computing the OR for a multi-unit (ie 10) increase in a continuous variable

[Robin Fatch]

Hello,

I have a coding question.

For example, I have a logistic regression model, with a continuous explanatory variable "age". The OR in the output represents the OR for a one-unit (one-year) increase in age.

Is there a way to calculate the odds for a 10-unit increase in age? (rather than 1-unit change)

Thanks,
Robin

[Clyde Schechter]

So, if the OR for a one unit change is x, then the OR for a 10 unit change is x^10.

[Robin Fatch]

Thank you!

Just to be sure:

Let's say I have a logistic regression model with both sex and age as explanatory variables.

The OR for men vs. women is z.
The OR for a one-unit increase in age is x.

Reporting the OR for a 10-unit increase in age (OR = x^10) doesn't affect how I would report the OR of z for sex, correct?

[Clyde Schechter]

Correct.

[Robin Fatch]

Hi again Clyde,

A slightly belated follow-up for this question... what about 95% CI? Do I need to do something different to calculate them for a 10-unit increase? Or does the same logic apply?

Sorry I am so fuzzy on these details.

Thanks,
Robin

[Clyde Schechter]

Yes, the transformation of the OR estimate can be applied in exactly the same way to the lower and upper confidence limits.

[Rich Goldstein]

just to note: a possibly simpler way is to make a new variable:

gen age_10 = age/10

that will automatically give you what you want (and you don't have to remember the formula above)

[Bruce Weaver]

I generally use the approach Rich suggested in #7. But if a situation arose where I wanted to report the OR for 2 or more increments (e.g., for both 5-year and 10-year increments in age), I might use lincom instead. E.g.,

use http://www.stata-press.com/data/r13/lbw.dta, clear

generate age5 = age/5
generate age10 = age/10
generate age20 = age/20

logit low age5, or
logit low age10, or
logit low age20, or

logit low age, or
lincom _b[age]*5,  or // OR for a  5-year increment in age
lincom _b[age]*10, or // OR for a 10-year increment in age
lincom _b[age]*20, or // OR for a 20-year increment in age

HTH.