 |
 |
Hi Liam, That's right. The answers have changed. That's why often it is worth answering again a similar question, instead simply pointing to old answers from a couple of years ago. Specifically on this topic, three changes over time are worth commenting on: 1) There used to be a consensus that between-subject effects was not possible using the simple GLM in repeated measures design. This changed when it became clear that a model with all the within-subject effects in the design could be constructed, leaving all the between subject effects of no interest (and not otherwise explained by nuisance variables) in the residuals, which then have the correct degrees of freedom, leading to accurate inferences. This changed responses given for these cases. 2) Also for repeated measures designs, given that not all observations are exchangeable, this used to be a no-no for permutation tests. However, with the implementation of exchangeability blocks in randomise, this became feasible. 3) The very idea of exchangeability blocks was later extended to accommodate multiple levels, such that designs that involve interactions of within and between-subject effects can now be analysed. This presently is only available in PALM, but may become available in randomise too eventually. There are more things coming and some answers will change again in the future... All the best! Anderson On Fri, 12 Oct 2018 at 05:42, Nestor, Liam J <[log in to unmask]> wrote: > Tony > > It seems that these models are constantly changing - there are different > ones proposed throughout the forum over the years - they can't all be > correct? This model was described this year while mine was in 2015. How do > you know any of them are correct - really? You are now having to try and > guess which parts of the model are computing the different effects. Perhaps > try implementing a 2x2 on your pre/post food addicted/non-food addicted > data first? If you find clusters, then extract signal change and run the > same model in SPSS to see if you see the same effects? > > Liam. > > > > Dr. Liam Nestor, > Neuropsychopharmacology Unit, > Imperial College London > > > -------- Original message -------- > From: Tony Goldstone <[log in to unmask]> > Date: 11/10/2018 20:34 (GMT+00:00) > To: [log in to unmask] > Subject: Re: [FSL] GLM for 3 (between subjects) x 2 (within subjects) > repeated measures ANOVA > > or is the overall effect of drug (independent of group) better modelled by > an F2 for C1-3 (rather than a C7 with 0.333 0.333 0.333 for EV1-3)? > Tony > > ######################################################################## > > To unsubscribe from the FSL list, click the following link: > https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1 > > ------------------------------ > > To unsubscribe from the FSL list, click the following link: > https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1 > ######################################################################## To unsubscribe from the FSL list, click the following link: https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1