Summary Health Economics week 6
6.1 Research methods
There are several study designs to establish a causation or correlation
When there is some variation in the treatment
a) We know where the variation is coming from b) We don’t know where variation is coming
from
- Randomization - Regression
- Matching
Natural experiments - Sibling, twin fixed effects
- Differences-in-differences - Individual fixed effects
- instrumental variables
- regression discontinuity
Trade-off between internal and external validity
Experiments and natural experiments are the most convincing in terms of establishing causality:
they have the highest internal validity (=we know where the variation comes from)
However, these experiments usually apply to a (very) specific group of people: this implies that the
most convincing methods usually have the lowest external validity (=to what extend can results be
generalized)
When a RCT study is not feasible (*ethical alcohol-mortality)
1. Differences-in-differences
compare before and after an event in time: study
trends. Problem: there might be other confounders. You
also need a control group without event
*decrease tax on alcohol: does the alcohol consumption
rise? Take Belgium as control
– Control for: pre-treatment differences + time
events (with a RCT)
* control group+ treatment group: offer event/treatment- do you see improvement
in treatment group?
2. Regression discontinuity
If there is a discontinuity at a variable of interest without
other discontinuities, you can see what has happened at the
tresshold
*More deaths by motor because of drinking age 21? Not
because they just started to drive (in America age 16)
– You have to check for more discontinuities (*increase in
mental health problems at 21?)
6.1 Research methods
There are several study designs to establish a causation or correlation
When there is some variation in the treatment
a) We know where the variation is coming from b) We don’t know where variation is coming
from
- Randomization - Regression
- Matching
Natural experiments - Sibling, twin fixed effects
- Differences-in-differences - Individual fixed effects
- instrumental variables
- regression discontinuity
Trade-off between internal and external validity
Experiments and natural experiments are the most convincing in terms of establishing causality:
they have the highest internal validity (=we know where the variation comes from)
However, these experiments usually apply to a (very) specific group of people: this implies that the
most convincing methods usually have the lowest external validity (=to what extend can results be
generalized)
When a RCT study is not feasible (*ethical alcohol-mortality)
1. Differences-in-differences
compare before and after an event in time: study
trends. Problem: there might be other confounders. You
also need a control group without event
*decrease tax on alcohol: does the alcohol consumption
rise? Take Belgium as control
– Control for: pre-treatment differences + time
events (with a RCT)
* control group+ treatment group: offer event/treatment- do you see improvement
in treatment group?
2. Regression discontinuity
If there is a discontinuity at a variable of interest without
other discontinuities, you can see what has happened at the
tresshold
*More deaths by motor because of drinking age 21? Not
because they just started to drive (in America age 16)
– You have to check for more discontinuities (*increase in
mental health problems at 21?)
