Interactive Interpretation of Generalized Linear Models

Description

This method implements proposals for the interpretation of models for binary outcomes made by Hoetker (2007) but works for other types of generalized linear models as well. The method is also suitable for generalized additive models fitted with gam (package gam and mgcv).

Usage

## S3 method for class 'glm'
fxInteractive(model, initial.values = as.list(NULL),
    preselect.var = NA, preselect.type = "link", preselect.groups = NULL,
    dev.height = 18, dev.width = 18, dev.width.legend = 8, dev.pointsize = 10,
    dev.defined = FALSE, ylim = NA, col = NA, lty = 1, lwd = 1, 
    main = NA, main.line = 1.5, xlab = NA, ylab = NA, 
    legend.add = TRUE, legend.space = legend.add, legend.only = FALSE,
    legend.pos = "center", legend.cex = 1, legend.width.factor = 1, 
    rug.ticksize = 0.02, rug.col = "black", vline.actual = TRUE, 
    pos.hlines = c(0, 0.5, 0), n.effects = 100, 
    autosave.plot = FALSE, snapshot.plot = FALSE, 
    graphics.filename = "LinRegIntPlot", graphics.numbering = !autosave.plot,
    graphics.type = "pdf", factor.sep = "|", level.sep = ".", 
    latex2console = FALSE, xtable.big.mark = ".", xtable.decimal.mark = ",", 
    xtable.digits = NULL, xtable.display = NULL, xtable.booktabs = FALSE,                 
    panel.title = "Generalized Linear Model", label.button = "Snapshot",
    label.slider.act = "Variable displayed: ", label.box.type = "Type",
    label.types = c("linear predictor", "response", "marginal effect"),
    label.box.groups = "Groups", 
    slider.width = 200, slider.height = 60, button.height = 30,
    box.type.height = 90, box.group.character.width = 7, 
    box.group.line.height = 28, dist.obj.width = 20,
    dist.obj.height = 10, ...)

Arguments

Details

The only mandatory argument of the function is a fitted-model object of class glm. For this object the following prerequisites must be met:

• The model must contain at least one metric covariate.

• The model must be specified with the formula interface and the data frame containing the variables must be passed with the data argument.

• The categorical variables must be factors (ordered or unordered).

Please refer to the documentation of fxInteractive or the vignette for more details.

Value

No object is returned, please refer to the documentation of fxInteractive or the vignette for more details.

Author(s)

Martin Meermeyer m.meermeyer@gmail.com

References

Hoetker, G. (2007). The use of logit and probit models in strategic management research: Critical issues. Strategic Management Journal, 28(4), 331-343.

Kleiber, C., Zeileis, A. (2008). Applied Econometrics with R. New York: Springer.

Examples

### Model specification ###
data("creditdata")

## Treat ordered factors as unordered factors.
options(contrasts=c("contr.treatment","contr.treatment"))

model.cd <- glm(credit ~ amount + I(amount^2) + age + duration*teleph + housing,
    family=binomial, data=creditdata)

## Not run: 

### Basic usage ###

## RStudio users may need to change the graphic device, see details.
options(device = "x11")

## Using defaults
fxInteractive(model.cd)

## Switch text output to LaTeX
fxInteractive(model.cd, latex2console = TRUE)

## Continental European number format in LaTeX output
fxInteractive(model.cd, latex2console = TRUE, decimal.mark = ",", big.mark = ".")

## Save plot as PDF to current working directory when 'Snapshot' is clicked
fxInteractive(model.cd, snapshot.plot = TRUE)

## Change color scheme and line types
fxInteractive(model.cd, col = rep(c(2,4),each = 3), lty = c(1,3,5))

## Change separation characters
fxInteractive(model.cd, factor.sep = "-", level.sep = ">")

## Suppress legend
fxInteractive(model.cd, legend.add = FALSE)

## Suppress rug plot
fxInteractive(model.cd, rug.ticksize = NA)

## Set initial values of metric covatiates 
fxInteractive(model.cd, initial.values = list(amount=5000, age=30, duration=24))

## Preselect covariate, plot type and groups
fxInteractive(model.cd, preselect.var = "duration", preselect.type = "response",
    preselect.groups = c(2,5))

## Preselect covariate and plot type and change axis annotations
fxInteractive(model.cd, preselect.var = "duration", preselect.type = "response", 
    xlab = "duration (months)", ylab = "probability of credit default")


