Message-ID: <6B810AFB14C606439FD57E5985E0379103831ACE@useagan1500p.GLOBAL.ECOLAB.CORP>
Date: 2009-04-28T18:18:38Z
From: Prew, Paul
Subject: effects package --- add abline to plot
In-Reply-To: <007601c9c816$b2a0fb20$17e2f160$@ca>
Dear John and David, thank you for your help. I apologize for not defining the analysis as an ordinal regression, or including a structure --- could have taken some of the guesswork out for you.
John --- for the ticks, I would still like to make this work for future analyses, but still not sure what specifically needs changing. Before initially posting, I did read ?effect, and several other searches around "tick" and "at", but couldn't find a workable description or example for how to use "at". The one example I did find I thought I had copied pretty closely with my command below.
plot(..., ticks=c(0.1,0.2,0.3,0.4,0.5,0.6))
I tried other combinations that probably look pretty silly...
plot(..., ticks(at=c(0.1,0.2,0.3,0.4,0.5,0.6)))
plot(..., ticks=c(0.1:0.6/0.1))
Just don't know how to properly populate this list ---
"ticks: a two-item list controlling the placement of tick marks on the vertical axis, with elements at and n. If at=NULL (the default), the program attempts to find `nice' locations for the ticks, and the value of n (default, 5) gives the approximate number of tick marks desired; if at is non-NULL, then the value of n is ignored."
Thanks again. Effects is a terrific package.
Paul
************** Model Specification ****************
Clean.label <- polr(Clean.lbl ~ City.Abbr, method="logistic", data=Safeway,
Hess=TRUE)
> str(Clean.label)
List of 18
$ coefficients : Named num [1:8] -1.0887 -1.1449 0.6923 0.0894 -0.8229 ...
..- attr(*, "names")= chr [1:8] "City.Abbr[T.Dublin]" "City.Abbr[T.Englwd]" "City.Abbr[T.Fairfax]" "City.Abbr[T.Falls Ch]" ...
$ zeta : Named num [1:4] -2.529 0.894 3.447 6.406
..- attr(*, "names")= chr [1:4] "Excellent|V.Good" "V.Good|Good" "Good|Fair" "Fair|Poor"
$ deviance : num 2327
$ fitted.values: num [1:1248, 1:5] 0.0568 0.0568 0.0568 0.0568 0.0568 ...
..- attr(*, "dimnames")=List of 2
.. ..$ : chr [1:1248] "1" "2" "3" "4" ...
.. ..$ : chr [1:5] "Excellent" "V.Good" "Good" "Fair" ...
$ lev : chr [1:5] "Excellent" "V.Good" "Good" "Fair" ...
$ terms :Classes 'terms', 'formula' length 3 Clean.lbl ~ City.Abbr
.. ..- attr(*, "variables")= language list(Clean.lbl, City.Abbr)
.. ..- attr(*, "factors")= int [1:2, 1] 0 1
.. .. ..- attr(*, "dimnames")=List of 2
.. .. .. ..$ : chr [1:2] "Clean.lbl" "City.Abbr"
.. .. .. ..$ : chr "City.Abbr"
.. ..- attr(*, "term.labels")= chr "City.Abbr"
.. ..- attr(*, "order")= int 1
.. ..- attr(*, "intercept")= int 1
.. ..- attr(*, "response")= int 1
.. ..- attr(*, ".Environment")=<environment: R_GlobalEnv>
.. ..- attr(*, "predvars")= language list(Clean.lbl, City.Abbr)
.. ..- attr(*, "dataClasses")= Named chr [1:2] "ordered" "factor"
.. .. ..- attr(*, "names")= chr [1:2] "Clean.lbl" "City.Abbr"
$ df.residual : num 1236
$ edf : num 12
$ n : num 1248
$ nobs : num 1248
$ call : language polr(formula = Clean.lbl ~ City.Abbr, data = Safeway, Hess = TRUE, method = "logistic")
$ method : chr "logistic"
$ convergence : int 0
$ niter : Named int [1:2] 62 22
..- attr(*, "names")= chr [1:2] "f.evals.function" "g.evals.gradient"
$ Hessian : num [1:12, 1:12] 31.9 0 0 0 0 ...
..- attr(*, "dimnames")=List of 2
.. ..$ : chr [1:12] "City.Abbr[T.Dublin]" "City.Abbr[T.Englwd]" "City.Abbr[T.Fairfax]" "City.Abbr[T.Falls Ch]" ...
.. ..$ : chr [1:12] "City.Abbr[T.Dublin]" "City.Abbr[T.Englwd]" "City.Abbr[T.Fairfax]" "City.Abbr[T.Falls Ch]" ...
$ model :'data.frame': 1248 obs. of 2 variables:
..$ Clean.lbl: Ord.factor w/ 5 levels "Excellent"<"V.Good"<..: 1 2 3 3 3 3 3 3 2 2 ...
..$ City.Abbr: Factor w/ 9 levels "DC","Dublin",..: 9 9 9 9 9 9 9 9 9 9 ...
