Documentation for methods for class frboot

1 2 3 4 5 6 7 8 9 10 11 12 13 | ```
## S3 method for class 'frboot'
print(x, ...)
## S3 method for class 'frboot'
confint(object, parm='all', level=0.95, ..., citypes='all')
## S3 method for class 'frboot'
plot(x, xlab=x$xvar, ylab=x$yvar, ...)
## S3 method for class 'frboot'
lines(x, all_lines=FALSE, tozero=FALSE, bootcol=1, bootalpha=1/sqrt(x$n_boot), ...)
## S3 method for class 'frboot'
drawpoly(x, ..., probs=c(0.025, 0.975), tozero=FALSE)
## S3 method for class 'frconf'
print(x, ...)
``` |

`x, object ` |
Output from a call to |

`parm ` |
A character vector. Which parameter to get CIs for? See Details. |

`level ` |
A numeric. The confidence limit for CIs. |

`citypes ` |
A character vector. What kind of CI? See Details. |

`all_lines ` |
A logical. Should the bootstrapped results be plotted? See Details. |

`tozero ` |
A logical. Should the line be drawn to the origin? See Details. |

`xlab ` |
Label for the x-axis. |

`ylab ` |
Label for the y-axis. |

`bootcol ` |
A valid colour for the bootstrapped lines. |

`bootalpha ` |
A numeric (0-1). A transparency for the (inevitably overlapping) lines. |

`probs ` |
Lower and upper tails for confidence interval polygons. See |

`... ` |
Other items passed to underlying functions. |

This documents standard methods for FRAIR objects of class `frboot`

. However, because standard naming conventions have been used, some undocumented utility functions might also work (e.g. `coefficients`

)

The code underlying `confint.frboot`

is quite complex and relies heavily on the excellent work done by Brian Ripley in `boot.ci`

. Some of the complexity of `boot.ci`

is hidden, but, like all FRAIR objects you can access the original method by passing the output directly (e.g. `boot.ci(object$fit)`

).

Like `print.bootci`

the `print()`

method for objects produced by `print.frboot`

will report potentially unstable intervals. However, these are calculated and returned by `confint.frboot`

, not when `print()`

is called (see Value, below). When calling `confint.frboot`

you can request (a combination of) different intervals. The default `'all'`

is equivalent to `c('norm', 'basic', 'stud', 'perc', 'bca')`

which are the Normal approximation, Basic, Studentised, Percentile and BCa intervals, respectively. Each has strengths and weaknesses which the user should be aware of.

`lines`

and `drawpoly`

only add lines or polygons to an existing plot, so an active graphics device needs to be present. By default `all`

is FALSE. The simple reason for this is because the code is a little slow (on some devices), so currently it is an 'opt-in' option.

`drawpoly`

draws empirical confidence intervals. The intervals are constructed by evaluating every set of bootstrapped coefficients at:

`seq(from=min(x$x), to=max(x$x), length.out = 50)`

.

and then calculating the empirical confidence limits at each value of *x* by:

`apply(val, 2, quantile, na.rm=T, probs=probs)`

Note that this is a rough approximation of a bootstrapped confidence interval and does not account for some of the intricacies (e.g. bootstrap bias) described in boot.ci.

Note also, that if `tozero`

is TRUE, then both `lines`

and `drawpoly`

attempt to draw to zero by evaluating every set of bootstrapped coefficients at:

`seq(from=0, to=max(x$x), length.out = 50)`

If the coefficients provided by a fit to the orginal data produce a value that is undefined at zero, then these functions will plot lines to a small, non-zero number (1e-04) instead (with a warning). However, this does not guarantee that all of the values produced by the bootstrapped coefficients will produce sensible values. Clearly the intention here is to provide a nice-looking representation of the fitted curve and it is up to the user to determine *why* their desired curve is undefined at zero.

`confint.frboot`

returns a nested list with m items at the top level and n items at the second level, where m is the number of coefficients and n is the number of types of confidence intervals. Each named object at the second level is a list containing:

`lower ` |
The upper limit. |

`upper ` |
The lower limit. |

`bootciout ` |
The output from |

and optionally:

`errors ` |
The error(s) encountered by |

`warnings ` |
The warning(s) encountered by |

`notes ` |
A comment on potential instability of intervals, if justified. |

These last two items combine 'true' warnings and the tests for interval stability described in `print.bootci`

.

All confidence intervals are calculated on the original scale. If you want to calculate intervals on a transformed scale, call `boot.ci`

directly using the `boot.ci(object$fit)`

syntax.

Daniel Pritchard

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 | ```
# This example is not run to save CRAN build server time...
## Not run:
data(gammarus)
# Holling's is the wrong fit for these data based on the experimental design
# But it runs more quickly, so is a useful demonstration
outhol <- frair_fit(eaten~density, data=gammarus, response='hollingsII',
start=list(a = 1, h = 0.08), fixed=list(T=40/24))
outholb <- frair_boot(outhol)
confint(outholb)
# Illustrate bootlines
plot(outholb, xlim=c(0,30), type='n', main='All bootstrapped lines')
lines(outholb, all_lines=TRUE)
points(outholb, pch=20, col=rgb(0,0,0,0.2))
# Illustrate bootpolys
plot(outholb, xlim=c(0,30), type='n', main='Empirical 95 percent CI')
drawpoly(outholb, col=rgb(0,0.5,0))
points(outholb, pch=20, col=rgb(0,0,0,0.2))
## End(Not run)
``` |

Questions? Problems? Suggestions? Tweet to @rdrrHQ or email at ian@mutexlabs.com.

Please suggest features or report bugs with the GitHub issue tracker.

All documentation is copyright its authors; we didn't write any of that.