[R-sig-ME] vif using GLMMadaptive

Ben Bolker bbo|ker @end|ng |rom gm@||@com
Mon May 4 00:04:04 CEST 2020


    Yes.  I misunderstood the issue, and particularly the meaning in 
this context of 'aliased' (which has now been well explained by you and 
Tom Philippi).  I would still stand by my first sentence ...

     cheers

     Ben Bolker

On 5/3/20 6:00 PM, Fox, John wrote:
> Hi Ben,
>
>> On May 3, 2020, at 4:34 PM, Ben Bolker <bbolker using gmail.com> wrote:
>>
>>      My advice would be to sort out the underlying problem (which is *not* at all related to mixed models)   and not move to mixed models in the hope that that will fix something.  I believe (but am not sure) that you're using vif() from the car package?
>>
>>     * you may be able to un-alias your variables by eliminating the 'main effect' terms in your model;  in R formula syntax,  A*B is equivalent to  1+A+B+A:B.  So dropping the main effects that are already included in the * terms, or switching from * to : for interactions, may solve your problem.
> I don't think so: That is, when the formula is processed, the redundancies introduced by the mistaken use of * should automatically be accounted for, as in the following:
>
>> D <- data.frame(a=factor(rep(c("a", "b"), 2)), b=factor(rep(c("A", "B"), each=2)))
>> D
>    a b
> 1 a A
> 2 b A
> 3 a B
> 4 b B
>
>> model.matrix(~1 + a + b + a*b, data=D)
>    (Intercept) ab bB ab:bB
> 1           1  0  0     0
> 2           1  1  0     0
> 3           1  0  1     0
> 4           1  1  1     1
> attr(,"assign")
> [1] 0 1 2 3
> attr(,"contrasts")
> attr(,"contrasts")$a
> [1] "contr.treatment"
>
> attr(,"contrasts")$b
> [1] "contr.treatment"
>
> The problem is probably more fundamental, as has already been suggested.
>
> Best,
>   John
>
>> On 5/1/20 2:12 AM, Matos, Grisenia wrote:
>>> I am a PhD student and am working on a school  project due over the weekend.  I ran the following regression:
>>> spending.REG <- glm.nb(spending_count ~ conservative + liberal + moderate + trust_gov + liberal*trust_gov + conservative*trust_gov + moderate*trust_gov + income + education + age + female + white + budget_difficult + democrat + republican, data = Trustdata1)
>>>
>>> I attempted to get a vif score and got this error in R Studio: there are aliased coefficients in the model
>>>
>>> The variables conservative, liberal and moderate are fixed effect where they are either 0 or 1.  The female variable is a 0 or 1.  There are three interactive variables:  moderate*trust_gov, liberal*trust_gov, and conservative*trust_gov.  Moreover, moderate and moderate*trust_gov are the base variables.
>>>
>>> I would like to calculate the vif for the regression equation.  First, in order to get rid of the error and thereafter calculate the vif scores, I attempted to use your code:
>>>
>>> library("GLMMadaptive")
>>>
>>> fm <- mixed_model(y ~ time + sex, random = ~ 1 | id, data = <your_data>,
>>> family = zi.negative.binomial(), zi_fixed = ~ sex, zi_random = ~ 1 | id)
>>>
>>> it returned an error: unexpected '=' in: "erate + trust_gov + liberal*trust_gov + conservative*trust_gov + moderate*trust_gov + income + education + age + female + white + budget_difficult + democrat + republican, random = 1 | id, dat
>>>                    +                   family ="
>>>
>>> Please provide guidance as to what I am doing incorrectly.  I appreciate your help.
>>>
>>> Thanks,
>>>
>>> Grisenia
>>>
>>>
>>>
>>>
>>>
>>> 	[[alternative HTML version deleted]]
>>>
>>> _______________________________________________
>>> R-sig-mixed-models using r-project.org mailing list
>>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>> _______________________________________________
>> R-sig-mixed-models using 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