search for: ssm_bin_ex

...al to the "observed" data. However, if I first do a logit transform of the proportions and then use ssm (or my version) with gaussian errors, the point estimates of the state vector are not nearly as close as those obtained with the binomial error structure (see my attached code "ssm_bin_ex.r"). I can understand why the variance of the estimates would differ, but why the means? If I fit some dummy models for non-time series models with glm, the estimates of the means using the 2 approaches just outlined are nearly identical (see my attached code "glm_bin_ex.r"). I...