Margaret,
(1) _i is the observation number and _j is the imputation number then
my imp=_j.
(2) Manually average over the imputed set. You will have 5 life-tables
then collapse them into 1 taking average. This is only for statistics,
for standard-errors you have to use Rubin's formula (webpage).
(3) The options were for my example only. su=survival rate and
h=hazard rate.
R.
----- Original Message -----
From: "Margaret Gassanov" <gassanov.1@osu.edu>
To: <statalist@hsphsun2.harvard.edu>
Sent: Saturday, June 24, 2006 8:52 AM
Subject: Re: st: Re: multiple imputation and life tables
Hi Rodrigo,
Thank you for your reply. I am having trouble understanding the formula
you've provided. More specifically, I'm not sure what the variable "imp"
is in my dataset.
After running the ice procedure, I have two new variables in my dataset:
_i -- which gives a value of 5 for each case
and
_j
tab _j
imputation |
number | Freq. Percent Cum.
------------+-----------------------------------
1 | 1,075 20.00 20.00
2 | 1,075 20.00 40.00
3 | 1,075 20.00 60.00
4 | 1,075 20.00 80.00
5 | 1,075 20.00 100.00
------------+-----------------------------------
Total | 5,375 100.00
Are you referring to one of these variables?
Do I manually average each of the values from each of the imputations to
obtain the correct final values?
Also, what are the options "su" and "h"?
Thank you for your help!
Margaret
> Margaret,
>
> Multiple Imputation (MI) allows you to deal with missing values.
> The "solution" is that you will change your missing values for
> several "possibles" values. For example if x has only one missing,
> then MI creates 5 possible values for this missing, that's the
> reason why you have N*5 number of observations.
>
> Then you change your N-1 dataset (assuming only 1 missing
> value) for N*5 dataset. The statistical support is based on
> Bayesian ideas and you can google MI to learn more on that.
>
> In your case you have to "run" the ltable command conditional
> to the set imputed. If imp is the variable that describes the
> # of imputation type
>
> forvalues i=1/5 {
> ltable... if imp==`i', su h
> }
>
> Then your statistics (beg total, deaths, lost and rates) should
> be averaged over the 5 tables. You should compute the
> standard errors for these numbers and the formula is in
> Joe Schafer's webpage: http://www.stat.psu.edu/~jls/mifaq.html
> (see the question: How do I combine the results across the
> multiply imputed sets of data?)
>
> I hope this helps you
> Rodrigo.
>
>
>
> ----- Original Message -----
> From: "Margaret Gassanov" <gassanov.1@osu.edu>
> To: <statalist@hsphsun2.harvard.edu>
> Sent: Friday, June 23, 2006 4:15 PM
> Subject: st: multiple imputation and life tables
>
>
> Hi,
>
> I'm new at multiple imputation (using the "ice" command), and I would
> appreciate any help you can offer.
>
> I've made 5 imputed datasets for an event-history analysis I am doing. I
> want to use the ltable command to get hazard rates and survival rates, but
> ltable is not supported by micombine. Thus, my output has the 1075*5 =
> 5375 cases instead of the 1075 cases in the original dataset.
>
> The rates would still be accurate (correct me if I'm wrong), but the
> "beginning total", "deaths", and "lost" numbers are not. Does anyone have
> suggestions on how to obtain the actual numbers for these three columns of
> data?
>
> Thank you,
> Margaret
> *
> * 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/
>
>
I try to take one day at a time -- but sometimes several days attack me at
once. -Jennifer Unlimited-
*
* 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/
*
* 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/