Add polynomial control variable to model

[Todd Motiwalla]

Hello, consider a simple regression model:

regress test_score i.gender age

However, upon inspection of my data, the relationship between test_score and age is not linear. How can I tell Stata to treat the age covariate as polynomial? In R, for example (which I am more familiar with), this would be done simply by doing: test_score ~ gender + I(age^2). But I am unsure of the equivalent in Stata. Thank you.

[Maarten Buis]

regress test_score i.gender c.age##c.age

[Daniel Shin]

This is probably what you're looking for:

 regress test_score i.gender c.age c.age#c.age

Edit: Or maybe not. Maarten's code gets you what you want I believe. My code adds both age and age^2 in the regression.

[Todd Motiwalla]

Ah okay, this is simpler than I thought. Thanks both--is the main difference between the two that ## will execute both the linear and polynomial forms of age? And does this translate to a quadratic (cubed) or squared term?

[Daniel Shin]

As far as I know, quadratic equations use squares, not cubes... The inclusion of both terms allow more flexibility to fit your data. You can assess what terms are meaningful from the model.

[Rich Goldstein]

the codes in #'s 2 and 3 will give you the exact same results; see

h fvvarlist

[Maarten Buis]

One # will give you just the interaction (product of variables). Two # will give you the interaction and main effects. The trick is that an interaction of a variable with itself is that variable squared. If you want to model a cubic relationship you would enter c.age##c.age##c.age