0

I'm learning Factor Analysis in SPSS. I faced the following problem. I am performing a PCA for factor extraction on a 49×5 dataset using SPSS. Separately I computed factor loadings by using the formula Loadings=(Orthonormal Eigenvectors) X (Square root of Absolute Eigen values) using spreadsheet (MS excel).

Two factors have been extracted in my study.

But loadings of my 1st factor in Excel is just of opposite sign of the result of SPSS. The loadings of the second factor matches exactly in both Excel & SPSS.

This mismatch is due the direction of orthonormal eigen vectors considered in both the cases. I can realize the fact that direction is not important but the absolute values are required to know which variables are loaded on which factor.

But the problem arises when I am going to calculate factor scores, which will be utilized in further regression analysis. If I consider SPSS out put as right then I will get different factor scores for the first factor and if I consider Excel out put as right then I will get a different factor scores for the first factor. Basically the factor scores in both the cases are of different sign but having same magnitude.

Which one should I accept in my research?

However , for the second factor, there is no ambiguity between SPSS result and Excel result.

user173611
  • 1
  • 2
    I can realize the fact that direction is not important - yes. So what is the problem? – amoeba Aug 14 '17 at 21:33
  • Actually I have plan to use the factor scores for further analysis using multiple regression. But change in sign will affect that substantially. This is why I am thinking which sign to accept? – user173611 Aug 14 '17 at 21:39
  • 2
    The change of sign will not affect anything in your regression analysis. – amoeba Aug 14 '17 at 21:40
  • "Which one (sign) should I accept in my research? " The one that makes your graphs look nicer. And no, it won't affect your multiple regression at all as the coefficient you will estimate will be dependent to the sign of your loadings anyway. – usεr11852 Aug 14 '17 at 21:41
  • 1
    Well. If it is then I have no problem to go for further analysis using multiple regression. Thank you for your quick clarification. – user173611 Aug 14 '17 at 21:44

0 Answers0