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: 95% CI calculation
From
Ronan Conroy <[email protected]>
To
"<[email protected]>" <[email protected]>
Subject
Re: st: 95% CI calculation
Date
Tue, 17 Sep 2013 09:33:37 +0000
On 2013 MFómh 16, at 21:20, Ubydul Haque wrote:
> Hello,
> In 2008 I had total 675 cases. In 2012 total cases were 168.
> Considering the base year 2008, 75% cases reduced in 2012. Please let
> me know how I can calculate 95% confidence interval on this estimate.
> Thank you.
Here are two suggestions. On the information given, it's as much as I can think of.
input year cases
0 675
1 168
end
poisson year [fw=cases], irr
Poisson regression Number of obs = 843
LR chi2(0) = 0.00
Prob > chi2 = .
Log likelihood = -438.9845 Pseudo R2 = 0.0000
------------------------------------------------------------------------------
year | IRR Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
_cons | .1992883 .0153754 -20.91 0.000 .1713208 .2318213
------------------------------------------------------------------------------
binreg year [fw=cases], rr
------------------------------------------------------------------------------
| EIM
year | Risk Ratio Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
_cons | .1992883 .0137583 -23.36 0.000 .1740673 .2281636
------------------------------------------------------------------------------
In both cases, the null hypothesis is really that there is a 50/50 split between years. The actual proportion in year 2 is given by the incidence rate ratio/relative risk. The confidence intervals are slightly different, and I prefer the Poisson because I think that you are observing a continuing process over a period of time rather than assessing a finite sample of trials.
Note that -binreg- is an alternative to logistic regression that outputs relative risks but which, for this reason, is not guaranteed to yield results every time, since the estimates thus produced may be outside the range (0,1) that defines probabilities.
Ronán Conroy
[email protected]
Associate Professor
Division of Population Health Sciences
Royal College of Surgeons in Ireland
Beaux Lane House
Dublin 2
*
* 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/