Model Selection and Multimodel Inference Based on (Q)AIC(c)

Functions to implement model selection and multimodel inference based on Akaike's information criterion (AIC) and the second-order AIC (AICc), as well as their quasi-likelihood counterparts (QAIC, QAICc) from various model object classes. The package implements classic model averaging for a given parameter of interest or predicted values, as well as a shrinkage version of model averaging parameter estimates or effect sizes. The package includes diagnostics and goodness-of-fit statistics for certain model types including those of 'unmarkedFit' classes estimating demographic parameters after accounting for imperfect detection probabilities. Some functions also allow the creation of model selection tables for Bayesian models of the 'bugs' and 'rjags' classes. Functions also implement model selection using BIC. Objects following model selection and multimodel inference can be formatted to LaTeX using 'xtable' methods included in the package.


News

####################################################################

THIS FILE DOCUMENTS CHANGES TO FUNCTIONS IN PACKAGE AICcmodavg

Author: Marc J. Mazerolle [email protected]

####################################################################

revision 2.1-1 (19 June 2017)

- created 'extractX' generic function to extract all predictors that appear

in a given list of candidate models and provide info on these variables

(added imports of getData from nlme and unmarked, and of formula for lme4)

- corrected minor bug for unmarkedFitPCO model class with 'modavgShrink'

to modify label of gamma depending on H0 (gamConst, gamAR, gamTrend)

- cleaned up modavgEffect

- added type = "certain" and "falsepos" as values for the 'parm.type' argument

of 'modavgPred', 'modavgEffect', 'modavgShrink', and 'modavg' methods for

objects of 'unmarkedFitOccuFP' class and added check for consistency of

parameter occurrence in different models

- added "iota" as a value for the 'parm.type' argument for 'modavgPred',

'modavgEffect', 'modavgShrink', and 'modavg' methods for objects of

'unmarkedFitPCO' class and added check for consistency of

parameter occurrence in different models

- added "detect" as a value for the 'parm.type' argument of 'modavgPred',

'modavgEffect', 'modavgShrink', and 'modavg' methods for objects of

'unmarkedFitDS' class

- added "detect" as a value for the 'parm.type' argument of 'modavgPred',

'modavgEffect', 'modavgShrink', and 'modavg' methods for objects of

'unmarkedFitGDS' class

####################################################################

revision 2.1-0 (17 November 2016)

- corrected minor bug in storing lower.CL in 'modavgPred' result

- modified 'fam.link.mer' to account for negative binomial with 'glmer.nb'

- corrected minor bug in rounding output of 'c_hat'

- corrected bug in 'countDist' with GDS objects including NA's

- modified ifelse( ) test when single site includes NA's in

'mb.chisq.unmarkedFitOccu' and 'mb.chisq.unmarkedFitColExt' (thanks to D. Linden)

- added drop = FALSE to preserve matrix in 'mb.chisq.unmarkedFitOccu' and

'mb.chisq.unmarkedFitColExt'

- added 'report' argument to print test statistic in terminal for functions

'mb.gof.test' and 'Nmix.gof.test'

- added 'checkConv' function to check convergence status of algorithm at

termination for different model classes

- added 'checkParms' function to identify parameters with high SE's in

different model classes

- modified gls method for 'predictSE' to extract model formula following

suggestion by C. R. Andersen

- corrected bug in creating final plot of theoretical distribution of sums of

chi-squares in mb.gof.test for dynamic occupancy models (thanks to M. Kéry)

- corrected bug in 'covDiag'

- added 'useBIC' and 'bictab' methods for different model objects based on

Bayesian Information Criterion

- added 'useBICCustom' function to compute BIC or QBIC from user-supplied

inputs (log-likelihood, K, n, c.hat)

- added 'bictabCustom' function to create model selection table using BIC or

QBIC from user-supplied inputs

- added methods for 'xtable' for objects of 'checkParms' and 'bictab' classes

- added functionality of 'evidence' with dictab and bictab objects

- added new 'useBIC' argument to 'modavgCustom' to conduct multimodel inference

based on BIC from user-supplied inputs

- removed functions that were moved to defunct during previous cycle of package

####################################################################

revision 2.0-4 (10 February 2016)

- updated email address

- modified 'mb.gof.test' to deal with dynamic occupancy data for entire

visits without data, and seasons with only a single visit (Thanks to G. Sadoti for comments)

