[R-sig-ME] Difference lme4 and nlme

Douglas Bates bates at stat.wisc.edu
Wed Feb 23 21:58:04 CET 2011 

On Wed, Feb 23, 2011 at 2:01 PM, Daniel <dmsilv at gmail.com> wrote:
> Yes,
> I'm try to fit a model of candidates campaign revenues using 3 levels model.
> My theoretical assumption is that parties is nested within districts,
> thus my model look likes:
>
> REVENUES ~ Incumbency + Gender + GOV + IPC + Partisan  +
> (1|DISTRICT:PARTY), data=data, REML = TRUE)
>
> where,
> REVENUES = total money received
> Level 1
> Incumbency = dummy
> Gender = dummy
>
> Level 2
> GOV = dummy if party runs state government
> IPC = Ordinal variable to Intra-party competition
>
> Level 3
> Partisan = total of partisan in the state
>
> However, this discussion  drove me to a importante question: am I
> account for possible main effect of PARTY or DISTRICT? If these points
> is plausible my model should be something like this:
>
> REVENUES ~ Incumbency + Gender + GOV + IPC + Partisan + (1|PARTY)  +
> (1|DISTRICT) + (1|DISTRICT:PARTY), data=data, REML = TRUE)

Assuming that the set of political parties is more-or-less fixed, I
would put PARTY in the fixed-effects and (1|DISTRICT) +
(1|DISTRICT:PARTY) in the random effects.
> What you think?
>
> Best,
>
> Daniel
>
>
>
> On Wed, Feb 23, 2011 at 4:04 PM, Douglas Bates <bates at stat.wisc.edu> wrote:
>> On Wed, Feb 23, 2011 at 12:19 PM, Daniel <dmsilv at gmail.com> wrote:
>>> This sound a controversial issue. If I change  "(1|J) + (1|J:PARTY)"
>>> for  "(1|PARTY) + (1|J:PARTY)" I get great different outcomes. So,
>>> first I need to place third level (J) and second PARTY nested within
>>> J, right?
>>>
>>> So, I take this opportunity to inform that scripts of "Linear Mixed
>>> Models: A Practical Guide Using Statistical Software" by Brady et al;
>>> perhaps are wrong. Scripts can be found at
>>> (http://www-personal.umich.edu/~bwest/chapter4.html)
>>
>> As Andrzej is one of the authors of that book I'll let him respond
>> about the scripts.
>>
>> Can you give us some background to the study - in particular, what
>> does  J represent and what does PARTY represent?
>>
>> This sort of confusion is, in my opinion, unnecessary.  If the factors
>> are defined sensibly - avoiding what I call "implicit nesting" - then
>> the model specification is straightforward.
>>
>>>
>>> Thanks,
>>> Daniel
>>>
>>> On Wed, Feb 23, 2011 at 12:55 PM, Andrzej Galecki <agalecki at umich.edu> wrote:
>>>>
>>>> Hello to everyone,
>>>>
>>>> Actually, the same type of mistake occurred in an earlier email.
>>>>
>>>>
>>>> "If you do indeed want to have PARTY nested within J then your call to
>>>> lmer should use the formula
>>>>
>>>> REVENUES ~ INCUMBENCY + (1|PARTY) + (1|J:PARTY)"
>>>>
>>>>
>>>> Preferred notation was used incorrectly. It should be:
>>>>
>>>> REVENUES ~ INCUMBENCY + (1|J) + (1|J:PARTY)
>>>>
>>>>
>>>> Thank you
>>>>
>>>> Andrzej Galecki
>>>> University of Michigan
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> On 2/23/2011 10:39 AM, Douglas Bates wrote:
>>>>>
>>>>> On Wed, Feb 23, 2011 at 8:39 AM, ONKELINX, Thierry
>>>>> <Thierry.ONKELINX at inbo.be>  wrote:
>>>>>>
>>>>>> "(1|PARTY) + (1|J:PARTY)" and "(1|J/PARTY)" are equal
>>>>>
>>>>> Actually (1|PARTY) + (1|J:PARTY) is equal to (1|PARTY/J).  It is easy
>>>>> to confuse these which is why I prefer not to use the (1|F/G)
>>>>> notation.
