hübner, M. (2012): “Do tuition fees affect enrollment behavior? Evidence from a ’natural experiment’ in Germany,” Economics of Education Review, 31, 949–960.
  • Huijsman et al. (1986) Huijsman, R., T. Kloek, D. Kodde, and J. Ritzen (1986): “An Empirical Analysis of College Enrollment in the Netherlands,” De Economist, 134, 181–190.
  • Kane (1994) Kane, T. J. (1994): “College Entry by Blacks since 1970 : The Role of College Costs , Family Background , and the Returns to Education,” Journal of Political Economy, 102, 878–911.
  • Leeb and Pötscher (2008) Leeb, H. and B. M. Pötscher (2008): “Can one estimate the unconditional distribution of post-model-selection estimators?” Econometric Theory, 24, 338–376.
  • MacKinnon and White (1985) MacKinnon, J. G. and H. White (1985): “Some heteroskedasticity-consistent covariance matrix estimators with improved finite sample properties,” Journal of Econometrics, 29, 305–325.
  • McPherson and Schapiro (1991) McPherson, M. S. and M. O. Schapiro (1991): “Does Student Aid Affect College Enrollment? New Evidence on a Persistent Controversy,” Economics of Education Review, 81, 309–318.
  • Meinshausen and Bühlmann (2010) Meinshausen, N. and P. Bühlmann (2010): “Stability selection,” Journal of the Royal Statistical Society: Series B (Statistical Methodology), 72, 417–473.
  • Mitze et al. (2015) Mitze, T., C. Burgard, and B. Alecke (2015): “The tuition fee ’shock’: Analysing the response of first-year students to a spatially discontinuous policy change in Germany,” Papers in Regional Science, 94, 385–419.
  • Neill (2009) Neill, C. (2009): “Tuition fees and the demand for university places,” Economics of Education Review, 28, 561–570.
  • Noorbakhsh and Culp (2002) Noorbakhsh, A. and D. Culp (2002): “The demand for higher education: Pennsylvania’s nonresident tuition experience,” Economics of Education Review, 21, 277–286.
  • Tibshirani (1996) Tibshirani, R. (1996): “Regression Selection and Shrinkage via the Lasso,” Journal of the Royal Statistical Society: Series B (Methodological), 58, 267–288.
  • Wagner et al. (2007) Wagner, G., J. Frick, and J. Schupp (2007): “The German Socio-Economic Panel Study (SOEP): Scope, Evolution and Enhancements,” Schmollers Jahrbuch - Journal of Applied Social Science Studies, 127, 139–170.
  • Wooldridge (2002) Wooldridge, J. M. (2002): Econometric analysis of cross section and panel data, Cambridge and London: MIT Press.