I just want to confirm if it is possible to reverse the sign (+/-) on the factor loadings of one factor. I run a PAF with Oblimin rotation. I obtained 6 factors but in one of them the items that load into the factor have negative values. I have read that this is not relevant in terms of interpreting within a factor so one could change all the signs of the factor to facilitate interpretation, for example, in the correlation matrix.

I am using SPSS. Maybe an image would help me to explain better. My question is in relation to factor 4.

enter image description here

  • $\begingroup$ It's hard to tell what is being asked here. Do you want to know how to negate all components of a vector? If so, please specify what software you are using and ask your question on StackOverflow. You might be interested in the related thread at stats.stackexchange.com/questions/34396/…, because it involves systematically negating principal components within a collection of related results. $\endgroup$ – whuber Feb 26 '13 at 17:41
  • $\begingroup$ @Alvaro Yes, you may invert sign of loadings of any factor to make it more pleasant for interpretation. But: if you compute factor scores don't forget to invert sign there as well; with oblique rotation also don't forget to change sign accordingly in the structure matrix and in the correlation-between-factors matrix - if you're going to report these results. $\endgroup$ – ttnphns Feb 26 '13 at 17:58
  • $\begingroup$ Thanks for the replies. I am using SPSS. Maybe an image would help me to explain better. My question is in relation to factor 4 in the image: docs.google.com/file/d/0B7iGQI85u8KlR0dadzZNT05TXzQ/… $\endgroup$ – Alvaro Carrasco Feb 26 '13 at 18:16
  • $\begingroup$ I inserted your pic right in the question $\endgroup$ – ttnphns Feb 26 '13 at 18:36
  • $\begingroup$ One way to persuade your software to cooperate--I can't guarantee it will work, because the signs of the components are after all arbitrary!--is to negate the four variables P1_2_SQ_029 through P1_2_SQ027 and re-run the calculation. If indeed the components turn out to be positive, all subsequent statistics will be consistent with this and you can interpret everything as originally intended. $\endgroup$ – whuber Feb 26 '13 at 18:52

Browse other questions tagged or ask your own question.