### Visualization of statistical concepts ###

## Nonparametric effect 
require("splines")
model.cd.bs <- glm(credit ~ bs(amount) + age + duration*teleph + housing,
    family = binomial, data = creditdata)
fxInteractive(model.cd.bs, preselect.var = "amount")

## Generalized additive model
require("mgcv")
model.cd.mgcv <- gam(credit ~ s(amount) + age + duration*teleph + housing,
family = binomial, data = creditdata)
fxInteractive(model.cd.mgcv)

## Interaction effect (directly)
fxInteractive(model.cd, preselect.var = "duration")

## Interaction effect (indirectly)
fxInteractive(model.cd, preselect.var = "age")
# manipulate slider for 'duration'

## Quasi-complete separation
# example from Kleiber, Zeileis (2008), p. 130ff 
require("AER")
data("MurderRates")
model.mr <- glm(I(executions > 0) ~ time + income + noncauc + lfp + southern,
    family = binomial, data = MurderRates)
fxInteractive(model.mr, preselect.var = "income", preselect.type = "response")


### Additional examples ###

## Customize device for printing
fxInteractive(model.cd, 
    dev.width = 6,
    dev.width.legend = 4,
    dev.height = 6,
    dev.pointsize = 6,
    col = c("darkred","red","salmon","darkblue","blue","lightblue"),
    legend.width.factor = 1.1,
    vline.actual = FALSE,
    snapshot.plot = TRUE,
    graphics.filename = "creddefault-termplot",
    mar = c(2.5,2.5,1,1)+0.1,
    mgp = c(1.5,0.5,0),
    tcl = -0.3)

## Save predefined plot automatically
fxInteractive(model.cd,
    initial.values = list(amount=5000, duration=24, age=30),
    preselect.var = "duration",
    preselect.type = "link",
    autosave.plot = TRUE,
    graphics.filename = "fig-creddefault-duration-link",
    legend.width.factor = 1.05)

## Modifications for models with many groups
# Increase space for legend and squeeze panel controls
model.cd.moregroups <- glm(credit ~ amount + I(amount^2) + age 
    + duration*teleph + housing + job, family = binomial, data = creditdata)
fxInteractive(model.cd.moregroups, 
    dev.width.legend = 9,
    legend.cex = 1,
    box.type.height = 90,
    box.group.character.width = 6,
    box.group.line.height = 25,
    dist.obj.height = 2)

# Squeeze legend and  panel controls
model.cd.manygroups <- glm(credit ~ amount + I(amount^2) + age 
    + duration*teleph + housing + intuse, family = binomial, data = creditdata)
fxInteractive(model.cd.manygroups, 
    dev.width.legend = 6,
    legend.cex = 0.54,
    box.type.height = 90,
    box.group.character.width = 6,
    box.group.line.height = 25,
    dist.obj.height = 2)
# Note that checkbox for groups grows beyond screen

# Preselect groups and specifiy color and line types directly
model.cd.manygroups <- glm(credit ~ amount + I(amount^2) + age 
    + duration*teleph + housing + intuse, family = binomial, data = creditdata)
index.groups <- c(1,11,21,31,41,51)
vec.col <- NULL
vec.col[index.groups] <- c(1:6)
vec.lty <- NULL
vec.lty[index.groups] <- rep(c(1,2), each = 3)
fxInteractive(model.cd.manygroups,
preselect.var = "amount",
preselect.groups = index.groups,
col = vec.col,
lty = vec.lty)

## End(Not run)

Results

R version 3.3.1 (2016-06-21) -- "Bug in Your Hair"
Copyright (C) 2016 The R Foundation for Statistical Computing
Platform: x86_64-pc-linux-gnu (64-bit)

R is free software and comes with ABSOLUTELY NO WARRANTY.
You are welcome to redistribute it under certain conditions.
Type 'license()' or 'licence()' for distribution details.

R is a collaborative project with many contributors.
Type 'contributors()' for more information and
'citation()' on how to cite R or R packages in publications.

Type 'demo()' for some demos, 'help()' for on-line help, or
'help.start()' for an HTML browser interface to help.
Type 'q()' to quit R.

> library(LinRegInteractive)
Error in library(LinRegInteractive) : 
  there is no package called 'LinRegInteractive'
Execution halted