We had participants perform a perceptual categorization task on phase plots of postural data, this produces a 'perceptual map' of the stimuli where each stimulus is represented by an x,y coordinate. We also obtained several quantitative indices for the same data, visual inspection suggests that the perceptual sort and quantitative measures vary systematically - would like to back this up statistically - does anyone have suggestions? (for context see the poster attached)
Best.
Hi,
No sure I understand your question here. But, don't you think a cluster analysis here would help you out?
Hi, in short I'm looking for an analysis that will allow me to correlate two sets of measures, one set derived from participant responses, and the other from quantitative analyses (non-linear/fractal). I will take a look at cluster analysis, thanks for the suggestion!
Hi Leonard,
I checked your poster twice this morning. I am not fully understand your idea by reading your poster. But, I think your question might involve multiple factors (PL, EA and etc.,). So, a factor analysis or as I mentioned before a cluster analysis might help. Good luck.
The PL, EA,... measures in the poster are the quantitative measures, what I want to do is see if the perceptual judgments map onto these measures - in the figures, each point is a phase plot (position, velocity) that can be represented by it's x and y axis coordinates - each point's position in the figure was determined by 'summing' participants' judgments, so the figures represent how people organized the phase plots. The key is that I believe that the x,y coordinates for each stimuli have to be analyzed as a unit, as it is the two points that specify each phase plot. - I think the factor analysis would treat them as independent measures, which may be inaccurate. I really appreciate you taking the time to ponder this with me!
Hi,
I am not sure whether I properly understand your problem, but if you want to correlate configurations (or so said a list of coordinates) you should use the bidimensional regression analysis approach (not to be mixed up with biVARIATE regression analysis!).
What is the big advantage over other approaches? With the bidim-approach the coordinates are still entangled and are not used independently from each other (which would not make sense).
I have written and published an open-source project on this topic. You can retrieve it from
Carbon, C. C. (2013). BiDimRegression: Bidimensional regression modeling using R. Journal of Statistical Software, Code Snippets, 52(1), 1-11.
Within the routine you can use Euclidean as w ell as Affine geometry.
You can also directly download it here:
http://www.jstatsoft.org/v52/c01/paper
All the best, hope this helps,
CCC
- Badges
- Science topic