gfitLogReg: Goodness of fit of a Logistic Regression

In Pakillo/pacotools: Miscellaneous tools I often need and forget about

Description

Make a goodness of fit plot (observed vs predicted values) in a logistic regression context. Optionally, a table with numerical results (including a chi-squared test) can be produced.

Usage

gfitLogReg(obs, pred, groups, print.table = FALSE, main = "Goodness of fit",
  xlab = "Observations", ylab = "Predictions", xlim = c(0, 1), pch = 16,
  ps = 2, cex.lab = 1.2, cex.axis = 1.1, las = 1, ...)

Arguments

obs	numeric vector with observations (either 0 or 1)
pred	numeric vector (same length as obs) with fitted model probabilities
groups	Number of groups (bins) to divide the data into. If missing, the default is 10.
print.table	Logical (default is FALSE).
...	other arguments (e.g. xlim, xlab, main, ...) to be passed to plot. Some default values have been chosen.

Value

A goodness of fit plot. Optionally (if print.table=T), also a table. See plotCalibration from package PredictABEL.

Author(s)

Paco, based on function plotCalibration from package PredictABEL.

See Also

plotCalibration from package PredictABEL

Examples

# Generate some data
x <- seq(1:1000)
yhat <- plogis(0.2 + 0.003*x)   # inverse logit
y <- rbinom(1000, 1, yhat)

# Fit model and get predicted probabilities
model <- glm(y~x, family="binomial")
ypred <- predict(model, type="response")

# Check goodness of fit
gfitLogReg(y, ypred)
gfitLogReg(y, ypred, print.table=TRUE)
gfitLogReg(y, ypred, xlim=c(0.5,1))
gfitLogReg(y, ypred, groups=5)

Pakillo/pacotools documentation built on May 7, 2019, 11:56 p.m.