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

# st: Incorporating error into a prevalence measure

 From Fred Wolfe To statalist@hsphsun2.harvard.edu Subject st: Incorporating error into a prevalence measure Date Sat, 18 Dec 2010 08:16:08 -0600

```After reviewing Stata programs and presentations relating to
misclassification and error, and searching on the web, I have not been
able to solve this problem. I wonder if any listers might be able to
help.

A soon to be published paper defines a binary outcome, cpesr, as being
present when 4 characteristics are present:
mdtjnt28 <= 1 [a continuous variable with a range of 0-28]  +
mdsjnt28 <= 1 [a continuous variable with a range of 0-28]  +
glb <= 1 <= 1 [a continuous variable with a range of 0-10]  +
(sex==0 & esr <30) | (sex == 1 & esr <20) [esr is a continuous
variable with a range of 0-150, sex is 1 for male, 0 for female)]

I calculate this as follows:
gen cpesr =  mdtjnt28 <=1 & mdsjnt28 <=1 & glb <=1 & ((sex==0 & esr
<30) | (sex == 1 & esr <20)) if !mi(mdtjnt28, mdsjnt28, glb, esr)

. ci cpesr if ruse

Variable |        Obs        Mean    Std. Err.       [95% Conf. Interval]
-------------+---------------------------------------------------------------
cpesr |       1478    .0568336    .0060243        .0450165    .0686506

However, mdtjnt28, mdsjnt28 and glb have reliabilities < .9, and above
results fail to account for reduced reliability.

My question is how I can incorporate the reliability information into
the estimate of cpesr to provide more accurate confidence intervals.

Thanks,

Fred

--
Fred Wolfe
National Data Bank for Rheumatic Diseases
Wichita, Kansas
NDB Office  +1 316 263 2125 Ext 0
Research Office +1 316 686 9195
fwolfe@arthritis-research.org

*
*   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/
```