Skip to content

one-sample t-test with correlated (clustered) observations

2 messages · Paul Artes, Ian Fiske

Paul Artes
# 
I would like to estimate the difference between two measurement techniques.
With both techniques, 4 measurements were obtained in each of 15
individuals. (These are not *repeated* measurements though - each of the 4
is of a different attribute).  The naive approach would be a paired t-test,
but of course this assumes that the 4 measures contributed by each
individual are not dependent (which they are), and would inflate the CI of
the differences.

I found t.test.cluster {Hmisc}, but this works for the 2-sample problem only
as far as I understand...

Could someone please point me in the right direction?

Many thanks!

Paul
Ian Fiske
# 
To handle the correlations, you can treat individuals as random blocks.  So
you have a mixed model with measurement technique crossed with measured
attribute and random intercepts for each individual.  You can fit this with
lmer() in the lme4 package.  Keep in mind there are a number of variations
on this... like whether or not to include a measurement*attribute
interaction, etc.

good luck,
ian
Paul Artes wrote: