modelsearch {lefko3}

Develop Best-fit Vital Rate Estimation Models for MPM Development

Description

Function modelsearch() runs exhaustive model building and selection for each vital rate needed to estimate a function-based MPM or IPM. It returns best-fit models for each vital rate, model table showing all models tested, and model quality control data. The final output can be used as input in other functions within this package.

Usage

modelsearch(
  data,
  stageframe = NULL,
  historical = TRUE,
  approach = "mixed",
  suite = "size",
  bestfit = "AICc&k",
  vitalrates = c("surv", "size", "fec"),
  surv = c("alive3", "alive2", "alive1"),
  obs = c("obsstatus3", "obsstatus2", "obsstatus1"),
  size = c("sizea3", "sizea2", "sizea1"),
  sizeb = c(NA, NA, NA),
  sizec = c(NA, NA, NA),
  repst = c("repstatus3", "repstatus2", "repstatus1"),
  fec = c("feca3", "feca2", "feca1"),
  stage = c("stage3", "stage2", "stage1"),
  matstat = c("matstatus3", "matstatus2", "matstatus1"),
  indiv = "individ",
  patch = NA,
  year = "year2",
  density = NA,
  test.density = FALSE,
  sizedist = "gaussian",
  sizebdist = NA,
  sizecdist = NA,
  fecdist = "gaussian",
  size.zero = FALSE,
  sizeb.zero = FALSE,
  sizec.zero = FALSE,
  size.trunc = FALSE,
  sizeb.trunc = FALSE,
  sizec.trunc = FALSE,
  fec.zero = FALSE,
  fec.trunc = FALSE,
  patch.as.random = TRUE,
  year.as.random = TRUE,
  juvestimate = NA,
  juvsize = FALSE,
  jsize.zero = FALSE,
  jsizeb.zero = FALSE,
  jsizec.zero = FALSE,
  jsize.trunc = FALSE,
  jsizeb.trunc = FALSE,
  jsizec.trunc = FALSE,
  fectime = 2,
  censor = NA,
  age = NA,
  test.age = FALSE,
  indcova = NA,
  indcovb = NA,
  indcovc = NA,
  random.indcova = FALSE,
  random.indcovb = FALSE,
  random.indcovc = FALSE,
  test.indcova = FALSE,
  test.indcovb = FALSE,
  test.indcovc = FALSE,
  test.group = FALSE,
  show.model.tables = TRUE,
  global.only = FALSE,
  accuracy = TRUE,
  data_out = FALSE,
  quiet = FALSE
)

Arguments

data

The vertical dataset to be used for analysis. This dataset should be of class hfvdata, but can also be a data frame formatted similarly to the output format provided by functions verticalize3() or historicalize3(), as long as all needed variables are properly designated.

stageframe

The stageframe characterizing the life history model used. Optional unless test.group = TRUE, in which case it is required. Defaults to NULL.

historical

A logical variable denoting whether to assess the effects of state in occasion t-1, in addition to state in occasion t. Defaults to TRUE.

approach

The statistical approach to be taken for model building. The default is "mixed", which uses the mixed model approach utilized in packages lme4 and glmmTMB. Other options include "glm", which uses generalized linear modeling assuming that all factors are fixed.

suite

Either a single string value or a vector of 14 strings for each vital rate model. Describes the global model for each vital rate estimation, and has the following possible values: full, includes main effects and all two-way interactions of size and reproductive status; main, includes main effects only of size and reproductive status; size, includes only size (also interactions between size in historical model); rep, includes only reproductive status (also interactions between status in historical model); age, all vital rates estimated with age and y-intercepts only; cons, all vital rates estimated only as y-intercepts. If approach = "glm" and year.as.random = FALSE, then year is also included as a fixed effect, and, in the case of full, included in two-way interactions. Order of models in the string vector if more than 1 value is used is: 1) survival, 2) observation, 3) primary size, 4) secondary size, 5) tertiary size, 6) reproductive status, 7) fecundity, 8) juvenile survival, 9) juvenile observation, 10) juvenile primary size, 11) juvenile secondary size, 12) juvenile tertiary size, 13) juvenile reproductive status, and 14) juvenile maturity status. Defaults to size.

