Any opinions expressed in this paper are those of the author(s) and not those of IZA. Research pu... more Any opinions expressed in this paper are those of the author(s) and not those of IZA. Research published in this series may include views on policy, but IZA takes no institutional policy positions. The IZA research network is committed to the IZA Guiding Principles of Research Integrity. The IZA Institute of Labor Economics is an independent economic research institute that conducts research in labor economics and offers evidence-based policy advice on labor market issues. Supported by the Deutsche Post Foundation, IZA runs the world's largest network of economists, whose research aims to provide answers to the global labor market challenges of our time. Our key objective is to build bridges between academic research, policymakers and society. IZA Discussion Papers often represent preliminary work and are circulated to encourage discussion. Citation of such a paper should account for its provisional character. A revised version may be available directly from the author.
This paper investigates the sensitivity of average wage gap decompositions to methods resting on ... more This paper investigates the sensitivity of average wage gap decompositions to methods resting on different assumptions regarding endogeneity of observed characteristics, sample selection into employment, and estimators' functional form. Applying five distinct decomposition techniques to estimate the gender wage gap in the U.S.\ using data from the National Longitudinal Survey of Youth 1979, we find that the magnitudes of the wage gap components are generally not stable across methods. Furthermore, the definition of the observed characteristics matters: merely including their levels (as frequently seen in wage decompositions) entails smaller explained and larger unexplained components than when including both their levels and histories in the analysis. Given the sensitivity of our results, we advise caution when using wage decompositions for policy recommendations.
This paper considers the evaluation of direct and indirect treatment effects, also known as media... more This paper considers the evaluation of direct and indirect treatment effects, also known as mediation analysis, when outcomes are only observed for a subpopulation due to sample selection or outcome attrition. For identification, we combine sequential conditional independence assumptions on the assignment of the treatment and the mediator, i.e. the variable through which the indirect effect operates, with either selection on observables/missing at random or instrumental variable assumptions on the outcome attrition process. We derive expressions for the effects of interest that are based on inverse probability weighting by specific treatment, mediator, and/or selection propensity scores. We also provide a brief simulation study and an empirical illustration based on U.S. Project STAR data that assesses the direct effect and indirect effect (via absenteeism) of smaller kindergarten classes on math test scores.
Based on empirical data from selected public universities in Khabarovsk, Russia, this paper compa... more Based on empirical data from selected public universities in Khabarovsk, Russia, this paper compares first- and fifth-year students regarding their attitudes towards corruption in general and university corruption in particular. Even after making both groups of students comparable with respect to a range of socio-economic characteristics by a matching approach, the results suggest that fifth-year students are more open to a range of informal and corrupt practices than first years. Our analysis therefore points to the possibility that the Russian higher education system might ‘favour’ compliance with corruption and informal practices, with potentially detrimental consequences for the Russian society as a whole.
This paper suggests a causal framework for disentangling individual level treatment effects and i... more This paper suggests a causal framework for disentangling individual level treatment effects and interference effects, i.e., general equilibrium, spillover, or interaction effects related to treatment distribution. Thus, the framework allows for a relaxation of the Stable Unit Treatment Value Assumption (SUTVA), which assumes away any form of treatment-dependent interference between study participants. Instead, we permit interference effects within aggregate units, for example, regions or local labor markets, but need to rule out interference effects between these aggregate units. Borrowing notation from the causal mediation literature, we define a range of policy-relevant effects and formally discuss identification based on randomization, selection on observables, and difference-indifferences. We also present an application to a policy intervention extending unemployment benefit durations in selected regions of Austria that arguably affected ineligibles in treated regions through general equilibrium effects in local labor markets. IfW Kiel, IZA (andreas.steinmayr@econ.lmu.de). Financial support by Deutsche Forschungsgemeinschaft through CRC TRR 190 is gratefully acknowledged. We thank Josef Zweimüller for his support with the empirical application. Joachim Winter provided helpful comments on the draft.
