True or False: If zero is not in the confidence interval then the coeficcient in a regression is significant?

[Dan theMan]

I'm having a bit of trouble analyzing some data on a regression I ran.
I have Z-scores and confidence intervals.

True or false: If a coefficient does not contain zero in its 95% confidence interval then it is significant at the 5% significance level.

I ask this because I have something with low z scores but that does not contain zero in its confidence interval.

[Rich Goldstein]

what kind of regression did you estimate? linear regression does not produce z-scores; some regressions that do produce z-scores, such as logistic regression, report coefficients as odds ratios and the null for that is 1.0 rather than 0

[Dan theMan]

what kind of regression did you estimate? linear regression does not produce z-scores; some regressions that do produce z-scores, such as logistic regression, report coefficients as odds ratios and the null for that is 1.0 rather than 0

Hi rich, thanks for the reply.
I did a logit regression

[Rich Goldstein]

if you used the -logit- command, without asking for odds ratios, then the null hypothesis is that the coefficient is 0; if however, you used the -logistic- command, or the -logit- command but asked for odds ratios, then the null hypothesis is that coefficient is 1.0; please read the FAQ which requests that you tell us, using CODE blocks (explained in the FAQ) exactly what you typed and exactly, what Stata gave you in return

[Carlo Lazzaro]

Dan the Man (https://en.wikipedia.org/wiki/Stan_t...ambiguation)):
please note the preference on this forum for real given and family names. Thanks.

[David Fisher]

Hi Dan (et al),

If I understand correctly, is the question about the difference between statistical significance and "practical" (or "clinical") significance? If a confidence interval excludes the null value, then yes, the effect is statistically significant. However, depending on the size of your dataset and the type of model, you may find that effects are very small yet statistically significant. Such effects may or may not be considered "practically" significant -- in other words, they may or may not have real-world value.

(Equally, in other situations you may find that you have an estimate that is large enough to be "practically" significant, but due to (e.g.) low power fails to reach statistical significance.)

Does that help?

Thanks,

David.

[Carlo Lazzaro]

David:
good (and too often overlooked) point.