ENMT_DIS#10_Regression_Discontinuity

Part I Basics on Regression Discontinuity Design

• What is it & How it works?

  • RDD is a quasi-experiment method that’s used to estimate causal inference. The idea is to compare the two groups of data (untreated vs. treated) between a cutoff point.

  • RDD involves estimating treatment effects by comparing outcomes of individuals just above and just below the threshold. The idea is that any abrupt change in outcomes at the threshold can be attributed to the treatment rather than other confounding factors.

• Are you convinced that RDD can give you causal inference?

  • Yes, if all assumption of RDD are met in a case. Then, I do believe RDD can isolate out the casual relationship between X and Y.

• When is RDD valid? / What are the assumptions of RDD

  • All observations around the cutoff point need to be similar/same in their relevant backgrounds.

  • Test subjects have no voluntary control over on the running variable. This means that they have cannot influence or manipulate their treament status, otherwise it could introduce bias in the regression

  • The cutoff point is not strategically placed to introduce bias.

  • Appropriate bandwidth is chose for the method. The correct bandwidth is subject to change in each individual case. But it should not be too narrow nor too wide.

  • Only one variable that acts as the running variable in which how the treatment is determined. There should not have multiple running variables that makes one qualifies for treatment or not.

• When does RDD become invalid?

  • RDD would become invalid in the following cases:

    • Observations does not share the same/similar backgrounds

    • Test subjects have direct control over their placement of the treatment

    • The Bandwidth is either too narrow or too wide

    • Multiple running variables exist in determining the placement of the treatment on test subjects

Part II Case study Using RDD

Journal : “The effectiveness of investment subsidies: evidence from a regression discontinuity design” - Decramer & Vanormelingen, 2016

• Purpose of the study:

  • to estimate the causal effect of the subsidy on our variables of interest investment, employment, output and productivity.

• Running variable:

  • In our context, a running variable is a determinant of whether a firm qualifies for governmental subsidies. The Flander government has outlined a set of criteria that firms must fulfill to be eligible for the grant, resulting in the presence of multiple running variables. Nevertheless, the researchers opted to focus on a specific variable, namely ‘cash flow,’ as their chosen running variable for the research.

• Sharp or Fuzzy RDD?

  • The criteria that a firm has to meet in order to successfully receiving the grant creates a sharp cutoff point. Therefore, this is study is using a sharp RDD.

• What would I done differently?

  • The researchers chose “Cash flow” as the running variable in their RDD, but this decision carries the risk of introducing bias to the study. It could be beneficial for the researchers to conduct separate RDD analyses with additional variables from the government subsidy criteria to gain a more comprehensive understanding of the relationship between subsidies and the future outlook of the firms

Additional Study Resources

These are the videos I watched to help me understand the concept of RDD

1. Causality: Regression Discontinuity Design
2. Fuzzy Regression Discontinuity (RD) Explained