We describe the R package causalweight for causal inference based on inverse probability weightin... more We describe the R package causalweight for causal inference based on inverse probability weighting (IPW). The causalweight package offers a range of semiparametric methods for treatment or impact evaluation and mediation analysis, which incorporates intermediate outcomes for investigating causal mechanisms. Depending on the method, identification relies on selection on observables assumptions or on instrumental variables when selection is on unobservables, approaches that may also be applied to tackle non-random outcome attrition and sample selection. Inference is based on the bootstrap.
We combine machine learning techniques with statistical screens computed from the distribution of... more We combine machine learning techniques with statistical screens computed from the distribution of bids in tenders within the Swiss construction sector to predict collusion through bid-rigging cartels. We assess the out of sample performance of this approach and find it to correctly classify more than 80% of the total of bidding processes as collusive or non-collusive. As the correct classification rate, however, differs across truly non-collusive and collusive processes, we also investigate tradeoffs in reducing false positive vs. false negative predictions. Finally, we discuss policy implications of our method for competition agencies aiming at detecting bid-rigging cartels.
This paper proposes semi-and nonparametric methods for disentangling the total causal effect of a... more This paper proposes semi-and nonparametric methods for disentangling the total causal effect of a continuous treatment on an outcome variable into its natural direct effect and the indirect effect that operates through one or several intermediate variables or mediators. Our approach is based on weighting observations by the inverse of two versions of the generalized propensity score (GPS), namely the conditional density of treatment either given observed covariates or given covariates and the mediator. Our effect estimators are shown to be asymptotically normal when the GPS is estimated by either a parametric or a nonparametric kernel-based method. We also provide a simulation study and an application to the Job Corps program.
When estimating local average and marginal treatment effects using instrumental variables (IV), mu... more When estimating local average and marginal treatment effects using instrumental variables (IV), multivalued endogenous treatments are frequently binarized based on a specific threshold in treatment support. However, such binarization introduces a violation of the IV exclusion if (i) the IV affects the multivalued treatment within support areas below and/or above the threshold and (ii) such IV-induced changes in the multivalued treatment affect the outcome. We discuss assumptions that satisfy the IV exclusion restriction with the binarized treatment and permit identifying the average effect of (i) the binarized treatment and (ii) unit-level increases in the original multivalued treatment among specific compliers. We derive testable implications of these assumptions and propose tests, which we apply to the estimation of the returns to (binary) college graduation instrumented by college proximity.
This paper examines how anti-corruption educational campaigns affect the attitudes of Russian uni... more This paper examines how anti-corruption educational campaigns affect the attitudes of Russian university students towards corruption and academic integrity. About 2,000 survey participants were randomly assigned to one of four different information materials (brochures or videos) about the negative consequences of corruption or to a control group. Using machine learning to detect effect heterogeneity, we find that various groups of students react to the same information differently. Those who commonly plagiarize, who receive excellent grades, and whose fathers are highly educated develop stronger negative attitudes towards corruption in the aftermath of our intervention. However, some information materials lead to more tolerant views on corruption among those who rarely plagiarize, who receive average or above average grades, and whose fathers are less educated. Therefore, policy makers aiming to implement anti-corruption education at a larger scale should scrutinize the possibility of (undesired) heterogeneous effects across student groups.
This paper investigates the effects of an information campaign about a
governmental r... more This paper investigates the effects of an information campaign about a governmental rural development program (RDP) in the Former Yugoslav Republic of Macedonia on the farmers’ intention to participate in the RDP. In the course of a survey among farmers, the treatment group received an information brochure with relevant details on selected RDP measures, while the control group received no information. Even though the intervention had been planned as experiment, randomization was not properly conducted, requiring sample adjustments and controlling for observed covariates in the estimation process. The results suggest that while the intervention succeeded in informing farmers, it had a negative, albeit marginally statically significant, effect on farmers’ reported possibility and intention to use RDP support in the near future. Evidence from further outcome variables suggests that this may be due to the information about administrative burden associated with RDP participation provided in the brochure. We also find that the negative effect is driven by the subsample of unprofitable farmers.
