Week 12: Quasi Experiments

STAT 235

Natural (i.e., QUASI) Experiments

Sometimes doing an RCT is not feasible. Instead, we can try to find an RCT out in the wild by observing

two groups that have arisen naturally.

Recall that the most important aspect of an RCT is that the treatment is randomly assigned. However,

for natural/quasi experiments, it is not possible for the researcher to implement random

assignment . Rather, groups are selected in an “effectively random” way.

For this course, you are responsible for understanding and applying two quasi experimental designs:

1) Difference-in-Differences (DiD)

2) Regression Discontinuity Design (RDD)

Difference- in- Differences (DID)

What we can do instead is combine the within-unit difference with the across-group growth

difference (i.e., take a difference-in-differences )

● measure change in traffic accident rates in both states before and after policy implementation

Regression Discontinuity Design

(RDD)

KEY IDEA:

● A threshold/cutoff that creates two groups that are otherwise similar to each other on average.

● There is no material difference between observations right before/right after the cutoff, except for

the treatment itself.

● Thus, at the threshold, we essentially have a randomized experiment and can find the causal effect

by comparing the outcomes of the two groups.

EXAMPLE:

● Once a driver in Texas turns 79, he/she must have their driver’s license renewed by successfully

passing a vision test. This policy is intended to reduce the risk of car accidents/other incidents.

● Drivers right below the age of 79 are not fundamentally different from drivers right above the age of

79. The only difference between the two groups is the treatment itself (renewing DL).

● Thus, by comparing the number of traffic violations for drivers in these two groups, we can estimate

the causal effect of DL renewal on safety outcomes.

RDD Visual Representations