Comments on: Evaluating returns to social media

By: J

Sat, 25 Jul 2009 00:57:08 +0000

I'm actually not sure - won't the line that fits the equation ln(y) = a + bx + e be the same as that which minimizes y = a * exp(bx + e)? The former equation is the transformation of the latter into a linear space, and they are equivalent. The distribution of the error has changed -- we'd need them to be normally distributed in the linear case for standard regression assumptions to hold; therefore they'd be exponentially distributed in the latter case and that regression should work as well. I think you'd get the same coefficients in either case.

By: Will Dwinnell

Will Dwinnell — Fri, 24 Jul 2009 11:41:29 +0000

"A linear regression of log-transformed data is the same as an exponential regression of the original data."

Strictly speaking, this is not true- at least not if both regressions minimize the squared error. Though I think it will make little practical difference in this case, transforming the dependent variable effectively transforms the errors as well.

By: J

Thu, 23 Jul 2009 12:35:21 +0000

Thanks Ian, that pretty much sums it up.

By: Ian Sohn

Ian Sohn — Thu, 23 Jul 2009 01:00:43 +0000

I'll give the shorter version of your post - there is simply no there there. This is a self-serving piece of fluff.

-Ian

By: R

Wed, 22 Jul 2009 16:05:09 +0000

very cool... you have too much free time... as opposed to me who has no free time...