ESRA logo
Tuesday 14th July      Wednesday 15th July      Thursday 16th July      Friday 17th July     

Friday 17th July, 09:00 - 10:30 Room: HT-101

Propensity score methods: methodological developments and innovative applications

Convenor Dr Bruno Arpino (Universitat Pompeu Fabra )

Session Details

Methods, such as matching, weighting and stratification, based on the propensity score have been increasingly used in many fields for estimating causal effects in observational studies. Still a lot of methodological work needs to be done to identify optimal ways of specifying the propensity score model, analyzing covariates balance, treat missing data and measurement error, etc. The panel will focus on recent developments in propensity score methods that address one or more of these issues. Submissions regarding innovative use of propensity score methods in applied works are also welcome.

Paper Details

1. Effectiveness of Sequences of Classroom Training for Welfare Recipients: What works best in West Germany?
Mrs Katharina Dengler (Institute for Employment Research)

Sequences of active labour market programmes (ALMPs) are very common among welfare recipients in Germany, but most studies only evaluate either single ALMPs or unemployed individuals’ first ALMP. Thus, I analyse the effects of different sequences of classroom training, unemployment benefit II (UB-II)-receipt and One-Euro-Jobs for West German men and women on different labour market outcomes. Using rich administrative data and a dynamic matching approach, I can control for dynamic selection problems that occur during a sequence.

2. Does the Participation in a Panel Study Really Change Respondents’ Behavior? Evidence of Panel Conditioning in a German Panel Study on Labor Market Outcomes
Mr Ruben Bach (PhD student at the University of Mannheim, Germany; Researcher at the Statistical Methods Department, Institute for Employment Research, Nuremberg, Germany)
Dr Stephanie Eckman (IAB)

Panel conditioning refers to the phenomenon where participation in repeated interviews changes behavior and/or changes the reporting of behavior. Using administrative data of respondents of a German panel survey on labor market outcomes (PASS), I analyze changes in labor market behavior that are unaffected by respondents’ reporting. Regarding selection and participation in three waves of PASS as a treatment, I estimate average treatment effects using propensity score weighting. Results show that PASS respondents participate in more active labor market policies than a sample of people who were also eligible for participation, but were not selected.

3. How to avoid misspecifying the propensity score equation – Evidence from a Monte Carlo simulation
Dr Gerhard Krug (University of Erlangen-Nuremberg)

Propensity score matching is a semi-parametric method to balance covariates to estimate the causal effect of a treatment, intervention, or action on a specific outcome. The propensity score is typically estimated with a logistic regression or similar parametric model for binary outcomes. Therefore, model specification still plays an important role even if the causal effect is estimated nonparametrically in the matched sample. Methodological research indicates that misspecifying the propensity score equation leads to biased estimates. We build on Dehejia and Wahba (2002) and propose a re-specification algorithm. It is shown to reduce bias in a Monte Carlo Simulation.

4. Propensity score matching with social network data
Dr Bruno Arpino (Universitat Pompeu Fabra, Spain)
Dr Luca De Benedictis (University of Macerata, Italy)
Dr Alessandra Mattei (University of Florence, Italy. )

In some causal inference studies, together with individual unit characteristics, also features of the social network in which units are embedded may be considered as confounders (i.e., variables that impact on both the probability of receiving the treatment and the outcome). Failing to adjust for these social networks factors may lead to biased estimates. We study how to specify the propensity score model when characteristics of the social network cannot be ignored for unconfoundedness to hold. As a motivating case study we consider estimating the effect of liberal trading order on bilateral trade in the XXth century.