>>>>>
>>>>>> "(1|PARTY) + (1|J:PARTY)" and "(J|PARTY)" are not equal
>>>>>>
>>>>>> ----------------------------------------------------------------------------
>>>>>> ir. Thierry Onkelinx
>>>>>> Instituut voor natuur- en bosonderzoek
>>>>>> team Biometrie&  Kwaliteitszorg
>>>>>> Gaverstraat 4
>>>>>> 9500 Geraardsbergen
>>>>>> Belgium
>>>>>>
>>>>>> Research Institute for Nature and Forest
>>>>>> team Biometrics&  Quality Assurance
>>>>>> Gaverstraat 4
>>>>>> 9500 Geraardsbergen
>>>>>> Belgium
>>>>>>
>>>>>> tel. + 32 54/436 185
>>>>>> Thierry.Onkelinx at inbo.be
>>>>>> www.inbo.be
>>>>>>
>>>>>> To call in the statistician after the experiment is done may be no more than asking him to perform a post-mortem examination: he may be able to say what the experiment died of.
>>>>>> ~ Sir Ronald Aylmer Fisher
>>>>>>
>>>>>> The plural of anecdote is not data.
>>>>>> ~ Roger Brinner
>>>>>>
>>>>>> The combination of some data and an aching desire for an answer does not ensure that a reasonable answer can be extracted from a given body of data.
>>>>>> ~ John Tukey
>>>>>>
>>>>>>
>>>>>>> -----Oorspronkelijk bericht-----
>>>>>>> Van: r-sig-mixed-models-bounces at r-project.org
>>>>>>> [mailto:r-sig-mixed-models-bounces at r-project.org] Namens Iker
>>>>>>> Vaquero Alba
>>>>>>> Verzonden: woensdag 23 februari 2011 15:30
>>>>>>> Aan: Daniel; Douglas Bates
>>>>>>> CC: R-sig-mixed-models at r-project.org
>>>>>>> Onderwerp: Re: [R-sig-ME] Difference lme4 and nlme
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>    Just a technical question: Would "(1|PARTY) + (1|J:PARTY)"
>>>>>>> be equal to "(1|J/PARTY)" and this to "(J|PARTY)"?
>>>>>>>
>>>>>>>    I've tried the last two ones and as long as I saw, I got
>>>>>>> the same results, but I might have overlooked something.
>>>>>>>
>>>>>>>    Thank you. Regards.
>>>>>>>
>>>>>>>    Iker
>>>>>>>
>>>>>>> --- El mié, 23/2/11, Douglas Bates<bates at stat.wisc.edu>  escribió:
>>>>>>>
>>>>>>> De: Douglas Bates<bates at stat.wisc.edu>
>>>>>>> Asunto: Re: [R-sig-ME] Difference lme4 and nlme
>>>>>>> Para: "Daniel"<dmsilv at gmail.com>
>>>>>>> CC: R-sig-mixed-models at r-project.org
>>>>>>> Fecha: miércoles, 23 de febrero, 2011 15:08
>>>>>>>
>>>>>>> Notice that the first model has 27 levels for J and the
>>>>>>> second model has 465 levels for PARTY %in% J.  That's the difference.
>>>>>>>
>>>>>>> If you do indeed want to have PARTY nested within J then your
>>>>>>> call to lmer should use the formula
>>>>>>>
>>>>>>> REVENUES ~ INCUMBENCY + (1|PARTY) + (1|J:PARTY)
>>>>>>>
>>>>>>>
>>>>>>> On Wed, Feb 23, 2011 at 6:27 AM, Daniel<dmsilv at gmail.com>  wrote:
>>>>>>>>
>>>>>>>> Hello list,
>>>>>>>>
>>>>>>>> I'm just try to find out how can I produce the results
>>>>>>>
>>>>>>> using both packages.
