Robust Covariance Matrix Estimates
Uses the Huber-White method to adjust the variance-covariance matrix of
a fit from maximum likelihood or least squares, to correct for
heteroscedasticity and for correlated responses from cluster samples.
The method uses the ordinary estimates of regression coefficients and
other parameters of the model, but involves correcting the covariance
matrix for model misspecification and sampling design.
Models currently implemented are models that have a
residuals(fit,type="score") function implemented, such as
coxph, and ordinary linear models (
The fit must have specified the
y=TRUE options for certain models.
Observations in different clusters are assumed to be independent.
For the special case where every cluster contains one observation, the
corrected covariance matrix returned is the "sandwich" estimator
(see Lin and Wei). This is a consistent estimate of the covariance matrix
even if the model is misspecified (e.g. heteroscedasticity, underdispersion,
wrong covariate form).
For the special case of ols fits,
robcov can compute the improved
(especially for small samples) Efron estimator that adjusts for
natural heterogeneity of residuals (see Long and Ervin (2000)
robcov(fit, cluster, method=c('huber','efron'))
- a fit object from the
- a variable indicating groupings.
clustermay be any type of vector (factor, character, integer). NAs are not allowed. Unique values of
clusterindicate possibly correlated groupings of observations. Note the data used in the fit
- can set to
"efron"for ols fits (only). Default is Huber-White estimator of the covariance matrix.
- a new fit object with the same class as the original fit,
and with the element
orig.varis the covariance matrix of the original fit. Also, the original
varcomponent is replaced with the new Huberized estimates.
ols fits do not have the corrected standard errors printed with
sqrt(diag(adjfit$var)) to get this, where
adjfit is the result of
- cluster sampling
- intra-class correlation
Huber, PJ. Proc Fifth Berkeley Symposium Math Stat 1:221--33, 1967.
White, H. Econometrica 50:1--25, 1982.
Lin, DY, Wei, LJ. JASA 84:1074--8, 1989.
Rogers, W. Stata Technical Bulletin STB-8, p. 15--17, 1992.
Rogers, W. Stata Release 3 Manual,
Long, JS, Ervin, LH. The American Statistician 54:217--224, 2000.
# A dataset contains a variable number of observations per subject, # and all observations are laid out in separate rows. The responses # represent whether or not a given segment of the coronary arteries # is occluded. Segments of arteries may not operate independently # in the same patient. We assume a "working independence model" to # get estimates of the coefficients, i.e., that estimates assuming # independence are reasonably efficient. The job is then to get # unbiased estimates of variances and covariances of these estimates. set.seed(1) n.subjects <- 30 ages <- rnorm(n.subjects, 50, 15) sexes <- factor(sample(c('female','male'), n.subjects, TRUE)) logit <- (ages-50)/5 prob <- plogis(logit) # true prob not related to sex id <- sample(1:n.subjects, 300, TRUE) # subjects sampled multiple times table(table(id)) # frequencies of number of obs/subject age <- ages[id] sex <- sexes[id] # In truth, observations within subject are independent: y <- ifelse(runif(300) <= prob[id], 1, 0) f <- lrm(y ~ lsp(age,50)*sex, x=TRUE, y=TRUE) g <- robcov(f, id) diag(g$var)/diag(f$var) # add ,group=w to re-sample from within each level of w anova(g) # cluster-adjusted Wald statistics # fastbw(g) # cluster-adjusted backward elimination plot(Predict(g, age=30:70, sex='female')) # cluster-adjusted confidence bands # Get design effects based on inflation of the variances when compared # with bootstrap estimates which ignore clustering g2 <- robcov(f) diag(g$var)/diag(g2$var) # Get design effects based on pooled tests of factors in model anova(g2)[,1] / anova(g)[,1] # A dataset contains one observation per subject, but there may be # heteroscedasticity or other model misspecification. Obtain # the robust sandwich estimator of the covariance matrix. # f <- ols(y ~ pol(age,3), x=TRUE, y=TRUE) # f.adj <- robcov(f)