frair (version 0.5)

frair_compare: Test the difference between two functional response fits

Description

Explicitly model, and then test, the difference between two functional response fits.

Usage

frair_compare(frfit1, frfit2, start = NULL)

Arguments

frfit1

An object of class frfit

frfit2

An object of class frfit

start

A named numeric list with starting values for the combined fit. See Details.

Value

Prints the results of the test to screen and invisibly returns on object of class frcompare inheriting from class(list) containing:

frfit1

The first FR fit, as supplied.

frfit2

The second FR fit, as supplied.

test_fit

The output of the re-fitted model.

result

Coefficients from the test that are otherwise printed to screen.

Details

This function provides a sensible test of the optimised coefficients between two functional responses fitted by frair_fit. This is achieved by explicitly modelling a 'difference' (delta) parameter for each optimised variable following the advice outlined in Juliano (2001). Briefly, consider the following Hollings type-II model:

a*X*T/(1+a*X*h)

the model containing delta parameters becomes:

(a-Da*grp)*X*T/(1+(a-Da*grp)*X*(h-Dh*grp))

where grp is a dummy coding variable and Da and Dh are the delta parameters. Here, the first functional response fit (frfit1) is coded grp=0 and the second fit (frfit2) is coded grp=1. ThereforeDa and Dh represent the difference between the two modelled coefficients and the standard output from the maximum likelihood fitting explicitly tests the null hypothesis of no difference between the two groups.

The difference model is re-fit to the combined data in frfit1 and frfit2 using the same maximum likelihood approaches outlined in frair_fit.

This function could be seen as a less computationally intensive alternative to bootstrapping but relies on mle2 successfully returning estimates of the standard error. mle2 does this by inverting a Hessian matrix, a procedure which might not always be successful.

Future versions of FRAIR will look to improve the integration between mle2 and allow users access to the various Hessian control parameters. In the meantime, the following delta functions are exported so users can 'roll their own' maximum likelihood implementation using this approach:

Original Function	Difference Function
Difference NLL Function	typeI
typeI_diff	typeI_nll_diff
hollingsII	hollingsII_diff
hollingsII_nll_diff	rogersII
rogersII_diff	rogersII_nll_diff
hassIII	hassIII_diff
hassIII_nll_diff	hassIIInr
hassIIInr_diff	hassIIInr_nll_diff
emdII	emdII_diff
emd_nll_diff	flexp
flexp_diff	flexp_nll_diff

References

Juliano SA (2001) Nonlinear curve fitting: Predation and functional response curves. In: Scheiner SM, Gurevitch J (eds). Design and analysis of ecological experiments. Oxford University Press, Oxford, United Kingdom. pp 178--196.

See Also

frair_fit

Examples

data(gammarus)

pulex <- gammarus[gammarus$spp=='G.pulex',]
celt <- gammarus[gammarus$spp=='G.d.celticus',]

pulexfit <- frair_fit(eaten~density, data=pulex, 
                response='rogersII', start=list(a = 1.2, h = 0.015), 
                fixed=list(T=1))
celtfit <- frair_fit(eaten~density, data=celt, 
                response='rogersII', start=list(a = 1.2, h = 0.015), 
                fixed=list(T=1))
# The following tests the null hypothesis that the 
# true difference between the coefficients is zero:
frair_compare(pulexfit, celtfit) # Reject null for 'h', do not reject for 'a'

## Not run: 
# # Provides a similar conclusion to bootstrapping followed by 95% CIs
# pulexfit_b <- frair_boot(pulexfit)
# celtfit_b <- frair_boot(celtfit)
# confint(pulexfit_b)
# confint(celtfit_b) # 'a' definitely overlaps
# ## End(Not run)