Hi I am trying to investigate the effect of topic (conversational context - 3 variants) on the linguistic production (code - 4 variants) of 16 speakers. I want to run a mixed effects model with (speaker - 16 variants) and (word 500+ variants) as random effects to see whether they improve the model's ability to predict the linguistic production (code). I also want to include the fixed effects of speaker occupation (2 variants: miner/non-miner) and speaker location (4 variants). I want to use log-likelihood tests on a model including topic (conversational context) vs a model discounting topic to gauge significance. Can anyone tell me where to start? I have tried using functions lme, lmer and lrtest, and the multinomial logit model but I'm not sure whether they are right for my data. Any help would be greatly appreciated. Thanks -- Thomas Devlin PhD candidate, Department of Language and Linguistic Science, University of York, York, YO10 5DD, UK tpd502@york.ac.uk [[alternative HTML version deleted]]