- corrected bug for negative binomial models and dispersion argument in

'modavg', 'modavgShrink', 'modavgPred', 'modavgEffect', 'multComp'

- corrected bug on parameter count when negative binomial formated to

glm object with glm.convert( )

- updated functionality for overdispersion correction of ZIP N-mixture

models in unmarked with 'modavgPred'

- added new estimators of overdispersion in 'c_hat' for GLM's (deviance,

Farrington, and Fletcher) and print method

- added comment on help page of 'mb.gof.test' to clarify that the table

only shows the observed detection histories, but that unobserved

detection histories are included in the computations, and that the

chi-square test statistic should be compared against bootstrap

distribution because low expected frequencies invalidate chi-square

assumption.

- modified 'modavgEffect' to allow multiple variables to vary and print

generic group names

- modified 'modavgShrink' to allow cases where variables do not appear

with the same frequency in models, but issues a warning

- corrected bug in 'modavgShrink' to deal with 'unmarkedFitGDS' objects

on lambda

- added support for model selection of 'lavaan' class objects

- added support for model selection of fitdistr class objects

(package MASS)

- added support for model selection of fitdist class objects

(package fitdistrplus)

- added functionality to request CI's (back-transformed when using

type = "response") with 'modavgPred' and modified 'xtable.modavgPred'

- added covariance diagnostic for lambda in N-mixture models based on

Dennis et al. 2015 Biometrics 71: 237-246

- added 'detHist' function to summarize information on detection histories

- added 'countHist' function to summarize information on count data

- added 'countDist' function to summarize information on count data

collected by distance sampling

- added xtable methods for objects of 'detHist' and 'countHist' classes

- added fat and body measurement data set

- added GPA data set and test scores

- moved deprecated functions to defunct

- cleaned up print method of functions

####################################################################

revision 2.0-3 (15 January 2015)

- modified CITATION file to use meta$Date and meta$Version as requested by CRAN

- added the Red-spotted newt CMR data set (Gill 1985)

####################################################################

revision 2.0-2 (5 January 2015)

- added support for c-hat argument for dynamic occupancy models fit with 'colext' and additional mixture

models fit from 'distsamp', 'gdistsamp', 'gmultmix', 'gpcount', 'multinomPois', and 'occuRN'

- made 'Nmix.chisq' and 'Nmix.gof.test' generic and added support for different N-mixture models

and updated the Rd file

- made 'mb.chisq' and 'mb.gof.test' generic for unmarkedFitOccu and unmarkedFitColExt models,

modified structure of output, added new arguments 'print.seasons' for 'mb.gof.test.unmarkedFitColExt',

and updated the Rd file

- added new function to estimate c-hat in lme4 models and made 'c_hat' function generic and

updated the Rd file

- clarified examples in Rd file of 'modavg' to clearly illustrate how to exclude models when

model-averaging the main effect of a variable also involved in an interaction

- corrected bug in 'boot.wt' when sorting models in ascending order of Akaike weight

- created a separate class for 'boot.wt' objects and print method to facilitate LaTeX formatting

- created methods for 'xtable' to format objects from 'aictab', 'boot.wt', 'dictab', 'modavg', 'modavgShrink',

'modavgPred', 'modavgEffect', 'multComp', 'mb.chisq' to LaTeX

- modified CITATION file to use meta$Date and meta$Version as requested by CRAN

####################################################################

revision 2.01 (9 October 2014)

- corrected bugs in 'modavg' and 'modavgShrink' for models of class 'unmarkedFitGMM' and 'unmarkedFitGPC'

- corrected bugs in 'dictab' to return deviance of all models in model selection table

- added logical test to identify cases with only a single cohort without missing values for

'mb.chisq'

- added custom versions of 'AICc', 'aictabCustom', and 'modavgCustom' when user manually supplies log-Likelihood,

K, estimates, and SE for model selection and multimodel inference when model type is not supported or

SE's are not available from predict methods (convenient for importing output from other software such as PRESENCE)

- added support for hurdle models for count data for model selection and model averaging parameter estimates

- added support for model selection of models of the 'gnls' class

- added support for model selection and multimodel inference of models of the 'betareg' class

- added support for model selection and multimodel inference of models of the 'survreg' class

- modified 'multComp' for computation of critical values with z value instead of t for certain model types

