Factor scores (component scores) weights in PCA

Chhatoq — Fri, 19 Jan 2018 23:04:58 GMT

Hello everyone,

I have a question regarding the weights used to calculate the factor scores in the FACTOR procedure. Since this procedure is calculating the optimal weights for each component, do I need to reverse a variable whose sign is negative in the rotated factor pattern table or do the optimal weights take care of that?

For example, in my data, component six is loaded with three variables as shown below. Item 3 has a negative sign, whereas Item 1 and 2 have positive signs. Does this mean I have to recode item 3 before calculating the factor scores? I know this is required when calculating factor-based scores, but I am not sure if that is the case for factor scores. The SAS procedure code I am using is below too.

Thank you so much for your assistance.

Rotated factor pattern table

	Component 5	Component 6
…	…
Item1	0.07	0.85	*
Item2	0.1	0.86	*
Item3	-0.12	-0.75	*

I used the following procedure to store the scores in a table called DATA_v2 :

proc factor data=WORK.DATA_v1
simple
method=prin
priors=one
nfact=6
plots=scree
rotate=varimax
round
flag=.40
out=DATA_v2;
var item1 item2 item3 item4 item5 item6 ;
run;

Re: Factor scores (component scores) weights in PCA

PaigeMiller — Sat, 20 Jan 2018 12:34:02 GMT

@Chhatoq wrote:

Hello everyone,

I have a question regarding the weights used to calculate the factor scores in the FACTOR procedure. Since this procedure is calculating the optimal weights for each component, do I need to reverse a variable whose sign is negative in the rotated factor pattern table

It means that in component 6, the third variable moves in the opposite direction of the other two. It means that in component 6, as the first and second variables increase, the third variable decreases.

topic Factor scores (component scores) weights in PCA in Statistical Procedures

Factor scores (component scores) weights in PCA

Re: Factor scores (component scores) weights in PCA