Message-ID: <51E75BD5.1060602@xtra.co.nz>
Date: 2013-07-18T03:07:01Z
From: Rolf Turner
Subject: R-squared and GLM
In-Reply-To: <1374071802773-4671754.post@n4.nabble.com>
On 18/07/13 02:36, Chris89 wrote:
> Dear users,
>
> I want to compute r-squared values from a glm regression using a gamma
> distribution and an "identity" link-function, but find no such thing when
> using the summary() or names() function. My next guess was to calculate it
> by "hand", i.e.
>
> r2 = (sum((estimate - xbar)^2) /sum((x-xbar)^2))
>
> but I am unsure if this is even allowed...
Who is going to disallow you? It's a free country. (I refer to Norway of
course; freer than most countries, in my understanding.)
But since you are maximizing a likelihood based on the Gamma distribution,
rather than doing least squares, what exactly is the relevance of R-squared
anyway?
cheers,
Rolf Turner