- corrected minor bug of 'multComp' for some models with a factor consisting of two levels

- added 'extractCN' function for models of 'unmarkedFit' classes

- modified namespace

####################################################################

revision 2.00 (14 July 2014)

- modified predictSE.lme( ) to handle the use of formulas specified externally from model call

suggested by Clark Andersen

- added support for named lists of candidate models and add generic model names when modnames vector is NULL

in model selection and model averaging functions

- replaced functions 'extract.LL', 'modavg.effect', 'modavgpred', 'modavg.shrink', 'mult.comp', with 'extractLL',

'modavgEffect', 'modavgPred', 'modavgShrink', and 'multComp' respectively, to avoid problems with generic functions

- 'extractLL', 'extractSE', 'AICc', 'aictab', 'modavg', 'modavgEffect', 'modavgShrink', 'modavgPred',

'multComp', and predictSE made generic

- changed order of some arguments to present the most often used first

- added a check in 'multComp' for missing values in data set, and added options to correct confidence intervals for multiple comparisons

- added support for presence-only models of maxlikeFit class

- added support for models of classes 'coxme' and 'lmekin'

- added support for models of classes 'vglm' (package VGAM) and 'zeroinfl' (package pscl)

- added support for models of 'unmarkedFitGMM', 'unmarkedFitMPois', and unmarkedFitGPC classes

- added support for model averaging availability parameter (phi) for models of 'unmarkedFitGMM',

'unmarkedFitMPois', and unmarkedFitGPC classes

- corrected minor bug in some cases of missing values in 'Nmix.chisq'

- corrected minor bug in 'mb.gof.test' that produced warning( )

- added aov method for all generic functions of the package

- simplified Rd files AICc.rd, aictab.Rd, boot.wt.Rd, DIC.Rd, dictab.Rd, importance.Rd, modavg.Rd, modavg.effect.Rd

- added Bullfrog occupancy data set

- added Spotted salamander capture-mark-recapture data set

- added Gopher Tortoise distance sampling data set

####################################################################

revision 1.35 (17 November 2013)

- corrected minor bug in chi-square for N-mixture models in 'Nmix.chisq'

- corrected minor bug in title display, now uses as.expression(substitute( ))

####################################################################

revision 1.34 (13 November 2013)

- added functionality to compute model-averaged estimates (means)

with 'mult.comp'

- added functionality for clogit models

- added one-way ANOVA turkey data set

- added two-way ANOVA iron data set

- modified 'predictSE.zip' to handle offsets and added c.hat argument

- added functionality of overdispersion correction for N-mixture models

in 'aictab', 'modavg', 'modavgpred', 'modavg.effect'

'modavg.shrink', 'predictSE.zip'

- added functions 'Nmix.gof.test' and 'Nmix.chisq' to assess GOF of

N-mixture models

####################################################################

revision 1.33 (12 September 2013)

- added 'mult.comp' function to compute multiple comparisons on factors

- added 'boot.wt' function to compute bootstrapped relative frequencies

- corrected minor bug in 'mb.chisq' for certain patterns of missing values

- added conditional test to avoid problems on CRAN with some packages unavailable

on certain platforms

- added the lizards habitat preference data set

####################################################################

revision 1.32 (5 July 2013)

- corrected minor bug in 'modavg' when model-averaging with models of classes 'clm' and 'clmm' when the 'exclude' argument was used

- corrected bug with certain types of missing values in mb.chisq( )

- added function to extract DIC and create model selection tables based on DIC (functions 'DIC' and 'dictab')

####################################################################

revision 1.31 (26 June 2013)

- added functions to work with new 'merMod' classes of package lme4

####################################################################

revision 1.30 (16 May 2013)

- modified example code of predictSE.zip( ) to avoid errors on some platforms during package building

####################################################################

revision 1.29 (22 April 2013)

- added 'mb.chisq' to implement MacKenzie and Bailey (2004) goodness-of-fit test on single-season occupancy models

and 'mb.gof.test' to compute P-values, c-hat, and histogram of bootstrapped test statistics

- corrected bug for offsets in predictSE.mer( )

####################################################################

revision 1.28 (21 March 2013)

- added functionality for zip models to compute SE's of predictions on abundance of N-mixture models from unmarked

- added functionality for 'occuFP' models (detections with false-positives)

