I want to ask your suggestions for the models that I tested.
My study consisted of 3 trials and each trial includes four different n-back conditions, 0-,1-,2-,3-back. Therefore, in total, each participant had 12 n-back conditions, in a different order. While they were performing n-back task, I have measured their dorsolateral prefrontal cortex activation via 16-channeled fNIR and obtained oxygenated hemoglobin measures from each of the 16 channels.
I included trials as repeated measures (trial1, trial2, trial3), like n-back conditions (0-,1-,2-,3-back). Because n-back conditions are nested within trials, I wonder whether the models that I construct are okey.
##compact model
library(nlme)
baseline = lme(Optode1 ~ 1,
random = ~ 1 | participant/NbackType/Trial,
data = oxyHbConditionandTrial,
na.action = "na.omit",
method = "ML")
summary (baseline)
##augmented model
library(nlme)
nbackModel = lme(Optode1 ~ NbackType,
random = ~ 1 | participant/Trial/NbackType,
data = oxyHbConditionandTrial,
na.action = "na.omit",
method = "ML")
summary (nbackModel)
##augmented model for two-way repeated measures with trial
library(nlme)
trialModel = lme(Optode1 ~ NbackType + Trial,
random = ~ 1 | participant/Trial/NbackType,,
data = oxyHbConditionandTrial,
na.action = "na.omit",
method = "ML")
summary (trialModel)
##full model for two-way repeated measures
library(nlme)
fullModel = lme(Optode1 ~ NbackType * Trial,
random = ~ 1 | participant/Trial/NbackType,
data = oxyHbConditionandTrial,
na.action = "na.omit",
method = "ML")
summary (fullModel)
Thank you for your suggestions.
I attach the most recent file on Multilevel modeling in R:
https://cran.r-project.org/doc/contrib/Bliese_Multilevel.pdf
Regards
You cannot test the full model as specified because you lack replicates within-subject for each combination of trial and n-back. You could do so if you treated trial as continuous to buy you back some degrees of freedom. You may want to test trial as a linear predictor and as a log(trial) predictor although with only three trials you may not be able to distinguish them. Another model to consider is to treat trial as continuous and varying from 1 to 12 for the 12 trials experienced (3 of each type).
The goal is to decrease your degrees of freedom and thus the flexibility of your model. With 16 DVs you are very prone to overfitting your data.
For growth models (repeated measure at different time points, what "trial" seems to indicate in your example), "trials" should be a random slope.
random = ~ Trial | NbackType / participant
See examples here: http://rpsychologist.com/r-guide-longitudinal-lme-lmer
Thank you all for your suggestions.
Now, I've changed my model as below. I've included gender interaction and removed Trial.
Because NbackType is categorical, I've wondered whether it is okey to test random slopes as in my model 1.
Last but not least, would you share your opinions about how to interpret this random slopes model?
# Random slopes model
- library(lme4)
library(lmerTest)
model1.nbackOpt1 = lmer (Optode1 ~ NbackType * gender +
(NbackType|participant),
na.action = na.exclude,
data=oxyHbConditionCellbyCell,
REML=FALSE)
summary(model1.nbackOpt1)
# Random intercepts model
- library(lme4)
library(lmerTest)
model2.nbackOpt1 = lmer (Optode1 ~ NbackType * gender +
(1|participant:NbackType),
na.action = na.exclude,
data=oxyHbConditionCellbyCell,
REML=FALSE)
summary(model2.nbackOpt1)
Yes, you can have a slope effect for a categorical predictor.
You are allowing participants to have different Optode sensitivities to NbackType and then asking if gender can help predict those differential sensitivities (if the interaction is significant) of if gender can predict individual differences in the level of activity at the Optode location.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS