Dear Statalist,
I'm writing my thesis on the relationship between general empathy and altruistic sharing.
I've used Polychoric correlation to transform my ordinal data to continuous data. After which i am trying to replicate a factor analysis of the underlying factors of empathy. In said factor analysis I find that only one factor has an eigenvalue of higher than 1. Therefore I force the factor analysis to calculate only one dimension. However when I look at how much of the variance is explained by this factor analysis the proportion of variance explained exceeds 1. Which doesn't make any sense. This is due to some lower ranking factors with negative eigenvalues. I'm not sure why these have negative eigenvalues though. In the polychoric correlations matrix there are no negative correlations present, a cronbach alpha test of the questions included in the analysis shows that all independent variables move in the same direction and my Kaiser-Meyer-Olkin measure of sampling adequacy is above 0.8.
My questions are :
- how my eigenvalues can be negative?
- how does a proportion of variance explained over 1 implicate the results from regression models that have this factor included as a variable?
- Is there any way i can interpret the negative eigenvalues as absolute values? And calculate the variance explained by Dividing the variance through the sum of the eigenvalues?
Any help would be greatly appreciated. I've included the code below and the output of the factor analysis below.
Thank you in advance
greetings from the Netherlands
Tim Wilmink
I'm writing my thesis on the relationship between general empathy and altruistic sharing.
I've used Polychoric correlation to transform my ordinal data to continuous data. After which i am trying to replicate a factor analysis of the underlying factors of empathy. In said factor analysis I find that only one factor has an eigenvalue of higher than 1. Therefore I force the factor analysis to calculate only one dimension. However when I look at how much of the variance is explained by this factor analysis the proportion of variance explained exceeds 1. Which doesn't make any sense. This is due to some lower ranking factors with negative eigenvalues. I'm not sure why these have negative eigenvalues though. In the polychoric correlations matrix there are no negative correlations present, a cronbach alpha test of the questions included in the analysis shows that all independent variables move in the same direction and my Kaiser-Meyer-Olkin measure of sampling adequacy is above 0.8.
My questions are :
- how my eigenvalues can be negative?
- how does a proportion of variance explained over 1 implicate the results from regression models that have this factor included as a variable?
- Is there any way i can interpret the negative eigenvalues as absolute values? And calculate the variance explained by Dividing the variance through the sum of the eigenvalues?
Any help would be greatly appreciated. I've included the code below and the output of the factor analysis below.
Thank you in advance
greetings from the Netherlands
Tim Wilmink
Code:
<edited for confidentiality> *EC: factortest q2 q4 q7 q12 q16 q17 q19 screeplot factormat r, n($N) factors (1) rotate, varimax horst
Comment