For a regression object, draw a plot of the response on the vertical axis versus
a linear combination \(u\) of regressors in the mean function on the horizontal
axis. Added to the plot are a smooth for the graph, along with
a smooth from the plot of the fitted values on \(u\). mmps is an alias
for marginalModelPlots, and mmp is an alias for marginalModelPlot.
marginalModelPlots(...)mmps(model, terms= ~ ., fitted=TRUE, layout=NULL, ask,
main, groups, key=TRUE, ...)
marginalModelPlot(...)
# S3 method for lm
mmp(model, variable, sd = FALSE,
xlab = deparse(substitute(variable)),
smoother = loessLine, smoother.args=list(span=2/3),
key=TRUE, pch, groups=NULL, ...)
# S3 method for default
mmp(model, variable, sd = FALSE,
xlab = deparse(substitute(variable)), smoother=loessLine,
smoother.args, key=TRUE, pch, groups=NULL,
col.line = palette()[c(4, 2)], col=palette()[1],
labels, id.method="y",
id.n=if(id.method[1]=="identify") Inf else 0,
id.cex=1, id.col=palette()[1], id.location="lr", grid=TRUE, ...)
# S3 method for glm
mmp(model, variable, sd = FALSE,
xlab = deparse(substitute(variable)), smoother=gamLine,
smoother.args=list(k=3), key=TRUE, pch, groups=NULL,
col.line = palette()[c(4, 2)], col=palette()[1],
labels, id.method="y",
id.n=if(id.method[1]=="identify") Inf else 0,
id.cex=1, id.col=palette()[1], id.location="lr", grid=TRUE, ...)
A regression object, usually of class either lm or glm,
for which there is a predict method defined.
A one-sided formula. A marginal model plot will be drawn for
each term on the right-side of this formula that is not a factor. The
default is ~ ., which specifies that all the terms in
formula(object) will be used. If a conditioning argument is given,
eg terms = ~. | a, then separate colors and smoothers are used for
each unique non-missing value of a. See examples below.
If the default TRUE, then a marginal model plot in the direction
of the fitted values or linear predictor of a generalized linear model will
be drawn.
If set to a value like c(1, 1) or c(4, 3), the layout
of the graph will have this many rows and columns. If not set, the program
will select an appropriate layout. If the number of graphs exceed nine, you
must select the layout yourself, or you will get a maximum of nine per page.
If layout=NA, the function does not set the layout and the user can
use the par function to control the layout, for example to have
plots from two models in the same graphics window.
If TRUE, ask before clearing the graph window to draw more plots.
Main title for the array of plots. Use main="" to suppress the title;
if missing, a title will be supplied.
Additional arguments passed from mmps to mmp and
then to plot. Users should generally use mmps, or equivalently
marginalModelPlots.
The quantity to be plotted on the horizontal axis. The
default is the predicted values predict(object). It can be any other
vector of length equal to the number of observations in the object. Thus the
mmp function can be used to get a marginal model plot versus any
regressor or predictor while the mmps function can be used only to get
marginal model plots for the first-order regressors in the formula. In
particular, terms defined by a spline basis are skipped by mmps, but
you can use mmp to get the plot for the variable used to define
the splines.
If TRUE, display sd smooths. For a binomial regression with all
sample sizes equal to one, this argument is ignored as the SD bounds don't
make any sense.
label for horizontal axis
the name of the smoother to use, selected from the choices
described at ScatterplotSmoothers. For linear models and
the default method, the default smoother is the function
is the function loessLine. For generalized linear models the default
is gamLine, using the gam package and using splines.
arguments passed to the smoother. For linear models the
defaults match the smoother used before September 2012, and may be changed
later. See ScatterplotSmoothers.
The name of a vector that specifies a grouping variable for
separate colors/smoothers. This can also be specified as a conditioning
argument on the terms argument.
If TRUE, include a key at the top of the plot, if FALSE omit the
key. If grouping is present, the key is only printed for the upper-left plot.
Arguments for labeling
points. The default id.n=0 suppresses labeling, and setting this
argument greater than zero will include labeling. See
showLabels for these arguments.
plotting character to use if no grouping is present.
colors for data and model smooth, respectively. Using the default palette, these are blue and red.
color(s) for the plotted points.
If TRUE, the default, a light-gray background grid is put on the graph
Used for its side effect of producing plots.
mmp and marginalModelPlot draw one marginal model plot against
whatever is specified as the horizontal axis.
mmps and marginalModelPlots draws marginal model plots
versus each of the terms in the terms argument and versus fitted values.
mmps skips factors and interactions if they are specified in the
terms argument. Terms based on polynomials or on splines (or
potentially any term that is represented by a matrix of regressors) will
be used to form a marginal model plot by returning a linear combination of the
terms. For example, if you specify terms = ~ X1 + poly(X2, 3) and
poly(X2, 3) was part of the original model formula, the horizontal
axis of the marginal model plot for X2 will be the value of
predict(model, type="terms")[, "poly(X2, 3)"]). If the predict
method for the model you are using doesn't support type="terms",
then the polynomial/spline term is skipped. Adding a conditioning variable,
e.g., terms = ~ a + b | c, will produce marginal model plots for a
and b with different colors and smoothers for each unique non-missing
value of c.
The smoothers used were changed in September 2012. For linear models, the
default smoother is still loess with the same smoothing parameters as were
used in the past, but these can be changed with the argument smoother.args.
For generalized linear models, the default smoother uses gamLine, fitting
a generalized additive model with the same family, link and weights as the fit of the
model. SD smooths are not computed for for generalized linear models.
For generalized linear models the default number of elements in the spline basis is
k=3; this is done to allow fitting for predictors with just a few support
points. If you have many support points you may wish to set k to a higher
number, or k=-1 for the default used by gam.
Cook, R. D., & Weisberg, S. (1997). Graphics for assessing the adequacy of regression models. Journal of the American Statistical Association, 92(438), 490-499.
Fox, J. and Weisberg, S. (2011) An R Companion to Applied Regression, Second Edition. Sage.
Weisberg, S. (2005) Applied Linear Regression, Third Edition, Wiley, Section 8.4.
c1 <- lm(infant.mortality ~ gdp, UN)
mmps(c1)
c2 <- update(c1, ~ poly(gdp, 4), data=na.omit(UN))
# plot against predict(c2, type="terms")[, "poly(gdp, 4)"] and
# and against gdp
mmps(c2, ~ poly(gdp,4) + gdp)
# include SD lines
p1 <- lm(prestige ~ income + education, Prestige)
mmps(p1, sd=TRUE)
# condition on type:
mmps(p1, ~. | type)
# logisitic regression example
# smoothers return warning messages.
# fit a separate smoother and color for each type of occupation.
m1 <- glm(lfp ~ ., family=binomial, data=Mroz)
mmps(m1)
Run the code above in your browser using DataLab