This paper investigates the finite sample properties of a range of inference methods for propensi... more This paper investigates the finite sample properties of a range of inference methods for propensity score-based matching and weighting estimators frequently applied to evaluate the average treatment effect on the treated. We analyse both asymptotic approximations and bootstrap methods for computing variances and condence intervals in our simulation design, which is based on large scale labor market data from Germany and varies w.r.t. treatment selectivity, eect heterogeneity, the share of treated, and the sample size. The results suggest that in general, the bootstrap procedures dominate the asymptotic ones in terms of size and power for both matching and weighting estimators. Furthermore, the results are qualitatively quite robust across the various simulation features.
We introduce a wild bootstrap algorithm for the approximation of the sampling distribution of pai... more We introduce a wild bootstrap algorithm for the approximation of the sampling distribution of pair or one-to-many propensity score matching estimators. Unlike the conventional iid bootstrap, the proposed wild bootstrap approach does not construct bootstrap samples by randomly resampling from the observations with uniform weights. Instead, it fixes the covariates and constructs the bootstrap approximation by perturbing the martingale representation for matching estimators. We also conduct a simulation study in which the suggested wild bootstrap performs well even when the sample size is relatively small. Finally, we provide an empirical illustration by analyzing an information intervention in rural development programs.
This study empirically evaluates the impact of the war in eastern Ukraine on the political attitu... more This study empirically evaluates the impact of the war in eastern Ukraine on the political attitudes and sentiments towards Ukraine and Russia among the population living close to the war zone on the territory controlled by the Ukrainian government. Exploiting unique survey data that were collected in early 2013 (13 months before the outbreak of the conflict) and early 2015 (11 months after the outbreak), we employ two strategies to infer how the war has affected two different groups defined by distance to the war zone. First, we apply a before-after analysis to examine intra-group changes in attitudes over time. Second, we use a difference-indifferences approach to investigate inter-group divergence over time. Under particular assumptions, the latter approach yields a lower absolute bound for the effect. We control for a range of observed characteristics and consider both parametric and semiparametric estimation based on inverse probability weight-ing. Our results suggest that one year of conflict negatively affected attitudes towards Russia, while mostly no statistically significant intra-or inter-group differences were found for sentiments towards Ukraine.
This paper proposes a difference-indifferences approach for disentangling a total treatment effec... more This paper proposes a difference-indifferences approach for disentangling a total treatment effect on some outcome into a direct impact as well as an indirect effect operating through a binary intermediate variable – or mediator – within strata defined upon how the mediator reacts to the treatment. We show under which assumptions the direct effects on the always and never takers, whose mediator is not affected by the treatment, as well as the direct and indirect effects on the compliers, whose mediator reacts to the treatment, are identified. We provide an empirical application based on the Vietnam draft lottery. The results suggest that a high draft risk due to the lottery leads to a relative increase in the support for the Republican Party and that this increase is mostly driven by those complying with the lottery outcome.
Changes in compulsory schooling laws have been proposed as an instrument for the endoge-nous choi... more Changes in compulsory schooling laws have been proposed as an instrument for the endoge-nous choice of schooling. It has been argued that raising minimum schooling exogenously increases the educational attainment of a subset of pupils without directly affecting later life outcomes such as income or health. Using the methods of Huber and Mellace (2015) and Kitagawa (2015) and data from the Survey of Health, Ageing and Retirement in Europe, we jointly test random instrument assignment, weak monotonicity of education in the instrument, and the instrument exclusion restriction. The satisfaction of these restrictions permits identifying the local average treatment effect of education on those choosing more schooling as a reaction to the law change. Our results do not point to the invalidity of the schooling law instrument, though we acknowledge that even asymptotically, testing cannot detect all possible violations of instrument validity.
