Occasion specific covariates

[simone77]

Hi all,

Two short questions:

1) I have 6 occasions and for each of them the measure of three covariates that could affect the survival (they represent the prevalence of some specific disease in the population). Is there any rule of thumb on the minumum number of the covariates' elements (i.e. in my case would be five, one for each interval) that make the analyses enough powerful to detect an effect on the survival?

2) Unfortunately, due to logistic reasons, the intervals among occasions are uneven (two, three, three, four and two months): do you believe that specifying correctly the intervals lengths at the beginning of the analysis in MARK would be enough? My doubt arises from the fact that the measures of each covariate would be affecting the survival during a different time interval (the time is varying between the measures).

Thanks in advance for any response,

Simone

[simone77]

Hi everybody,

I am afraid my questions were not clear enough, so I will try to elaborate more. I am really interested in getting an answer.

1) I know there are rules of thumb in statistics that allow you to find the minimum number of covariates needed to perform that analysis (i.e. in a least square regression, for continuous covariates there should be at least ten elements per covariates), the question is:
- What is the minimum ratio between occasion specific covariates and sample size, and what is the sample size in this case? the number of occasions? the number of unique encounter histories? number of individuals?

For sure somewhere in the GM there is an answer to this but I could not find it.

2) In my case the only way to test the effects of these covariates should be by considering them in an uneven intervals study. The intervals, as said in the previous post, are: two, three, three, four and two months. At the moment, I am looking at the monthly survival and the covariates represent the prevalence of some disease in the population that, unfortunately, have not been measured every x time units.
The problem arises from the fact that if I look at the monthly survival, the survival of the first two months is the root square of the first interval survival, the survival of the third, fourth and fifth month is the cubic square of the second interval survival and so forth. This way, the effect of the covariate would be calculated before on a two months interval, after on a three months, and so on...
Perhaps I should not consider the monthly survival...I am having some trouble to face it. Any suggestion?

Thanks for any help you may provide.

Simone