(Sorry if this is a repost, I got a bounce reply from the r-help server)
Hi,
I’m using the biglm() function to create some linear models for a very
large data set than lm() can’t fit due to memory issues (the problem is
with the number of interactions, I can fit the main effects model)
I need to determine if the 2-way interactions are necessary or not. Ideally
I’d like to use anova() to get an anova table and a p-value for the
interactions, however it appears that anova is not supported for biglm
objects.
So my next idea was to compare the main effects model with the 2-way
interaction model using a likelihood ratio test. I seem to be able to get
the deviance and residual DF from a biglm object, so I think I should be
able to calculate the LRT and get my p-value if I assume a chi-squared
distribution.
I was wondering if anyone sees any problems with this approach (or would be
kind enough to confirm it)? Or has any better suggestions, ideas or
comments?
Thankyou
Chris Howden B.Sc. (Hons) GStat.
Founding Partner
Evidence Based Strategic Development, IP Commercialisation and Innovation,
Data Analysis, Modelling and Training
(mobile) 0410 689 945
(fax) +612 4782 9023
chris@trickysolutions.com.au
[[alternative HTML version deleted]]
