st: Re: strange world

[Ronan Conroy <rconroy@rcsi.ie>]

There's another consideration too. Logical operators are often found  
in complex expressions. While sometimes you have to guard against  
missing values, some expressions depend on all variable being  
nonmissing, while others do not.

Consider

. gen underweight = (bmi1 < 19 ) | (bmi2 < 19) | (bmi3 < 19)
. lab var underweight "At least one body mass index below 19"

The variable thus defined can be calculated even when one or two of  
the bmi variables are missing. If that's fine by you, then Stata  
should not stand in your way.

The user might specify

. egen bmi_missing = rowmiss(bmi1 bmi2 bmi3)
. gen underweight = (bmi1 < 19 ) | (bmi2 < 19) | (bmi3 < 19) if  
bmi_missing < 2

which would allow the expression to be evaluated if there were at  
least two BMI measurements. But the choice of how many missing  
measurements to tolerate has to be a scientific one.

For this reason, I think that the user is the only person who knows  
under what circumstances a logical expression should evaluate to  
missing. It's unfortunate that Stata, SAS and SPlus/R have different  
ways of handling missing data in logical expressions, but I don't  
think that switching to the S philosophy that x < NA evaluates to NA  
is going to be any easier.

A happy new year, belatedly, to one and all!

Ron�n


*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/