Stata The Stata listserver
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: Re: sample size calculation

From   Joseph Coveney <>
To   Statalist <>
Subject   st: Re: sample size calculation
Date   Thu, 30 Sep 2004 12:28:31 +0900

It seems as Ricardo's authors are assaying nicotine or its metabolites in some 
tissue (placenta, perhaps).  In pharmacokinetics, variation often *is* 
expressed in terms of the coefficient of variation, as Clint suggests.  If 
Ricardo's authors intended to say that the standard deviation is 15% instead of 
15 ng/mg, and if they contemplated using Bonferroni's inequality to adjust 
alpha for two comparisons among the three groups, then 

-sampsi 100 115, alpha(0.025) power(0.8) sd1(15) sd2(15)-

gives 43.  Here, I'm assuming that power, not beta, is 80%.  It's possible that 
Ricardo's authors' article is peppered with unintentional errors and omissions 
in describing things, such as power and the scale for variation.

In planning clinical studies, it's common to supplement the estimated sample 
size in anticipation of participants being lost to follow-up or withdrawing 
informed consent, and in anticipation of those whose data might be considered 
ineligible for analysis due to after-the-fact discovery of protocol violations, 
and so on.  With such compensation for potential drop-out cases, the sample 
size could have been increased to 50, or so, for each group.

Joseph Coveney


Clint Thompson wrote:

I'm no authority, but there is a small section in Van Belle's
"Statistical Rules of Thumb" that discusses sample size calculations in
terms of relative changes, e.g. percent.  Van Belle frames the sample
size calculation in terms of 'proportionate change in means' and
'coefficient of variation' which would require a percent expression of
the variability.  If possible, you may want to consider reading Van
Belle's discussion on this topic.  

>>> 9/29/2004 12:04:04 PM >>>
Dear all,

In an article that I am reading the authors write:

"it was decided a priori to include a minimum quantity
of mother, which could allow the statistical power
(alpha=0.05, beta=0.80) for assessing
differences of 10% in nicotine concentration by group
with an expected standard deviation of 15 ng/mg. These
criteria led to the analysis of 150 mothers (50 for
each of the 3 groups)" 

They claim that they did this and all analysis using
Stata 5.0. They do not provide group means or any
discriptive statistics on which the sample size was

Can someone please tell me how this could have been
done. And is there a program to do this in Stata?

Thank you in advance,

*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index