|
|
Monte Carlo for survival data |
|
Message/Author |
|
Hanjoe Kim posted on Wednesday, January 20, 2016 - 6:33 pm
|
|
|
I am trying to understand the HAZARDC command. First I thought that this command is used to specify the censoring proportion in data but I learned that the value given here is the lambda value for an exponential parameter. From Asparouhov and Muthen (2014), if lambda=lambda1 (where lambda is defined within Model Population T#1*lambda with GENERATE = T(s) and lambda1 is the value used in HAZARDC), about 50% of the observations will be censored. Can you give me guidance to adjust the censoring proportion in Monte Carlo generated data? For example, if T#1*.1, then HAZARDC=T(.1) will give about 50% censoring. Given that the lambda value is fixed at .1, what would the lambda1 values be to get 10% or 30% censoring? Thank you in advance! |
|
|
We generate the survival variable T and independently censoring variable C with exponential distribution with parameter given by the HAZARDC command. If C > T the variable is not censored. If C < T the T variable is censored at time C. If C~Exp(a) and T~Exp(b) then Prob(C<T)=a/(a+b), see http://www.math.wm.edu/~leemis/chart/UDR/PDFs/ExponentialLaplace.pdf If a=b you get 50% censoring. If a=(3/7)*b you get 30% censoring. If a=(1/9)*b you get 10% censoring. HAZARDC=T(.043) will give you 30% HAZARDC=T(.011) will give you 10% |
|
Hanjoe Kim posted on Thursday, January 21, 2016 - 1:35 pm
|
|
|
This is crystal clear! Thank you Dr. Asparouhov! Hanjoe. |
|
Back to top |
|
|