Paik MC (2004) Nonignorable missingness in matched case-control data analyses. Biometrics 60:306-314, Table 5

What's wrong with this one?

There are far too many digits (as indicated by the estimated standard errors, which you can't see immediately; take the square roots of the variances---that you have to do that is another problem), and ending zeros were dropped. If the digits in 0.02229 were meaningful, then 0.02100 should be given rather than 0.021.

What should have been done?

One should essentially always cite standard errors rather than variances; they are in the appropriate scale. Use the standard errors to guide you regarding which digits to present; don't include digits that are just noise. Don't drop ending zeros when they are meaningful.


Last modified: Wed May 7 21:43:07 2008