I'm using Cox-Regression to analyse the risk of loan portfolios.

I found out that with increasing sample size, my regressors are getting more and more significant (p-values are shrinking) due to shrinking standard errors.

How valuable are these p-value statistics in the end? I'm concerned that these p-values are only "good" due to my large data set. Is there maybe like a limit of how big the data set should be?

  • $\begingroup$ From my basic knowledge I would say that if your assumptions are not violated (there can be hidden dependencies within your large dataset), pvalues are reasonably get lower for larger sample sizes and there is no reason to doubt their significance. $\endgroup$ – German Demidov Oct 9 '17 at 13:33
  • 2
    $\begingroup$ Look at the hazard ratios, regardless of their significance: are they relevant ? For example, a hazard ratio of 1.001 can be statistically significant, but probably not relevant. $\endgroup$ – ocram Oct 9 '17 at 13:36