Dear all,

I am trying to determine directions of biased (upward or downward) estimates in my study but unfortunately I am still stuck with it. Thus, I would be grateful if someone here can give me a clue on how to determine directions of bias.

Specifically, I am trying to examine the effect of providing caregiving for family members on ones' health. My data is a two-wave panel data and the health outcome is self-rated health (1=good health; 0=poor health). I would expect that providing care would have negative impacts on caregivers' health.

For analysis procedures, I first use propensity score matching to create two balanced groups (caregivers and non-caregivers). I then, after matching, run regression with self-rated health onto an indicator of caregiving (1=caregivers; 0=non-caregivers) and other covariates. I also use weight produced by the matching as probability weight in the regression analysis. Interestingly, I found that providing care, in fact, have positive on health. As I controlled for a set of covariates that could explain individuals' decision to provide care, plus lagged variables of caregiving status and self-rated health at the first wave so I think that my estimates could be suffered from inverse casualty, but I could not figure out what bias direction my estimates could be (upward or downward bias).

Thanks!