This paper provides a review of methodological advancements in the evaluation of heterogeneous tr... more This paper provides a review of methodological advancements in the evaluation of heterogeneous treatment effect models based on instrumental variable (IV) methods. We focus on models that achieve identification by assuming monotonicity of the treatment in the IV and analyze local average and quantile treatment effects for the subpopulation of compliers. We start with a comprehensive discussion of the binary treatment and binary IV case as for instance relevant in randomized experiments with imperfect compliance. We then review extensions to identification and estimation with covariates, multi-valued and multiple treatments and instruments, outcome attrition and measurement error, and the identification of direct and indirect treatment effects, among others. We also discuss testable implications and possible relaxations of the IV assumptions, approaches to extrapolate from local to global treatment effects, and the relationship to other IV approaches.
Using a sequential conditional independence assumption, this paper discusses fully nonpara-metric... more Using a sequential conditional independence assumption, this paper discusses fully nonpara-metric estimation of natural direct and indirect causal effects in causal mediation analysis based on inverse probability weighting. We propose estimators of the average indirect effect of a binary treatment, which operates through intermediate variables (or mediators) on the causal path between the treatment and the outcome, as well as the unmediated direct effect. In a first step, treatment propensity scores given the mediator and observed covariates or given covariates alone are estimated by nonparametric series logit estimation. In a second step, they are used to reweigh observations in order to estimate the effects of interest. We establish root-n consistency and asymptotic normality of this approach as well as a weighted version thereof. The latter allows evaluating effects on specific subgroups like the treated, for which we derive the asymptotic properties under estimated propensity scores. We also provide a simulation study and an application to an information intervention about male circumcisions.
This paper investigates regional differences in the perception of corruption and informal practic... more This paper investigates regional differences in the perception of corruption and informal practices among Ukrainian firms. Using two different data sets from Ukraine we show that perceived corruption differs significantly across regions, even when taking into account the size, industry, workforce composition, and other characteristics of the firms based on propensity score matching. In particular, perceived corruption is highest in the eastern areas and lowest in the western region, which points to distinct business practices that may be rooted in the different political, cultural, and historical development of Ukrainian regions.
This paper evaluates the effect of a voucher award system for assignment into vocational training... more This paper evaluates the effect of a voucher award system for assignment into vocational training on the employment outcomes of unemployed voucher recipients in Germany, along with the causal mechanisms through which it operates. It assesses the direct effect of voucher assignment net of actual redemption, which may be driven by preference shaping/learning about (possibilities of) human capital investments or simply by costs of information gathering. Using a mediation analysis framework based on sequential conditional independence assumptions and semiparametric matching estimators, our results suggest that the negative short term and positive long term employment effects of voucher award are mainly driven by actual training participation. However, also the direct effect of just obtaining the voucher is negative in the short run. This points to potential efficiency losses of voucher award systems if individuals decide not to redeem, as employment chances are lower than under non-award in the short run and under redemption in the long run, making nonredemption the least attractive option.
Uploads
Papers by Martin Huber
Applying five distinct decomposition techniques to estimate the gender wage gap in the U.S.\ using data from the National Longitudinal Survey of Youth 1979, we find that the magnitudes of the wage gap components are generally not stable across methods. Furthermore, the definition of the observed characteristics matters: merely including their levels (as frequently seen in wage decompositions) entails smaller explained and larger unexplained components than when including both their levels and histories in the analysis. Given the sensitivity of our results, we advise caution when using wage decompositions for policy recommendations.
compliers. We derive testable implications of these assumptions and propose tests, which we apply to the estimation of the returns to (binary) college graduation instrumented by college proximity.
governmental rural development program (RDP) in the Former Yugoslav
Republic of Macedonia on the farmers’ intention to participate in the RDP. In the
course of a survey among farmers, the treatment group received an information
brochure with relevant details on selected RDP measures, while the control group
received no information. Even though the intervention had been planned as
experiment, randomization was not properly conducted, requiring sample
adjustments and controlling for observed covariates in the estimation process.
The results suggest that while the intervention succeeded in informing farmers, it
had a negative, albeit marginally statically significant, effect on farmers’ reported
possibility and intention to use RDP support in the near future. Evidence from
further outcome variables suggests that this may be due to the information about
administrative burden associated with RDP participation provided in the brochure.
We also find that the negative effect is driven by the subsample of unprofitable
farmers.