Paul Carreno <PCARRENO@ACSERV.AD.COWAN.EDU.AU> writes: > Hi SASlers, > > I was hoping that someone can point me in the right direction... > I have been asked to do a Factor Analysis on some data.... > It seems that the Proc Factor procedure maybe the appropriate one to > use... > My problem is that I am not a statistician but a computer person so > I dont know how to prepare the data to put it thru the test.... > > I have 3 very similar medical tests for each of 120 patients: > > TEST 1....20 questions with answers from 0 to 5 > 0 indicates "No Problem" > 1 to 5 indicate "Various degree of the problem" > > TEST 2....22 questions with answers from 0 to 4 > 0 indicates "No Problem" > 1 to 4 indicate "Various degree of the problem" > > TEST 3....25 questions with answers from 0 to 5 > 0 indicates "No Problem" > 1 to 5 indicate "Various degree of the problem" > > Each test has a total score by adding all the scores for each > question. > > ***The client needs to know what problems go together so to speak...*** > > 1: DO i use each test total scores for the factor analysis? > > 2: Do I just have 2 variables per question, ie. 0= No Problem; 1=Problem? > > 3: Do I just run the factor analysis with variables as they are? > > Hpw do I prepare the data to run the proc factor procedure? > > Thanks for your assistance... > PC > ******************************************************************* > Paul Carreno > Research Analyst-Measurement, Assessment and Evaluation Laboratory > Edith Cowan University, Perth - Western Australia > E_mail: p.carreno@cowan.edu.au Ph: 61-09-273 8384 > ******************************************************************* > * I used to be uncertain of myself but now... I'm not so sure. * > *******************************************************************

Paul: the way you currently have the data entered (ordinal scores from 1 to 5) should be just fine for entering into proc factor. Factor analysis assumes that you have a interval/ratio scale, but I have often seen ordinal scales used in factor analysis.

Without more information on what types of measures are constituted by each "Test", I can't tell you what I would recommend for how many factor analyses you would want to conduct. It seems from the number of questions you've got for each test, you may want to run one factor analysis for each "test". But you should think carefully about what it is you want to find out from your factor analysis first.

Which brings me to the most important part of my response: do you have a good idea about why you want to use factor analysis for this research? Factor analysis can be difficult to interpret and explain to those unfamiliar with the approach. Additionally, I think doing some more straightforward analyses: crosstabs and measures of agreement might yield more intuituve results from the get-go. Then later you could consider why a factor analysis might or might not be appropriate.

Robin Way Director, Market Research Northwest Natural Gas Portland, Oregon email: rew@gasco.com