In social or medical science, we sometimes have different treatments (likely to affect the same outcome) which have been applied simultaneously, i.e. some units have received treatment A, others have received treatment B, others have received treatment A and B and others have received nothing?
is there a rigorous way to measure ex post the relative effectiveness of treatments A and B, in the form of Average Treatment Effect or more sophisticated estimands capturing the heterogeneity of impacts? e.g. if there are n interventions, i can use 2^n control groups to study all the interactions.
How it will be quasi experiment if you are having intervention and control group
If you have treated and control groups that you have somehow made sure they are nearly identical pre-treatment (matching, for example), can you just have a post-matching regression with your treatments and the the interactions between the relevant treatments?
Kind of like Y=b1Ta+b2Tb+b3(Ta)(Tb)+e ?
Dear Sebastian
A study can be defined as quasi-experimental if there is no control group/condition or no random allocation of the participants into the groups (otherwise it is called RCT). Thus, a quasi with groups is possible (but in this case no randomization has been performed). The statistical analysis for comparing the effectiveness of the different interventions is no different from RCTs. In this case, the two most popular statistical approaches are a) a two-way mixed ANOVA or b) ANCOVA using as covariates the baseline scores.
Hope this helps
Thanks to the three of you!
I was indeed thinking of control groups obtained after matching on observable variables (the outcomes i am measuring are confounded by many variables independently of the treatment=environmental policies).
Therefore, my research question is i) does treatment A (or ii) a combination of treatment A and treatment B) is more effective than no treatment at all? i am not just comparing the outcomes of treatment A and B, i want to know which combination has been the more effective.
Adrian's suggestion seems to be what i was looking for, although i would probably rewrite it to include a difference-in-difference estimator. I will try to figure out if i have enough data to have other estimators able to take into account temporal dynamics and heterogeneity of treatment effects.
Hi Sebastian,
A combination of matching and DiD is usually a good idea. Do get in touch if you want to discuss it further (I'm also a good friend of Esteve), I am always happy to read/hear about these topics.
Cheers.
- Badges
- Science topic
- Similar topics
- Physical Sciences
- Physical Phenomena
- Luminescence
- Fluorescence