Princiapl component analysis

[Olive Bat]

Hello Nick Sir... need some help on pca to form an index... the commands are somehwat like this..
global xlist savingacc fd recurringdeposits chitty ProvidentFund microfinance others landinvest buildings construction businessenterprise share mutualfunds humancapital bonds gold otherspecify
describe $xlist
summarize $xlist
corr $xlist
pca $xlist,mineigen(1)
screeplot, yline(1)
pca $xlist,comp(7)blanks(.3)
rotate, varimax
rotate, varimax blanks(.3)
rotate,clear
rotate,promax
rotate,promax blanks(.3)
rotate,clear
loadingplot
scoreplot
estat loadings
predict assetindex
br assetindex
tab assetindex
estat kmo
...is this correct

[Karl Tjensvoll]

Hi Olive, it would be very helpful if you put your code in the using the code wrapper, it makes the code more readable. And also, what happens when you do the code? Do you get an error? Or are the graphs not as you would expect?

[Olive Bat]

i am able to create an index .. just wanted to be sure on the code.. and apologies.. will use code wrapper from next time

[Nick Cox]

I can't tell what advice you want here. If you want an index, with nothing else said, PC1 is where to start. Whether to rotate and how to rotate are matters of taste and convention.

As a personal opinion, PCA is vastly oversold unless you are dealing with a bundle of very similarly defined variables. Even with a default of using a correlation matrix expecting to find latent structure intelligibly and reliably in a haphazard mix of this and that is over-optimistic. You're likely to be better off using judgment and subject-matter knowledge to pick a single variable that is pertinent to the research goals -- or not to seek a fix-all-problems single variable at all.

[Phil Bromiley]

You might enjoy Scott Armstrong's "Derivation of theory by means of factor analysis or Tom Swift and his electric factor analysis machine".

[Olive Bat]

do i need to standardize the variables before running pca..my all values are in INR .. and they vary between 0 and some thousand

[Nick Cox]

The same units of measurement do allow use of the covariance matrix. So, standardizing isn't compulsory.

[Olive Bat]

Thank you i just checked this..