fr_flexpnr: Scaling Exponent Response, not assuming replacement
In dpritchard/frair: Tools for Functional Response Analysis

Description Usage Arguments Details Author(s) References See Also Examples

Scaling exponent response (not assuming replacement) based on ideas dating back to Real (1977, at least)

    flexpnr_fit(data, samp, start, fixed, boot=FALSE, windows=FALSE)
    flexpnr_nll(b, q, h, T, X, Y)
    flexpnr(X, b, q, h, T)

`data`	A dataframe containing X and Y.
`samp`	A vector specifying the rows of data to use in the fit. Provided by `boot()` or manually, as required.
`start`	A named list. Starting values for items to be optimised. Usually 'a' and 'h'.
`fixed`	A names list. 'Fixed data' (not optimised). Usually 'T'.
`boot`	A logical. Is the function being called for use by `boot()`?
`windows`	A logical. Is the operating system Microsoft Windows?
`b, q, h`	The search coefficient (b), scaling exponent (q) and the handling time (h). Usually items to be optimised.
`T`	T, the total time available.
`X`	The X variable. Usually prey density.
`Y`	The Y variable. Usually the number of prey consumed.

This combines a type-II non-replacement functional response (i.e. a Roger's random predator equation) with a scaling exponent on the capture rate (a). This function is generalised from that described in flexp relaxing the assumption that prey are replaced throughout the experiment.

The capture rate (a) follows the following relationship:

a = b*X^q

and then (a) is used to calculate the number of prey eaten (Ne) following the same relationship as rogersII:

Ne=N0*(1-exp(a*(Ne*h-T)))

where b is a search coefficient and other coefficients are as defined in rogersII. Because Ne appears on both side of the equation, the solution is found using Lambert's transcendental equation. FRAIR uses the lambertW0 function from the lamW package and the internal function is:

Ne <- X - lambertW0(a * h * X * exp(-a * (T - h * X)))/(a * h)

where X = N0. When q = 0, then a = b and the relationship collapses to traditional type-II Rogers' random predator equation. There is, therefore, a useful test on q = 0 in the summary of the fit.

None of these functions are designed to be called directly, though they are all exported so that the user can call them directly if desired. The intention is that they are called via frair_fit, which calls them in the order they are specified above.

flexpnr_fit does the heavy lifting and also pulls double duty as the statistic function for bootstrapping (via boot() in the boot package). The windows argument if required to prevent needless calls to require(frair) on platforms that can manage sane parallel processing.

The core fitting is done by mle2 from the bbmle package and users are directed there for more information. mle2 uses the flexpnr_nll function to optimise flexpnr.

Further references and recommended reading can be found on the help page for frair_fit.

Daniel Pritchard

Real LA (1977) The Kinetics of Functional Response. The American Naturalist 111: 289–300.

frair_fit, flexp.

# A 'type-II' example
data(gammarus)

rogfit <- frair_fit(eaten~density, data=gammarus, 
                response='rogersII', start=list(a = 1.2, h = 0.015), 
                fixed=list(T=40/24))
expofit <- frair_fit(eaten~density, data=gammarus, 
                response='flexpnr', start=list(b = 1.2, q = 0, h = 0.015), 
                fixed=list(T=40/24))
## Plot
plot(rogfit)
lines(rogfit)
lines(expofit, col=2)

## Inspect
summary(rogfit$fit)
summary(expofit$fit) # No evidence that q is different from zero...
AIC(rogfit$fit)
AIC(expofit$fit) # The exponent model is *not* preferred

# A 'type-III' example
data(bythotrephes)

rogfit <- frair_fit(eaten~density, data=bythotrephes, 
                response='rogersII', start=list(a = 1.2, h = 0.015), 
                fixed=list(T=12/24))
expofit <- frair_fit(eaten~density, data=bythotrephes, 
                response='flexpnr', start=list(b = 1.2, q = 0, h = 0.015), 
                fixed=list(T=12/24))
## Plot
plot(rogfit)
lines(rogfit)
lines(expofit, col=2)

## Inspect
summary(rogfit$fit)
summary(expofit$fit) # Some evidence that q is different from zero...
AIC(rogfit$fit)
AIC(expofit$fit) # The exponent model is preferred