bestfit

A variable indicating the model selection criterion for the choice of best-fit model. The default is AICc&k, which chooses the best-fit model as the model with the lowest AICc or, if not the same model, then the model that has the lowest degrees of freedom among models with \Delta AICc <= 2.0. Alternatively, AICc may be chosen, in which case the best-fit model is simply the model with the lowest AICc value.

vitalrates

A vector describing which vital rates will be estimated via linear modeling, with the following options: surv, survival probability; obs, observation probability; size, overall size; repst, probability of reproducing; and fec, amount of reproduction (overall fecundity). May also be set to vitalrates = "leslie", which is equivalent to setting c("surv", "fec") for a Leslie MPM. This choice also determines how internal data subsetting for vital rate model estimation will work. Defaults to c("surv", "size", "fec").

surv

A vector indicating the variable names coding for status as alive or dead in occasions t+1, t, and t-1, respectively. Defaults to c("alive3", "alive2", "alive1").

obs

A vector indicating the variable names coding for observation status in occasions t+1, t, and t-1, respectively. Defaults to c("obsstatus3", "obsstatus2", "obsstatus1").

size

A vector indicating the variable names coding for the primary size variable on occasions t+1, t, and t-1, respectively. Defaults to c("sizea3", "sizea2", "sizea1").

sizeb

A vector indicating the variable names coding for the secondary size variable on occasions t+1, t, and t-1, respectively. Defaults to c(NA, NA, NA), in which case sizeb is not used.

sizec

A vector indicating the variable names coding for the tertiary size variable on occasions t+1, t, and t-1, respectively. Defaults to c(NA, NA, NA), in which case sizec is not used.

repst

A vector indicating the variable names coding for reproductive status in occasions t+1, t, and t-1, respectively. Defaults to c("repstatus3", "repstatus2", "repstatus1").

fec

A vector indicating the variable names coding for fecundity in occasions t+1, t, and t-1, respectively. Defaults to c("feca3", "feca2", "feca1").

stage

A vector indicating the variable names coding for stage in occasions t+1, t, and t-1. Defaults to c("stage3", "stage2", "stage1").

matstat

A vector indicating the variable names coding for maturity status in occasions t+1, t, and t-1. Defaults to c("matstatus3", "matstatus2", "matstatus1").

indiv

A text value indicating the variable name coding individual identity. Defaults to "individ".

patch

A text value indicating the variable name coding for patch, where patches are defined as permanent subgroups within the study population. Defaults to NA.

year

A text value indicating the variable coding for observation occasion t. Defaults to year2.

density

A text value indicating the name of the variable coding for spatial density, should the user wish to test spatial density as a fixed factor affecting vital rates. Defaults to NA.

test.density

Either a logical value indicating whether to include density as a fixed categorical variable in linear models, or a logical vector of such values for 14 models, in order: 1) survival, 2) observation, 3) primary size, 4) secondary size, 5) tertiary size, 6) reproductive status, 7) fecundity, 8) juvenile survival, 9) juvenile observation, 10) juvenile primary size, 11) juvenile secondary size, 12) juvenile tertiary size, 13) juvenile reproductive status, and 14) juvenile maturity status. Defaults to FALSE.

sizedist

The probability distribution used to model primary size. Options include "gaussian" for the Normal distribution (default), "poisson" for the Poisson distribution, "negbin" for the negative binomial distribution (quadratic parameterization), and "gamma" for the Gamma distribution.

sizebdist

The probability distribution used to model secondary size. Options include "gaussian" for the Normal distribution, "poisson" for the Poisson distribution, "negbin" for the negative binomial distribution (quadratic parameterization), and "gamma" for the Gamma distribution. Defaults to NA.

sizecdist

The probability distribution used to model tertiary size. Options include "gaussian" for the Normal distribution, "poisson" for the Poisson distribution, "negbin" for the negative binomial distribution (quadratic parameterization), and "gamma" for the Gamma distribution. Defaults to NA.

fecdist

The probability distribution used to model fecundity. Options include "gaussian" for the Normal distribution (default), "poisson" for the Poisson distribution, "negbin" for the negative binomial distribution (quadratic parameterization), and "gamma" for the Gamma distribution.

size.zero

A logical variable indicating whether the primary size distribution should be zero-inflated. Only applies to Poisson and negative binomial distributions. Defaults to FALSE.

