print.rpinpall {winputall}

Fit Input Allocation Random Parameters Model

Description

Designed to fit a random parameters input allocation model proposed in Koutchade et al., (2024) https://hal.science/hal-04318163. It provides crops input cost for each individual at each time and can account for Weighted Panel Data.

Usage

## S3 method for class 'rpinpall'
print(x, error = FALSE, ...)

## S3 method for class 'rpinpall'
summary(object, ...)

## S3 method for class 'rpinpall'
plot(x, ...)

rpinpallEst(
  data,
  id_time,
  total_input,
  crop_acreage,
  crop_indvar = NULL,
  crop_rp_indvar = NULL,
  weight = NULL,
  distrib_method = c("lognormal", "normal", "censored-normal"),
  sim_method = c("map_imh", "mhrw", "marg_imh", "mhrw_imh", "nuts", "variat",
    "lapl_approx"),
  calib_method = c("cmode", "cmean", "rscd", "estim-sim"),
  saem_control = list(),
  par_init = list()
)

Arguments

x

An object produced by the function rpinpallEst, to be displayed

error

logical. If TRUE, residuals are considered in variable input prediction

...

Other arguments

object

An object produced by the function rpinpallEst, to be displayed

data

name of the data frame or matrix containing all the variables included in the model.

id_time

first (individual) and second (time) level variables allowing characterizing panel data.

total_input

variable (name) containing the total input used at farm level per ha to be allocated to the different crops.

crop_acreage

list of variables containing the acreage of the different crops.

crop_indvar

optional list of vector of (time-varying) variables specific to each crop used to control for observed (individual and/or temporal) characteristics in the estimation process. Default=NULL.

crop_rp_indvar

optional list of vector of (time-constant) variables specific to each crop used to control for observed time-constant characteristics in the estimation process. Default=NULL.

weight

optional variable containing weights of individual sample farms. Default=NULL (equal weight is given to each farm). Default=NULL.

distrib_method

assumption on the distribution of input use per crop (x_kit): "normal", "lognormal" or "censored-normal". Default="lognormal".

sim_method

method used to draw the random parameters in the simulation step of the SAEM algorithm in the estimation process: "map_imh" (independant Metropolis Hasting with Laplace approximation as proposal distribution), "marg_imh" (independant Metropolis Hasting with marginal distribution of random pararameters as proposal distribution), "mhrw" (Metropolis Hasting Random Walk), "mhrw_imh" (combined "imh" and "mhrw"), "nuts","variat" and"lapl_approx". Default= "map_imh".

calib_method

method used: "cmode" (conditional mode), "cmean" (conditional mean). Default="cmode".

saem_control

list of options for the SAEM algorithm. See 'Details

par_init

list of some parameters' initialization.

Details

An SAEM algorithm is used to perform the estimation of input uses per crop. Different options can be specified by the user for this algorithm in the saem_control argument. The saem_control argument is list that can supply any of the following component.

• nb_burn_saem: Number of iterations of the burn-in phase where individual parameters are sampled from their conditional distribution using sim_method and the initial values for model parameters without update these parameters. Default=20.

• nb_SA: Number of iterations in the exploration phase where algorithm explore parameters space without memory. The parameter that controls the convergence of the algorithm is set to 1. Default=200.

• nb_smooth: Number of iterations in the smoothing phase.Default=200 and the parameter that controls the convergence of the algorithm is set to 0.85 by default.

• nb_RS: Number of iterations where tempering approach is used

• tol: Tolerance value for the convergence. Default 1.10-3.

• estim_rdraw: Number of random draws using in the estimation process. Default=100

• calib_rdraw: Number of random draws using in the calibration process. Default=100

• stde_rdraw: Number of random draws using for computation of estimation standard errors. Default=100

• p_SA: Parameter determining step sizes in the Stochastic Approximation (SA) step. Must be comprise between 0 and 1. Default=0.85

• doParallels: Logical.If TRUE a parallel processing is used when more than 2 cores are available. Default=FALSE

• doTempering: Logical. If TRUE the tempering approach proposed by (Allassonnière and Chevallier, 2021) is used to avoid convergence to local maxima. Default=TRUE

• doDiagEps: = "2",

• showProgress: Logical. If TRUE the evolution of the estimation process is displayed graphically at the bottom of the screen. Default=TRUE

• showIterConvLL: Logical. If TRUE iteration number and convergence value are displayed during the estimation process. Default=FALSE

Value

Distribution of estimated crop input uses.

This function returns a list with the following components:

• xit_pred: matrix of predicted crop input used per ha.

• xit_pred_with_error: matrix of predicted crop input used per ha.

• yit_predict: vector of predicted totat input used.

• est_pop list of results of estimation: estimated parameters.

• est_stdelist of parameters standard errors.

• call: a copy of the function call.

• opt: a list of saem algorithm control parameters.

• conv_ind_cll: vector of convergence indicator.

• data_list: a list of individual data used for estimation.

Functions

• print(rpinpall): Displays the distribution of estimated crop input uses accounting for error by default

• summary(rpinpall): Displays a summary of estimated parameters

• plot(rpinpall): Plot the "global" convergence indicator

References

Koutchade, O. P., Carpentier A. and Femenia F. (2024).

Examples

data(my_winputall_data)
mydata <- my_winputall_data
fit <- rpinpallEst(data = my_winputall_data,
                   id_time = c("id","year"),
                   total_input = "tx",
                   crop_acreage = c("s_crop1","s_crop2","s_crop3"),
                   distrib_method = "lognormal",
                   sim_method = "map_imh",
                   calib_method = "cmode",
                   saem_control = list(nb_SA = 10, nb_smooth = 10, estim_rdraw = 10))
print(fit)
plot(fit)
summary(fit)
head(fit$xit_pred)