# Regression Discontinuity Design

#### Timole

##### New Member
Hey,

I am new here and hope that you can help me with my questions as I cannot find an answer myself. For a paper I am working on, I did some research about Regression Discontinuity Design. I was wondering what exactly the difference to a simple dummy variable is? For example, if I have a change in regulations on the 12 of May 2018, then I can use Regression Discontinuity Design to analyse the effect of the change in the regulatory framework. But to my understanding, the only thing that I am doing under the Regression Discontinuity Design is to implement a dummy variable that is "1" if the observation is after the 12th of May 2018 and "0" if it is before. So my questions is: Is Regression Discontinuity Design in this context really just inserting a dummy variable that specifies whether it is before or after the cutoff date?

Thank you very much for taking the time to answer the question.

All the best,
Timo

#### noetsi

##### No cake for spunky
The purpose of RDD is totally different than regression, it serves for those who want experimental results when experimental results are not possible (for example when random assignment is not possible). It is called a semi-experimental method (it is actually one of the best if you can meet its assumptions which often you can not).

Normally when you run a regression you accept the validity of correlation research and are not interested per se in causation or meeting semi experimental requirements.

#### Timole

##### New Member
Thank you for the answer. I am aware that Regression Discontinuity Design is a semi-experimental method. However, I am still wondering how the application itself (so the regression equation) differs from a regression with a dummy variable? I am aware that the underlying idea of RDD is different from a simple regression with a dummy variable.
To come back to the example with the change in the regulatory framework. Hausmann & Rapson (2018) compare in their paper "Regression Discontinuity in Time: Considerations for Empirical Applications" different studies that used Regression Discontinuity in Time to evaluate for example the effects of new regulations. So in this case, the random assignment comes from the date the new regulations are implemented. All observations before the date are allocated to one group and all after the date are allocated to the other group. So what I was wondering is how the regression equation in the practical application differs from a simple regression with a dummy variable that allocates a "1" to one group and a "0" to the other group?

#### Dason

##### Ambassador to the humans
Regression Discontinuity Design refers to the design of the "experiment". There are a few different ways to analyze these. The method you mentioned is a common approach and yes it is as simple as adding the dummy/indicator variable into the model. There are other approaches to analyzing this kind of experiment though.

#### hlsmith

##### Less is more. Stay pure. Stay poor.
The use of a time based intervention may also be used in interrupted time series. Where you can have a dummy for the period and perhaps also a control group, though you need to control for autocorrelation in data values next to each other. I agree, that the design you are describing seems as simple as adding a dummy variable (indicator). This is given random treatment allocation can be assumed. If subjects aren't comparable between groups, additional approaches may need to be implemented to balance groups. When I think of RDD, I think them as a perk of dichotomania, where someone instituted a decision based on a data split of a continuous variable and given measurement variability and random error, the people on right next to the threshold can be comparable on any given day. Given this, they can be compare and the intervention can be considered randomized.

#### noetsi

##### No cake for spunky
It depends on how interrupted time series is done. Often one looks qualitatively at change in which case autocorrelation is not meaningful. Interrupted time series has trouble with controlling for confounds or impact which gradual rather than immediate.

RDD Is very useful for many public organization where random assignment is commonly illegal.