[R] R: anova.Design

[Frank E Harrell Jr]

John Smith wrote:
> Dear All:
> 
> I tried to replicate a case study described by Prof. Harrell in Chapter 7 of
> his Regression Modeling Strategies book, but failed on using anova.Design to
> reproduce his table 7.1,  Following is the code:
> 
> rm(list=ls())
> library(Hmisc)
> library(Design)
> 
> getHdata(counties)
> counties$older <- counties$age6574 + counties$age75
> label(counties$older) <- '% age >= 65, 1990'
> counties$pdensity <- log10(counties$pop.density+1)
> label(counties$pdensity) <- 'log 10 of 1992 pop per 1990 miles^2'
> 
> dd <- datadist(counties)
> options(datadist='dd')
> 
> f <- ols(democrat ~ rcs(pdensity,4) + rcs(pop.change,3) + rcs(older,3) +
> crime + rcs(college,5)
>          + rcs(income,4) + rcs(college,5) %ia% rcs(income,4) + rcs(farm,3) +
> rcs(white,5) +
>          rcs(turnout,3), data=counties)
> f
> 
> an <- anova(f)
> 
> 
> and the error message reads as:
> 
> Error in solvet(cov[idx, idx], coef[idx], tol = tol) :
>     apparently singular matrix
> 
> Does anyone have a clue?

add the tol argument to the anova call, e.g. tol=1e-13

Frank




-- 
Frank E Harrell Jr   Professor and Chair           School of Medicine
                      Department of Biostatistics   Vanderbilt University