I am conducting a study on autistic traits in typically developed subjects and I am wondering what would be a better approach to examine neural differences in this population:
1. looking at the scores (the AQ questionnaire for autistic traits) as a continuum
2. looking at the two extreme quartiles
Does anyone have a pro\cons suggestions for each option?
Can you please recommend a relevant article that deals with the topic of working with the extreme quartiles as opposed to the continuum?
Thank you all for helping,
Leehe.
Hello Leehe,
Generally speaking, I think that categorising data by dividing it into quartiles (or perhaps tertiles) and comparing only the outer two categories is frowned upon. Most methodologists who comment on this situation argue that important information is lost in the process, and also that sometimes it is possible to overlook some really important features within your data and therefore reach quite invalid conclusions. Incidentally, what you are planning to do is sometimes referred to as extreme groups analysis or EGA.
Check this article by Preacher et al. out:
http://quantpsy.org/pubs/preacher_rucker_maccallum_nicewander_2005.pdf
In essence, I think you'll find out that there is little support for selecting the two extreme categories and then comparing them. Having said that, a lot of researchers do EGA. But then, a lot of researchers do things that aren't above board. Whether you want to join them or not is your choice!
Robt.
The paper Robert suggests is very good but his EGA is actually a perfectly valid strategy for detecting effects - and it is worth re-reading the paper. EGA increases statistical power to detect interactions and linear effects but can't estimate standardised effect sizes very well (or rather it can but you need to make a statistical adjustment to the calculation). However, as the standardised effect size is rarely of real interest (and can be readily recovered with a bit of arithmetic) this is a marginal loss.
For non-linear effects there are better approaches - but in general EGA is an excellent strategy for detecting interactions and so forth.
Are you using the AQ to screen subjects to choose which to use in the imagining phrase of the study?
btw, for the distribution of AQ for women and men, see end of the following paper https://www.researchgate.net/publication/221882649_Measuring_Empathizing_and_Systemizing_with_a_Large_US_Sample?ev=prf_pub
Article Measuring Empathizing and Systemizing with a Large US Sample
Thank you so much for all your suggestions and remarks. I deeply appreciate your recommendations. It is important for me to note that I analyzed my data both as a continuum and with EGA in order to conclude what each approach has to offer. I will pass this information on to my lab students and other colleagues.
Kind Regards,
Leehe.
Dear Leehe,
I think that your analysing the data both ways is a good idea. I suspect that many researchers analyse their data in a number of ways until they can finally attain statistical significance - which is a dodgy strategy. However, if there are two or more ways in which the data might be validly analysed, conducting those analyses might throw some interesting light on what is going on.
I have seen some beautiful and impressive research in top journals in which the researchers openly revealed how they analysed their data in more than one way - and that they obtained different results depending on the analysis chosen. Being honest and open about analyses is one way in which knowledge can be advanced.
All the best with your work.
Robt.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Applied Statistics
- Survey Design
- Questionnaire