[R] Problems using allEffects() (package effect)

Joris Meys jorismeys at gmail.com
Fri Jun 18 11:39:15 CEST 2010


That's about the worst fit possible if you ask me :-)
Check the fitted parameters, very likely indeed that there is your
problem. You'll definitely have to reconsider your method. an lm is
far from optimal on this dataset. Plus, you should use serie as a
random or nesting factor. I'd suggest you look for a statistician
close to you and get advice on that one.

Cheers
Joris

On Fri, Jun 18, 2010 at 11:10 AM, Jonas Mandel <jonas.mandel at curie.fr> wrote:
> I have no error when fitting the model. The summary and anova seems to
> be fine :
>
>> summary(lm1)
> Call:
> lm(formula = Volume ~ temps + Traitement:temps + Série, data = data)
>
> Residuals:
>     Min       1Q   Median       3Q      Max
> -34.3134 -14.9527  -0.9835  12.9174  49.5546
>
> Coefficients:
>                 Estimate Std. Error t value Pr(>|t|)
> (Intercept)        24.827      5.128   4.841 4.93e-06 ***
> temps              -3.786      3.057  -1.239   0.2185
> Série8             11.680      5.203   2.245   0.0271 *
> temps:Traitement    3.162      1.277   2.477   0.0150 *
> ---
> Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
>
> Residual standard error: 18.86 on 96 degrees of freedom
> Multiple R-squared: 0.08465,    Adjusted R-squared: 0.05604
> F-statistic: 2.959 on 3 and 96 DF,  p-value: 0.03617
>
>> anova(lm1)
> Analysis of Variance Table
> Response: Volume
>                 Df Sum Sq Mean Sq F value  Pr(>F)
> temps             1    762  762.09  2.1417 0.14661
> Série             1    214  214.18  0.6019 0.43976
> temps:Traitement  1   2183 2182.74  6.1340 0.01501 *
> Residuals        96  34161  355.84
> ---
> Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
>
>
> But the data is not at all balanced :
>> table(data$Série, data$Traitemen)
>
>     1  2  3
>  7  0 28 31
>  8 24 17  0
>
> In batch 7 there is no mouse treated with treatment 1, and in batch 8
> there is no mouse treated with treatment 3. Maybe the error come from
> here, even if lm() is fine with this ?
>
> Cheers,
> Jonas
>
>
> Joris Meys a écrit :
>> Could you then at least give the summary of the lm? It's really
>> impossible to tell where your problem is without any information about
>> the fit. It's a strange error, I never had it in this context, and
>> this kind of errors often indicate that what goes in is wrong. Your
>> lm() object is not well formed, so the problem should come from the
>> fit itself. And as I can't see any obvious errors in your model
>> formula, it has to be a problem with your data. Did you get no warning
>> at all when fitting the model?
>>
>> Cheers
>> Joris
>>
>> On Fri, Jun 18, 2010 at 10:19 AM, Jonas Mandel <jonas.mandel at curie.fr> wrote:
>>> Hello,
>>>
>>> Sorry for the lack of clarity, I thought I was clear about the package :
>>> it's the function allEffects from the package effects.
>>>
>>> Here is my example again (unfortunately I can't give the data) :
>>> The data concerns effects of 3 treatments on the tumoral volume of mice.
>>>
>>>> head(data)
>>>           Id Série Traitement vTum temps Volume
>>>  55656.1 55656     7          3    1     1  14.03
>>>  55805.1 55805     7          3    3     1  59.92
>>>  55829.1 55829     7          3    1     1  15.31
>>>  55806.1 55806     7          3    2     1  28.01
>>>  50725.1 50725     7          3    2     1  34.48
>>>  55737.1 55737     7          3    2     1  43.56
>>>
>>> vTum is a factor coding for the qualitative initial volume, from small
>>> to big, temps is the time (integer) in month since beginning of
>>> treatment, and Série is a factor coding for the batch. Data is unbalanced.
>>>
>>> I fit a linear model as follows:
>>>> lm1=lm(Volume ~  temps + Traitement:temps + Série , data)
>>> And then I try to use allEffects to compute (and then plot) the effects
>>> of the interaction:
>>>> eff.lm1 <- allEffects(mod=lm1, xlevels=list(temps=c(1:3), Traitement=1:3))
>>> And here is the error I get:
>>> Error in apply(mod.matrix[, components], 1, prod) : index out of range
>>>
>>> In the example given in the help page there are factor variables and
>>> numeric variables so I don't think the problem comes from here. Any help
>>> is welcome here.
>>>
>>> Cheers,
>>> Jonas
>>>
>>>
>>> Joris Meys a écrit :
>>>> Please, read the posting guide:
>>>> - provide a minimal example
>>>> - tell us from which package the function is coming
>>>>
>>>> I guess you used the function allEffects() from the package effects, and did :
>>>> eff.lm1 <- allEffect(lm1)
>>>>
>>>> Then I guess that there's something wrong with the variables you put
>>>> into the model. Sure they're factor variables and not numeric?
>>>>
>>>> Cheers
>>>> Joris
>>>>
>>>>
>>>>
>>>> On Thu, Jun 17, 2010 at 6:03 PM, Jonas Mandel <jonas.mandel at curie.fr> wrote:
>>>>> Dear R users,
>>>>>
>>>>> I have some trouble using the allEffects() function to compute and
>>>>> display effect plots for a linear model.
>>>>> My data is quite simple, it concerns effects of 3 treatments on the
>>>>> tumoral volume of mice. vTum codes for the qualitative initial volume,
>>>>> from small to big, temps is the time in month since beginning of
>>>>> treatment, and Série codes for the batch. Data is unbalanced.
>>>>>
>>>>>> head(data)
>>>>>           Id Série Traitement vTum temps Volume
>>>>> 55656.1 55656     7          3    1     1  14.03
>>>>> 55805.1 55805     7          3    3     1  59.92
>>>>> 55829.1 55829     7          3    1     1  15.31
>>>>> 55806.1 55806     7          3    2     1  28.01
>>>>> 50725.1 50725     7          3    2     1  34.48
>>>>> 55737.1 55737     7          3    2     1  43.56
>>>>>
>>>>>
>>>>> The linear model is also quite simple :
>>>>>> lm1=lm(Volume ~  temps + Traitement:temps + Série , data)
>>>>> But when I try to use allEffects() I get this error:
>>>>>> eff.lm1 <- allEffect(mod=lm1,
>>>>> Error in apply(mod.matrix[, components], 1, prod) : index out of range
>>>>>
>>>>> I read the help and the example but I don't understand the reason why it
>>>>> doesn't work. Can you help me ?
>>>>> Thanks
>>>>>
>>>>>
>>>>> --
>>>>> Jonas Mandel
>>>>>
>>>>> ______________________________________________
>>>>> 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.
>>>>>
>>>>
>>>>
>>
>>
>>
>



-- 
Joris Meys
Statistical consultant

Ghent University
Faculty of Bioscience Engineering
Department of Applied mathematics, biometrics and process control

tel : +32 9 264 59 87
Joris.Meys at Ugent.be
-------------------------------
Disclaimer : http://helpdesk.ugent.be/e-maildisclaimer.php



More information about the R-help mailing list