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]

st: From: Steve Samuels <[email protected]>


From   [email protected]
To   [email protected]
Subject   st: From: Steve Samuels <[email protected]>
Date   Mon, 18 Nov 2013 05:25:30 -0500

Lilliana:

Sorry, I pushed the send button prematurely.


I have concerns about your proposed analysis. You would have helped
potential responders if you had, as the FAQ ask, shown "exactly" what you
typed and what Stata did.

• Some issues:

1. I doubt that -sttocc- did the matching you describe, matching a case
only to a control of the same age at the time of interview. -sttocc-
samples controls from risk sets of a cohort. For a case who dies at time
t, -sttocc- will select as a control someone who was alive and at risk
at that "time", namely the time dimension specified in -stset-. Note
that a control can be selected repeatedly and can die at a later date.

2. Further matching on your "mortality" indicator would have matched
deaths with deaths and controls with controls.

3. With imputed data you need -mi stset-, not -stset-.

4. -sttocc- must be applied separately to each imputation (that is the
cause of the -clogit- error message, I believe).

5. The analysis does not consider potential correlation induced by
studying children in the same family.

6. If you have only age in integer years, then there will be only five
possible ages. In that case, a grouped analysis is probably more
appropriate than an analysis that matches individuals.


• Before I can say more, I need to know:

0. Why you chose a case-control design and not a cohort analysis.
If you don't really need one (see Remarks and Examples, first paragraph
in the Manual entry for -sttocc-), a cohort analysis will be much easier.


1. A description of your study population and how you selected your
analysis sample.

2. If you have survey sample data, a description of the sampling design.

3. The exact -stset- statement (-mi stset-) and an explanation of the
variables if that is not completely clear.

4. Whether you time-varying predictors, for example family size at each
age. I'd actually be interested in a description of the determinants you
intend to study.

5. How dates and ages were ascertained. Do you have only years of age at
death and interview? Or, do you have more exact information, such as
months from date of birth.


Steve 
[email protected]



> On Nov 14, 2013, at 9:59 AM, liliana wrote:

> I am trying to run a conditional logistic regression. The analysis consists
> of studying the determinants of child mortality. Before running the
> conditional logistic regression I should randomly select pairs from each
> sample, consisting of one death reported during five years preceding the
> survey and one birth that still survived at the survey time. Moreover, I
> have also imputed the missing values of an explanatory variables, which
> means that before the stata command I should insert the mi estimate: option. 
> 
> What I have done in Stata is 
> 
> set seed 8021
> 
> then I have declared my data to be survival data and then
> 
> sttocc, match(mortality), where mortality is a dummy variable assuming value
> 0 if the child died during five years preceding the survey. 
> 
> Then I go with:
> 
> mi estimate: clogit y x1 x2, group(_set)
> 
> Stata tells me that 
> 
> (regular variables _st _d _t _t0 unregistered because not in m=0)
> groups (strata) are not nested within clusters
> an error occurred when mi estimate executed clogit on m=1
> r(459);
> 
> 
> May you tell me what's wrong with these commands, please? 
> 
> Thank you very much.
> 
> Best, 
> Liliana
> 
> 



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


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