[R] Mininum number of resamples required to do BCa bootstrap?

[Robert A. LaBudde]

I'm using R 2.15.1 on a 64-bit machine with Windows 7 Home Premium
and package 'boot'.

I've found that using a number of bootstrap resamples in boot() that
is less than the number of data results in a fatal error. Once the
number of resamples meets or exceeds the number of data, the error disappears.

Sample problem (screwy subscripted syntax is a relic of edited down a

N <- 25
s <- rlnorm(N, 0, 1)
require("boot")

Loading required package: boot

v <- NULL # hold sample variance estimates
i <- 1
v[i] <- var(s) # get sample variance
nReal <- 10
varf <- function (x,i) { var(x[i]) }
fabc <- function (x, w) { # weighted average (biased) variance

+ sum(x^2 * w) / sum(w) - (sum(x * w) / sum(w))^2
+ }

p <- c(.25, .75, .2, .8, .15, .85, .1, .9, .05, .95, .025, .975,

.005, .995)

cl <- c(0.5, 0.6, 0.7, 0.8, 0.9, 0.95, 0.99)
b <- boot(s, varf, R = nReal) # bootstrap
bv <- NULL # hold bootstrap mean variance estimates
bias <- NULL #hold bias estimates
bv[i] <- mean(b$t) # bootstrap mean variance
bias[i] <- bv[i] - v[i] # bias estimate
bCI90 <- boot.ci(b, conf = 0.90)

Error in bca.ci(boot.out, conf, index[1L], L = L, t = t.o, t0 = t0.o, :
estimated adjustment 'a' is NA
In addition: Warning messages:
1: In norm.inter(t, (1 + c(conf, -conf))/2) :
extreme order statistics used as endpoints
2: In boot.ci(b, conf = 0.9) :
bootstrap variances needed for studentized intervals
3: In norm.inter(t, alpha) : extreme order statistics used as endpoints

nReal <- 25
b <- boot(s, varf, R = nReal) # bootstrap
bv[i] <- mean(b$t) # bootstrap mean variance
bias[i] <- bv[i] - v[i] # bias estimate
bCI90 <- boot.ci(b, conf = 0.90)

Warning messages:
1: In boot.ci(b, conf = 0.9) :
bootstrap variances needed for studentized intervals
2: In norm.inter(t, adj.alpha) :
extreme order statistics used as endpoints

The problem is that doing 10 resamples generates an NA in the
estimation of the 'acceleration' in the function abc.ci(), but doing
25 resamples does not. This implies a connection between the number
of resamples and the 'acceleration' which should not exist.
('Acceleration' should be obtained from the original sample via
jackknife as 1/6 the coefficient of skewness.)

The script apparently works correctly if the number of resamples
equals or exceeds the number of original data, but not otherwise.


================================================================
Robert A. LaBudde, PhD, PAS, Dpl. ACAFS e-mail: ral at lcfltd.com
Least Cost Formulations, Ltd. URL: http://lcfltd.com/
824 Timberlake Drive Tel: [phone]
Virginia Beach, VA 23464-3239 Fax: [phone]

"Vere scire est per causas scire"

[Robert A. LaBudde]

I'm using R 2.15.1 on a 64-bit machine with Windows 7 Home Premium
and package 'boot'.

I've found that using a number of bootstrap resamples in boot() that
is less than the number of data results in a fatal error. Once the
number of resamples meets or exceeds the number of data, the error disappears.

Sample problem (screwy subscripted syntax is a relic of edited down a

N <- 25
s <- rlnorm(N, 0, 1)
require("boot")

Loading required package: boot

v <- NULL # hold sample variance estimates
i <- 1
v[i] <- var(s) # get sample variance
nReal <- 10
varf <- function (x,i) { var(x[i]) }
fabc <- function (x, w) { # weighted average (biased) variance

+ sum(x^2 * w) / sum(w) - (sum(x * w) / sum(w))^2
+ }

p <- c(.25, .75, .2, .8, .15, .85, .1, .9, .05, .95, .025, .975,

.005, .995)

cl <- c(0.5, 0.6, 0.7, 0.8, 0.9, 0.95, 0.99)
b <- boot(s, varf, R = nReal) # bootstrap
bv <- NULL # hold bootstrap mean variance estimates
bias <- NULL #hold bias estimates
bv[i] <- mean(b$t) # bootstrap mean variance
bias[i] <- bv[i] - v[i] # bias estimate
bCI90 <- boot.ci(b, conf = 0.90)

Error in bca.ci(boot.out, conf, index[1L], L = L, t = t.o, t0 = t0.o, :
estimated adjustment 'a' is NA
In addition: Warning messages:
1: In norm.inter(t, (1 + c(conf, -conf))/2) :
extreme order statistics used as endpoints
2: In boot.ci(b, conf = 0.9) :
bootstrap variances needed for studentized intervals
3: In norm.inter(t, alpha) : extreme order statistics used as endpoints

nReal <- 25
b <- boot(s, varf, R = nReal) # bootstrap
bv[i] <- mean(b$t) # bootstrap mean variance
bias[i] <- bv[i] - v[i] # bias estimate
bCI90 <- boot.ci(b, conf = 0.90)

Warning messages:
1: In boot.ci(b, conf = 0.9) :
bootstrap variances needed for studentized intervals
2: In norm.inter(t, adj.alpha) :
extreme order statistics used as endpoints

The problem is that doing 10 resamples generates an NA in the
estimation of the 'acceleration' in the function abc.ci(), but doing
25 resamples does not. This implies a connection between the number
of resamples and the 'acceleration' which should not exist.
('Acceleration' should be obtained from the original sample via
jackknife as 1/6 the coefficient of skewness.)

The script apparently works correctly if the number of resamples
equals or exceeds the number of original data, but not otherwise.


================================================================
Robert A. LaBudde, PhD, PAS, Dpl. ACAFS e-mail: ral at lcfltd.com
Least Cost Formulations, Ltd. URL: http://lcfltd.com/
824 Timberlake Drive Tel: [phone]
Virginia Beach, VA 23464-3239 Fax: [phone]

"Vere scire est per causas scire"