Hi
I second Jeremy, EFA is the way to go and also would want to add that based on the number of items and factors you expect, an ideal first (because you will surelly need a second one as well) development sample should be 300-400 participants minimum. But again if you look in the literature people never agree on sample sizes and prefer to go with rules of thumb.
Best
Panos
-----Original Message-----
From: Jeremy Miles
Sent: 6 Mar 2013 17:15:36 GMT
To: [log in to unmask]
Subject: Re: advice regarding factor analysis
Exploratory.
Confirmatory factor analysis will not fit a 5 factor, 25 item scale
when you first develop it. It will probably never fit, even after
several rounds of refinement. So jump to the EFA. (Long established
scales designed to measure on conceptually clear thing with a small
number of items don't fit in CFA - e.g. the Beck Depression
Inventory).
Sample size of 144 is low. Think about it this way. If you have a
correlation of 0.3, then the 95% CIs of the correlation are 0.14, 0.44
(I calculated that here: http://www.vassarstats.net/rho.html ). So
your correlation might be low, or might be high.
But you don't have just one correlation, you have 25 * (25 - 1) / 2 =
300 correlations. That's a lot of data from not very many people.
You might get away with it, but no one is going to get excited about
the results.
J
On 6 March 2013 01:28, Rebecca Craig <[log in to unmask]> wrote:
> Hi all,
>
> I am looking to conduct some form of factor analysis on a new questionnaire
> I am developing as part of my PhD. It is an entirely new questionnaire but
> is designed to have a five factor structure based on the existing literature
> for the topic in question.
>
> My first question is whether I should perform an exploratory or confirmatory
> factor analysis or both? I have read conflicting advice from sources on the
> internet. Some advocate exploratory FA when the scale is completely new,
> whilst others seem to say that if the factor structure is assumed a priori
> confirmatory FA is more appropriate. In an ideal world it seems as if I
> would do both approaches, but again I am unsure which way around is best.
> Some say to do exploratory FA followed by confirmatory FA in another sample
> (which I am not able to do within the confines of my PhD), whilst others say
> that confirmatory FA that fails to find the expected factor structure can
> then be followed up with an exploratory FA. If anyone can shed some light on
> this, or suggest papers which demonstrate this kind of new scale
> development, it would be much appreciated.
>
> My other question is whether anyone knows of any definitive guidelines
> concerning sample size for either form of FA? Again the literature available
> online is very confusing. I have a 25 item scale which is assumed to break
> down into 5 factors and a current sample size of 144 - is FA appropriate?
>
> Many thanks for your help,
> Rebecca
|