Including Control Variables

[Saba Khan]

I'm running a regression of per capita consumption expenditure on income. I have a set of controls for household size, religion, and number of males in the household. I'm a bit confused why stata reports different regression coefficients depending on how I specify the set on controls (just put them as independent variables or put them as $controls. For instance, if I run the regression

reg consumption income householdsize religion numbermale

I get a different coefficient for income than if I run the following code:

local controls "household size religion numbermale"
reg consumption income $controls

I do not understand why this happens. Can anyone provide any insight? Thanks!

[William Lisowski]

It seems likely to me that if you look closely at your regression output from

local controls "household size religion numbermale"
reg consumption income $controls

you will see that none of your control variables were included in the regression. That is because $controls is a global macro but you used the local command which created a local macro.

Perhaps

global controls "household size religion numbermale"
reg consumption income $controls

will give you the same results as your initial regression.

[Clyde Schechter]

Or, even better, since global macros are an unsafe programming practice:

local controls householdsize religion numbermale
reg consumption income `controls'

By the way, your original code is inconsistent in that your first regression command has a single variable named householdsize, and your second approach has two separate variables, household and size. In #2 William Lisowski goes with two separate variables. In my code here, I'm going with a single variable. In any case, this is independent of the global vs local issue, whether it's two variables or one.

[Saba Khan]

Thank you both so much. That clears things up a lot (the household size thing was a typo. It should have said householdsize)!