 |
 |
Hi Mina First, principal components analysis followed by varimax is (just about) the worst way to do factor analysis. You should read 'Repairing tom swift's electric factor analysis machine', by Preacher and MacCallum (type the title into Google and you'll find it). Second, that's what you expect to happen. It's when you rotate a smaller number of factors, it's got to divide them differently. The component numbers are arbitrary - there's nothing special about component 4 that means it won't be component 7 next time. Jeremy 2009/10/1 Asimina Vasalou <[log in to unmask]>: > Thank you Jeremy. I am sorry for the confusion - I omitted some information. > 10 components had eigenvalues more than 1 out of which I retained 4. These 4 > are interpretable. The rotation is Varimax. > > However, when I rerun the analysis by setting/constraining the components in > SPSS to 4, the components change. What used to be component 3 and 4, is now > grouped together as component 3 and a 4th new component emerges. All 4 > components are again interpretable. > > Mina > > On 1 Oct 2009, at 00:10, Jeremy Miles wrote: > >> First, if you are doing principal components analysis, they are >> components, not factors. PCA and EFA are different things, with >> different aims (and it's the cause of massive amounts of confusion >> that they have the same name). >> >> Second, if you extract a different number of factors, you'll get >> different solutions. Factor 1 in a 4 factor solution and factor 1 in >> a five factor solution won't be the same. >> >> I'm not sure what you mean when you say you retained 4 factors from 10 >> factors. If you have 39 items, then you have (potentially) 39 >> components. So you've retained either 4 or 10 from 39. >> >> You should first decide how many factors to extract. Then extract >> them and interpret. If you don't like it (because, as in your case, >> they are uninterpretable), extract fewer, and try again. >> >> You also don't mention anything about rotation. >> >> Jeremy >> >> >> >> 2009/9/30 Mina Vasalou <[log in to unmask]>: >>> >>> Dear all, >>> I have conducted principle component analysis on 39 items. I retained 4 >>> factors from a total >>> of 10 factors (the six remaining factors are not easy to interpret as >>> items load highly on >>> multiple factors). Andy Field's book chapter on PCA mentions re-running >>> the analysis a >>> second time, this time constraining the number of factors - in my case >>> this would be 4. >>> When I set the number of factors to 4, factors 3 and 4 now both load on >>> the third factor and >>> a new fourth factor emerges. Can anyone point me towards more information >>> on if and why >>> this step is necessary? >>> >>> Many thanks >>> Mina Vasalou >>> >> >> >> >> -- >> Jeremy Miles >> Psychology Research Methods Wiki: www.researchmethodsinpsychology.com > > -- Jeremy Miles Psychology Research Methods Wiki: www.researchmethodsinpsychology.com