sci.stat.consult readers: sorry for the duplication. Many sci.stat.edu don't read multiple groups, and this was separately posted in both.
On 7 May 2004 16:11:26 -0700, [EMAIL PROTECTED] (mu) wrote: On 7 May 2004 16:12:58 -0700, [EMAIL PROTECTED] (mu) wrote: > I am analyzing a survey dataset where respondents are asked to rate > the importance of 20+ statements for factor analysis that will be used > for segmentations. I just found that all ratings are highly correlated > with all else, examining the data I found respondents seem to be > pretty much rating everything either all important or all not > important, therefore the factor solutions I got was very crappy and > can not go on for segmentations, which is the key objective of the > study. > > Is there anything I can do to the data to make up the correlation > problem, The ratings is on a 5 point scale. Please help! If each of the subjects rated everything high, or rated everything low ... I guess that makes two segments. When the 1st principal component accounts for all the usable variance, which is what you seem to be saying, I guess the blame goes to whoever drew up the items. On the other hand, I have seen 'bad solutions' most often when the sample size was too small, and the cure for that was (1) to wait for more data, or (2) to drop some of the items. Do you have more than 100 subjects (too low)? more than 300 (should be plenty, if the items were decent)? -- 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/ . =================================================================
