Message-ID: <44CDA06C-2868-4477-A663-3EECDC1EB28C@comcast.net>
Date: 2011-11-21T04:57:56Z
From: David Winsemius
Subject: logistic regression by glm
In-Reply-To: <555802ef.a61.133c38486fe.Coremail.tujchl02@163.com>
On Nov 20, 2011, at 7:26 PM, ???? wrote:
> Thank you very much :)
> I search on net and find sometimes response value in logistic model
> can have more than 2 values, and the way of this kinds of regression
> is called "Ordinal Logistic Regression". and even we can caculate it
> by the same way I mean glm in R.
> here are some references:
> 1. http://en.wikipedia.org/wiki/Ordered_logit
> 2. http://www.stat.ubc.ca/~rollin/teach/643w04/lec/node62.html
> above two tell us what is "Ordinal Logistic Regression".
> 3. http://www.ats.ucla.edu/stat/r/dae/ologit.htm
> this show that we can use glm to model it
When I looked through the UCLA code it appeared they were using the
Design package (now superseded by the `rms` package) and that the
function was `lrm` rather than `glm`. In addition to Harrell's
excellent text which has a full chapter on this topic you might also
want to look at Laura Thompson's Companion to Agresti's text:
https://home.comcast.net/~lthompson221/Splusdiscrete2.pdf
--
David.
>
> ?? 2011-11-21 00:56:33??"Uwe Ligges" <ligges at statistik.tu-
> dortmund.de> ??????
>>
>>
>> On 20.11.2011 17:27, ???????? wrote:
>>> I worried it too, Do you have idear that what tools I can use?
>>
>>
>> Depends on your aims - what you want to do with the fitted model.
>> A multinomial model, some kind of discriminant analysis (lda, qda),
>> tree
>> based methods, svm and so son come to mind. You probably want to
>> discuss
>> this on some statistics mailing list/forum or among local experts
>> rather
>> than on the R list. Since this is actually not that R releated.
>>
>> Uwe Ligges
>>
>>
>>
>>>
>>>
>>>
>>>
>>> ?? 2011-11-21 00:13:26??"Uwe Ligges"<ligges at statistik.tu-
>>> dortmund.de> ??????
>>>>
>>>>
>>>> On 20.11.2011 16:58, ???????? wrote:
>>>>> Thank you Ligges :)
>>>>> one more question:
>>>>> my response value "diagnostic" have 4 levels (0, 1, 2 and 3), so
>>>>> I use it like this:
>>>>> "as.factor(diagnostic) ~ as.factor(7161521) +as.factor(2281517)"
>>>>> Is it all right?
>>>>
>>>>
>>>> Uhh. 4 levels? Than I doubt logistic regression is the right tool
>>>> for
>>>> you. Please revisit the theory first: It is intended for 2
>>>> levels...
>>>>
>>>>
>>>> Uwe Ligges
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> ?? 2011-11-20 23:45:23??"Uwe Ligges"<ligges at statistik.tu-dortmun
>>>>> d.de> ??????
>>>>>>
>>>>>>
>>>>>> On 20.11.2011 12:46, tujchl wrote:
>>>>>>> HI
>>>>>>>
>>>>>>> I use glm in R to do logistic regression. and treat both
>>>>>>> response and
>>>>>>> predictor as factor
>>>>>>> In my first try:
>>>>>>>
>>>>>>> *******************************************************************************
>>>>>>> Call:
>>>>>>> glm(formula = as.factor(diagnostic) ~ as.factor(7161521) +
>>>>>>> as.factor(2281517), family = binomial())
>>>>>>>
>>>>>>> Deviance Residuals:
>>>>>>> Min 1Q Median 3Q Max
>>>>>>> -1.5370 -1.0431 -0.9416 1.3065 1.4331
>>>>>>>
>>>>>>> Coefficients:
>>>>>>> Estimate Std. Error z value Pr(>|z|)
>>>>>>> (Intercept) -0.58363 0.27948 -2.088 0.0368 *
>>>>>>> as.factor(7161521)2 1.39811 0.66618 2.099 0.0358 *
>>>>>>> as.factor(7161521)3 0.28192 0.83255 0.339 0.7349
>>>>>>> as.factor(2281517)2 -1.11284 0.63692 -1.747 0.0806 .
