bootstrap {FSA}R Documentation

Case bootstrapping and associated S3 methods.

Description

The bootCase function was added to FSA to maintain backward compatability (because bootCase was removed from car), mostly for users of the Introductory Fisheries Analyses with R book. bootCase is largerly a wrapper to Boot from car with method="case". It is suggested that Boot from car be used instead. S3 methods are also provided to construct non-parametric bootstrap confidence intervals, predictions with non-parametric confidence intervals, hypothesis tests, and plots of the parameter estimates for bootCase objects.

Usage

bootCase(object, f. = stats::coef, B = R, R = 999)

## S3 method for class 'bootCase'
confint(
  object,
  parm = NULL,
  level = conf.level,
  conf.level = 0.95,
  plot = FALSE,
  err.col = "black",
  err.lwd = 2,
  rows = NULL,
  cols = NULL,
  ...
)

## S3 method for class 'bootCase'
predict(object, FUN, conf.level = 0.95, digits = NULL, ...)

## S3 method for class 'bootCase'
htest(
  object,
  parm = NULL,
  bo = 0,
  alt = c("two.sided", "less", "greater"),
  plot = FALSE,
  ...
)

## S3 method for class 'bootCase'
hist(
  x,
  same.ylim = TRUE,
  ymax = NULL,
  rows = round(sqrt(ncol(x))),
  cols = ceiling(sqrt(ncol(x))),
  ...
)

## S3 method for class 'bootCase'
plot(x, ...)

Arguments

object, x

A regression object of type lm, glm or class nls for bootCase or a bootCase object for the S3 methods.

f.

A function that will be applied to the updated regression object to compute the statistics of interest. The default is coef, to return to regression coefficient estimates.

B, R

Number of bootstrap samples.

parm

A number or string that indicates which column of object contains the parameter estimates to use for the confidence interval or hypothesis test.

level

Same as conf.level.

conf.level

A level of confidence as a proportion.

plot

A logical that indicates whether a plot should be constructed. If confint then a histogram of the parm parameters from the bootstrap samples with error bars that illustrate the bootstrapped confidence intervals will be constructed. If codehtest then a histogram of the parm parameters with a vertical line illustrating the bo value will be constructed.

err.col

A single numeric or character that identifies the color for the error bars on the plot.

err.lwd

A single numeric that identifies the line width for the error bars on the plot.

rows

A single numeric that contains the number of rows to use on the graphic.

cols

A single numeric that contains the number of columns to use on the graphic.

...

Additional items to send to functions. See details.

FUN

The function to be applied for the prediction. See the examples.

digits

A single numeric that indicates the number of digits for the result.

bo

The null hypothesized parameter value.

alt

A string that indicates the “direction” of the alternative hypothesis. See details.

same.ylim

A logical that indicates whether the same limits for the y-axis should be used on each histogram. Defaults to TRUE. Ignored if ylmts is non-null.

ymax

A single value that sets the maximum y-axis limit for each histogram or a vector of length equal to the number of groups that sets the maximum y-axis limit for each histogram separately.

col

A named color for the histogram bars.

Details

confint finds the two quantiles that have the (1-conf.level)/2 proportion of bootstrapped parameter estimates below and above. This is an approximate 100conf.level% confidence interval.

predict applies a user-supplied function to each row of object and then finds the median and the two quantiles that have the proportion (1-conf.level)/2 of the bootstrapped predictions below and above. The median is returned as the predicted value and the quantiles are returned as an approximate 100conf.level% confidence interval for that prediction. Values for the independent variable in FUN must be a named argument sent in the ... argument (see examples). Note that if other arguments are needed in FUN besides values for the independent variable, then these are included in the ... argument AFTER the values for the independent variable.

In htest the “direction” of the alternative hypothesis is identified by a string in the alt= argument. The strings may be "less" for a “less than” alternative, "greater" for a “greater than” alternative, or "two.sided" for a “not equals” alternative (the DEFAULT). In the one-tailed alternatives the p-value is the proportion of bootstrapped parameter estimates in object$coefboot that are extreme of the null hypothesized parameter value in bo. In the two-tailed alternative the p-value is twice the smallest of the proportion of bootstrapped parameter estimates above or below the null hypothesized parameter value in bo.

Value

If object is a matrix, then confint returns a matrix with as many rows as columns (i.e., parameter estimates) in object and two columns of the quantiles that correspond to the approximate confidence interval. If object is a vector, then confint returns a vector with the two quantiles that correspond to the approximate confidence interval.

htest returns a two-column matrix with the first column containing the hypothesized value sent to this function and the second column containing the corresponding p-value.

hist constructs histograms of the bootstrapped parameter estimates.

plot constructs scatterplots of all pairs of bootstrapped parameter estimates.

predict returns a matrix with one row and three columns, with the first column holding the predicted value (i.e., the median prediction) and the last two columns holding the approximate confidence interval.

Author(s)

Derek H. Ogle, derek@derekogle.com

References

S. Weisberg (2005). Applied Linear Regression, third edition. New York: Wiley, Chapters 4 and 11.

See Also

Boot in car.

Examples

fnx <- function(days,B1,B2,B3) {
  if (length(B1) > 1) {
    B2 <- B1[2]
    B3 <- B1[3]
    B1 <- B1[1]
  }
  B1/(1+exp(B2+B3*days))
}
nl1 <- nls(cells~fnx(days,B1,B2,B3),data=Ecoli,
           start=list(B1=6,B2=7.2,B3=-1.45))
nl1.bootc <- bootCase(nl1,coef,B=99)  # B=99 too few to be useful
confint(nl1.bootc,"B1")
confint(nl1.bootc,c(2,3))
confint(nl1.bootc,conf.level=0.90)
confint(nl1.bootc,plot=TRUE)
predict(nl1.bootc,fnx,days=1:3)
predict(nl1.bootc,fnx,days=3)
htest(nl1.bootc,1,bo=6,alt="less")
hist(nl1.bootc)
plot(nl1.bootc)
cor(nl1.bootc)


[Package FSA version 0.8.26.9000 Index]