TMod: Comparison Table For Linear Models

Description

Collect the coefficients and some qualifying statistics of linear models and organize it in a table for comparison and reporting. The function supports linear and general linear models.

Usage

TMod(..., FUN = NULL)
ModSummary(x, ...)
# S3 method for lm
ModSummary(x, conf.level = 0.95, ...)
# S3 method for glm
ModSummary(x, conf.level = 0.95, use.profile = TRUE, ...)
# S3 method for TMod
plot(x, terms = NULL, intercept = FALSE, ...)
# S3 method for TMod
print(x, digits = 3, na.form = "-", ...)

Arguments

a (general) linear model object.

…

a list of (general) linear models.

conf.level

the level for the confidence intervals.

FUN

function with arguments est, se, tval, pval, lci, uci to display the coefficients. The default function will display the coefficient and significance stars for the p-values.

terms

a vector with the terms of the model formula to be plotted. By default this will be all of them.

use.profile

logical. Defines if profile approach should be used, which normally is a good choice for small datasets. Calculating profile can however take ages for large datasets and not be necessary there. So we can fallback to normal confidence intervals.

intercept

logical, defining whether the intercept should be plotted (default is FALSE).

digits

integer, the desired (fixed) number of digits after the decimal point. Unlike formatC you will always get this number of digits even if the last digit is 0.

na.form

character, string specifying how NAs should be specially formatted. If set to NULL (default) no special action will be taken.

Value

character table

Details

In order to compare the coefficients of linear models, the user is left to his own devices. R offers no support in this respect. TMod() jumps into the breach and displays the coefficients of several models in tabular form. For this purpose, different quality indicators for the models are displayed, so that a comprehensive comparison of the models is possible. In particular, it is easy to see the effect that adding or omitting variables has on forecast quality.

A plot function for a TMod object will produce a dotchart with the coefficients and their confidence intervals.

Examples

Run this code

# NOT RUN {
r.full <- lm(Fertility ~ . , swiss)
r.nox <- lm(Fertility ~ . -Examination - Catholic, swiss)
r.grp <- lm(Fertility ~ . -Education - Catholic + CutQ(Catholic), swiss)
r.gam <- glm(Fertility ~ . , swiss, family=Gamma(link="identity"))
r.gama <- glm(Fertility ~ .- Agriculture , swiss, family=Gamma(link="identity"))
r.gaml <- glm(Fertility ~ . , swiss, family=Gamma(link="log"))

TMod(r.full, r.nox, r.grp, r.gam, r.gama, r.gaml)

# display confidence intervals
TMod(r.full, r.nox, r.gam, FUN = function(est, se, tval, pval, lci, uci){
  gettextf("%s [%s, %s]",
           Format(est, fmt=Fmt("num")),
           Format(lci, digits=3),
           Format(uci, digits=2)
           )
})


# cbind interface is not supported!!
# d.titanic <- reshape(as.data.frame(Titanic),
#                       idvar = c("Class","Sex","Age"),
#                       timevar="Survived",
#                       direction = "wide")
#
# r.glm0 <- glm(cbind(Freq.Yes, Freq.No) ~ 1, data=d.titanic, family="binomial")
# r.glm1 <- glm(cbind(Freq.Yes, Freq.No) ~ Class, data=d.titanic, family="binomial")
# r.glm2 <- glm(cbind(Freq.Yes, Freq.No) ~ ., data=d.titanic, family="binomial")

d.titanic <- Untable(Titanic)

r.glm0 <- glm(Survived ~ 1, data=d.titanic, family="binomial")
r.glm1 <- glm(Survived ~ Class, data=d.titanic, family="binomial")
r.glm2 <- glm(Survived ~ ., data=d.titanic, family="binomial")

TMod(r.glm0, r.glm1, r.glm2)

# plot OddsRatios
d.pima <- MASS::Pima.tr2

r.a <- glm(type ~ npreg + bp + skin + bmi + ped + age, data=d.pima, family=binomial)
r.b <- glm(type ~ npreg + glu + bp + skin, data=d.pima, family=binomial)
r.c <- glm(type ~ npreg + age, data=d.pima, family=binomial)

or.a <- OddsRatio(r.a)
or.b <- OddsRatio(r.b)
or.c <- OddsRatio(r.c)


# create the model table
tm <- TMod(m_A=or.a, m_B=or.b, m_C=or.c)
# .. and plotit
plot(tm, main="ORs for Models A, B, C", intercept=FALSE,
     pch=15, col=c(hred, hblue, horange), 
     panel.first=abline(v=1, col="grey30"))
# }