sizeb.zero

A logical variable indicating whether the secondary size distribution should be zero-inflated. Only applies to Poisson and negative binomial distributions. Defaults to FALSE.

sizec.zero

A logical variable indicating whether the tertiary size distribution should be zero-inflated. Only applies to Poisson and negative binomial distributions. Defaults to FALSE.

size.trunc

A logical variable indicating whether the primary size distribution should be zero-truncated. Only applies to Poisson and negative binomial distributions. Defaults to FALSE. Cannot be TRUE if size.zero = TRUE.

sizeb.trunc

A logical variable indicating whether the secondary size distribution should be zero-truncated. Only applies to Poisson and negative binomial distributions. Defaults to FALSE. Cannot be TRUE if sizeb.zero = TRUE.

sizec.trunc

A logical variable indicating whether the tertiary size distribution should be zero-truncated. Only applies to Poisson and negative binomial distributions. Defaults to FALSE. Cannot be TRUE if sizec.zero = TRUE.

fec.zero

A logical variable indicating whether the fecundity distribution should be zero-inflated. Only applies to Poisson and negative binomial distributions. Defaults to FALSE.

fec.trunc

A logical variable indicating whether the fecundity distribution should be zero-truncated. Only applies to the Poisson and negative binomial distributions. Defaults to FALSE. Cannot be TRUE if fec.zero = TRUE.

patch.as.random

If set to TRUE and approach = "mixed", then patch is included as a random factor. If set to FALSE and approach = "glm", then patch is included as a fixed factor. All other combinations of logical value and approach lead to patch not being included in modeling. Defaults to TRUE.

year.as.random

If set to TRUE and approach = "mixed", then year is included as a random factor. If set to FALSE, then year is included as a fixed factor. All other combinations of logical value and approach lead to year not being included in modeling. Defaults to TRUE.

juvestimate

An optional variable denoting the stage name of the juvenile stage in the vertical dataset. If not NA, and stage is also given (see below), then vital rates listed in vitalrates other than fec will also be estimated from the juvenile stage to all adult stages. Defaults to NA, in which case juvenile vital rates are not estimated.

juvsize

A logical variable denoting whether size should be used as a term in models involving transition from the juvenile stage. Defaults to FALSE, and is only used if juvestimate does not equal NA.

jsize.zero

A logical variable indicating whether the primary size distribution of juveniles should be zero-inflated. Only applies to Poisson and negative binomial distributions. Defaults to FALSE.

jsizeb.zero

A logical variable indicating whether the secondary size distribution of juveniles should be zero-inflated. Only applies to Poisson and negative binomial distributions. Defaults to FALSE.

jsizec.zero

A logical variable indicating whether the tertiary size distribution of juveniles should be zero-inflated. Only applies to Poisson and negative binomial distributions. Defaults to FALSE.

jsize.trunc

A logical variable indicating whether the primary size distribution in juveniles should be zero-truncated. Defaults to FALSE. Cannot be TRUE if jsize.zero = TRUE.

jsizeb.trunc

A logical variable indicating whether the secondary size distribution in juveniles should be zero-truncated. Defaults to FALSE. Cannot be TRUE if jsizeb.zero = TRUE.

jsizec.trunc

A logical variable indicating whether the tertiary size distribution in juveniles should be zero-truncated. Defaults to FALSE. Cannot be TRUE if jsizec.zero = TRUE.

fectime

A variable indicating which year of fecundity to use as the response term in fecundity models. Options include 2, which refers to occasion t, and 3, which refers to occasion t+1. Defaults to 2.

censor

A vector denoting the names of censoring variables in the dataset, in order from occasion t+1, followed by occasion t, and lastly followed by occasion t-1. Defaults to NA.

age

Designates the name of the variable corresponding to age in time t in the vertical dataset. Defaults to NA, in which case age is not included in linear models. Should only be used if building Leslie or age x stage matrices.

test.age

Either a logical value indicating whether to include age as a fixed categorical variable in linear models, or a logical vector of such values for 14 models, in order: 1) survival, 2) observation, 3) primary size, 4) secondary size, 5) tertiary size, 6) reproductive status, 7) fecundity, 8) juvenile survival, 9) juvenile observation, 10) juvenile primary size, 11) juvenile secondary size, 12) juvenile tertiary size, 13) juvenile reproductive status, and 14) juvenile maturity status. Defaults to FALSE.