..- attr(*, "terms")=Classes 'terms', 'formula' length 3 Clean.lbl ~ City.Abbr
.. .. ..- attr(*, "variables")= language list(Clean.lbl, City.Abbr)
.. .. ..- attr(*, "factors")= int [1:2, 1] 0 1
.. .. .. ..- attr(*, "dimnames")=List of 2
.. .. .. .. ..$ : chr [1:2] "Clean.lbl" "City.Abbr"
.. .. .. .. ..$ : chr "City.Abbr"
.. .. ..- attr(*, "term.labels")= chr "City.Abbr"
.. .. ..- attr(*, "order")= int 1
.. .. ..- attr(*, "intercept")= int 1
.. .. ..- attr(*, "response")= int 1
.. .. ..- attr(*, ".Environment")=<environment: R_GlobalEnv>
.. .. ..- attr(*, "predvars")= language list(Clean.lbl, City.Abbr)
.. .. ..- attr(*, "dataClasses")= Named chr [1:2] "ordered" "factor"
.. .. .. ..- attr(*, "names")= chr [1:2] "Clean.lbl" "City.Abbr"
$ contrasts :List of 1
..$ City.Abbr: chr "contr.Treatment"
$ xlevels :List of 1
..$ City.Abbr: chr [1:9] "DC" "Dublin" "Englwd" "Fairfax" ...
- attr(*, "class")= chr "polr"
Paul Prew | Statistician
651-795-5942?? |?? fax 651-204-7504
Ecolab Research Center | Mail Stop ESC-F4412-A
655 Lone Oak Drive | Eagan, MN 55121-1560
-----Original Message-----
From: John Fox [mailto:jfox at mcmaster.ca]
Sent: Tuesday, April 28, 2009 10:34 AM
To: 'David Winsemius'
Cc: r-help at r-project.org; Prew, Paul
Subject: RE: [R] effects package --- add abline to plot
Dear David,
> -----Original Message-----
> From: r-help-bounces at r-project.org [mailto:r-help-bounces at r-project.org] On
> Behalf Of David Winsemius
> Sent: April-28-09 10:12 AM
> To: Prew, Paul
> Cc: r-help at r-project.org
> Subject: Re: [R] effects package --- add abline to plot
>
>
> On Apr 28, 2009, at 12:00 AM, Prew, Paul wrote:
>
> > Hello, I am not having success in a simple task. Using the effects
> > package, I would like to add reference lines at probability values
> > of 0.1 ? 0.6 on a plot of the effects.
>
> I have concerns that you are considering these probabilities. They are
> not going to be probabilities. They are effects.
>
> > The plot command works, but following up with an abline command
> > produces the message ?plot .new has not been called yet?, and of
> > course the reference lines were not added.
> >
> > Looking through past R help lists, there was a similar request for
> > help --- trying to add an abline but ?got the error "plot.new has
> > not been called yet".
> >
> > The help list reply was
> >
> > ? ?abline: "This function adds one or more straight lines through
> > the
> > current plot.", i.e. the already existing *current plot*.
> >
> > So plot your data (e.g. with plot(x, y)) before adding a regression
> > line.?
> >
> > I interpreted the above to suggest the following ---
> >
> > plot(allEffects(Clean.label),ask=FALSE, alternating = TRUE,
> > ylab="Probability of Rating", xlab="City",main="Cleanliness Ratings
> > by City",
> > factor.names=FALSE, ticks=c(0.1,0.2,0.3,0.4,0.5,0.6))
> > abline(h=c(0.1,0.2,0.3,0.4,0.5,0.6))
>
> I do not know why that is happening and you have not provided a
> minimal executable example. The vectorized use of abline does succeed
> in a simper example:
>
> > plot(.5,.5)
> > abline(h=c(0.1,0.2,0.3,0.4,0.5,0.6))
>
> .... so the problem may lie in how the effects package completes its
> plot function for this particular object. You ought to provide at a
> minimum the results of str on that object. Perhaps it executes a
> device call and then turns off the device? However I loaded the
> effects package and ran that abline call after the example:
>
> > mod.cowles <- glm(volunteer ~ sex + neuroticism*extraversion,
> + data=Cowles, family=binomial)
> > eff.cowles <- allEffects(mod.cowles, xlevels=list(neuroticism=0:24,
> + extraversion=seq(0, 24, 6)))
> > eff.cowles
>
>
> I did not get what I expected, which would have been a single
> horizontal line at 0.4 but rather got four lines roughly at 0.351,
> 0.378, 0.408, 0.439. Even then, I would have expected one more line
> before the upper limits of that plot, which makes me think these four
> lines were the results of arguments 0.3 ,0.4, 0.5, 0.6. Most R
> plotting is done in the coordinate system rather than with absolute
> coordinates, but perhaps the mixture of base graphics with lattice
> graphis is ht eproblem
The plot() methods in the effects package make lattice graphs which in most instances will have more than one panel. For a binomial GLM, the default is to plot on the scale of the linear predictor (e.g., the logit scale) but to label the response axis on the scale of the response (i.e., the probability scale). To draw a line on the graph, even if you could do it, would require that you translate to the scale of the linear predictor [e.g., for a logit model, log(p/(1 - p))].
> >
> >
> > Error in int_abline(a = a, b = b, h = h, v = v, untf = untf, ...) :
> > plot.new has not been called yet
> >
> > Less bothersome is the fact that the tick marks weren?t modified to
> > 0.1, 0.2, 0.3, etc.