GESIS Training Courses
user_jsdisabled
Search

Scientific Coordination

Alisa Remizova

Administrative Coordination

Noemi Hartung

Propensity Score Matching: Computation and Balance Estimation for two and more groups in R

About
Location:
Online via Zoom
 
General Topics:
Course Level:
Format:
Software used:
Duration:
Language:
Fees:
Students: 220 €
Academics: 330 €
Commercial: 660 €
 
Keywords
Additional links
Lecturer(s): Julian Urban, Markus Feuchter

About the lecturer - Julian Urban

About the lecturer - Markus Feuchter

Course description

Are you currently engaged in quasi-experimental field research? Or are you concerned about the possible impact of a covariate on the internal validity of your experiment or observational study? Perhaps you want to conduct group comparisons with your observational (survey) data? For example, you may want to evaluate an intervention (e.g., training, medication, etc.) or assess social disparities (e.g., socio-economic groups, neighborhood, political orientation, etc.). In doing so, you will likely encounter significant group differences in covariates (e.g., age, health, job experience, intelligence) between your target groups, that require post hoc statistical control. If you are struggling to address these types of biases in your data, you may be interested in learning (more) about Propensity Score Matching, a statistical method that also allows for the estimation of causal effects while controlling for uneven numbers of participants per group!
Propensity scoring helps address the imbalance in covariates between your target groups by matching participants from different groups based on similarity in covariates (e.g., background variables). Thus, it enhances the quality of your data and analyses by controlling bias caused by cofounds or unequal group sizes. Traditionally, only one grouping variable with two groups could be considered for matching. Recent developments have expanded this method to accommodate more than two groups (e.g., one intervention group and two control groups) as well as more than one grouping variable (e.g., 2x2 designs).
This comprehensive two-day online course covers the following topics:
  • An introduction into the statistical and substantive problems of imbalanced group designs for social-science research;
  • A review of the rationale, benefits, and drawbacks of propensity score matching in estimating causal effects in various study designs;
  • Different techniques and estimation methods for matching participants via propensity scores ;
  • Utilizing R for the process of computing propensity scores and matching participants;
  • A framework for assessing pre- and post-matching balance;
  • Guidelines for effectively reporting your matching approach in a manuscript.
  • Throughout the workshop, you will benefit from the instructors' experience, conveyed through lectures and practical exercises. To this end, we will provide a simulated data set in R (though you are encouraged to bring your own data for practicing on it). Moreover, there will be the opportunity to discuss specific issues related to your own research with one of the two instructors during the afternoon session on the second day.
     
    Organizational structure of the course
    The workshop has four daily sessions. Each session lasts 90 minutes. The sessions include both lectures and coding labs. The first session of the day starts at 9:00am and the last session ends at 4:30pm. Two sessions take place in the morning and are separated by a coffee break. After these two sessions, there will be a one-hour lunch break. In the afternoon of the first day, there will be two sessions separated by a coffee break. On the second day, the second afternoon session will be dedicated to individual consultations, allowing you to reflect on and discuss in depth your own research questions in depth with one of the lectures.


    Target group

    You will find the course useful if:
    • you conduct quasi-experimental field research.
    • you conduct statistical analyses with groups based on observational (survey) data.
    • you seek to estimate causal effects using propensity score matching.
    • you seek to correct experimental or observational data for imbalanced covariates.
    • you face the challenge of balancing data with more than two relevant groups.
    • you face the challenge of balancing data with more than one grouping variable.
    • you would like to estimate and report post-matching balance.


    Learning objectives

    By the end of the course, you will:
    • Understand the rationale and current discussion on propensity score matching;
    • Be able to conduct propensity score matching in R using R packages, such as twang and MAGMA.R;
    • Be able to apply propensity score matching in their own research projects.


    Prerequisites

  • Basic knowledge of R (opening and viewing data, manipulating objects, working with functions and packages, ideally worked with R in earlier projects).
  • Basic knowledge of (quasi-)experimental/observational study designs and understanding the differences between these designs (randomization, possibility to estimate causal effects).
  • Basic knowledge of inference statistics (statistical tests such as t-test, (M)ANOVA, correlations).
  • Basic knowledge of effect sizes (Cohen's d, ideally Hedges' g).
  •  
    Software requirements
    You need to have the recent R (and R Studio) version installed.


    Schedule

    Recommended readings