Is GLM appropriate for longer sounds?

[Victoria Lishak]

Hi Dr. Bach,

We've analyzed a task with GLM model and were wondering if our results were valid. We looked at participants reactivity to sounds (positive, negative, neutral). Overall, we found that the response was higher for negative sounds however, the data is not normally distirubted (a lot of outliers on both ends). Does this happen?

Also, we are aware that GLM is best suited for sound duration of 1 second. However, because we used the International Affective Digital Sound System, our sound duration was 5 seconds. The results seem to make sense though. Do you think we need to reanalyze with DCM?

Thank you for your help,

Vicky

[Dominik Bach]

Hi Vicky

we don't have much experience with GLM for longer events. But there are a few sanity checks you could do to confirm that it is suitable:

• plot the average response across all trials/conditions/subjects and make sure it has a somewhat similar time course (in particular, peak latency) as the SCRF. If there is a delay of several seconds, GLM is problematic
• look at the estimated/reconstructed response per condition which should be positive or zero across subjects. If a condition has a significantly negative response, there is usually a problem with the response timing (if a response occurs later then modelled, then falling flank of the modelled response and the rising flank of the true response can coindide - and this often causes negative parameter estimates)

Hope this helps
Dominik