Computer Model & Past data humans solve syllogisms.
Condition IV (Lesion location: control, parietal, temporal),
Conclusion Congruency IV (belief-congruent, belief-incongruent) & these 2 can be split further on whether the correct answers are "valid" or "invalid".
DV (x2): Reaction Time, Accuracy.
DV data I have are an average across all of its problem class (ex. Congruent accuracy is mean accuracy over all congruent problems solved, congruent-valid accuracy is accuracy over all of such problems, etc. such that I have 6 columns x2 DV of data REALLY I'm focused on)
Comparisons of interest: parietal v temporal, control v parietal, control v temporal, control v parietal+temporal.
Question of how to analyze while maintaining any needed factors independence?
All data gathered all at one session
Question of looking at DVs in separate analysis or together
Why? Reaction time (RT) and accuracy (acc) not linearly related.
Fast RT can be from healthy accurate network, or a damaged system causing some process to be ignored or too powerful and making an early decision (which could be right or wrong)
Slow RT can be healthy system taking its time or damaged system struggling to compute. So no expected relations between DV overall which should have some power loss
So... if I was looking at congruent v incongruent Mixed (M)ANOVA seems OK?; between-subject factors lesion location (3 levels), within factors (2 levels congruent, incongruent), and 1 or 2 DV depending how I go about that...
But I keep confusing myself on how to include congruent subgroups (valid/invalid) as if it were 2-way, or if to do those separately in another test for independence of IVs / reduce overlap...or if I can just 2-way mixed (M)ANOVA it all at once somehow.
Almost definitely doing a separate test set involving comparing "multiple-model" VS "single model" problems too but that's another matter.
Computational modelling rare and unclear - a bit rusty with analysis.
But anyway doing all of this twice, one for model data one for past data, and having a spearman correlation for estimate of fit (considered equivalence test too but...) that part is off-topic though.
ANOVA is not appropriate for accuracy data (see e.g. Jaeger 2008 in Journal of Memory and Language). A mixed-effects model would be more appropriate. Even for the reaction time data I think it would be more appropriate, since presumably you tested multiple items as well as multiple participants.
Yes 64 participants total in real data (42 control; 6 parietal; 16 temporal lesion), and model I usually ran ~500 simulated patients per group which does make significance a bit too easy. . .
Multiple items- yes-
Congruent category 10 syllogisms per patient, 5 valid-type 5 invalid-type.
Incongruent 9 syllogisms per patient 3 valid, 6 invalid (unfortunately a bit lopsided here) only congruent-valid group failed levene equality of error variance (p = .004)
For single vs multiple model representation maybe 7 multiple model & 12 single model (typically) problems. so accuracy / RT score is an average across all trials as model can vary from data on particular syllogisms, but is a better representative of results of groupings as a whole.
Mixed effects model.. thanks
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis