RRPP (version 2.1.2)

pairwise.model.Z: Pairwise comparisons of model effects

Description

Function generates pairwise statistics for comparing model fits and returns important statistics for hypothesis tests.

Usage

pairwise.model.Z(
  ...,
  nsamp = NULL,
  two.tailed = TRUE,
  predictor = NULL,
  tol = NULL,
  pc.no = NULL,
  gls.null = FALSE
)

Value

A list containing the following

sample.z

A vector of model effect sizes.

pairwise.z

A matrix of pairwise test statistics comparing model effect sizes.

pairwise.P

A matrix of pairwise significance levels.

tails

Number of tails used for P-value calculation.

Arguments

...

Either an object of class model.comparison, or several objects of class lm.rrpp. If the former, arguments of type = 'Z' and verbose = TRUE are required. If the latter, a model comparsion analysis will first be performed with these arguments.

nsamp

An optional vector containing the sample sizes for each model fit

two.tailed

A logical value to indicate whether a two-tailed test (typical and default) should be performed.

predictor

An optional vector to be passed to model.comparison, if used.

tol

An optional value to be passed to model.comparison, if used.

pc.no

An optional value to be passed to model.comparison, if used.

gls.null

An optional logical value to be passed to model.comparison, if used.

Author

Dean Adams and Michael Collyer

Details

The function statistically compares the effect sizes of two or more models fit and evaluated using RRPP. Input for the function is a list of fitted models of the class 'model.comparison', whose options included type = 'Z' and verbose = TRUE when the models were compared with that function.

A two-sample test is performed on each pair of models, comparing the strength of model fits to one another (Collyer and Adams 2025). This might be used to compare the strength of fit of the data to differing statistical models (as in model selection) or for comparing the fit across differing datasets for the same model to determine whether the strength of fit in one dataset is greater than that found in another (see Collyer and Adams 2025). In the latter case, one is advised to include a vector containing the sample sizes of each dataset, so that two-sample tests may account for differences in sample size.

References

Collyer, M.L., and D.C. Adams. 2025. Permutational Biometry. Volume 1: Univariate Data. Iowa State University Digital Press. (Forthcoming).

Examples

if (FALSE) {
 data(Pupfish)
 Pupfish$logSize <- log(Pupfish$CS)
 fit1 <- lm.rrpp(coords ~ logSize, data = Pupfish,
 print.progress = FALSE)
 fit2 <- lm.rrpp(coords ~ Pop, data = Pupfish,
 print.progress = FALSE)
 fit3 <- lm.rrpp(coords ~ Sex, data = Pupfish, 
 print.progress = FALSE)
 fit6 <- lm.rrpp(coords ~ logSize + Sex * Pop, data = Pupfish, 
 print.progress = FALSE)
 Mod.C <- model.comparison(fit1, fit2, fit3, fit6,
 pc.no = 4, type = "Z", verbose = TRUE)
 res <- pairwise.model.Z(Mod.C)
 summary(res, stats.table = TRUE)
 summary(res, stats.table = FALSE)
 }