Hi Following simulation in spss shows the effect of dichotomizing variables on factor analysis. With the original continuous variables, factor picks up the correct 3 factor solution automatically most of the time. Dichotomizing, produces more factors most of the time, with last some not that interpretable. Asking for 3 factors with the dichotomous data, recovers the correct structure. So one thing to try might be to specify how many factors you think there are in the data [/criteria = factor(#)] and see if the results are interpretable?
Jim input program. loop o = 1 to 200. end case. end loop. end file. end input program. list. compute z1 = rv.norm(0,1). compute z2 = rv.norm(0,1). compute z3 = rv.norm(0,1). do repeat a = a1 to a8. compute a = .707*z1 + .707*rv.norm(0,1). end repeat. do repeat b = b1 to b8. compute b = .707*z2 + .707*rv.norm(0,1). end repeat. do repeat c = c1 to c8. compute c = .707*z3 + .707*rv.norm(0,1). end repeat. factor /vari = a1 to c8. rank a1 to c8 /ntiles(2). factor /vari = na1 to nc8. factor /vari = na1 to nc8 /crit fact(3). Jim Clark Professor & Chair of Psychology U Winnipeg Room 4L41A 204-786-9757 204-774-4134 Fax ________________________________________ From: Annette Taylor [[email protected]] Sent: June-18-13 2:32 PM To: Teaching in the Psychological Sciences (TIPS) Subject: [tips] factor analysis I am coming to the statistical well one more time. Sigh. Other than what I can figure out from SPSS with my colleague, we are at a loss on what we can do with factor analysis--we understand the basic premises. The problem is how to carry it out with SPSS. Or perhaps we have done it correctly and there truly is nothing here :( We had hoped to find some factors and had a couple of possible ways we thought the items might cluster together. We have a data set with over 200 participants and a questionnaire with 23 items. The items were coded as 0 = incorrect response and 1 = correct response in a 2-choice forced-choice format. We entered the 0,1 data set for these participants into a factor analysis using principal components analysis with a varimax rotation method with Kaiser normalization that gives what we understand to be an "orthogonal" analysis. We have 10 factors for the 23 items, the largest has 5 items, then there a bunch of 3, 2, 1 item factors :( We repeated this with a principal components analysis with a quatrimax rotation with Kaiser normalization which gave us what we think is a "correlated" analysis. Except for the precise component values the factors were 100% exactly the same. Unfortunately, the factors seem weird to us and not at all what we might have predicted in our scenario. Does someone with more factor analysis knowledge have some suggestions for us? Thanks in advance! Annette Annette Kujawski Taylor, Ph. D. Professor, Psychological Sciences University of San Diego 5998 Alcala Park San Diego, CA 92110 [email protected] --- You are currently subscribed to tips as: [email protected]. To unsubscribe click here: http://fsulist.frostburg.edu/u?id=13251.645f86b5cec4da0a56ffea7a891720c9&n=T&l=tips&o=26137 or send a blank email to leave-26137-13251.645f86b5cec4da0a56ffea7a89172...@fsulist.frostburg.edu --- You are currently subscribed to tips as: [email protected]. To unsubscribe click here: http://fsulist.frostburg.edu/u?id=13090.68da6e6e5325aa33287ff385b70df5d5&n=T&l=tips&o=26141 or send a blank email to leave-26141-13090.68da6e6e5325aa33287ff385b70df...@fsulist.frostburg.edu
