Dear all,
We are trying to analyze eye-tracking data using mixed multinomial logit regression. The data comes from the experiment where participants were looking at images of faces - rated in separate evaluation study - on a scale from looking as nationals to looking as foreigners. We are interested in the chance that the first fixation lands on the eyes, nose or mouth depending on how the image was rated. Therefore for every participant (osoba variable) and for every image presented (image variable), we have a rating variable as well ROI alternatives defined (Region of Interest variable with eyes, nose and month) and a Hit (depended variable) which is 1 if first fixation landed in given ROI and zero otherwise. The data structure was modelled after this Stata example: https://www.stata.com/features/overv...ta-mixed-logit,
but instead of t (time) we have an image (also repeated measures variable) and instead of alt we have ROI. In effect my panel is defined as follows:
I was trying to use cmxtmixlogit to model my data, but it turns out that it won't work as I do not have alternative specific independent variable. Therefore if I want to see if faces evaluated as nationals get more fixations in any given ROI the analysis reports that the rating variable is omitted as it has no variance across ROIs, which is true as the rating pertains to the image, not ROI.
Can you please advise me on the proper statistical model to use, provided my data structure and a research question? We already did mixed logit analysis treating each ROI as a separate IV, but this is suboptimal as those choices are not independent (if the first fixation landed into the eyes ROI it automatically did not land in the nose or mouth ROI), hence the idea of using multinomial approach.
Best regards
Mike
We are trying to analyze eye-tracking data using mixed multinomial logit regression. The data comes from the experiment where participants were looking at images of faces - rated in separate evaluation study - on a scale from looking as nationals to looking as foreigners. We are interested in the chance that the first fixation lands on the eyes, nose or mouth depending on how the image was rated. Therefore for every participant (osoba variable) and for every image presented (image variable), we have a rating variable as well ROI alternatives defined (Region of Interest variable with eyes, nose and month) and a Hit (depended variable) which is 1 if first fixation landed in given ROI and zero otherwise. The data structure was modelled after this Stata example: https://www.stata.com/features/overv...ta-mixed-logit,
but instead of t (time) we have an image (also repeated measures variable) and instead of alt we have ROI. In effect my panel is defined as follows:
I was trying to use cmxtmixlogit to model my data, but it turns out that it won't work as I do not have alternative specific independent variable. Therefore if I want to see if faces evaluated as nationals get more fixations in any given ROI the analysis reports that the rating variable is omitted as it has no variance across ROIs, which is true as the rating pertains to the image, not ROI.
Can you please advise me on the proper statistical model to use, provided my data structure and a research question? We already did mixed logit analysis treating each ROI as a separate IV, but this is suboptimal as those choices are not independent (if the first fixation landed into the eyes ROI it automatically did not land in the nose or mouth ROI), hence the idea of using multinomial approach.
Best regards
Mike
Attached Files
