I want to test two fixed factors while considering assessors (third factor) as random effect, and I'm not sure how to write correctly the R script. What about interactions between assessor and fixed factors?
OK, so I just looked very briefly at your data (enough to construct these), so caveats here apply, here are some calls you can use:
dat = read.table('dataset.txt',header=TRUE)
# lme4:
library(lme4)
lmeFit = lmer(a ~ factor1*factor2 + (1+factor1+factor2|factorRandom),data=dat)
summary(lmeFit)
# BayesFactor:
library(BayesFactor)
bf = generalTestBF(a ~ factor1*factor2 + factor1:factorRandom+factor2:factorRandom + factorRandom,data=dat)
bf
There are several ways to do this. If you'd like a classical approach, use the lme4 package and the function lmer(). Your call will look something like:
lmer(dv ~ ff1*ff2 + (1+ff1+ff2|rf), data = yourdataframe)
If you'd like a Bayesian approach, use the BayesFactor package and the function generalTestBF(). Your call will look something like this:
generalTestBF(dv ~ ff1*ff2 + rf + rf*ff1 + rf*ff2, whichRandom="rf", data=yourdataframe)
(in the above, ff1 and ff2 are your fixed factors, rf is your random factor, and dv is your dependent variable). The difference between the two is that lmer gives you a traditional "per factor/interaction" analysis (testing significance of individual effects) and BayesFactor does model comparison, allowing you to select the best combination of the factors (including ones that have null effects, something a classical approach cannot do).
Without R code to load your data, I can't really tell you more. If you attach your data and an R script, and describe your desired analysis, then I can help you more.
Hi Richard, thank you very much for your answer!
I'm attaching you a txt file with my dataset. You'll find factor 1 and 2 which are fixed, and the random factor. Letters from "a" to "q" are the dependent variables to test. This is the result of a wine sensory test. Factors 1 and 2 are two treatments applied to the grapes and the random factor is for the judges (the persons tasting the wine and giving a score for each variable, i.e. color intensity, bitterness, etc.).
I have no experience with Bayes factor analysis. Usually, I'd have considered the judges as replicate in a sensory session and testing just the two fixed factors, but since I read the paper "Næs T., and Ø. Langsrud. 1998. Fixed or random assessors in sensory profiling? Food Quality and Preference, Vol.9, N.3, 145-152." , I'm curious to know the very difference between the approaches and what results I can observe.
OK, so I just looked very briefly at your data (enough to construct these), so caveats here apply, here are some calls you can use:
dat = read.table('dataset.txt',header=TRUE)
# lme4:
library(lme4)
lmeFit = lmer(a ~ factor1*factor2 + (1+factor1+factor2|factorRandom),data=dat)
summary(lmeFit)
# BayesFactor:
library(BayesFactor)
bf = generalTestBF(a ~ factor1*factor2 + factor1:factorRandom+factor2:factorRandom + factorRandom,data=dat)
bf
Basically I agree with Richard Morey.
I usually add the Anova after the lmeFit.
In a usual test where users are random factors the full example would be the following.
library(lme4) # For the mixed model
library(phia) # For post-hoc analysis
formula
Try this page
http://star.psy.ohio-state.edu/coglab/Demos/r.anova.html
You can find any examples
BIC criterion is a Bayes Factor approach and is usefull for comparing models.
http://www.stat.cmu.edu/~kass/papers/bayesfactors.pdf
On my experience I used the solution given by Trembley
http://cran.r-project.org/web/packages/LMERConvenienceFunctions/LMERConvenienceFunctions.pdf
pamer.fnc function give to you something similar to a classical ANOVA. And multcomp package is usefull for post-hoc analysis.
http://stats.stackexchange.com/questions/14078/post-hoc-test-after-anova-with-repeated-measures-using-r
I borrow you an example
#######################################################
# Be carefull this script may install packages in your R
#######################################################
if(!require(lme4)){
install.packages("lme4")
}
if(!require(LMERConvenienceFunctions)){
install.packages("LMERConvenienceFunctions")
}
if(!require(multcomp)){
install.packages("multcomp")
}
if(!require(languageR)){
install.packages("languageR")
}
#--------------------------------------
# Data selection
#--------------------------------------
data(danish)
danish[danish$LogRT==max(danish[ danish$Subject != "2s22", ]$LogRT),]
danish[danish$LogRT==min(danish[ danish$Subject != "2s16", ]$LogRT),]
datos
When I was learning mixed effects modelling "Zuur, Alain, et al. Mixed effects models and extensions in ecology with R. Springer, 2009" was my best friend. Concise, friendly and full of examples
Richard, shouldn't this line:
bf = generalTestBF(a ~ factor1*factor2 + factor1:factorRandom+factor2:factorRandom + factorRandom,data=dat)
define whichRandom, so that it reads:
bf = generalTestBF(a ~ factor1*factor2 + factor1:factorRandom+factor2:factorRandom + factorRandom,whichRandom="factorRandom",data=dat)
?
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics