Hello,
The context of my analysis is a longitudinal complex survey. Variable information is updated every 2 years from 1994/95 through 2010/11 providing up to 8 time points for each survey participant for analysis. Sample size is roughly 12000, weighted to represent the Canadian population at the beginning of the survey in 1994. I am interested in the extent to which arthritis (analyzed as a time varying exposure) is associated with time to first occurrence of heart disease adjusting for time-invariant and time-varying covariates, and the extent to which the effect of arthritis on incident heart disease is mediated through effects of disability (also analyzed as a time-varying covariate). Given the panel nature of the data I use discrete time-survival analysis with data in a person-period format. Preliminary analysis point to an interaction between the exposure arthritis and mediator disability.
1. I use a logit link to run the discrete time survival analysis. Is it correct to assume that in this context, I can apply the package paramed in stata to estimate the total direct and indirect effects following the example below slightly modifying the help instructions with the installation file Binary outcome, binary mediator, a binary treatment coded 0 and 1, covariates, interaction between treatment and mediator, bootstrap standard errors with 500 replications
paramed y_bin, avar(treat) mvar(m_bin) cvars(var1 var2...) a0(0) a1(1) m(1) yreg(logistic) mreg(logistic) boot reps(500)
2. Is there a way to incorporate survey weights [pweight=...] and bootstrap standard errors using pre-generated bootstrap replication values provided by Statistics Canada? If it is not possible to incorporate survey parameters, would it be reasonable to proceed using the standard estimation procedure specifying an additional assumption that the sampling design is non-informative for the present case and will therefore be ignored.
Thanking you in advance
Orit
The context of my analysis is a longitudinal complex survey. Variable information is updated every 2 years from 1994/95 through 2010/11 providing up to 8 time points for each survey participant for analysis. Sample size is roughly 12000, weighted to represent the Canadian population at the beginning of the survey in 1994. I am interested in the extent to which arthritis (analyzed as a time varying exposure) is associated with time to first occurrence of heart disease adjusting for time-invariant and time-varying covariates, and the extent to which the effect of arthritis on incident heart disease is mediated through effects of disability (also analyzed as a time-varying covariate). Given the panel nature of the data I use discrete time-survival analysis with data in a person-period format. Preliminary analysis point to an interaction between the exposure arthritis and mediator disability.
1. I use a logit link to run the discrete time survival analysis. Is it correct to assume that in this context, I can apply the package paramed in stata to estimate the total direct and indirect effects following the example below slightly modifying the help instructions with the installation file Binary outcome, binary mediator, a binary treatment coded 0 and 1, covariates, interaction between treatment and mediator, bootstrap standard errors with 500 replications
paramed y_bin, avar(treat) mvar(m_bin) cvars(var1 var2...) a0(0) a1(1) m(1) yreg(logistic) mreg(logistic) boot reps(500)
2. Is there a way to incorporate survey weights [pweight=...] and bootstrap standard errors using pre-generated bootstrap replication values provided by Statistics Canada? If it is not possible to incorporate survey parameters, would it be reasonable to proceed using the standard estimation procedure specifying an additional assumption that the sampling design is non-informative for the present case and will therefore be ignored.
Thanking you in advance
Orit
Comment