Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

RE: st: troubleshooting code for simulating survival data


From   "Nick Cox" <n.j.cox@durham.ac.uk>
To   <statalist@hsphsun2.harvard.edu>
Subject   RE: st: troubleshooting code for simulating survival data
Date   Thu, 29 May 2008 17:57:00 +0100

As I understand your program, you will get no bootstrap results without
at least one covariate, as no scalars are ever defined otherwise. 

Somewhere you need to define in your program

local X <list of covariates> 

Alternatively, you can use -indeplist- 

indeplist, local 
foreach var of local locnames { 
	return scalar `var' = ...
} 

Except that seems a bad idea. You would then get scalars that had
idiosyncratic names depending on your covariates. It might be better to
do something more like this: 

indeplist, local 
local nx : word count `locnames' 
return scalar nx = `nx' 
tokenize "`locnames'" 
forval j = 1/`nx' { 
	return scalar b`j' = exp(_b[``j''])
} 

That way, your -bootstrap- can pick up whichever of r(b1), r(b2), etc.
you want. 

Michael McCulloch

Thank you Nick for pointing out areas of programming I don't yet know
well.
Three replies:
1. The purpose of my program is to simulate survival data. I want to 
run this program with -simulate-. (My broader aim is to use these 
simulated data to validate a Marginal Structural Models program which 
I'm using on real data.)
2. I wanted -indeplist- because once the model with no covariates 
works, I'll add them.
3. As for local macro X, that bit of code:
	foreach var of varlist `X' {
		return scalar `var' = exp(_b[`var'])
	}
is supposed to give me the hazard ratios for independent variable(s) 
of the line:
	stcox a
so, how would I modify this to address the "X" issue?
Many thanks,
Michael



>This trace does not add to what you know already. One alternative is to
>increase the trace depth, but the program is sufficiently short that
>line-by-line scrutiny pays dividends.
>
>Your bootstrap command expects to see a scalar r(a).
>
>This will be produced if, and only if, the local macro X contains a.
>
>But 
>
>First, Svend pointed out that local macro X is never defined by your
>program. Nor is it defined by -indeplist-. -indeplist, local- defines a
>local locnames, as its help indicates. You do not address this point in
>your reply.
>
>Second, you are calling -stcox- without covariates, so I see no point
in
>firing up      -indeplist- to find out what your covariates are. One of
>those things is either wrong or irrelevant. 
>
>I don't understand exactly what your program is trying to do but it
does
>appear buggy. 
>
>You may also be getting confused between a and `a', which are quite
>different beasts.
>
>The variable `id' does nothing in your program. I don't know if that
>matters. My guess is not.
>
>Nick
>n.j.cox@durham.ac.uk
>
>Michael McCulloch
>
>Thanks Svend. I've re-run the program preceded by
>     . set tracedepth 1
>     . set trace on
>and here is what is returned, but I'm not sure how to understand the
>error messages given.
>
>********* begin command *************
>. capture program drop sim_surv
>
>.         program define sim_surv, rclass
>    1.         tempname lambda a id t
>    2.         set obs 10 //create 10 observations
>    3.         gen `id'=_n //Create ID numbers
>    4.         set seed 1235 //set random number seed for
reproducibility
>    5.         gen `a'=0 + int(2*uniform()) //Generate treatment:
>discrete 0/1 uniform random variable
>>   s
>    6.         gen `lambda'=-2
>    7.         gen `t'=(ln(uniform())/(`lambda'*exp(-0.8*`a'))) //
>using Bender, 2005
>    8.         stset `t'
>    9.         stcox `t'
>   10.                 indeplist, local
>   11.                 foreach var of varlist `X' {
>   12.                         return scalar `var' = exp(_b[`var'])
>   13.                 }
>   14.
>. end
>
>. set seed 12358
>
>. bootstrap a=r(a), eform reps(10): sim_surv
>
>********* begin response *************
>
>-----------------------------------------------------------------------
-
>------
>begin bootstrap ---
>    - version 9, missing
>    - local version : di "version " string(_caller()) ", missing:"
>    - if _caller() < 9 {
>      capture syntax [anything] using [, * ]
>      if !c(rc) {
>      if _by() {
>      error 190
>      }
>      `version' bootstrap_8 `0'
>      exit
>      }
>      }
>    - capture _on_colon_parse `0'
>    = capture _on_colon_parse a=r(a), eform reps(10): sim_surv
>    - if c(rc) | `"`s(after)'"' == "" {
>    = if c(rc) | `" sim_surv"' == "" {
>      gettoken old : 0 , qed(qed)
>      if `qed' {
>      `version' bootstrap_8 `0'
>      exit
>      }
>      else if "`old'" != "" {
>      capture which `old'
>      if !c(rc) {
>      `version' bootstrap_8 `0'
>      exit
>      }
>      capture program list `old'
>      if !c(rc) {
>      `version' bootstrap_8 `0'
>      exit
>      }
>      }
>      capture syntax [, * ]
>      if !c(rc) {
>      if _caller() < 9 {
>      `version' bootstrap_8 `0'
>      exit
>      }
>      if "`e(prefix)'" != "bootstrap" {
>      error 301
>      }
>      `version' Display `0'
>      exit
>      }
>      else {
>      `version' bootstrap_8 `0'
>      exit
>      }
>      }
>    - preserve
>    - `version' BootStrap `0'
>    = version 9.2, missing: BootStrap a=r(a), eform reps(10): sim_surv
>(running sim_surv on estimation sample)
>'r(a)' evaluated to missing in full sample
>
>-----------------------------------------------------------------------
-
>--------
>end bootstrap ---
>r(322);
>
>
>
>>Michael McCulloch wrote:
>>
>>I've written the following do-file to simulate survival data, in the
>>case of exponential distribution with one treatment variable (a) and
>>no covariates. The code works when not in a program, and returns the
>>specified Beta-coefficient (-0.8). May I ask for help in fine-tuning
>>the program? I'm getting error message: "  'r(a)' evaluated to
>>missing in full sample".
>>
>>clear
>>capture program drop sim_surv
>>	program define sim_surv, rclass
>>	tempname lambda a id t
>>	set obs 10 //create 10 observations
>>	gen `id'=_n //Create ID numbers
>>	set seed 1235 //set random number seed for reproducibility
>>	gen `a'=0 + int(2*uniform()) //Generate treatment: discrete
>>0/1 uniform random variables
>>	gen `lambda'=-2
>>	gen `t'=(ln(uniform())/(`lambda'*exp(-0.8*`a'))) // using
>>Bender, 2005
>>	stset `t'
>>	stcox `t'
>>		indeplist, local
>>		foreach var of varlist `X' {
>>			return scalar `var' = exp(_b[`var'])
>>		}
>>
>>end
>>set seed 12358
>>bootstrap a=r(a), eform reps(10): sim_surv
>>
>>==============================================================
>>
>>It seems that contents of the local macro `X' is not defined within
>>the program.  That may explain why the code works when not in a
>>program, but not in this program.
>>
>>To examine such problems try:
>>
>>      set tracedepth 1
>>      set trace on
>>
>>Other advice: Tell which command in the program that gave rise to
>>the error message, by copying and pasting both command and response
>>to you mai..

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index