No. of parameters when using CLogit

[rcurrey]

I am using a parameter specific cumulative logit link function (CLogit) to model declining survival (see 7.8.1 sidebar in Cooch & White) and just wanted to confirm the appropriate number of parameters that should appear in the results browser.

I am estimating survival over 14 periods but the model derives only 3 different survival estimates (i.e. survival remains constant for a number of years then declines). Should the number of parameters reflect the number that are able to be estimated (14) or the number that are actually estimated (3)? The sidebar in 7.8.1 appears to suggest the latter but I just wanted to confirm this. I realise the number of parameters in the results browser will also include those associated with estimating recapture probability.

[cooch]

I am using a parameter specific cumulative logit link function (CLogit) to model declining survival (see 7.8.1 sidebar in Cooch & White) and just wanted to confirm the appropriate number of parameters that should appear in the results browser.

I am estimating survival over 14 periods but the model derives only 3 different survival estimates (i.e. survival remains constant for a number of years then declines). Should the number of parameters reflect the number that are able to be estimated (14) or the number that are actually estimated (3)? The sidebar in 7.8.1 appears to suggest the latter but I just wanted to confirm this. I realise the number of parameters in the results browser will also include those associated with estimating recapture probability.

Good question - the text in the book should have dealt with it (next revision). The number of parameters is the number that actually show up in the constraint - not the number of potential parameters. The constraint, if you will, is the ordinal sequence. The number of separate parameters that are estimated subject to that constraint is what you're after. So, if you had

S(1) <= S(2) <= S(3) <= S(4)

as the constraint, and if the ML estimate under that constraint yields

S(1) <= S(2) = S(3) < S(4)

then you have 3 parameters estimated.

[rcurrey]

Good question - the text in the book should have dealt with it (next revision). The number of parameters is the number that actually show up in the constraint - not the number of potential parameters. The constraint, if you will, is the ordinal sequence. The number of separate parameters that are estimated subject to that constraint is what you're after. So, if you had

S(1) <= S(2) <= S(3) <= S(4)

as the constraint, and if the ML estimate under that constraint yields

S(1) <= S(2) = S(3) < S(4)

then you have 3 parameters estimated.

Thanks very much, that clears it up. By the way, the book has been invaluable. Cheers.