gof {qfasar}

Goodness-of-fit for modeled predator signatures

Description

The function gof uses estimated diet compositions and bootstrap resampling of the prey library to construct a statistic that may conservatively indicate predator fatty acid signatures that were not accurately modeled during diet estimation.

Usage

gof(
  prey_sigs,
  prey_loc,
  mean_sigs,
  diet_est,
  conv,
  obj_func,
  dist_meas = 1,
  gamma = 1,
  boot_gof = 500
)

Arguments

prey_sigs

A matrix of prey signatures in the optimization space used for diet estimation. Intended to be the object prey_sigs returned by the function est_diet.

prey_loc

A matrix giving the first and last locations of the signatures of each prey type within prey_sigs. Intended to be the object loc returned by the function prep_sig if diets were estimated using an unpartitioned prey library or make_prey_part if diets were estimated using a partitioned library.

mean_sigs

A numeric matrix of mean prey-type signatures in the optimization space used for diet estimation. Intended to be the object mean_sigs returned by the function est_diet.

diet_est

A numeric matrix of estimated diet compositions. Intended to be the object est_ind returned by the function est_diet.

conv

A logical vector indicating whether the optimization function successfully converged during diet estimation. Intended to be the object conv returned by the function est_diet.

obj_func

A numeric vector of the value of the minimized objective function for each predator. Intended to be the object obj_func returned by the function est_diet.

dist_meas

An integer indicator of the distance measure used for diet estimation. Default value 1.

gamma

The power parameter of the chi-square distance measure. Default value 1.

boot_gof

The number of bootstrap replications to use. Default value 500.

Value

A list containing the following elements:

gof_ss

The number of diet estimates that converged for each predator, therefore producing a simulated value of the objective function.

p_val

The proportion of the simulated objective function values that exceeded the value produced during diet estimation.

err_code

An integer error code (0 if no error is detected).

err_message

A string contains a brief summary of the execution.

Details

Diet estimation involves modeling an observed predator fatty acid signature as a mixture of prey signatures. However, methods to assess how well predator signatures are modeled have received little attention in the literature (but see Bromaghin et al. 2015).

One byproduct of diet estimation is the value of the distance measure that is minimized during diet estimation (est_diet), called the objective function. If a predator signature is accurately modeled, the value of the objective function will be relatively small. Conversely, the more poorly the signature is approximated, the larger the objective function will tend to be. However, what value of the objective function to use as a warning flag for a potentially poor fit is not clear.

The function gof represents one attempt to answer this question. The algorithm is based on the following logic. First, we assume that a predator consumes the mixture of prey specified by its estimated diet composition. Given that assumption, the expected value of the objective function is, in a sense, fixed (Bromaghin 2015). Large values of the objective function are then most likely to occur when variation in a predator signature, which results from the selection of individual prey within prey types, is maximized. Within the framework of simulating predator signatures, variation in the signatures is maximized when the bootstrap sample sizes of the prey signatures used to construct a predator signature are minimized (Bromaghin et al. 2016).

Implementing the above logic, gof randomly samples a single prey signature from each prey type and weights the resulting signatures with a predator's estimated diet composition to construct a modeled signature. The modeled signature is then used to estimate diet. If the optimization function converges, the value of the objective function obtained with the modeled signature is compared to the value of the objective function obtained while estimating diet with the observed signature (argument obj_func. This is repeated boot_gof times and the proportion of the simulated objective function values that exceed the observed objective function value is computed. gof therefore constructs a statistic similar to a p-value, with small values being suggestive of a predator signature that was not closely approximated during diet estimation.

NOTE: the method implemented in gof is at this point only an idea whose performance has not been explored. It has been included in qfasar to support future research on this topic.

References

Bromaghin, J.F. 2015. Simulating realistic predator signatures in quantitative fatty acid signature analysis. Ecological Informatics 30:68-71.

Bromaghin, J.F., S.M. Budge, and G.W. Thiemann. 2016. Should fatty acid signature proportions sum to 1 for diet estimation? Ecological Research 31:597-606.

Bromaghin, J.F., K.D. Rode, S.M. Budge, and G.W. Thiemann. 2015. Distance measures and optimization spaces in quantitative fatty acid signature analysis. Ecology and Evolution 5:1249-1262.

Examples

gof(prey_sigs = matrix(c(0.06, 0.09, 0.31, 0.54,
                         0.05, 0.09, 0.30, 0.56,
                         0.03, 0.10, 0.30, 0.57,
                         0.08, 0.07, 0.30, 0.55,
                         0.09, 0.05, 0.33, 0.53,
                         0.09, 0.06, 0.34, 0.51,
                         0.09, 0.07, 0.34, 0.50,
                         0.08, 0.11, 0.35, 0.46,
                         0.06, 0.14, 0.36, 0.44), ncol = 9),
    prey_loc = matrix(c(1, 4, 7, 3, 6, 9), ncol=2),
    mean_sigs = matrix(c(0.047, 0.093, 0.303, 0.557,
                         0.087, 0.050, 0.323, 0.530,
                         0.077, 0.106, 0.350, 0.467), ncol = 3),
    diet_est = matrix(c(0.394, 0.356, 0.250,
                        0.336, 0.365, 0.299), ncol = 2),
    conv = c(TRUE, TRUE),
    obj_func = c(1.13, 2.24),
    dist_meas = 1,
    boot_gof = 10)