95%CI for Rsquared in linear regression

Question

[toyan.j.p]

Hi,

I used SPSS to check R squared for associated SNPs in a model with sex and age as covariates and R squared change from step wise regression to find individual contribution of other SNPs on the most prominent SNP by removing each of the associated SNPs.My query is that is it possible to compute 95% CI of R squared/R squared change obtained through linear regression of SNPs with a quantitative trait, if yes how?

Thanks in advance for any help

Regards

Toyanji

[toyan.j.p]

Thank you very much Sean for your help. I tried bootstrapping from

http://www.statmethods.net/advstats/bootstrapping.html in R,

However there are types of confidence interval returned there are options from "norm", "basic", "stud", "perc", "bca" and "all". Though they describe bca for R-squared for their mt cars data. Which type is to be reported since my original R-squared were got through SPSS with Sex and Age as covariates. There is no option here to give covariates.

thanks for the reply

Answer 1

[Sean Davis]

Getting CI for nearly any statistic of interest can often be accomplished using bootstrapping. Bootstrapping conditions in a complicated experimental design can be challenging to get right, though.

You may be better served by using penalized regression methods such as lasso or elasticnet which are designed to deal with the problem you describe in a formalized way.