[Oded Mcdossi]

Joan, it's hard to follow your questions because it is not always clear what you mean. If the goal is to copy the output of factor analysis into excel worksheet, probably all you need is putexcel. For example:

factor a-c
rotate
putexcel A1=(e(title)) B2=matrix(e(r_L)) using factor_results,  sheet("loading") replace​

[joan marc]

Hi Nick Cox and Oded Mcdossi

My apologies if I'm not clear enough the problem is at the same time I don't always follow your comments if they are very brief so my questions might sound strange in turn. Let me reformulate and stop bothering you

your comment in #10: When performing l in 1/6, sepby(clinid) I got sevaral output in columns. As I was asking in #12: are the six variables (b2cost1- bgcost6) the numbers on the left? I would then understand the role of the three factors (weights of each variable), but not the variable bg2cost (the last column on the right). The variables are b2cost1- bgcost6, but suddenly you created a new one? What is that?

your comment in #16. This is what I was looking for. Can I tell this table to (1) only display weights above a certain threshold and (2) be saved in a certain sheet of a certain excel file (as far as I have read putexcel does not allow for sheetreplace as export excel does

Thank you a lot Nick Cox and Oded Mcdossi . Best regards

Joan

[Oded Mcdossi]

joan marc, you are not bothering us at all, the opposite is true. The bottom line of this forum is to give you direct and helpful answer in short time. Although it's a kind of investment, the idea is not let you read all the Stata documentation.

All the code in #10 is not relevant any more since you want to export the result into Excel and not import it into your data file.
list in 1/6 is to list all the variables of the first 6 observations.
in putexcelyou have an option for replacing the sheet sheet("sheetname", replace) or replace to override the file.

[Nick Cox]

joan: I agree with Oded. What you want is becoming impossible to follow. Anyone coming new to the thread would have to juggle several posts in their head to work out what your questions are.

Please start afresh. Post an example of your data as requested in the FAQ (http://www.statalist.org/forums/help#stata) or, alternatively, follow Oded's lead in his post #10 and recast your problem in terms of a dataset we can all access.

Start at the beginning, explaining your problem directly in terms of what you have and what extra you want.

[joan marc]

@Oded Mcdossi

in putexcelyou have an option for replacing the sheet sheet("sheetname", replace) or replace to override the file.

You are right. I would only add that I had to include the modify option. Let me end the discussion here so it is sufficiently long as @Nich Cox mentions

All the code in #10 is not relevant any more since you want to export the result into Excel and not import it into your data file.

It is because I suspect it will be a much more efficient way to work rather than excel
So as far as I see. You create the matrix and then generate variables out of this matrix. I don't understand the rest of it

I would like to identify the values of the factors individually (for example, now factor1 can take two values). I need to have one weight per variable, independently on which nuber of the factor, only those above the mentioned threshold (.3, to say)

[soumava basu]

Hi,
Is it possible to do a PCA which is not mean-centered in STATA?

Thanks,
Soumava

[Arto Arman]

The variance of each component will instead be equal to the corresponding eigenvalue (again, with minimal rounding errors). The components are typically most useful when left that way

Dear Clyde
In what way are they most useful when left that way? I'm currently doing a PCA analysis and would really like to be able to interpret my treatment effect on my PCA component, which has mean zero, but a std. dev of 1.2. Thus, I cannot interpret the magnitude of my effect sizes in standard deviations. Do you know how you could interpret it, if left the way it is? And if I re-standardized my component, would I then be able to interpret my effect size as std. deviations?
Thanks a lot for he help!

[Clyde Schechter]

In what way are they most useful when left that way?

I made that comment because I am opposed, in most circumstances,* to using standardized variables in analyses. That is, I am generally not supportive of expressing effect sizes in standard deviations in the first place. But if that is what you want to do, and if there is a good reason to do it, then, by all means, rescale your component for the purpose.

That said, you don't even really have to do all the calculations over with a rescaled component. If you know that the standard deviation of the component is 1.2, whatever the effect of a unit difference is, the effect of a standard deviation difference in the component will be 1.2 times larger.

*However, one of the circumstances in which using standardized variables can, in my opinion, be helpful is when the variable is an artificial construct with no natural scale, and components extracted from a PCA often fit that description.