indcova

Vector designating the names in occasions t+1, t, and t-1 of an individual covariate. Defaults to NA.

indcovb

Vector designating the names in occasions t+1, t, and t-1 of a second individual covariate. Defaults to NA.

indcovc

Vector designating the names in occasions t+1, t, and t-1 of a third individual covariate. Defaults to NA.

random.indcova

A logical value indicating whether indcova should be treated as a random categorical factor, rather than as a fixed factor. Defaults to FALSE.

random.indcovb

A logical value indicating whether indcovb should be treated as a random categorical factor, rather than as a fixed factor. Defaults to FALSE.

random.indcovc

A logical value indicating whether indcovc should be treated as a random categorical factor, rather than as a fixed factor. Defaults to FALSE.

test.indcova

Either a logical value indicating whether to include the indcova variable as a fixed categorical variable in linear models, or a logical vector of such values for 14 models, in order: 1) survival, 2) observation, 3) primary size, 4) secondary size, 5) tertiary size, 6) reproductive status, 7) fecundity, 8) juvenile survival, 9) juvenile observation, 10) juvenile primary size, 11) juvenile secondary size, 12) juvenile tertiary size, 13) juvenile reproductive status, and 14) juvenile maturity status. Defaults to FALSE.

test.indcovb

Either a logical value indicating whether to include the indcovb variable as a fixed categorical variable in linear models, or a logical vector of such values for 14 models, in order: 1) survival, 2) observation, 3) primary size, 4) secondary size, 5) tertiary size, 6) reproductive status, 7) fecundity, 8) juvenile survival, 9) juvenile observation, 10) juvenile primary size, 11) juvenile secondary size, 12) juvenile tertiary size, 13) juvenile reproductive status, and 14) juvenile maturity status. Defaults to FALSE.

test.indcovc

Either a logical value indicating whether to include the indcovc variable as a fixed categorical variable in linear models, or a logical vector of such values for 14 models, in order: 1) survival, 2) observation, 3) primary size, 4) secondary size, 5) tertiary size, 6) reproductive status, 7) fecundity, 8) juvenile survival, 9) juvenile observation, 10) juvenile primary size, 11) juvenile secondary size, 12) juvenile tertiary size, 13) juvenile reproductive status, and 14) juvenile maturity status. Defaults to FALSE.

test.group

Either a logical value indicating whether to include the group variable from the input stageframe as a fixed categorical variable in linear models, or a logical vector of such values for 14 models, in order: 1) survival, 2) observation, 3) primary size, 4) secondary size, 5) tertiary size, 6) reproductive status, 7) fecundity, 8) juvenile survival, 9) juvenile observation, 10) juvenile primary size, 11) juvenile secondary size, 12) juvenile tertiary size, 13) juvenile reproductive status, and 14) juvenile maturity status. Defaults to FALSE.

show.model.tables

If set to TRUE, then includes full modeling tables in the output. Defaults to TRUE.

global.only

If set to TRUE, then only global models will be built and evaluated. Defaults to FALSE.

accuracy

A logical value indicating whether to test accuracy of models. See Notes section for details on how accuracy is assessed. Defaults to TRUE.

data_out

A logical value indicating whether to append all subsetted datasets used in model building and selection to the output. Defaults to FALSE.

quiet

May be a logical value, or any one of the strings "yes", "no", or "partial". If set to TRUE or "yes", then model building and selection will proceed with most warnings and diagnostic messages silenced. If set to FALSE or "no", then all warnings and diagnostic messages will be displayed. If set to "partial", then only messages related to transitions between different vital rate models will be displayed. Defaults to FALSE.

