Basically, I want to know how accurate my participants are, but it would be good to take participants and items into consideration as random effects rather than having to average them out or risk pseudo replication.

I don’t have a fixed effect so I can’t use a LMM – but is there something else I can use? Or, alternatively, would it be best to just use average participant accuracy score and conduct a one-sample t-test?


If all you have are repeated measurements within participants, then you can fit a linear mixed model. This would look something like:

Y ~ (1 | participant)

where the model will estimate a variance at the participant level along with a residual variance.

Such a model is known as a variance components model, because it splits up the total variance into that attributal to participants, and what is left.

  • $\begingroup$ Thank you so much Robert! If the items are also repeated could I then do Y ~ (1 | participant) + (1 | item) ? $\endgroup$ Jun 7 at 6:09
  • $\begingroup$ Sorry, another question: can I still use lmer, because I get an error message "Error in eval(predvars, data, env) : object 'accuracy' not found"? $\endgroup$ Jun 7 at 6:19
  • $\begingroup$ And for lme I get the error message "Error in eval(predvars, data, env) : object 'participant' not found"? $\endgroup$ Jun 7 at 6:28
  • $\begingroup$ Yes. That shluld be fine. As for those errors, are they from predict ? You need to pass a data.frame containing both variables. $\endgroup$ Jun 7 at 7:54

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.