On 2 Apr 2003 20:06:58 -0800, [EMAIL PROTECTED] (Zina Taran) wrote: > Dear members, > > I have a problem: > > in a leadership class in another school, respondents to a survey have > answered open-ended questions asking them to name 5 leaders (actually, three > sets of 5). The respondents then formed groups of 4-5 persons in each, and > developed new lists, this time as a group. So now for each person, there is > a list of their answers and a list of group answers. How would you approach > getting a measure of some dependancy of the group answers and the individual > answers? I thought to look for %% of agreement, that is how many respondents > in the group name a certain leader (named as a group) individually. But the > problem then, the lists can vary wildly, and might be altogether different > not even having any common names, or remarkably similar. > > There are several pieces of knowledge one could gain from comparing group > and individual lists, for example, if one member of the group > disproportionately swayed the group opinion; if some majority rules were > used, etc. > > I really appreciate any pointers.
Here's something to start with. I will ignore the fact that each person created several lists, since you don't say (at all) how that fits in. How much does one person's list overlap with the individual lists of others? You can get a count of "appearances" for each nominee on each list. A person who nominates 5 (or 4 or 3) who are highly popular is a good nominator. That must confound the conclusions about "good persuaders." - Somewhere in the counts, there's a measure of how good each list was; perhaps, the sum of the overlaps for the best two? sum of all? - What will make a good measure? You don't say how much overlap there tends to be, or how many people are contributing. You do something different if most of the nominations are unique, compared to what you do if few are unique. Okay, do the group-nominees match the most-frequent? If that is perfect, conclude, "Democratic", and there is little else to pursue. The "discrepancies" comprise the information that might show individual persuasiveness. Are there enough discrepancies to give you any statistical power for drawing any conclusions? -- Rich Ulrich, [EMAIL PROTECTED] http://www.pitt.edu/~wpilib/index.html . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