>>>>>>> as.factor(2281517)3 -0.02286 0.80708 -0.028 0.9774
>>>>>>> ---
>>>>>>> Signif. codes: 0 ??***?? 0.001 ??**?? 0.01 ??*??
>>>>>>> 0.05 ??.?? 0.1 ?? ?? 1
>>>>>>>
>>>>>>> (Dispersion parameter for binomial family taken to be 1)
>>>>>>>
>>>>>>> Null deviance: 678.55 on 498 degrees of freedom
>>>>>>> Residual deviance: 671.20 on 494 degrees of freedom
>>>>>>> AIC: 681.2
>>>>>>>
>>>>>>> Number of Fisher Scoring iterations: 4
>>>>>>> *******************************************************************************
>>>>>>>
>>>>>>> And I remodel it and *want no intercept*:
>>>>>>> *******************************************************************************
>>>>>>> Call:
>>>>>>> glm(formula = as.factor(diagnostic) ~ as.factor(2281517) +
>>>>>>> as.factor(7161521) - 1, family = binomial())
>>>>>>>
>>>>>>> Deviance Residuals:
>>>>>>> Min 1Q Median 3Q Max
>>>>>>> -1.5370 -1.0431 -0.9416 1.3065 1.4331
>>>>>>>
>>>>>>> Coefficients:
>>>>>>> Estimate Std. Error z value Pr(>|z|)
>>>>>>> as.factor(2281517)1 -0.5836 0.2795 -2.088 0.0368 *
>>>>>>> as.factor(2281517)2 -1.6965 0.6751 -2.513 0.0120 *
>>>>>>> as.factor(2281517)3 -0.6065 0.8325 -0.728 0.4663
>>>>>>> as.factor(7161521)2 1.3981 0.6662 2.099 0.0358 *
>>>>>>> as.factor(7161521)3 0.2819 0.8325 0.339 0.7349
>>>>>>> ---
>>>>>>> Signif. codes: 0 ??***?? 0.001 ??**?? 0.01 ??*??
>>>>>>> 0.05 ??.?? 0.1 ?? ?? 1
>>>>>>>
>>>>>>> (Dispersion parameter for binomial family taken to be 1)
>>>>>>>
>>>>>>> Null deviance: 691.76 on 499 degrees of freedom
>>>>>>> Residual deviance: 671.20 on 494 degrees of freedom
>>>>>>> AIC: 681.2
>>>>>>>
>>>>>>> Number of Fisher Scoring iterations: 4
>>>>>>> *******************************************************************************
>>>>>>>
>>>>>>> *As show above in my second model it return no intercept but
>>>>>>> look this:
>>>>>>> Model1:
>>>>>>> (Intercept) -0.58363 0.27948 -2.088 0.0368 *
>>>>>>> Model2:
>>>>>>> as.factor(2281517)1 -0.5836 0.2795 -2.088 0.0368 **
>>>>>>>
>>>>>>> They are exactly the same. Could you please tell me what happen?
>>>>>>
>>>>>> Actually it does not make sense to estimate the model without an
>>>>>> intercept unless you assume that it is exactly zero for the
>>>>>> first levels
>>>>>> of your factors. Think about the contrasts you are interested
>>>>>> in. Looks
>>>>>> like not the default?
>>>>>>
>>>>>> Uwe Ligges
>>>>>>
>>>>>>
>>>>>>>
>>>>>>> Thank you in advance
>>>>>>>
>>>>>>>
>>>>>>> --
>>>>>>> View this message in context: http://r.789695.n4.nabble.com/logistic-regression-by-glm-tp4088471p4088471.html
>>>>>>> Sent from the R help mailing list archive at Nabble.com.
>>>>>>>
>>>>>>> ______________________________________________
>>>>>>> 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.
>
> [[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.
David Winsemius, MD
West Hartford, CT