####################################################################

revision 1.27 (7 February 2013)

- implemented model-averaging estimates and predictions for 'gdistsamp' models

- added support for nonlinear mixed models of class 'nlme'

####################################################################

revision 1.26 (30 August 2012)

- added support for robust regression models of class 'rlm' - AICc, aictab, importance, modavg, modavg.effect, modavg.shrink, modavgpred

- added support for ordinal logistic regression models of classes 'clm' and 'clmm' - AICc, aictab, importance, modavg, modavg.effect, modavg.shrink

- modified modavg.unmarked( ) to enable treatment of gamma parameter in pcountOpen models

with different formulations (Const, Trend, AR)

- added reference of Luckacs et al. 2010 about model averaging with shrinkage in modavg.shrink.Rd file

- added references Burnham et al. 2011 for using model selection as alternative to multiple comparisons in modavg.effect.Rd file

- removed white space from formulas to reduce chance of reaching character limit in parse( ) in predictSE.mer( ), predictSE.lme( ), and predictSE.gls( )

- changed treatment of parm with modavg.unmarked to add object parm.strip (line 35) to enable functionality with regexpr( )

####################################################################

revision 1.25 (7 June 2012)

- added Cox proportional hazards model support

- added modavg.effect functions to compute model-averaged effect

sizes for true experiments as an information-theoretic analogue

to multiple comparisons

####################################################################

revision 1.24 (6 February 2012)

- corrected minor bug in AICc.nls( ) to determine number of observations

automatically when "second.ord = TRUE"

####################################################################

revision 1.23 (27 January 2012)

- corrected minor bug following revision 1.22 on predictSE.mer( )

to avoid problems with factors

####################################################################

revision 1.22 (19 January 2012)

- modified example in modavg( ) help file

- added error message for models without intercept with modavg( )

- added support for offset variable in predictSE.mer( ) for Poisson GLMM

- added support for 'distsamp' and 'gdistsamp' models from unmarked package

(Poisson or negative binomial abundance with log link)

- added example in modavg( ) and modavgpred( ) for 'distsamp' models

- added example for Poisson GLMM with log link and offset

####################################################################

revision 1.21 (4 November 2011)

- corrected error in computing sample size for binomial model with events/trials syntax in AICc.glm( ) and AICc.mer( )

####################################################################

revision 1.20 (3 October 2011)

- added functionality of 'nobs' argument for objects created

in unmarked package

####################################################################

revision 1.19 (14 September 2011)

- created function 'predictSE.gls' to compute predicted values

and SE's from 'gls' model fits and added functionality of

'modavgpred' with 'gls' model fit

- added documentation with examples for 'gls'

####################################################################

revision 1.18 (26 August 2011)

- added NAMESPACE

- modified functions 'extractSE.mer' and streamlined extraction of estimate

labels of 'mer' objects in 'modavg.mer', 'modavg.shrink.mer', 'importance'

####################################################################

revision 1.17 (17 June 2011)

- fixed minor bug for external calls to AICc to assign 'unmarkedFit' classes

- added functionality of 'AICc', 'aictab', 'evidence', 'importance', 'confset' with 'nls' fits

########################################################

revision 1.16 (27 May 2011)

- added model selection and multimodel inference functionality with

unmarked package for objects of classes unmarkedFitOccu,

unmarkedFitColExt, unmarkedFitPCount, unmarkedFitPCO, unmarkedFitOccuRN

for models incorporating detectability

- modifications of functions 'aictab', 'importance', 'confset', 'evidence',

'modavg', 'modavgpred', 'modavg.shrink', and extract.LL.unmarked utility

function added

- added new argument parm.type, to specify class of parameters for objects

of unmarkedFit classes

- added several examples for dealing with objects of 'unmarkedFit' classes

- modified 'modavg', 'modavg.shrink', 'importance', and 'reverse.exclude'

functions to strip white spaces within elements entered inadvertently

by user in the parm and exclude arguments

########################################################

revision 1.15 (21 April 2011)

- in 'modavg' and related functions, added functionality of dealing with specification of interaction terms (e.g., A:B, B:A, AB, BA)

by adding two new utility functions to recognize the terms when user is not consistent in his model syntax

- created function 'modavg.shrink' for the shrinkage version of model-averaging estimates to reduce model selection bias when

the same number of models include/exclude a variable

########################################################

revision 1.14 (11 March 2011)

