reference coding

[noetsi]

A very basic issue for someone unused to class statements. I worked with proc reg mainly in the past. I have data that is coded 1 if you are in a specific group (say age 25-44) and 0 otherwise. These are of course dummy variables. I do  in part.

CLASS

"Age 25 to 44"n (ref ="0")

and get  this parameter estimate. 

If I understand correctly level 0 in the original data is the reference level and the mean difference (controlling for other variables) between level one and level 0 is 3014.736, that is level 1 is higher than level 0 (a terrible way to refer to a slope I know, but I just want to be sure I understand how the coding works). 

[StatDave · Accepted solution]

That is correct. Equivalently, you could just not specify that variable in the CLASS statement.

[Reeza]

If you had an age variable that was in categories it's usually normal to have an AGE variable with the different categories in a single variable and then use the CLASS statement. It doesn't make sense to create a bunch of 0/1 variables to use in the CLASS statement.

[noetsi]

The variables were created by the federal agency we report to. They have split age into a series of categories all of which are dummy variables and I have to run analysis using their variable schema. That said I don't understand what this entails.

"If you had an age variable that was in categories it's usually normal to have an AGE variable with the different categories in a single variable and then use the CLASS statement."

The organization in question says they are doing lots of things that puzzle me such as doing fixed effect regression (as per the Allison book) with no panel data. 🙂

[Reeza]

Typically you don't usually have variables with the names Age14_20 Age21_25 that are 0/1 to include in a class statement. Instead you would have a single variable Age, that has the values: "14 to 20", "21 to 25" etc and specify a single reference level in your CLASS statement.