Value

This function yields an object of class lefkoMod, which is a list in which the first 14 elements are the best-fit models for survival, observation status, primary size, secondary size, tertiary size, reproductive status, fecundity, juvenile survival, juvenile observation, juvenile primary size, juvenile secondary size, juvenile tertiary size, juvenile transition to reproduction, and juvenile transition to maturity, respectively. This is followed by 14 elements corresponding to the model tables for each of these vital rates, in order, followed by a data frame showing the order and names of variables used in modeling, followed by a single character element denoting the criterion used for model selection, and ending on a data frame with quality control data:

survival_model

Best-fit model of the binomial probability of survival from occasion t to occasion t+1. Defaults to 1.

observation_model

Best-fit model of the binomial probability of observation in occasion t+1 given survival to that occasion. Defaults to 1.

size_model

Best-fit model of the primary size metric on occasion t+1 given survival to and observation in that occasion. Defaults to 1.

sizeb_model

Best-fit model of the secondary size metric on occasion t+1 given survival to and observation in that occasion. Defaults to 1.

sizec_model

Best-fit model of the tertiary size metric on occasion t+1 given survival to and observation in that occasion. Defaults to 1.

repstatus_model

Best-fit model of the binomial probability of reproduction in occasion t+1, given survival to and observation in that occasion. Defaults to 1.

fecundity_model

Best-fit model of fecundity in occasion t+1 given survival to, and observation and reproduction in that occasion. Defaults to 1.

juv_survival_model

Best-fit model of the binomial probability of survival from occasion t to occasion t+1 of an immature individual. Defaults to 1.

juv_observation_model

Best-fit model of the binomial probability of observation in occasion t+1 given survival to that occasion of an immature individual. Defaults to 1.

juv_size_model

Best-fit model of the primary size metric on occasion t+1 given survival to and observation in that occasion of an immature individual. Defaults to 1.

juv_sizeb_model

Best-fit model of the secondary size metric on occasion t+1 given survival to and observation in that occasion of an immature individual. Defaults to 1.

juv_sizec_model

Best-fit model of the tertiary size metric on occasion t+1 given survival to and observation in that occasion of an immature individual. Defaults to 1.

juv_reproduction_model

Best-fit model of the binomial probability of reproduction in occasion t+1, given survival to and observation in that occasion of an individual that was immature in occasion t. This model is technically not a model of reproduction probability for individuals that are immature, rather reproduction probability here is given for individuals that are mature in occasion t+1 but immature in occasion t. Defaults to 1.

juv_maturity_model

Best-fit model of the binomial probability of becoming mature in occasion t+1, given survival to that occasion of an individual that was immature in occasion t. Defaults to 1.

survival_table

Full dredge model table of survival probability.

observation_table

Full dredge model table of observation probability.

size_table

Full dredge model table of the primary size variable.

sizeb_table

Full dredge model table of the secondary size variable.

sizec_table

Full dredge model table of the tertiary size variable.

repstatus_table

Full dredge model table of reproduction probability.

fecundity_table

Full dredge model table of fecundity.

juv_survival_table

Full dredge model table of immature survival probability.

juv_observation_table

Full dredge model table of immature observation probability.

juv_size_table

Full dredge model table of primary size in immature individuals.

juv_sizeb_table

Full dredge model table of secondary size in immature individuals.

juv_sizec_table

Full dredge model table of tertiary size in immature individuals.

juv_reproduction_table

Full dredge model table of immature reproduction probability.

juv_maturity_table

Full dredge model table of the probability of an immature individual transitioning to maturity.

paramnames

A data frame showing the names of variables from the input data frame used in modeling, their associated standardized names in linear models, and a brief comment describing each variable.

criterion

Character variable denoting the criterion used to determine the best-fit model.

qc

Data frame with five variables: 1) Name of vital rate, 2) number of individuals used to model that vital rate, 3) number of individual transitions used to model that vital rate, 4) parameter distribution used to model the vital rats, and 5) accuracy of model, given as detailed in Notes section.

subdata

An optional list of data frames, each of which is the data frame used to develop each model in the lefkoMod object, in order.

Notes

When modelsearch() is called, it first trims the dataset down to just the variables that will be used (including all response terms and independent variables). It then subsets the data to only complete cases for those variables. Next, it builds global models for all vital rates, and runs them. If a global model fails, then the function proceeds by dropping terms. If approach = "mixed", then it will determine which random factor term contains the most categories in the respective subset, and drop that term. If this fails, or if approach = "glm", then it will drop any two-way interactions and run the model. If this fails, then the function will attempt to drop further terms, first patch alone, then year alone, then individual covariates by themselves, then combinations of these four, and finally individual identity. If all of these attempts fail and the approach used is mixed, then the function will try running a glm version of the original failed model. Finally, if all attempts fail, then the function displays a warning and returns 1 to allow model building assuming a constant rate or probability.

