Number of Parameters in Known Fate Analysis

[jlaskowski]

I am running a known fate analysis to assess daily survival across 147 encounter occasions (days) for 51 individuals. When I run a time dependent model (via PIM or the Design Matrix) Mark reports only 24 parameters in the result browser. Mark provides 147 real survival estimates, only 24 of which are less than 1.0. I think this is due to a combination of our small sample size and frequently censored individuals. If I manually change the number of parameters in the result browser to 146, the model ranking drastically changes. I'm wondering if anyone has come across this? Thank you.

[AdamGreen]

That is common when parameters aren't being estimated well - usually when estimates are near the boundaries. If you look at the parameter estimates, a lot of the SEs should be either 0 or very large, indicating that they weren't estimated well. I would suggest a simpler model - maybe a linear or quadratic trend. But you should correct the number of parameters in the results browser to reflect the correct number of parameters (which you can get from the beta estimate output). As for the AIC values, that will be different because the "2k" term in the equation.

Adam Green

[jlaskowski]

Thanks for your response. This is really helpful.