Covariate balancing PS applied to female labor force supply


The aim of the project is develop a new estimator to analyze the case of a universal policy intervention when control groups are unavailable. The estimator will be applied to quantify the effect of child support instrument on the labor supply of men and women. We will verify a hypothesis that monetary non-equivalent transfer reduces the labor supply of the second earner in a household, ceteris paribus. We will separate between the effects on the breadwinner and the second earner in the aftermath of the large family transfer program in Poland. Since in Poland second earner is typically a woman, we will also compare the labor supply reaction of married women and single earner households with woman as a head of the household.

This project combines applied labor economics with theoretical econometrics. The estimation of the effects of the unconditional non-equivalent transfer program program on labor supply falls into the category of program evaluation econometrics, but most of the estimators require a valid control group. We propose to apply a novel approach: a difference-in-difference (DID) estimator with weights derived from the Covariate Balancing Propensity Score (CBPS) estimator by Imai and Ratkovic (2014). The strategy based on DID exploits the quasi-natural experiment character of program, whereas the weighting scheme based on CBPS will assure proper adjustment of the control group to the treated group. Utilizing data for Poland (labor force survey and household budget survey) we will estimate a range of local treatment effects, to provide reliable boundaries for the total effect.


Źródło finansowania | financingNarodowe Centrum Nauki, PRELUDIUM 12

Projekt realizowany | Timeline: 03/2017 – 03/2020

Budżet łączny | Total budget: 101 997 zł

  • wynagrodzenia dla podstawowych wykonawców | compensation to researchers: 36 000 zł
  • komputery i oprogramowanie | hardware and software: 7 700 zł
  • konferencje i inne wyjazdy | conferences: 24 900 zł
  • Dane |  Data: 9 000 zł
  • książki | books: 2 880 zł
  • materiały | consumables: 2 880 zł
  • koszty pośrednie dla FAME | overheads for FAME: 15 716 zł

Our research proposal consists of two important contributions. First, methodological, we propose a novel way to estimate the causal effect of policy instrument, in a situation, when the instrument design invalidates other estimation methods. Second, this novel estimator will be applied to provide an evaluation of the large scale policy instrument in Poland, effects of child support instrument on household labor supply. Additionally, will develop a statistical package for CBPS in Stata environment. The package will be distributed free of charge on the project website and on user forums. As the CBPS method serves for calculation of covariate balancing propensity score in any context, the scope of potential usage is very broad.

Opublikowane | Published

  • Evaluating the 500+ child support program in Poland | Gospodarka Narodowa

    We investigate immediate effects of a large scale child benefit program introduction on labor supply of the household members in Poland. Due to nonrandom eligibility and universal character of the program standard evaluation estimators are likely to be inconsistent. In order to address this issues we propose a novel approach which combines difference-in-difference (DID) propensity score based methods with covariate balancing propensity score (CBPS) by Imai and Ratkovic (2014). The DID part solves potential problems with non-parallel outcome dynamics in treated and non-treated subpopulations resulting from non-experimental character of the data, whereas CBPS is expected to reduce significantly bias from the systematic differences between treated and untreated subpopulations. We account also for potential heterogeneity among households by estimating a range of local average treatment effects which jointly provide a reliable view on the overall impact. We found that the program has a minor impact on the labor supply in periods following its introduction. There is an evidence for a small encouraging effect on hours worked by treated mothers of children at school age, both sole and married. Additionally, the program may influence the intra-household division of duties among parents of the youngest children as suggested by simultaneous slight decline in participating mothers' probability of working and a small increase in treated fathers' hours worked.


W toku | Work in progress

  • Estimating the effects of universal transfers: new ML approach and application to labor supply reaction to child benefits

    This paper evaluates effects of introduction of a universal child benefit program on female labor supply. Large scale government interventions affect economic outcomes through different channels of various magnitude and direction of the effects. In order to account for this feature, I develop a model in which a woman decides whether to participate in the labor market in a given period. I show how to use the resulting decision rules to explain flows in aggregate labor supply and simulate counterfactual paths of labor force. My framework combines flexibility of reduced form approaches with an appealing structure of dynamic discrete choice models. The model is estimated nonparametrically using recent advances in machine learning methods. The results indicate a 2-4 percentage points drop in labor force among the eligible females, mainly driven by changes in women's perceived trade-offs and beliefs that discouraged inflows.

    In addition to this study, I also present a variety of sensitivity analyses. With the development of statistical theory behind the machine learning algorithms, they are becoming an important tool in the empirical economists' toolbox. By construction, they rely on a set of pre-specified hyper parameters governing the architecture of the algorithm chosen arbitrarily by a researcher. In this note, I show that the economic interpretation of the estimates (obtained via Generalized Random Forest by Susane Athey, Julie Tibshirani, and Stefan Wager ) is robust to different choices of the hyper parameters. This is an encouraging result suggesting that despite their complexity, the machine learning algorithms are likely to become a part of applied econometricians' toolbox.

    Download the note.


Since the seminal paper of Rosenbaum and Rubin, propensity score (PS) has played a significant role in the causal inference framework. It aims to indicate similar units that will be matched or to provide appropriate weights. PS has gained its great popularity by dramatically reducing dimensionality in estimation. Further development of related methods has turned the attention of researchers to the dual nature of PS as a covariate balancing score and conditional probability of treatment assignment. Imai and Ratkovic (2014) exploit the aforementioned duality by deriving a set of appropriate moment conditions and thereby introduce a PS estimator that optimizes the covariate balance—covariate balancing propensity score (CBPS). The paper introduces a new Stata user-written function CBPS that implements the CBPS method within a generalized method of moments framework. The short description of the estimator and the function is presented. Additionally, an empirical exercise that concerns a relationship between a woman's fertility and her labor supply using the exogenous variation due to twin births (Rosenzweig and Wolpin 1980; Angrist and Evans 1998) is provided. The paper also compares the CBPS method with classical PS estimators in unfavorable data environment of a high degree of heterogeneity among women, low fraction of twin births, and exogeneity of the treatment variable with respect to covariates. Moreover, to my knowledge, this is the first paper that concerns the labor supply of Polish women using the first-birth twins identification strategy.

Read the background paper here

Download codes here

See our presentation here