topic Confidence Interval in Statistical Procedures

Confidence Interval

KafeelBasha — Fri, 25 Nov 2016 14:05:47 GMT

Hello

How confidence intervals are used to decide significant and insignificant variables in regression analysis?

Can I get an example?.

Thanks

Re: Confidence Interval

Reeza — Fri, 25 Nov 2016 14:11:43 GMT

Confidence intervals are a function of a confidence level, usually denoted alpha. Ie 95% is (100-alpha)

This is then equivalent to comparing the pvalue to the alpha level.

Confidence intervals are easy to denote significance in some cases, not all.

If the expected value is 0 or 1 then you can see if the interval contains that value. But in a linear regression there isn't necessarily some easily known value to compare to.

Re: Confidence Interval

Rick_SAS — Fri, 25 Nov 2016 23:52:22 GMT

I assume you are asking about confidence intervals for the regression coefficients in OLS regression. If a parameter estimate is small, you might as k whether it is significantly different from zero. If the CI for the parameter contains zero, then the parameter estimate is not significantly different from zero.

For example, the PROC REG documentation uses the Fitness data set. If you analyze it like this

ods select parameterestimates;
proc reg data=fitness plots=none;
model Oxygen=Age Weight RunTime RunPulse RestPulse MaxPulse / clb;
quit;

you see that the parameters that have large p-values also have CIs that contain zero.

Re: Confidence Interval

KafeelBasha — Sat, 26 Nov 2016 05:40:30 GMT

Yes, what if CI contains 1.

Thanks

Re: Confidence Interval

Rick_SAS — Sat, 26 Nov 2016 11:14:21 GMT

It depends on the parameter that you are estimating. If you are estimating an odds ratio, then the OR is not significant if the CI contains 1. See the article on "Odds ratio plots" for an example.