Is there a way to make any inference on the impact of treatment on a continuous variable using a non randomized group of subjects that switch on and off treatment? Can a simple comparison of the impact while on and off treatment determine if the treatment itself is impacting the outcome?
switchdirectionality.pdf (4.8 KB)
You are essentially asking if you can get the inference that a randomized crossover study gives without randomization. I think the answer is no. An exception would be in the pharmacokinetics setting where response are rapid and one can get causal inference by temporal alignment.
Could a case-crossover design be useful?
The most fundamental limitation of a case-crossover study is inherent in the question it is designed to answer, “Was this event triggered by anything unusual that happened just before?” Answering this question is seldom sufficient. If one also wants to test hypotheses concerning chronic exposures, a traditional case control study must be conducted simultaneously or the case-crossover study must be nested in a cohort study
maclure2000.pdf (358.3 KB)