I have 14 different type of issues reported by three different mediums (research, newspapers & community). All of them have stressed on few common and few different issues based on frequency or the number of issues recorded in them. Now I wish to compare them to derive most significant issues. But since the total number of data collected for each category is different (either high or low). How do I normalize, compare and then prioritize these data sets proving their statistical significance of major issues along the way?
Hello Riyan, It's not completely clear to me what sort of comparison (or, more accurately, what specific hypothesis) you're trying to evaluate. If I understand correctly, you have: (a) a common set of 14 issues; that are (b) referred to different numbers of times based on the 3 mediums used to elicit these referrals/mentions. The simplest approach I could see would be to rank the issues from most to least frequently mentioned separately within each medium. That adjusts for the potentially large disparity in total number of referrals/mentions across media/sources. You could then compute a concordance statistic that would quantify how consistently specific issues were ranked across media (if that is of interest). Kendall's coefficient of concordance would be an appropriate statistic. The question of whether rankings differed significantly across media could be addressed by running a Friedman two-way anova by ranks (rows = issues, columns = media), an analog to a repeated measures anova. For this to work, however, you'd have to skip the "normalizing" step; which may result in the values being of little utility, especially if one medium yields far more (or fewer) referrals than do others. Good luck with your work.
- Badges
- Science topic