residuals.ols {rms}

Residuals for ols

Description

Computes various residuals and measures of influence for a fit from ols.

Usage

## S3 method for class 'ols'
residuals(object, 
      type=c("ordinary", "score", "dfbeta", "dfbetas", 
             "dffit", "dffits", "hat", "hscore", "influence.measures",
             "studentized"), ...)

Arguments

object

object created by ols. Depending on type, you may have had to specify x=TRUE to ols.

type

type of residual desired. "ordinary" refers to the usual residual. "score" is the matrix of score residuals (contributions to first derivative of log likelihood). dfbeta and dfbetas mean respectively the raw and normalized matrix of changes in regression coefficients after deleting in turn each observation. The coefficients are normalized by their standard errors. hat contains the leverages — diagonals of the “hat” matrix. dffit and dffits contain respectively the difference and normalized difference in predicted values when each observation is omitted. The S lm.influence function is used. When type="hscore", the ordinary residuals are divided by one minus the corresponding hat matrix diagonal element to make residuals have equal variance. When type="influence.measures" the model is converted to an lm model and influence.measures(object)$infmat is returned. This is a matrix with dfbetas for all predictors, dffit, cov.r, Cook's d, and hat. For type="studentized" studentized leave-out-one residuals are computed. See the help file for influence.measures for more details.

...

ignored

Value

a matrix or vector, with places for observations that were originally deleted by ols held by NAs

Author(s)

Frank Harrell
Department of Biostatistics
Vanderbilt University
[e-mail]

See Also

lm.influence, ols, which.influence

Examples

set.seed(1)
x1 <- rnorm(100)
x2 <- rnorm(100)
x1[1] <- 100
y <- x1 + x2 + rnorm(100)
f <- ols(y ~ x1 + x2, x=TRUE, y=TRUE)
resid(f, "dfbetas")
which.influence(f)
i <- resid(f, 'influence.measures') # dfbeta, dffit, etc.