Setting suite = "cons" prevents the inclusion of size and reproductive status as fixed, independent factors in modeling. However, it does not prevent any other terms from being included. Density, age, individual covariates, individual identity, patch, and year may all be included.

The mechanics governing model building are fairly robust to errors and exceptions. The function attempts to build global models, and simplifies models automatically should model building fail. Model building proceeds through the functions lm() (GLM with Gaussian response), glm() (GLM with Poisson, Gamma, or binomial response), glm.nb() (GLM with negative binomial response), zeroinfl() (GLM with zero-inflated Poisson or negative binomial response), vglm() (GLM with zero-truncated Poisson or negative binomial response), lmer() (mixed model with Gaussian response), glmer() (mixed model with binomial, Poisson, or Gamma response), and glmmTMB() (mixed model with negative binomial, or zero-truncated or zero-inflated Poisson or negative binomial response). See documentation related to these functions for further information. Any response term that is invariable in the dataset will lead to a best-fit model for that response represented by a single constant value.

Exhaustive model building and selection proceeds via the dredge() function in package MuMIn. This function is verbose, so that any errors and warnings developed during model building, model analysis, and model selection can be found and dealt with. Interpretations of errors during global model analysis may be found in documentation for the functions and packages mentioned. Package MuMIn is used for model dredging (see dredge()), and errors and warnings during dredging can be interpreted using the documentation for that package. Errors occurring during dredging lead to the adoption of the global model as the best-fit, and the user should view all logged errors and warnings to determine the best way to proceed. The quiet = TRUE and quiet = "partial" options can be used to silence dredge warnings, but users should note that automated model selection can be viewed as a black box, and so care should be taken to ensure that the models run make biological sense, and that model quality is prioritized.

Exhaustive model selection through dredging works best with larger datasets and fewer tested parameters. Setting suite = "full" may initiate a dredge that takes a dramatically long time, particularly if the model is historical, individual covariates are used, or a zero-inflated distribution is assumed. In such cases, the number of models built and tested will run at least in the millions. Small datasets will also increase the error associated with these tests, leading to adoption of simpler models overall. Note also that zero-inflated models are processed as two models, and so include twice the assumed number of parameters. If suite = "full", then this function will switch to a main effects global model for the zero-inflated parameter models if the total number of parameters to test rises above the limits imposed by the dredge() function in package MuMIn.

Accuracy of vital rate models is calculated differently depending on vital rate and assumed distribution. For all vital rates assuming a binomial distribution, including survival, observation status, reproductive status, and juvenile version of these, accuracy is calculated as the percent of predicted responses equal to actual responses. In all other models, accuracy is actually assessed as a simple R-squared in which the observed response values per data subset are compared to the predicted response values according to each best-fit model. Note that some situations in which factor variables are used may result in failure to assess accuracy. In these cases, function modelsearch() simply yields NA values.

Care must be taken to build models that test the impacts of state in occasion t-1 for historical models, and that do not test these impacts for ahistorical models. Ahistorical matrix modeling particularly will yield biased transition estimates if historical terms from models are ignored. This can be dealt with at the start of modeling by setting historical = FALSE for the ahistorical case, and historical = TRUE for the historical case.

This function handles generalized linear models (GLMs) under zero-inflated distributions using the zeroinfl() function, and zero- truncated distributions using the vglm() function. Model dredging may fail with these functions, leading to the global model being accepted as the best-fit model. However, model dredges of mixed models work for all distributions. We encourage the use of mixed models in all cases.

The negative binomial and truncated negative binomial distributions use the quadratic structure emphasized in Hardin and Hilbe (2018, 4th Edition of Generalized Linear Models and Extensions). The truncated negative binomial distribution may fail to predict size probabilities correctly when dispersion is near that expected of the Poisson distribution. To prevent this problem, we have integrated a cap on the overdispersion parameter. However, when using this distribution, please check the matrix column sums to make sure that they do not predict survival greater than 1.0. If they do, then please use either the negative binomial distribution or the zero-truncated Poisson distribution.

