Hello,
I have repeated observations on people and am interested in estimating risk ratios, which can be done in a log-binomial model (or a Poisson model, but the Poisson estimates will generally have larger standard errors). There are important unmeasured confounders, so the planned approach is a person-level fixed-effects model to control for all time-invariant confounding between individuals.
I've read about several packages (xtlogit, xtgee, binreg, reghdfe, xtpoisson). It looks to me like xtpoisson is the only one that includes a fixed-effects estimator and can be used to estimate risk ratios. Is anyone aware of another package that will do this, preferably using log-binomial regression? If so, I would love to hear about that package and any trade-offs between it and xtpoisson.
I'd prefer to avoid logistic regression. I'd like to be able to adjust the standard errors for clustering within person and to estimate average marginal effects, both of which xtpoisson will do.
Thank you!
Alan
I have repeated observations on people and am interested in estimating risk ratios, which can be done in a log-binomial model (or a Poisson model, but the Poisson estimates will generally have larger standard errors). There are important unmeasured confounders, so the planned approach is a person-level fixed-effects model to control for all time-invariant confounding between individuals.
I've read about several packages (xtlogit, xtgee, binreg, reghdfe, xtpoisson). It looks to me like xtpoisson is the only one that includes a fixed-effects estimator and can be used to estimate risk ratios. Is anyone aware of another package that will do this, preferably using log-binomial regression? If so, I would love to hear about that package and any trade-offs between it and xtpoisson.
I'd prefer to avoid logistic regression. I'd like to be able to adjust the standard errors for clustering within person and to estimate average marginal effects, both of which xtpoisson will do.
Thank you!
Alan
