Significance of correlation in large datasets

Hey everyone,

I have a dataset consisting of more than N=160k samples and would like to quantify the correlation of two variables within that. The correlation is 0.554 (r^2=0.31). As to the significance test (t-test), since the p-value scales with sqrt(n) it will always be large - does that mean that the correlation coefficient is always significant for large datasets? Or does the concept of the t-test at all make sense here?

Thanks for help!

Dennis

[José-Luis]

It doesn't scale linearly. It is asymptotic. More data are generally good. It just means that the test is more robust to individual outliers.

[Von Duesenberg]

In practice, because of the very nature of statistical inference, you easily get significant p-values with large datasets. What matters most is whether, as an expert in your own field, you think the magnitude of the correlation coefficient is sufficient for your goals.