If density dependence is explored through function modelsearch(), then the interpretation of density is not the full population size but rather the spatial density term included in the dataset.

Users building vital rate models for Leslie matrices must set vitalrates = c("surv", "fec") or vitalrates = "leslie" rather than the default, because only survival and fecundity should be estimated in these cases. Also, the suite setting can be set to either age or cons, as the results will be exactly the same.

Users wishing to test age, density, group, or individual covariates, must include test.age = TRUE, test.density = TRUE, test.group = TRUE, or test.indcova = TRUE (or test.indcovb = TRUE or test.indcovc = TRUE, whichever is most appropriate), respectively, in addition to stipulating the name of the variable within the dataset. The default for these options is always FALSE.

Examples

data(lathyrus)

sizevector <- c(0, 4.6, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 1, 2, 3, 4, 5, 6, 7, 8,
  9)
stagevector <- c("Sd", "Sdl", "Dorm", "Sz1nr", "Sz2nr", "Sz3nr", "Sz4nr",
  "Sz5nr", "Sz6nr", "Sz7nr", "Sz8nr", "Sz9nr", "Sz1r", "Sz2r", "Sz3r", 
  "Sz4r", "Sz5r", "Sz6r", "Sz7r", "Sz8r", "Sz9r")
repvector <- c(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1)
obsvector <- c(0, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1)
matvector <- c(0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1)
immvector <- c(1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0)
propvector <- c(1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 
  0)
indataset <- c(0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1)
binvec <- c(0, 4.6, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 
  0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5)

lathframeln <- sf_create(sizes = sizevector, stagenames = stagevector, 
  repstatus = repvector, obsstatus = obsvector, matstatus = matvector, 
  immstatus = immvector, indataset = indataset, binhalfwidth = binvec, 
  propstatus = propvector)

lathvertln <- verticalize3(lathyrus, noyears = 4, firstyear = 1988,
  patchidcol = "SUBPLOT", individcol = "GENET", blocksize = 9, 
  juvcol = "Seedling1988", sizeacol = "lnVol88", repstracol = "Intactseed88",
  fecacol = "Intactseed88", deadacol = "Dead1988", 
  nonobsacol = "Dormant1988", stageassign = lathframeln, stagesize = "sizea",
  censorcol = "Missing1988", censorkeep = NA, NAas0 = TRUE, censor = TRUE)

lathvertln$feca2 <- round(lathvertln$feca2)
lathvertln$feca1 <- round(lathvertln$feca1)
lathvertln$feca3 <- round(lathvertln$feca3)

lathmodelsln3 <- modelsearch(lathvertln, historical = TRUE, 
  approach = "mixed", suite = "main", 
  vitalrates = c("surv", "obs", "size", "repst", "fec"), juvestimate = "Sdl",
  bestfit = "AICc&k", sizedist = "gaussian", fecdist = "poisson", 
  indiv = "individ", patch = "patchid", year = "year2",year.as.random = TRUE,
  patch.as.random = TRUE, show.model.tables = TRUE, quiet = "partial")

# Here we use supplemental() to provide overwrite and reproductive info
lathsupp3 <- supplemental(stage3 = c("Sd", "Sd", "Sdl", "Sdl", "mat", "Sd", "Sdl"), 
  stage2 = c("Sd", "Sd", "Sd", "Sd", "Sdl", "rep", "rep"),
  stage1 = c("Sd", "rep", "Sd", "rep", "Sd", "mat", "mat"),
  eststage3 = c(NA, NA, NA, NA, "mat", NA, NA),
  eststage2 = c(NA, NA, NA, NA, "Sdl", NA, NA),
  eststage1 = c(NA, NA, NA, NA, "Sdl", NA, NA),
  givenrate = c(0.345, 0.345, 0.054, 0.054, NA, NA, NA),
  multiplier = c(NA, NA, NA, NA, NA, 0.345, 0.054),
  type = c(1, 1, 1, 1, 1, 3, 3), type_t12 = c(1, 2, 1, 2, 1, 1, 1),
  stageframe = lathframeln, historical = TRUE)

lathmat3ln <- flefko3(year = "all", patch = "all", stageframe = lathframeln, 
  modelsuite = lathmodelsln3, data = lathvertln, supplement = lathsupp3, 
  reduce = FALSE)