>>>>>>>>
>>>>>>>> Perhaps I'm using different equation. Trailer model are
>>>>>>>
>>>>>>> consistent to
>>>>>>>>
>>>>>>>> Stata output using (tmixed REVENUES INCUMBENCY || J: || PARTY:)
>>>>>>>>
>>>>>>>> lme2<-
>>>>>>>
>>>>>>> lmer(REVENUES~INCUMBENCY+(1|J)+(1|PARTY),data=data,na.action =
>>>>>>>>
>>>>>>>> "na.omit", REML=TRUE)
>>>>>>>>
>>>>>>>> Linear mixed model fit by REML
>>>>>>>> Formula: REVENUES ~ INCUMBENCY + (1 | J) + (1 | PARTY)
>>>>>>>>  Data: data
>>>>>>>>  AIC   BIC logLik deviance REMLdev
>>>>>>>> 78123 78153 -39057    78154   78113
>>>>>>>> Random effects:
>>>>>>>> Groups   Name        Variance   Std.Dev.
>>>>>>>> J        (Intercept) 9.6263e+08  31026 PARTY    (Intercept)
>>>>>>>
>>>>>>> 1.7502e+09
>>>>>>>>
>>>>>>>> 41836 Residual             3.0534e+10 174741 Number of obs: 2894,
>>>>>>>> groups: J, 27; PARTY, 27
>>>>>>>>
>>>>>>>> Fixed effects:
>>>>>>>>           Estimate Std. Error t value
>>>>>>>> (Intercept)    34244      11657   2.938 INCUMBENCY    211495
>>>>>>>> 9536  22.178
>>>>>>>>
>>>>>>>> Correlation of Fixed Effects:
>>>>>>>>          (Intr)
>>>>>>>> INCUMBENCY -0.097
>>>>>>>>
>>>>>>>> lme3<- lme(REVENUES~INCUMBENCY, random=~1
>>>>>>>> |J/PARTY,data=data,na.action = "na.omit", REML=TRUE)
>>>>>>>>
>>>>>>>> Linear mixed-effects model fit by REML
>>>>>>>>  Data: data
>>>>>>>>  Log-restricted-likelihood: -39078.07
>>>>>>>>  Fixed: REVENUES ~ INCUMBENCY
>>>>>>>> (Intercept)  INCUMBENCY
>>>>>>>>   52469.19   220521.74
>>>>>>>>
>>>>>>>> Random effects:
>>>>>>>>  Formula: ~1 | J
>>>>>>>>        (Intercept)
>>>>>>>> StdDev:    25424.31
>>>>>>>>
>>>>>>>>  Formula: ~1 | PARTY %in% J
>>>>>>>>        (Intercept) Residual
>>>>>>>> StdDev:     45574.5 173465.7
>>>>>>>>
>>>>>>>> Number of Observations: 2894
>>>>>>>> Number of Groups:
>>>>>>>>           J PARTY %in% J
>>>>>>>>          27          465
>>>>>>>>
>>>>>>>> --
>>>>>>>> Daniel Marcelino
>>>>>>>> Skype: dmsilv
>>>>>>>> http://sites.google.com/
>>>>>>>>
>>>>>>>>        [[alternative HTML version deleted]]
>>>>>>>>
>>>>>>>> _______________________________________________
>>>>>>>> R-sig-mixed-models at r-project.org mailing list
>>>>>>>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>>>>>>>>
>>>>>>> _______________________________________________
>>>>>>> R-sig-mixed-models at r-project.org mailing list
>>>>>>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>       [[alternative HTML version deleted]]
>>>>>>>
>>>>>>>
>>>>> _______________________________________________
>>>>> R-sig-mixed-models at r-project.org mailing list
>>>>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>>>>>
>>>>>
>>>>
>>>> _______________________________________________
>>>> R-sig-mixed-models at r-project.org mailing list
>>>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>>>
>>>
>>>
>>> --
>>> Daniel Marcelino
>>> Skype: dmsilv
>>> http://sites.google.com/
>>>
>>
>
>
>
> --
> Daniel Marcelino
> Skype: dmsilv
> http://sites.google.com/
>

More information about the R-sig-mixed-models mailing list