validate: Reject Implausible Statistics

Description

Simulation output may contain rogue values due to idiosyncracies of model fitting. For example, nonidentifiability due to inadequate data can result in spurious extreme `estimates' of the sampling variance. Undue influence of rogue replicates can be reduced by using the median as a summary field rather than the mean. This function is another way to deal with the problem, by setting to NA selected statistics from replicates for which some `test' statistic is out-of-range.

Usage

validate(x, test, validrange = c(0, Inf), targets = test, quietly = FALSE)

Value

An object of class c(`selectedstatistics', secrdesign', `list') with the same structure and header information as the input, but possibly with some values in the `output' component converted to NA.

Arguments

x: object that inherits from `selectedstatistics'
test: character; name of statistic to check
validrange: numeric vector comprising the minimum and maximum permitted values of `test', or a matrix (see details)
targets: character vector with names of one or more statistics to set to missing (NA) when test is out-of-range
quietly: logical; if TRUE messages are suppressed

Details

Values of `test' and `targets' should be columns in each component `replicate x statistic' matrix (i.e., scenario) of x$output. You can check for these with find.stats.

If validrange is a matrix its first and second columns are interpreted as scenario-specific bounds (minima and maxima), and the number of rows must match the number of scenarios.

If all non-missing values of `test' are in the valid range, the effect is to force the target statistics to NA wherever `test' is NA.

The default is to change only the test field itself. If the value of `test' does not appear in `targets' then the test field is unchanged.

If targets = "all" then all columns are set to NA when the test fails.

Examples

Run this code


if (FALSE) {

## generate some data
scen1 <- make.scenarios(D = c(5,10), sigma = 25, g0 = 0.2)
traps1 <- make.grid()
tmp1 <- run.scenarios(nrepl = 5, trapset = traps1, scenarios = scen1,
    fit = TRUE, extractfn = trim)
tmp2 <- predict(tmp1)
tmp3 <- select.stats(tmp2, 'D', c('estimate','RB','RSE','COV'))

## just for demonstration --
## apply scenario-specific +/- 20% bounds for estimated density
## set RB, RSE and COV to NA when estimate is outside this range
permitted <- outer(tmp3$scenarios$D, c(0.8,1.2)) 
permitted   ## a 2 x 2 matrix
tmp4 <- validate(tmp3, 'estimate', permitted, c('RB', 'RSE','COV'))

## what have we done?!
tmp4$output
summary(tmp4)

}