1) What should I choose between Univariate ANOVA for repeated measures,
MANOVA, random mixed models and general estimating equations, and why ?
2) What are the differences between random mixed models and GEE, if any ?
Mixed models are generally more appropriate for analyzing longitudinal panel
data than ANOVA/MANOVA approaches. This is particularly true if there are
more then two waves of data collection. ANOVA/MANOVA assumes that all
observations are independent (which is not the case in a longitudinal study)
and results in underestimating error variance. Mixed models are also more
flexible in their handling of missing data (ANOVA/MANOVA only includes
complete case data in the analysis). GEE is also suitable for longitudinal
data, but as you indicated, tends to provide more conservative results
compared to mixed models. Also, mixed models is a "subject specific"
approach to analysis (i.e., each case can have its own random intercept
and/or random regression coefficients) whereas GEE is a "population
averaged" approach where the emphasis is on group effects.
3) I will have to perform a sample-size analysis as well: what I have is
baseline mean and SD of the outcome (by previous study) and I should be able
to get mean and SD at the second assessment (by previous pilot study), might
you drive me on how to perform a power analysis in this setting ?
There are no established methods (i.e., software) that I'm aware of for
estimating power for a mixed model (though there have been some research on
estimating power for GEE models, see references below). The best way, I
think, is to conduct a series of simulations in which you repeatedly
generate hypothesized sample data of varying N and perform the desired
statistic over several (e.g., 1,000) iterations to estimate power.
Rochon, J. (1998). Application of GEE procedures for sample size
calculations in repeated measures experiments. Statistics in Medicine,
17(14), 1643-1658.
Jung, S. H., & Ahn, C. (2003). Sample size estimation for GEE method for
comparing slopes in repeated measurements data. Statistics in Medicine,
22(8), 1305-1315.
Liu, G., & Liang, K. Y. (1997). Sample Size Calculations for Studies with
Correlated Observations. Biometrics, 53(3), 937-947.
Barth
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/