[R] confidence intervals for nls or nls2 model

Gabor Grothendieck ggrothendieck at gmail.com
Wed May 16 05:28:39 CEST 2012


On Tue, May 15, 2012 at 11:20 PM, Gabor Grothendieck
<ggrothendieck at gmail.com> wrote:
> On Tue, May 15, 2012 at 8:08 PM, Francisco Mora Ardila
> <fmora at oikos.unam.mx> wrote:
>> Hi all
>>
>> I have fitted a model usinf nls function to these data:
>>
>>> x
>>  [1]   1   0   0   4   3   5  12  10  12 100 100 100
>>
>>> y
>>  [1]  1.281055090  1.563609934  0.001570796  2.291579783  0.841891853
>>  [6]  6.553951324 14.243274230 14.519899320 15.066473610 21.728809880
>> [11] 18.553054450 23.722637370
>>
>> The model fitted is:
>>
>> modellogis<-nls(y~SSlogis(x,a,b,c))
>>
>> It runs OK. Then I calculate confidence intervals for the actual data using:
>>
>> dataci<-predict(as.lm(modellogis), interval = "confidence")
>>
>> BUt I don´t get smooth curves when plotting it, so I want to get other "confidence
>> vectors" based on a new x vector by defining a new data to do predictions:
>>
>> x0 <-  seq(0,15,1)
>> dataci<-predict(as.lm(modellogis), newdata=data.frame(x=x0), interval = "confidence")
>>
>> BUt it does not work: I get the same initial confidence interval
>>
>> Any ideas on how to get tconfidence and prediction intervals using new X data on a
>> previous model?
>>
>
> as.lm is a linear model between the response variable and the gradient
> of the nonlinear model and as we see below x is not part of that
> linear model so x can't be in newdata when predicting from the tangent
> model.  We can only make predictions at the original x points.   For
> other x's we could use Interpolation. See ?approx  (?spline can also
> work in smooth cases but in the example provided the function has a
> kink and that won't work well with splines.)
>
>> as.lm(modellogis)$model
>              y          a             b             c  (offset)
> 1   1.281055090 0.06601796 -4.411829e-01  1.168928e+00  1.397153
> 2   1.563609934 0.04798815 -3.268846e-01  9.766080e-01  1.015584
> 3   0.001570796 0.04798815 -3.268846e-01  9.766080e-01  1.015584
> 4   2.291579783 0.16311227 -9.767241e-01  1.597189e+00  3.451981
> 5   0.841891853 0.12203013 -7.665928e-01  1.512752e+00  2.582551
> 6   6.553951324 0.21464369 -1.206154e+00  1.564573e+00  4.542552
> 7  14.243274230 0.74450055 -1.361047e+00 -1.455630e+00 15.756031
> 8  14.519899320 0.59707858 -1.721353e+00 -6.770205e-01 12.636107
> 9  15.066473610 0.74450055 -1.361047e+00 -1.455630e+00 15.756031
> 10 21.728809880 1.00000000 -2.943955e-13 -9.073765e-12 21.163223
> 11 18.553054450 1.00000000 -2.943955e-13 -9.073765e-12 21.163223
> 12 23.722637370 1.00000000 -2.943955e-13 -9.073765e-12 21.163223
>

Also regarding your comment about not getting smooth curves be sure
that you order the x's (and permute the y's the same way) before
plotting.


-- 
Statistics & Software Consulting
GKX Group, GKX Associates Inc.
tel: 1-877-GKX-GROUP
email: ggrothendieck at gmail.com



More information about the R-help mailing list