[R] [GGplot] Geom_smooth with formula "power"?

[Bert Gunter]

See section 8.4-8.5 of MASS 4th Ed. (the book) and the use of
profile.nls() and friends for profiling the log likelihhood surface.

Warning: standard F,t approximations may be poor if the log likelihood
is not nearly enough quadratic. The whole issue of what df to use is
also contentious/unresolved.

-- Bert

Bert Gunter

"Data is not information. Information is not knowledge. And knowledge
is certainly not wisdom."
   -- Clifford Stoll


On Tue, Nov 10, 2015 at 9:47 AM, David Winsemius <dwinsemius at comcast.net> wrote:
>
>> On Nov 9, 2015, at 7:19 AM, Catarina Silva <bolseiro.raiz.csilva at portucelsoporcel.com> wrote:
>>
>> I've tried others initial solutions and the adjustement was done to power model in ggplot - geom_smooth.
>> But, with "nls" I can't do the confidence interval with ggplot - geom_smooth? I read that with "nls" we have to force "se=FALSE". Is this true?
>
> Well, sort of. Setting `se = FALSE` prevents the ggplot2 functions from trying to force nls and nls.predict to do something that is not in their design.
>
>> How can I draw confidence interval in the plot?
>>
>> I've done this:
>>> ggplot(data,aes(x = idade,y = v_mt)) +
>> +   geom_point(alpha=2/10, shape=21,fill="darkgray", colour="black", size=3) +
>> +   geom_smooth(method = 'nls', formula = y ~ a * x^b, start = list(a=1,b=2),se=FALSE)
>>>
>> And then I don't have the confidence interval.
>>
>> If I do:
>>> ggplot(data,aes(x = idade,y = v_mt)) +
>> +   geom_point(alpha=2/10, shape=21,fill="darkgray", colour="black", size=3) +
>> +   geom_smooth(method = 'nls', formula = y ~ a * x^b, start = list(a=1,b=2))
>> Error in pred$fit : $ operator is invalid for atomic vectors
>>>
>>
>> Return error…
>>
> Read the help page for nls. The ‘se.fit' parameter is set to FALSE and efforts to make it TRUE will be ignored. So `predict.nls` simply does not return std-error estimates in the typical manner of other predict.* functions. I believe this is because the authors of `nls` did not think there was a clear answer to the question of what confidence bounds should be returned.
>
>  If you want to add confidence bounds to an nls, then you need to decide what bounds to add, and then use the ggplot2 line-drawing functions to overlay them on your own. I found posts in Rhelp that pointed me to the ‘nls2' package, but when I tried to run the code I got messages saying that the `as.lm` function could not be found.
>
> http://markmail.org/message/7kvolf5zzpqyb7l2?q=list:org%2Er-project%2Er-help+as%2Elm+is+a+linear+model+between+the+response+variable+and+the+gradient
>
>> require(nls2)
> Loading required package: nls2
> Loading required package: proto
>> fm <- nls(demand ~ SSasympOrig(Time, A, lrc), data = BOD)
>> predict(as.lm(fm), interval = "confidence")
> Error in predict(as.lm(fm), interval = "confidence") :
>   could not find function "as.lm"
>> getAnywhere(as.lm)
> no object named ‘as.lm’ was found
>
>
> I also found a couple of posts on R-bloggers pointing me to the ‘propagate' package which has two different methods for constructing confidence intervals.
>
> http://rmazing.wordpress.com/2013/08/14/predictnls-part-1-monte-carlo-simulation-confidence-intervals-for-nls-models/
> http://rmazing.wordpress.com/2013/08/26/predictnls-part-2-taylor-approximation-confidence-intervals-for-nls-models/
>
>
>
> —
> David.
>
>> Ty,
>> Catarina Silva
>>
>> -----Original Message-----
>> From: Jeff Newmiller [mailto:jdnewmil at dcn.davis.CA.us]
>> Sent: sábado, 7 de Novembro de 2015 01:09
>> To: bolseiro.raiz.csilva at portucelsoporcel.com; R mailling list
>> Subject: Re: [R] [GGplot] Geom_smooth with formula "power"?
>>
>> Does  [1] help?
>>
>> [1] http://stackoverflow.com/questions/10528631/add-exp-power-trend-line-to-a-ggplot
>> ---------------------------------------------------------------------------
>> Jeff Newmiller                        The     .....       .....  Go Live...
>> DCN:<jdnewmil at dcn.davis.ca.us>        Basics: ##.#.       ##.#.  Live Go...
>>                                      Live:   OO#.. Dead: OO#..  Playing
>> Research Engineer (Solar/Batteries            O.O#.       #.O#.  with
>> /Software/Embedded Controllers)               .OO#.       .OO#.  rocks...1k
>> ---------------------------------------------------------------------------
>> Sent from my phone. Please excuse my brevity.
>>
>> On November 6, 2015 2:41:18 AM PST, Catarina Silva <bolseiro.raiz.csilva at portucelsoporcel.com> wrote:
>>> Hi,
>>>
>>> It's possible to use ggplot and geom_smooth to adjust a power curve to
>>> the data?
>>>
>>> Initially i have done the adjustement with nls and the formula 'a*x^b',
>>> but resulted the singular matrix error for start solution.
>>> Alternatively I used the log transformation and i had correct results,
>>> but I can't draw a power curve on the graphic.
>>>
>>> Someone know how to solve this problem?
>>>
>>>
>
> David Winsemius
> Alameda, CA, USA
>
> ______________________________________________
> R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see
> 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.