# Thread: Factor analysis with multiple constructs

1. ## Factor analysis with multiple constructs

Hi all,

Have a question related to factor analysis and how to execute it. For instance, in a survey you measure 3 seperate constructs, and use 5 seperate statements for each of those constructs.

When you conduct factor analysis to analyse the validity of these constructs, do you execute this factor analysis on a per-construct basis?

Meaning, if I have 3 constructs, I execute the factor analysis 3 times, incorporating only the 5 statements beloning to this construct? Or do I put all statements in one pool, and then conduct a factor analysis (which ideally would then result in 3 factors).

Some examples I have seen do the analysis seperately, some don't. When I put them all together I noticed that sometimes very different construct-statements load on the same factor -- while they are not related in a sensible way. In any case there is a difference in doing this at once or doing this seperately, that is what I noticed in all cases.

2. ## Re: Factor analysis with multiple constructs

I find that when people do construct-by-construct analysis they engage in the bad practice of assuming that their cross-loadings are zero yet they never test for it. If you have a scale or questionnaire that measures multiple constructs and you artificially separate each subsection as if it were and independent section artificially stacks the odds in favour of the researchers' hypothesis by preventing potential sources from misfit from appearing.

So unless you plan on each subsection to be administered separately and independently, to different people, it makes no sense to me as far as why someone would choose to do their analyses in this way.

3. ## Re: Factor analysis with multiple constructs

Originally Posted by spunky
I find that when people do construct-by-construct analysis they engage in the bad practice of assuming that their cross-loadings are zero yet they never test for it. If you have a scale or questionnaire that measures multiple constructs and you artificially separate each subsection as if it were and independent section artificially stacks the odds in favour of the researchers' hypothesis by preventing potential sources from misfit from appearing.

So unless you plan on each subsection to be administered separately and independently, to different people, it makes no sense to me as far as why someone would choose to do their analyses in this way.
Thanks spunky for your answer! I have read several sources on the internet and they were not clear, but to me your answer makes perfect sense. By grouping everything together you have a more robust way of finding a real research result and makes sure you are not favoring your hypothesis by already excluding potential cross-loadings. Thanks!!

 Tweet

#### Posting Permissions

• You may not post new threads
• You may not post replies
• You may not post attachments
• You may not edit your posts