Hi guys!
As a part of the preliminary analysis, I want to show that the items used to measure attitude are valid and can be computed into a single factor score.
(NOTE: I know they are valid from other studies, I just want to present it in the paper as a double-check. It is really not the main purpose of the paper, but it is a good option to learn for me).
I've been told to run a principal components analysis and report the explained % of the variance of the first component row in the "total variance explained" box.
An example of the written output should be something like:
"Factor score of attitudes (single-factor principal component analysis explained xx% of the variance) was computed by averaging the item score (reliability: Cronbach's alpha =x)"
Rather than just writing as I've been told, I really want to understand this. However, I have yet to find any explanations or examples of this in other research papers.
So I hope some of you clever people could help me
Kind regards
In principal component analysis rather than just reporting the percentage contribution of the first you need to report the cumulative percentage of the principal components involved in the approximation of the overall function. That is, for example, the first three principal components if that will be enough to approximate the overall function.
Your question is not so clear. If you want to find the potential of any variable individually, you can report the percentage of variance covered by it as the first independent variable.
Hello Ida,
The premise that a set of variables is unidimensional would imply that the proportion of variance explained by the first extracted factor or component (though you're being advised to use PCA, I would recommend factor analysis with a principal axis or maximum likelihood extraction method instead) was: (a) "substantial," with (b) good item-factor (or item-component) loadings for all items, and (c) subsequent factors/components, if extracted, were "minor" relative to the first. The first part (a) is the reason you were given the suggestion to look only at the variance accounted for for the first extracted component.
I would suggest that a better approach to verifying that the measure structure held with your sample would be to run a confirmatory factor analysis (for the declared one-factor model).
Good luck with your work.