Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: or option for oldie does not work after mi stimate


From   "Sheena Sullivan" <[email protected]>
To   <[email protected]>
Subject   Re: st: or option for oldie does not work after mi stimate
Date   Sat, 26 Mar 2011 11:29:49 -0700

Thanks, Yulia, for the syntax tips and for the reassurance that eform is
appropriate for a log-normal AFT model.
Sheena 

________________________________________
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] 
Re: st: or option for oldie does not work after mi stimate
________________________________________
From
  [email protected] (Yulia Marchenko, StataCorp LP)
To
  [email protected]
Subject
  Re: st: or option for oldie does not work after mi stimate
Date
  Fri, 25 Mar 2011 17:32:51 -0500
________________________________________
"Sheena Sullivan" <[email protected]> asks about the use of
-streg-'s
-tr- option with -mi estimate: streg-:

> Typically, to get time ratios I would type the following: 
> streg Y X Z, dist(lnormal) frailty(gamma) sh(G) tr
> Where the -tr- option produces the time ratio. 
> 
> If I use mi estimate, I can type the same thing:
> mi estimate: streg Y X Z, dist(lnormal) frailty(gamma) sh(G) tr
> But -tr- is ignored and only the coefficients are displayed.  
>
> I followed the solution for Roland to type the option in the mi estimate
> command, i.e:  mi estimate, tr: streg Y X Z, dist(lnormal) frailty(gamma)
> sh(G) But the tr option is not allowed.
>
> If I instead use the eform option I do get my exponeniated coefficients:
>
> mi estimate, eform: streg Y X Z, dist(lnormal) frailty(gamma) sh(G)
>
> But I wanted to ask the Stata community whether there are any issues I
> should know about when using this option for accelerated failure time
> models?  As far as I understand, the time ratios are just exponentiated
> coefficients, so I assume it should be okay.
>
> Is the -tr- option not allowed because it can be achieved with the eform
> option?  Or is it unwise to exponentiate these coefficients and their
> corresponding confidence intervals?

The reason option -tr- is not supported by -mi estimate- is because, in
general, it is not purely a reporting option.  For example, with exponential
and Weibull models, this option not only exponentiates coefficients but more
importantly changes the default estimation metric of -streg-, proportional 
hazards (PH), the the accelerated failure time (AFT).  In other words,
with these models option -tr- implies option -time-, an estimation
option which specifies that exponential or Weibull model is fit in the
AFT metric and the results are displayed as coefficients.

It is fine to use -eform- with -mi estimate- to obtain time ratios when the
-streg- model is fit in the AFT metric.  Time ratios are simply
exponentiated coefficients.  In fact, you can use option -eform("Tm.
Ratio")- 
to also match the column heading of time ratios as reported by -streg, tr-.

To obtain time ratios with -mi estimate: streg- for the exponential or
Weibull
model use:

  . mi estimate, eform("Tm. Ratio"): streg ... , time

or

  . mi estimate, eform("Tm. Ratio"): streg ... , tr


> As an addendum, to get both coefficients and ratios, I saved the
estimates,
> which seemed to work
> 
> . mi estimate, saving(mi_aft_ests.ster, replace) : streg Y X Z,
>     dist(lnormal) frailty(gamma) sh(G)
> . mi estimate using mi_aft_ests.ster, eform

Let me note that the -eform- option can also be used with -mi estimate- on
replay.  Sheena could have typed

  . mi estimate, saving(mi_aft_ests.ster, replace) : streg ...
  . mi estimate, eform

to see exponentiated coefficients.

However, Sheena followed a good general strategy:

1.  Use -mi estimate, saving()- to fit a model on multiply-imputed data once
    and store the estimation results.

2.  Use -mi estimate using- to perform any further analyses which do not
    require refitting the model.

This strategy can save you time when the estimation task is time consuming
and
you are planning to perform postestimation analyses such as multiple
hypothesis testing.

-- Yulia
[email protected]


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/


© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index