Close Home Forum Sign up / Log in

principal component analysis

A

Hi all! Just as i approach that point where I feel like throwing my computer out the window and myself along with it, does anyone know of any good programs or ways of doing PCA?! I tried on SPSS (horrible experience) and on SIMCA P+, but its not doing too good, I can't figure out a way of getting a covariate analysis yet. I did try looking at R to use it, but the whole computer syntax thing just blows my mind, I may as well be reading Russian for all I understand of it... I'm trying to write a paper to be published, and part of the results section will live or die on my stats so I really need to make sure I do a good job. However, all 3 of my supervisors are non-statisticians, and my mian sup has actually told me he doesnt want to see anything but the final result of the tests cos he actually hates stats, so I'm kinda stuck with no-one here to help me out! :-( so if anyone knows anything please fill me in!!

L

Try Minitab. You can get a 30 day free trial, and after that you can but a student licence for 6 month or a year. And its not expensive!
It uses drop down menus so its very user friendly.

S

I feel your pain. My sup doesn't do maths either. And I am avoiding syntax. I know I will have to get with it some time - just now now TYVM. Now I'm not exactly a mega-expert but I've done most of my PCAs in SPSS - what was your 'horrible experience' ?

A

omg! thanks v much! ul never believe it but i just figured it out in spss! i had one textbook that was just doing it in an awfully complicated way and it would never work, but then i tried a new textbook which was sooo easy! im v chuffed! however, i get different results depending on if i choose the correlation matrix or covariance matrix, i have two groups (aiming to see if there is any difference between the groups) so i was planning on sticking with the covariance matrix, am i right here?

Try looking at this paper

E Ferguson, T Cox - Exploratory Factor Analysis: A Users' Guide, International Journal of Selection and Assessment, 1993

Its on google scholar. either that or Andy Field explains it very well http://www.statisticshell.com/factor.pdf

S

Note though that in factor analysis (as opposed to principal components analysis) you can use either kind of matrix as FA outcome does not depend on the units of your variables being the same. In PCA you will get different outcomes if your variables have different units - different units will render a covariance matrix meaningless in PCA. But check that out and don't take my word for it - it's been a while since I had to sort this out for myself.:p

T

http://folk.uio.no/ohammer/past/ I use this sorftware for doing PCA and MDS of my data. It was put together for palaeontological data, but I use it for analysing multivariate 2D gel data and it works fairly well to produce PCA and MDS plots. It is a small free program which always helps :o)

F

Hey. Judging by your name are you doing something ecological? I would use CANOCO if you have access to it, or C2 (which you can get a version of for free if you don't have a license). SPSS is not good for complex data sets. If you need to know more let me know.

A

hey all!

just wanted to say thanks for all the replies, it has really helped! It's just an initial study, with sampling locations restrained by practical factors so it's not a very complex dataset, i managed to get it going on spss, and i have arranged to learn R (the basics anyway) and my very nice supervisor finally admitted that he knows someone who can help with more indepth stuff, so it's on the right track again! and my computer escaped unscathed, whihc is more than i can say for my brain! so thanks guys!!

11224