LogisticDx (version 0.1)

plotLogiDx: Diagnostic plots for a logistic regression

Description

Common diagnostic plots for a logistic regression model

Usage

plotLogiDx(x, noPerPage = 6,
    cols = c("deepskyblue", "dodgerblue"), cex = 2,
    pch = 21, inches = 0.25, identify = FALSE,
    extras = FALSE, width = NULL, height = NULL)

Arguments

x

A logistic regression model of class glm

noPerPage

Number of plots per page (for initial plots). Will be used as guidance and optimised for ease of display

cols

Colours. As used by graphics::points

cex

Cex Character expansion. See ?graphics::plot.default

pch

Plotting character. See ?graphics::points

inches

Width of circles for bubble plot. See ?graphics::symbols

identify

If TRUE will give option to identify individual points on a number of the plots produced. The number which appears next to the point corresponds to the relevant row as given by logiGOF

extras

If TRUE produces additional plots, detailed below

width

Width of screen(display device) in pixels

height

Height of screen(display device) in pixels

Value

• The following are plotted, for each covariate group:
• p_X_lev
  Probability of $y=1$ for this group by leverage (diagonal of hat matrix, a measure of influence)
• p_X_dXsq
  Probability as above by dXsq change in Pearson chi-square statistic with deletion of this group
• p_X_dBhat
  Probability by dBhat change in Bhat; the difference in the maximum likelihood estimators Beta for model coefficients with all subjects included vs those with this group, standardized by the estimated covariance matrix of Beta
• p_X_dDev
  Probability by dDev, the change in deviance when this group is excluded
• bubbleplot
  Probability by dXsq, with area of circle proportional to dBhat
• lev_X_dXsq
  Leverage by dXsq, the change in the Pearson chi-square statistic when this group is excluded
• lev_X_dBhat
  Leverage by dBhat, the difference in the maximum likelihood estimators Beta for model coefficients with all subjects included vs those when this group is excluded. This is standardized by the estimated covariance matrix of Beta
• lev_X_dDev
  Leverage by dDev, the change in deviance when this group is excluded
• ROC
  Receiver Operator Curve
• Additional plots are given when extras=TRUE:
• influenceplot
  See ?car::influencePlot
• sr_X_hat
  Studentized residual by hat values. Studentized residual = residual / estimate of standard deviation of residual
• slp
  Spread-level plot. See from ?car::spreadLevelPlot
• qqPlot
  quantile-quantile plot vs Normal for residuals. See ?stats::qqplot
• iip
  Influence-index plot. Gives Cooks distance, studentized residual and hat values for each observation
• pairs
  Pairs plot for the measures of influence dBhat, dXsq and dDev. See ?graphics::pairs
• crPlots
  Component + residual plots. See ?car::crPlots
• avPlots
  Added-variable plots. See ?car::avPlots
• mmps
  Marginal model plots. These require that the data.frame used to fit the model be present in the current environment. See ?car::mmps

Examples

set.seed(1)
### generate up to 8x covariate patterns
mod1 <- genLogiDf(b=3, f=0, c=0, n=50)$model
plotLogiDx(mod1, cex=8, noPerPage=1)
plotLogiDx(mod1, cex=3, noPerPage=6, extras=TRUE)
df1 <- genLogiDf(b=0,f=0,c=2,n=50, model=FALSE)
g1 <- glm(y ~ ., family=binomial("logit"), data=df1)
plotLogiDx(g1)