Re: Multivariable regression (t-stat)

Tech-Archive recommends: Repair Windows Errors & Optimize Windows Performance

From: Jerry W. Lewis (post_a_reply_at_no_e-mail.com)
Date: 07/05/04 

Date: Mon, 05 Jul 2004 12:10:47 GMT

I assume that you are are using the Regression tool in the Analysis ToolPak.

For each coefficient, you get the estimate and the standard error of
that estimate (also given by LINEST with the stats argument set to
TRUE). The t statistic is the ratio of the estimate (est) divided by
the standard error (se). This provides a test of whether the model
coefficient is actually zero. To test whether the model coefficient is
equal to some other pre-specified hypothesized value (hv), use
    t = (est-hv)/se
The p-value is calculated by
    TDIST(ABS(t),df,2)
is the probability of getting a t-statistic at least as extreme as was
observed with your data, if the hypothesized value (zero by default)
were known to be true. If that p-value is too small (often considered
to be <0.05, but that value is not set in stone) then the corresponding
coefficient is considered "statistically significant". The logic, is
that you choose to believe that the hypothesized value is wrong, rather
than believing that you have seen something unusual.

The ATP Regression tool also gives confidence intervals for the
coefficients, which may be more informative than a simple significance
test. The confidence interval indicates how much uncertainty is
associated with the estimate. If the 95% confidence interval for a
coefficient includes zero, then that coefficient is statistically
significantly different from zero at the 0.05 level (p-value <=0.05)

This is a once over lightly of some concepts that are easily abused.
You would be well served to take a class or at least read a text on
statistics before assuming that you know much.

Jerry

anonymous@discussions.microsoft.com wrote:

> Im trying to run a multivariable regression using excel,
> and was wondering what the t-stat variable represents.
>
> How do I use it to know if that certain variable's
> coefficient is statistically significant?
>
> I read on some website that i have to compare it with (n-
> k), but Im not sure how to compare it.
>
> Thanks for the help.
>
>
>

Relevant Pages

  • Re: F&T stats from regression
    ... > Are these stats computed for the residuals, covariance,... ... Statistics Toolbox. ... coefficients except the constant term could be zero. ... It also computes a t statistic for each coefficient estimate. ...
    (comp.soft-sys.matlab)
  • Re: interaction term in linear regression with a dummy coded predictor
    ... of the response vs. X1 or response vs. X2 is not a reasonable thing to ... model coefficient is very large relative to the value of the ... are "zero" with a large interaction. ... Their analysis had been presented to the plant manager and his direct ...
    (sci.stat.math)
  • Re: interaction term in linear regression with a dummy coded predictor
    ... of the response vs. X1 or response vs. X2 is not a reasonable thing to ... model coefficient is very large relative to the value of the ... are "zero" with a large interaction. ... Their analysis had been presented to the plant manager and his direct ...
    (sci.stat.math)
  • Re: Stepwise regress: final model with intercept term
    ... solution to have a statistically significant constant term?. ... STEPWISEFIT always includes a constant coefficient ... STEPWISEFIT doesn't consider looking at the standard ... error of the intercept to determine if it is not statistically ...
    (comp.soft-sys.matlab)
  • Re: Stepwise regress: final model with intercept term
    ... solution to have a statistically significant constant term?. ... STEPWISEFIT always includes a constant coefficient ... STEPWISEFIT doesn't consider looking at the standard ... error of the intercept to determine if it is not statistically ...
    (comp.soft-sys.matlab)