3
$\begingroup$

I am trying to compute effect size (SS predictor / SS total) is it possible to have a REALLY small effect size (0.02 no not 0.2 but 0.02) but have the effect be statistically significant?

$\endgroup$
  • 2
    $\begingroup$ Yes. Large sample size. $\endgroup$ – HelloWorld Apr 11 '17 at 23:01
  • 2
    $\begingroup$ It might help the intuition to realize that in a complete census of a (finite) population, any nonzero effect size will be "significant". $\endgroup$ – whuber Apr 11 '17 at 23:13
8
$\begingroup$

Yes. For any non-zero effect size as n approaches $\infty$, p approaches 0. This is because for Wald type tests, the test statistic is $\frac{\theta}{s_{\theta}}$, and $s_{\theta}$ typically looks something like $\frac{s}{\sqrt{n}}$, where $s$ is a standard deviation (e.g., of one's data for the t test).

Further elaboration on significance and tiny effect sizes from the perspective of equivalence and relevance.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.