- fixed minor bug in 'modavg.mer' regarding exclusion of interaction terms as 'formula' includes spaces within interaction terms

########################################################

revision 1.13 (10 January 2011)

- fixed minor bug in 'AICc.lme' on the computation of n in certain lme's with several levels

#######################################################

revision 1.12 (22 November 2010)

- fixed minor bug in 'modavgpred.lme' that generated error message when second.ord = FALSE

- fixed minor bug by dispatching 'mer' class objects to 'AICc.mer'

- fixed minor bug with incorrect computation of n for 'mer' models with binomial distribution using

cbind(incidence, size - incidence) or binary syntax

#######################################################

revision 1.11 (8 October 2010)

- fixed minor bug in 'modavgpred.glm' that generated an error message for Poisson, binomial, and negative

binomial distributions when 'c.hat' = 1

- added additional example in 'predictSE.mer' with gaussian distribution but with log link

#######################################################

revision 1.10 (7 October 2010)

- fixed minor bug in 'modavgpred.glm' for SE's of predictions from gaussian model

- fixed minor bug in 'modavg.glm' for SE's of beta estimates from negative binomial glm

#######################################################

revision 1.09 (16 September 2010)

- added function 'predictSE.mer' to compute predictions and SE's on predictions

accounting only for fixed effects (mean population prediction) from objects

of 'mer' class

- added functionality to 'modavgpred' to include computation of model-averaged

predictions from objects of class 'mer'

- added a check for link function to ensure same link function was used for all models in 'modavg' when

model averaging beta estimates

- created a utility function to determine family and link function for objects of 'mer' class called by

'modavg', 'modavgpred', and 'predictSE.mer'

- added a warning when some models have an identical (Q)AIC(c) and are possibly redundant in 'aictab'

- corrected minor bug in number of observations in 'modavgpred'

- added a check for binomial family with events/trials syntax to use sum of weights as number of observations

in 'AICc.mer' when nobs = NULL

#######################################################

revision 1.08 (16 August 2010)

- added Suggests "lme4" in DESCRIPTION file

#######################################################

revision 1.07 (24 May 2010)

- modified modavg to include 'nobs' argument in call to modavg.lme and modavg.gls

- added support for generalized linear mixed models of the mer class for AICc, aictab, modavg, evidence, importance, confset functions

- updated information on support for 'mer' class in help files

- added examples in some of the help files with 'mer'

#######################################################

revisions 1.06 (6 February 2010)

-added a 'polr' class check to 'AICc' function;

-added function 'predictSE.lme' to make predictions based on fixed effects for objects

of class 'lme' (mean population predictions) and compute associated SE's;

-added function 'modvagpred.lme' to compute model-averaged predictions and unconditional

standard errors for 'lme' models;

-added statement in Rd files for 'aictab', 'confset', 'evidence', 'importance', 'modavgpred',

'modavg', and generics 'nobs = NULL' defaults to total number of observations in

computations of information criteria;

-added NEWS file to document changes across versions of AICcmodavg package;

-added check that same response variable is used throughout model set for 'aictab' and generics

with lapply(X = cand.set, FUN = function(b) formula(b)[2]);

-added functionality of 'AICc', 'aictab', 'confset', 'evidence', 'importance', 'modavg' with

objects of 'gls' class;

-added additional flexibility to 'evidence' function when model selection table provided is not

sorted ('evidence' now sorts table internally) and set default comparison to top-ranked model

vs second-ranked model;

-added check/error message when no dispersion parameter (i.e., 'gamdisp') is specified for 'modavg'

and 'modavgpred' and generics with gamma regression models;

-added additional check/error message when c.hat < 1 in 'AICc.glm';

#######################################################

revisions 1.05 (3 December 2009):

-modified number of digits to print with 'importance' to be consistent with revision 1.04;

-added parenthesis where one was missing in aictab.Rd under value at line for (Q)AIC(c);

-corrected example in evidence.Rd for faulty name of model.low - should have been

'model.low = raw density';

-deleted CumWt when 'sort = FALSE' in 'aictab' - by modifying argument in 'print.aictab';

-for 'AICc.glm', 'AICc.mult', c.hat > 1 && c.hat <= 4 now gives warning;

-added a column for logLik, quasi-logLik, or restricted logLik in 'aictab';

