BIOSTAT 212 Forum

We often get students asking questions about when and how much to round the numbers that Stata produces. It's worth considering this carefully. Here are some rules of thumb to consider:

1) Unless there's a good reason not to, round to 2 "significant" digits (or "sig figs", as they are often referred to; e.g., 36543 could be rounded to 37000; .0033422 could be rounded to .0033, etc). It's much easier for the human brain to absorb and compare numbers with only 2 digits.

2) Sometimes there IS a good reason to use more the 2 sig figs. For example, when your estimate is very precise and you are providing a confidence interval that is very tight, it might be useful to reflect this. Here's an example:

Regression coefficient (Lower 95%CI - Upper 95%CI)

From Stata: 1.032134 (1.0294765 - 1.033435)

Options for how to present:
1) 1.0 (1.0 - 1.0)
2) 1.03 (1.03 - 1.03)
3) 1.032 (1.029 - 1.033)
4) 1.0321 (1.0295 - 1.0334)
5) 1.03213 (1.02948 - 1.03344)

Options 1 and 2 really don't show any "space" between the two sides of the confidence interval, and don't effectively communicate the precision of your estimate. Option 3 is probably a better bet, which minimizes the number of digits while still communicating the degree of precision. If you are presenting only the point estimate (1.032), it's a good idea to use the same rule of thumb, though rounding even further is also reasonable if 1.03 is, for all intents and purposes, the same as 1.0 in interpretation.