0
$\begingroup$

The format of my studies is testing 10 individuals in 5 conditions 100 times per condition (this is a behavioral test) - resulting in around 5000 raw data points.

I am currently using 3 statistical method "families" for my project. These are:

  • t-tests
  • ANOVAa
  • Linear Models

I am wondering for which of these methods it would be more beneficial to analyze ALL my data in raw form (this was my initial impulse for all methods, but apparently that inflates my significance at least for some) and for which I should only take the means of every individual as a data point.

Can you help me out? Presently I am using t-tests and ANOVA for per-participant means and linear models on my raw data, though that is only because I have a couple of botched trials and paired t-tests and ANOVA don't deal well with missing data. But maybe it would be the right thing to do even for full data sets?

$\endgroup$
2
  • $\begingroup$ It looks like what you need are linear mixed-effects models (aka random-effects models, multilevel models), which account for the repeated measures within subjects. These models also handle missing data on the dependent variable without any kind of adjustment or imputation (or stated differently, they handle unbalanced designs). $\endgroup$ Commented Nov 16, 2013 at 23:59
  • $\begingroup$ I am currently using lme4:lmer() for my linear modelling, as seen here. Is that appropriate? Also, what about whenever I use t-tests and ANOVA - should I take means or input the raw data there? (supposeing I have a complete dataset) $\endgroup$
    – TheChymera
    Commented Nov 17, 2013 at 0:18

0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.