dprime_posthoc: Post-hoc estimates and tests for multiple discrimination...

In perbrock/sensR: Thurstonian Models for Sensory Discrimination

Post-hoc estimates and tests for multiple discrimination experiments.

Description

This function provides estimates and p-values for post-hoc tests such as pairwise comparisons. p-values are (by default) adjusted for multiplicity.

Usage

posthoc(x, ...)

## S3 method for class 'dprime_compare'
posthoc(x, alpha = 0.05,
    test = c("pairwise", "common", "base", "zero"), base = 1,
    alternative = c("two.sided", "less", "greater"),
    statistic = c("likelihood", "Wald"),
    padj.method = c("holm", "bonferroni", "none"), ...)

## S3 method for class 'dprime_test'
posthoc(x, alpha = 0.05,
    test = c("pairwise", "common", "base", "zero"), base = 1,
    alternative = c("two.sided", "less", "greater"),
    statistic = c("likelihood", "Wald"),
    padj.method = c("holm", "bonferroni", "none"), ...)

Arguments

x	an object of class dprime_compare or dprime_test.
alpha	the significance level for tests and confidence intervals.
test	the type of post-hoc tests performed. Se the details section for further details.
base	when test = "base", the experiment against which to provide pairwise comparisons.
alternative	direction of the hypothesis test.
statistic	The test statistic used - currently there is only partial support for statistic = "likelihood".
padj.method	controls the method by which p-values are adjusted for multiplicity. Any one of the values in p.adjust.methods (currently "holm" "hochberg" "hommel" "bonferroni" "BH" "BY" "fdr" "none") may be specified, cf. p.adjust.
...	currently not used.

Details

The test argument specifies the type of test performed. "pairwise" performs all pairwise comparisons and produces a compact letter display indicating groups of experiments that different/not-different. "common" tests, for each experiment in turn, if the by-experiment d-prime is different from a common d-prime computed from the remaining experiments. "base" provides pairwise comparisons to a single experiment indicated by the separate argument base. If test = "zero" all d-primes are tested versus zero. As a final option a numeric value can be supplied, e.g. test = 1 in which case all d-primes are tested versus one. Note that test = 0 gives the same test as test = "zero".

When test = "pairwise" a compact letter display is provided and it is determined from the p-values after adjustment of these for multiplicity.

The dprime_compare and dprime_test methods a have (common) print method.

Value

an object of class c(paste0("posthoc.", class(x)), class(x)) with the following elements from the original object, x and :

posthoc	coefficient table for the post-hoc tests.
test	the value of the test argument.
alternative	the value of the alternative argument.
padj.method	the method used to adjust p-values with.
base	the value of the base argument.
posthoc.stat	name of the statistic for the post-hoc tests.
Letters	if test = "pairwise" the compact letter display, otherwise NULL.
dprime0	unless test = "pairwise" or "common" the value of d-prime under the null hypothesis.

Author(s)

Rune Haubo B Christensen

See Also

dprime_test, dprime_table, dprime_compare.

Examples

## Make some fake data:
n <- rep(40, 4)
x <- c(25, 25, 30, 35)
protocol <- c("triangle", "duotrio", "threeAFC", "twoAFC")
## Look at the data table with d-primes etc.:
dprime_table(x, n, protocol)

## 'any differences' test:
## ML estimation and test with likelihood statistic:
(dpc <- dprime_compare(x, n, protocol))

posthoc(dpc, alpha=.1) ## test="pairwise"

## Test if each d' is different from the common d' estimated from the
## remaining experiments:
posthoc(dpc, test="common")

## Test if d' from experiment 2 is different from the others (with
## adjustment for multiplicity):
posthoc(dpc, test="base", base=2)

## Test if each d' is different from 2 (with Bonferroni adjustment for
## multiplicity) using the Wald statistic:
posthoc(dpc, test=2, stat="Wald", padj.method="bonferroni")

perbrock/sensR documentation built on Nov. 5, 2023, 10:41 a.m.