Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.

# Re: st: interpretation for negative and positive slope combination of interaction term

 From David Hoaglin To statalist@hsphsun2.harvard.edu Subject Re: st: interpretation for negative and positive slope combination of interaction term Date Fri, 10 May 2013 07:09:49 -0400

```Dear Nahla,

If the dependent variable is y, my basic suggestion is to use log(y)
instead as the dependent variable, because you are interested in
ratios.  In a simple example, if the model is
log(y) = b0 + b1x
and x is an indicator variable (i.e., 0 for one group and 1 for the
other), then the difference between the mean of log(y) in the two
groups is b1.  In the original scale of the data, that difference
translates into a ratio, the "antilog" of b1.  I usually use logs base
10 when data are transformed, so the ratio would be 10^b1.  For logs
base e, it would be exp(b1).

Whether the continuous explanatory variables should be transformed (to
the log scale or some other scale) is a question of choosing a scale
in which their contributions to log(y) are linear.

David Hoaglin

On Fri, May 10, 2013 at 5:20 AM, Nahla Betelmal <nahlaib@gmail.com> wrote:
> Hi David, thank you for your reply, but can you kindly explain more or
> give me a link or reference how having a log function can solve the
> issue and how to do it plz. Do you mean taking the log of the
> dependent variable only or both the dependent and all continuous
> independent variables? And how can this help plz
>