Re: factor loadings and item-total correlations

Posted by Kirill Orlov on Aug 12, 2013; 6:05pm
URL: http://spssx-discussion.165.s1.nabble.com/factor-loadings-and-item-total-correlations-tp5721548p5721551.html

Factor loading (correlation item b/w item and factor) and correlation b/w item and the sum of the other items need not be comparable or agreeable to each other. The Loading is one of possible construct validity coefficients. The corrected correlation is one of construct reliability, homogeineity, coefficients. Note that the notion of validity always implies correlating with an external independent criterion; here, factor is such criterion - despite it had been extracted even from those variables as a "construct". The notion of reliability implies correlating within a group of variables that are assumed to measure the same thing as if reasonable "dublicates".

12.08.2013 19:57, Maguin, Eugene пишет:

What can be made of a comparison of factor loadings and item-total correlations? I understand the computation of each but I’ve never thought of the two sets of value in relation to each other. To focus comments here are the data for items A, B, C, D. I-T are the item-total correlations and FL are the factor loadings.

 

              A           B           C            D           I-T         FL

A           1.00      .70        .38        .34        .57        .83

B           .70        1.00      .50        .32        .62        .90

C           .38        .50        1.00      .60        .63        .49

D           .34        .32        .60        1.00      .52        .33

 

The surprising result—to me, is similarity of the item-total correlations compared to the factor loadings. I think that can be accounted for by the fact that items A are good proxies for each other and that the same is true for items C and D but to a slightly lesser extent. And also that items B and C are pretty good proxy for each other as well. Is this too much to be made of the data? Can more be said?

 

Thanks, Gene Maguin