I'm finalizing an article where I found useful to used PCA to shows interaction between variables. However I found several papers using that tool and as many version to communicate on PCA results (with and without eigenvalues, with and without correlation coefficient of variable and their correspondent p-values etc.)
Information on how many Principal Components should be written ?
[Data are concerning bacteria physiology/viability and different response to stress. Is a bit like this work : http://www.ncbi.nlm.nih.gov/pubmed/20452079]
Thanking you in advance.
You need to write the number of items used. Your software and version used for PCA.
The assessed suitability of your data for PCA is to be included. Inspection of the correlation matrix and the correlation coeff. you discarded (e.g. < .3 or 4) and if your Kaiser-Meyer-Olkin value exceeded .6. How many factors your PCA revealed with an Eigenvalue >.1 and how much variance they explained and if you had special reasons to exclude one component or decrease your number of components. Then you write which rotation method you used. Then give a reference if the interpretation of your components are consistent with previous research. Give references for everything you have done and for the theory your results support. You shall include the pattern Matrix and the Structure Matrix. This can be done in one big Table. Sometimes you can be asked to include unroted loadings. Common references are Bartlett, 1954; Catell ,1966 and Thurnstone, 1947.
Here as a start, other contributors give you more.
Dear Cyril, I had something simmilar. We had some 10 bacterial species and about 10 oils (to kill them).
There are several things that You can do: analysis by rows (to see what bacterias are similar) or by columns (to see what substances are similar)
Similarity, well, read axis by axis on PCA plot and see the groupping of variables.
If all of them are close together try to plot 2nd vs 3rd instead 1st and 2nd,
or apply different rotation method (none is PCA, varimax is something like ICA in practice, and promax is good for very close samples).
If You are trying to separate for example Gram + and Gram - try with LDA.
For any further questions feel free to cotact me.
You need to write the number of items used. Your software and version used for PCA.
The assessed suitability of your data for PCA is to be included. Inspection of the correlation matrix and the correlation coeff. you discarded (e.g. < .3 or 4) and if your Kaiser-Meyer-Olkin value exceeded .6. How many factors your PCA revealed with an Eigenvalue >.1 and how much variance they explained and if you had special reasons to exclude one component or decrease your number of components. Then you write which rotation method you used. Then give a reference if the interpretation of your components are consistent with previous research. Give references for everything you have done and for the theory your results support. You shall include the pattern Matrix and the Structure Matrix. This can be done in one big Table. Sometimes you can be asked to include unroted loadings. Common references are Bartlett, 1954; Catell ,1966 and Thurnstone, 1947.
Here as a start, other contributors give you more.
In French, you can find useful examples and a modern syllabus of PCA, among many other multivariate techniques, including aids to interpretation for variables and observations, with:
Statistique exploratoire multidimensionnelle :
Ludovic Lebart , Alain Morineau , Marie Piron, Dunod, 2004.
and
Probabilites Analyse des Données et Statistique
Gilbert Saporta, Technip, 2011.
@Aleksandar : thank you for your instructions, would certainly help.
@Béatrice : Thank you very much. You point out some points which were missing in my analysis (and missing in some published publication also).
@Dominique : Thank you for your references ! In french, that could be my summer readings ! :)
Regards,
Cyril
- Badges
- Science topic