unequal time intervals and polynomial constraints

[Lyndon]

Setting time intervals adjusts survival estimates to an underlying rate (e.g., daily, monthly). I assume that to fit a linear trend to survival a sequence like 1,2,3,4,... in the DM would be correct whether equal or unequal time intervals were specified. Does this apply to all parameters, e.g., lambda's or f's in Pradel models, pents in POPAN models, or in my present case, transition probabilities in a multistate model? Are these inherently estimated on an occasion to occasion basis and would it be necessary to adjust a sequence like 1,2,3,4...according to the time intervals to get an underlying linear trend?

[cooch]

Setting time intervals adjusts survival estimates to an underlying rate (e.g., daily, monthly). I assume that to fit a linear trend to survival a sequence like 1,2,3,4,... in the DM would be correct whether equal or unequal time intervals were specified. Does this apply to all parameters, e.g., lambda's or f's in Pradel models, pents in POPAN models, or in my present case, transition probabilities in a multistate model? Are these inherently estimated on an occasion to occasion basis and would it be necessary to adjust a sequence like 1,2,3,4...according to the time intervals to get an underlying linear trend?

As long as you specify the appropriate time intervals, then you shouldn't need to do more than specify the trend in the DM.

Unequal intervals in a MS model are more problematic though, given the dimensionality of state uncertainty if (say) you miss an occasion. In fact, the preferred approach is to code the unequal interval as a missing occasion in the input file (using the 'dot' notation), and then proceeding from there.

[Lyndon]

Thanks Evan. The survey was originally designed to produce data for a robust design multistate model but transitions were observed to occur within primary samples. I'm now trying to fit a CJS mulitstate model in which the time intervals are very unequal, e.g., 1,1,19,1,1,... I ran some models with the time intervals specified either on a daily or monthly basis and found that the psi's were not changed by this. Also, the trends seemed just too smooth given the wide variation in interval lengths. Hence my question.