Yes, this is one of the advantages of using mixed-effects models compared to classical rmANOVA. But there are all sorts of subtle issues with controlling co-variates: http://www.johnmyleswhite.com/notebook/2016/02/25/a-variant-on-statistically-controlling-for-confounding-constructs-is-harder-than-you-think/ (and make sure to read the Westfall and Yarkoni paper linked there) http://www.johnmyleswhite.com/notebook/2017/04/06/covariate-based-diagnostics-for-randomized-experiments-are-often-misleading/ although I do recommend including these covariates in general and especially in psycholinguistic studies: https://arxiv.org/abs/1602.04565 Note, however, that lme is in the nlme package, while lmer is the corresponding function from lme4. See https://stats.stackexchange.com/q/5344/26743 for a comparison. Phillip
On 05/12/2017 04:05 PM, Alexandre Obert wrote:
Dear all, I'm working on ERPs data from a language comprehension study and I'm confronting to some problems with items' features. Briefly, participants saw words on a screen and have to decide if they're meaningful or not. Some of them were meaningful (condition 1), others not (condition 2) and others were ambiguous (condition 3). Using classical ANOVA, I observed significant differences. However, words came with characteristics such as frequency, number of letters and so on...that I would like to control for. In other words, I would like to test the effect of the condition and control the words' features in a same analysis. I think that lme from lme4() could compute such analysis, is-this right? Regards,