[R-sig-ME] [R] Major difference in the outcome between SPSS and R statisticalprograms
Douglas Bates
bates at stat.wisc.edu
Fri Aug 1 22:44:45 CEST 2008
On Fri, Aug 1, 2008 at 10:56 AM, Doran, Harold <HDoran at air.org> wrote:
> First off, Marc Schwartz posted this link earlier today, read it.
>
> http://cran.r-project.org/doc/FAQ/R-FAQ.html#Why-are-p_002dvalues-not-di
> splayed-when-using-lmer_0028_0029_003f
>
> Second, your email is not really descriptive enough. I have no idea what
> OR is, so I have no reaction.
Perhaps OR is "odds ratio". In a generalized linear model or a
generalized linear mixed model for binary responses and using the
logit link, the exponentials of the coefficients are scale factors for
the odds ratio.
> Third, you're comparing estimates from different methods of estimation.
> lmer will give standard errors that account for the correlation of
> individuals within similar units whereas the SPSS procedure will not.
> The lmer standard errors better capture the true sampling variance of
> the parameters and SPSS doesn't.
>
>
>
>> -----Original Message-----
>> From: Draga, R. [mailto:R.Draga at umcutrecht.nl]
>> Sent: Friday, August 01, 2008 11:45 AM
>> To: Doran, Harold
>> Subject: RE: [R] Major difference in the outcome between SPSS
>> and R statisticalprograms
>>
>> Thanks for the reaction
>>
>> I know, I would not expect the outcomes to be the same.
>> But, I have never before encountered such a difference in
>> outcomes between SPSS and R; mostly the OR's and p-values
>> differed a little bit.
>>
>> Strange is that R shows a OR of 10,176 and 95% CI of
>> 6,295-14,056. Then the p-value must be <0.05 doesn't it?
>> For age the OR's differ dramatically between SPSS and R,
>> 0.985 and 0.003.
>>
>> I just can not explain it.
>>
>> Ronald
>>
>> -----Oorspronkelijk bericht-----
>> Van: Doran, Harold [mailto:HDoran at air.org]
>> Verzonden: vrijdag 1 augustus 2008 17:36
>> Aan: Draga, R.; r-help at r-project.org
>> Onderwerp: RE: [R] Major difference in the outcome between
>> SPSS and R statisticalprograms
>>
>>
>> The biggest problem is that SPSS cannot fit a generalized linear mixed
>> model but lmer does. So, why would you expect the GLM in SPSS and the
>> GLMM in lmer to match anyhow?
>>
>> > -----Original Message-----
>> > From: r-help-bounces at r-project.org
>> > [mailto:r-help-bounces at r-project.org] On Behalf Of Draga, R.
>> > Sent: Friday, August 01, 2008 10:19 AM
>> > To: r-help at r-project.org
>> > Subject: [R] Major difference in the outcome between SPSS and
>> > R statisticalprograms
>> >
>> > Dear collegues,
>> >
>> > I have used R statistical program, package 'lmer', several
>> > times already.
>> > I never encountered major differences in the outcome between
>> > SPSS and R.
>> > ...untill my last analyses.
>> >
>> > Would some know were the huge differences come from.
>> >
>> > Thanks in advance, Ronald
>> >
>> > In SPSS the Pearson correlation between variable 1 and
>> > variable 2 is 31% p<0.001.
>> >
>> >
>> >
>> > In SPSS binary logistic regression gives us an OR=4.9 (95% CI
>> > 2.7-9.0), p<0.001, n=338.
>> >
>> > OR lower upper
>> >
>> > gender 1,120 0,565 2,221
>> >
>> > age 0,985 0,956 1,015
>> >
>> > variable 2 4,937 2,698 9,032
>> >
>> >
>> >
>> > In R multilevel logistic regression using statistical
>> package 'lmer'
>> > gives us an OR=10.2 (95% CI 6.3-14), p=0.24, n=338, groups:
>> group 1,
>> > 98; group 2 84.
>> >
>> > OR lower upper
>> >
>> > gender 2,295 -2,840 7,430
>> >
>> > age 0,003 -70,047 70,054
>> >
>> > variable 2 10,176 6,295 14,056
>> >
>> >
>> >
>> > The crosstabs gives us:
>> >
>> > variable A
>> >
>> > Var B 0 1
>> >
>> > 0 156 108
>> >
>> > 1 17 57
>> >
>> >
>> >
>> > Would somebody know how it is possible that in SPSS we get
>> > p<0.001 and in R we get p=0.24?
>> >
>> >
>> > [[alternative HTML version deleted]]
>> >
>> > ______________________________________________
>> > R-help at 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-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>
More information about the R-sig-mixed-models
mailing list