choose_order {regclass}

Choosing order of a polynomial model

Description

This function takes a simple linear regression model and displays the adjusted R^2 and AICc for the original model (order 1) and for polynomial models up to a specified maximum order and plots the fitted models.

Usage

choose_order(M,max.order=6,sort=FALSE,loc="topleft",...)

Arguments

M

A simple linear regression model fitted with lm()

max.order

The maximum order of the polynomial model to consider.

sort

How to sort the results. If TRUE, "R2", "r2", "r2adj", or "R2adj", sorts from highest to lowest adjusted R^2. If "AIC", "aic", "AICC", "AICc", sorts by AICc.

loc

Location of the legend. Can also be "top", "topright", "bottomleft", "bottom", "bottomright", "left", "right", "center"

...

Additional arguments to plot(), e.g., pch

Details

The function outputs a table of the order of the polynomial and the according adjusted R^2 and AICc. One strategy for picking the best order is to find the highest value of R^2 adjusted, then to choose the smallest order (simplest model) that has an R^2 adjusted within 0.005. Another strategy is the find the lowest value of AICc, then to choose the smallest order that has an AICc no more than 2 higher.

The scatterplot of the data is provided and the fitted models are displayed as well.

Author(s)

Adam Petrie

References

Introduction to Regression and Modeling

Examples

  data(BULLDOZER)
	M <- lm(SalePrice~YearMade,data=BULLDOZER)
  #Unsorted list, messing with plot options to make it look alright
	choose_order(M,pch=20,cex=.3)
	
	#Sort by R2adj.  A 10th order polynomial is highest, but this seems overly complex
	choose_order(M,max.order=10,sort=TRUE)

	#Sort by AICc.  4th order is lowest, but 2nd order is simpler and within 2 of lowest
	choose_order(M,max.order=10,sort="aic")