On 5 Jul 2003 12:17:02 -0700, [EMAIL PROTECTED] (Doreen) wrote: > I've created an attitudinal survey using a 5-point Likert scale and a > "Not Applicable" category. Originally I selected "Not Applicable" as > a missing value but now believe that was not correct...it > substantially reduces the cases for inter-item correlations. > 1. What is the normal practice for coding such a "Not Applicable" > scale option?
In the first place, you avoid them like the very devil. Then, if you have to persist, you test particular items to see whether the people with NA are peculiar. Are there patterns of missing data (missing jointly? correlated with something else?). > > I will then conduct a reliability and factor analysis test. Practice > I've read in the literature is to first conduct reliability on the > entire survey (say 50 to 60 items) then factor analysis to determine > number of factors in survey. After reading about Cronbach's alpha I'm > not sure that practice of reliability then factor analysis is correct. Well you are supposed to use the first reliability to tell you which items to re-write so you can collect them better. If you have rotten items on hand, then you throw them away. > Seems the order should be factor analysis then reliability within > designated factors. Reliability seems to refer to reliability of a > group of common items (a factor), not groups of common items (as in > reliability run on the entire survey of say 50 to 60 items). Likert scales were originally *designed* to reflect a single attitude, with most frequent endorsement of the middle, neutral scores. If you are not following that example, what you SHOULD do next... depends. On all those other details. > > 2. What is common practice with reliability and factor analysis? -- Rich Ulrich, [EMAIL PROTECTED] http://www.pitt.edu/~wpilib/index.html "Taxes are the price we pay for civilization." Justice Holmes. . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
