I fit a model having 12 parameters on different samples by MLE. For each sample, I retrieve the standard error of each parameter with the hessian inverse.

I’m writing a thesis and I’m wondering : how many decimal place should keep when I report the estimate of these parameters ? Intuitively, I thought that I could use the standard error of each parameter and round the estimate to the decimal place of that standard error, plus 1. For example, if an estimate is 3.456 and the standard error is 0.1, then I round to 2 decimal places : 3.46.

However I found nowhere people using this rule so I’m wondering if it’s a good / common practice on research papers. If not, why ? is there any standard rule to choose how many decimal place one should keep for a parameter estimate ?

  • $\begingroup$ I found the dups by searching the site for significant digits. You can find more about this topic by investigating the other posts that show up. $\endgroup$
    – whuber
    Apr 30, 2023 at 15:02


Browse other questions tagged or ask your own question.