-added options in 'print.aictab' to select items to display;

-added a check for 'exclude =' in 'modavg' and generics to warn if exclude is not a list:

if(!is.null(exclude)) {

if(!is.list(exclude)) stop(cat("\n'exclude' should be a list\n"))

}

-documented 'aictab' with new LL, Quasi.LL, Res.LL elements

-used 'vcov' and argument 'dispersion =' to scale SE's in a single step instead of adding

more code to scale it on separate lines - works with Poisson, Binomial, Gamma;

-in 'modavg.polr' and 'modavg.mult', added a logical test to distinguish between 'coef'

or zeta which are the intercepts;

-modified 'print.aictab' so that c_hat is printed only when c_hat appears in AICc table

(i.e, when any(names(x)=="c_hat"));

-modified 'print.modavg' so that c_hat is printed only when c_hat appears in AICc table

(i.e, when any(names(oldtab)=="c_hat")) AND deleted Cum.Wt from x$Mod.avg.table;

-added a check for the order for aictab and modavg and generics so that LL and other

measures are displayed properly in table and sorted at the very end (placed line of sort

and cumsum at very end of code) with Results[rev(order(Results[,5])), ];

-added Cum.Wt in last column of Results;

-changed order of elements with colnames so that prints nicely - this affects 'importance',

'confset', and 'evidence' - number of columns used might differ depending on what is

requested;

-in 'modavg' and generics, changed 'regexpr' to determine whether parameter appears alone

in the model and added 'fixed = TRUE', to find matches "as is" to avoid problems with

'mult' (e.g., "well|mild");

#######################################################

revision 1.04 (11 November 2009):

-added the new equation for unconditional SE based on equation 6.12 of Burnham and

Anderson (2002) in 'modavg' and 'modavgpred' and new argument 'uncond.se' with default

value "revised" - 'uncond.se = old' uses equation 4.19;

-added a check to determine whether all link functions are identical for 'modavg'

as averaging over beta estimates under different link functions is not appropriate

and returns error);

-added a check to determine whether all link functions are identical for 'modavgpred'

when linear predictor is used;

-changed default to display only 2 digits after decimal point for all output 'print'

generic functions;

#######################################################

revision 1.03 (21 October 2009):

-replaced 'regexpr' in 'modavg.glm', 'modavg.lme', 'modavg.mult' with 'match';

-add functionality for 'polr';

-cleaned up code with indentations;

-added 'exclude' argument to manually exclude models with interaction terms/quadratic

terms in 'modavg';

-added 'warn' argument (to suppress warning that variables with similar names appear

in a given model) to 'modavg' function. This is a fundamental difference with the

previous versions of the function as now the user must specify the variables (as a list)

to exclude models for the computation of model-averaged estimates (case when

interaction/polynomial terms are present). This modication also solved the

problem of variables with similar names;

#######################################################

revision 1.02 (29 September 2009):

-corrected small bug for computation of AICc (in 'AICc.glm') for binomial models with

events/trials syntax to use the correct sample size;

-added function to compute (Q)AIC(c) for multinomial models and updated 'aictab'

function accordingly;

-added citation file;

-changed 'Depends' to 'Suggests' for packages MASS, nnet, nlme;

-modified 'importance' to include 'multinom' objects;

-removed dependencies with nlme to avoid potential problems with lme4;

#######################################################

revision 1.01:

-cleaned up documentation file, added a few more sentences and added space within formulas

for the example to display properly in pdf;

Reference manual

It appears you don't have a PDF plugin for this browser. You can click here to download the reference manual.

install.packages("AICcmodavg")

2.1-1 by Marc J. Mazerolle, 2 years ago


Browse source code at https://github.com/cran/AICcmodavg


Authors: Marc J. Mazerolle <[email protected]>.


Documentation:   PDF Manual  


GPL (>= 2) license


Imports methods, stats, graphics, lattice, MASS, Matrix, nlme, stats4, survival, unmarked, VGAM, xtable

Suggests betareg, coxme, fitdistrplus, lavaan, lme4, maxlike, nnet, ordinal, pscl, R2jags, R2OpenBUGS, R2WinBUGS


Imported by gasfluxes, lmSupport, temperatureresponse.

Suggested by AMModels, GSIF, PSIMEX, RSA, bbmle.


See at CRAN