I have 100 subjects who were scored on their performance in a series of trials (binary outcome for each trial: "hit" or "miss"). I want to find out whether overall there are more "hit" than "miss" trials. Since the trials are repeated for each individual (each subject has 4 trials), then the data may not be independent. If so, do I still use a binomial test? Or is there are other analysis I can use?

Thank you very much. Any help is deeply appreciated.

  • $\begingroup$ Are you comparing groups or controlling for covariates? $\endgroup$ Jan 29 '16 at 17:05
  • $\begingroup$ Nope I am not. It's a pilot study so we just want to find out if frequency of hits and misses differ. $\endgroup$
    – Gooper
    Jan 29 '16 at 17:14
  • $\begingroup$ You refer to "hits" & "misses". Was this a signal detection study? $\endgroup$ Jan 29 '16 at 17:31

It sounds like you could add a random effect for subject in a GLM framework. For subject s, trial t let

$Y_{s,t} \sim Bernoulli(p_s), \quad p_s = logit(\theta + \beta_s), \quad \beta_s \sim N(0,\sigma_s^2)$.

Fitting this model would be pretty simple (using the R package lme4, for example).


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.