{smcl}
{* 13dec2007}{...}
{hline}
help for {hi:hetred}
{hline}
{title:Heterogeneity reducing algorithms}
{p 4 14 10}{cmd:hetred} {it:#logEffect} {it:#SElogEffect} {cmd:,} {cmdab:i2h:(}{it:#}{cmd:)} {cmdab:i2l:(}{it:#}{cmd:)} {cmdab:ID:(}{it:varname}{cmd:)} [random tuples max]
{p 4 14 10} where {p_end}
{p 4 14 10}{it:#logEffect} is the ln of the effect size estimate {p_end}
{p 4 14 10}{it:#SElogEffect} is the standard error of ln of the effect size estimate {p_end}
{title:Description}
{p 4 10 10}{cmd:hetred} reduces the statistical heterogeneity of meta-analysis as measured by I-squared. It uses 2 differnt algorithms. A sequential one (default) and a combinatorial one ({it:tuples} option).
For further explanation please see {it:Reference} section below. Using the combinatorial algorithm can demand extreme CPU resources in case where more than 6-7 studies have to be dropped.
Always run the sequential algorithm (default option) to see how many studies have been ommitted. The difference between the two is that the combinatorial one might find a better combination ommitting fewer studies.
{title:Options}
{p 4 14 10}{hi:i2h} is the level of initial I-squared we want to evaluate, e.g. 50 {p_end}
{p 4 14 10}{hi:i2l} is the level under which we want to drop the I-squared , e.g. 25 {p_end}
{p 4 14 10}{hi:ID} is a variable that uniquely identifies the studies {p_end}
{p 4 14 10}{hi:random} uses the random effects model to synthesize the data. Fixed effects model is used by default. {p_end}
{p 4 14 10}{hi:tuples} uses the combinatorial algorithm as well. Drops 2, 3, etc studies together at each step till i2l level is reached. {p_end}
{p 4 14 10}{hi:max} returns the result that drops I-squared below i2l but has the largest possible heterogeneity. {p_end}
{title:Examples}
If I-squared >=75% then try to drop it under 25%
. hetred lnES selnES, i2h(75) i2l(25) ID(study)
The same as above but using random effects model (affects only the summary estimate and respective CIs)
. hetred lnES selnES, i2h(75) i2l(25) ID(study) random
Using combinatorial algorithm as well
. hetred lnES selnES, i2h(75) i2l(25) ID(study) random tuples
Asking for the maximum possible I-squared below the i2l (25%)
. hetred lnES selnES, i2h(75) i2l(25) ID(study) max
{title:Author}
{p 4 4 2}Nikolaos A Patsopoulos, Department of Hygiene and Epidemiology
University of Ioannina School of Medicine {p_end}
{title:Reference}
{p 4 4 2}Patsopoulos NA, Evangelou E, Ioannidis JPA. {it:Sensitivity of between-study heterogeneity in meta-analysis: Proposed metrics and empirical evaluation}, Submitted{p_end}
{title:Support}
{p 4 12}{browse "mailto:npatsop@cc.uoi.gr?subject=info hetred":npatsop@cc.uoi.gr}{p_end}
{p 4 12}{browse "mailto:npatsop@gmail.com?subject=info hetred":npatsop@gmail.com}{p_end}
{title:Also see}
{p 0 19}On-line: help for {help metan}{p_end}