What your saying is true. The sequential/marginal difference can account for the discrepancy in p values but not necessarily the coefficients. One thing I've found that can lead to differences in coefficients and p values between R and SPSS is whether or not you specify that a variable is a factor (e.g., group2 <- factor(group) ).
Marty Teicher On Aug 13, 2010, at 10:58 PM, Ben Bolker wrote: > Yes, but ... the original poster said the coefficients differed too. > (The blog post > you refer to deals with ANOVA (i.e. linear models) rather than GLMs > (generalized > linear models): it is true that the sequential/marginal > distinction still applies, but I don't think that can be the *only* > thing going on here.) > > > On Fri, Aug 13, 2010 at 10:50 PM, Martin Teicher > <martin_teic...@hms.harvard.edu> wrote: >> R usesType I sequential SS, not the default Type III marginal SS reported by >> SPSS. There is a good blog post explaining this difference along with some >> interesting comments -- >> http://myowelt.blogspot.com/2008/05/obtaining-same-anova-results-in-r-as-in.html >> >> Best Wishes, >> >> Martin H. Teicher >> Dept of Psychiatry >> McLean Hospital / Harvard Medical School >> Belmont MA 02478 >> >> >> On Aug 13, 2010, at 10:32 PM, Ben Bolker wrote: >> >>> Leo Vorthoren <L.Vorthoren <at> nioo.knaw.nl> writes: >>> >>>> I have been using generalized linear models in SPSS 18, in order to build >>>> models and to calculate the P values. When I was building models in Excel >>>> (using the intercept and Bs from SPSS), I noticed that the graphs differed >>>> from my expectations. When I ran the dataset again in R, I got totally >>>> different outcomes for both the P values as well as the Bs and the >>>> intercepts. The outcomes of R seem much more likely to be the correct ones, >>>> but I really cannot explain the differences. >>> >>> I appreciate/assume that you're asking on the off chance that someone >>> else has tried something very similar and gone to the trouble of figuring >>> out the differences between R's and SPSS's default setup, but you're >>> unlikely to get an answer without more detailed information. >>> >>> My best guess is that SPSS and R are using different contrasts >>> and/or different baseline levels. R uses treatment contrasts by default, >>> and assumes that the first (alphabetical) level of a factor is the >>> baseline level. >>> >>> It's conceivable that you have a dataset where the results are >>> numerically unstable and sensitive to small details in the algorithms >>> used. >>> >>> ______________________________________________ >>> R-help@r-project.org mailing list >>> https://stat.ethz.ch/mailman/listinfo/r-help >>> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html >>> and provide commented, minimal, self-contained, reproducible code. >> >> ______________________________________________ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.