And now it comes to this: thesis data analysis. I am doing both qualitative analysis of the interviews and quantitative analysis for the eye-tracking, mostly. However, I will also quantify some of the interview coding and “qualify” the eye-tracking data, mainly while I analyze the paths and orders in which people view the images.
So now the questions become, what exactly am I looking for, and how do I find evidence of it? I have some hypotheses, but they are pretty general at this point. I know that I’m looking for differences between the experts and the non-experts, and among the levels of scaffolding for the non-experts in particular. For the interviews, that means I expect experts will 1) have more correct answers than the non-experts, 2) have different answers from the non-experts about how they know the answers they give, 3) be able to answer all my questions about the images, and 4) have basically similar meaning-making across all levels of scaffolding. This means I have a general idea of where to start coding, but I imagine my code book will change significantly as I go.
With the eye-tracking data, I’ll also be trying to build the model as I go, especially as this analysis is new to our lab. With the help of a former graduate student in the Statistics department, I’ll be starting at the most general differences, again whether the number of fixations (as defined by a minimum dwell time in a maximum diameter area) differ significantly: 1) between experts and non-experts overall with all topics included and all images, 2) between supposedly-maximally-different unscaffolded vs. fully-scaffolded images but with both populations included, and 3) experts looking at unscaffolded vs. non-experts looking at fully-scaffolded images. At this point, I think that there should be significant differences in cases 1 and 2, but hope that, if significant, at least the value of the difference should be smaller in 3, indicating that the non-experts are indeed moving closer to the patterns of experts when given scaffolding. However, this may not reveal itself in the eye-tracking as the populations could make similar meaning as reflected in the interviews but not have the same patterns of eye-movements; that is, it’s possible that the non-experts might be less efficient than experts but still eventually arrive at a better answer with scaffolding than without.
As for the parameters of the eye-tracking, the standard minimum dwell time for a fixation included in our software is 80 ms, and the maximum diameter is 100 pixels, but again, we have no standard for this in the lab so we’ll play around with this and see if results hold up over smaller dwell times or at least smaller diameters, or if they appear. My images are only 800×600 pixels, so a minimal diameter of 1/6th to 1/8th of the image seems rather large. Some of this will be mitigated by the use of areas of interest drawn in the image, where the distance between areas could dictate a smaller minimum diameter, but at this point, all of this remains to be seen and to some extent, the analysis will be very exploratory.
That’s the plan at the moment; what are your thoughts, questions, and/or suggestions?
