felogit: This function implements the estimators of the bounds on the...

In cgaillac/MarginalFElogit: Estimation of Average Marginal Effects in Fixed Effect Logit Models

This function implements the estimators of the bounds on the AME/ATE proposed in DDL.

Description

The function summary_felogit applied to the output of this felogit function prints the table containing the estimation results.

Usage

felogit(
  data,
  formul = NULL,
  Option = "quick",
  compute_X = "all",
  compute_T = "all",
  cluster = NULL,
  alpha = 0.05,
  CIOption = "CI2",
  nbCores = 4,
  varY = "Y",
  varX = NULL
)

Arguments

data	is an environment variable containing the data. It can either be in long format or in wide format. If in long format, data should be a data frame with individual indexes in the first columns, periods in the second column, and in the remaining columns the value of each variable at the corresponding period for the corresponding individual. All columns should be named. If in wide format, data should be a list or environment variable containing: - data$Y is a matrix of size n x Tmax containing the values of the dependent variable Y for each individual at each period. NAs represent missing observations. - data$X is an array of size n x Tmax x dimX containing the values of the covariates X for each individual at each period. NAs represent missing observations. - data$clusterIndexes is a vector of size n x 1 that specifies the cluster each observation pertains to. If it does not exist, the function enforces the default setting of i.i.d. observations - the parameter takes value 1:n so that each observation is its own cluster.
formul	(default NULL) is a formula used when the data is in long format to specify which variable to use. It must be of the form formula("Y ~ X_1 + X_2") where Y is the name of the binary variable of interest and X_1, X_2, etc are the variables to use in the logit model. Set formul to NULL to indicate the data is already in wide format.
Option	(default "quick") Estimation method to be used. If "quick" or "outer" (case-insensitive) the outer bounds are computed. Otherwise, the sharp bounds are computed. If any of the variable with respect to which the effect is being measured is binary (so that an ATE must be computed), we always switch to the quick method for efficiency reasons. In general, we recommend using the outer bounds method if the number of covariates is at least three, if the number of periods observed is four or more, or if the sample size is small (less than 500) or large (more than 10^4).
compute_X	(default "all") is a vector containing all the variables with respect to which the AME/ATE must be computed. It can either contain the variable names, as given in the arguments formul or varX, or their rank, e.g. 3 for the third variable appearing in formul/varX or, for lack theoreof, the third column in data$X.
compute_T	(default "all") is a vector containing all periods at which the AME/ATE must be computed. Alternatively, it can be "all", in which case the AME/ATE will be computed successively at every period and, on top of that, the average AME/ATE across all periods will also be computed using the function compute_average_AMTE. Also note that non-positive values will be counted backwards from the last period at which each individual is observed, as in an event-study. If NULL, the first period is used. Periods MUST be specified by their rank (1 for first period, etc) and NOT by value (e.g. 1980 for the year).
cluster	is clustering
alpha	(default 0.05) desired asymptotic level of the estimated confidence intervals
CIOption	(default "CI2") When the outer bounds method is being used, specifies which confidence interval should be used. If "CI2", the CI2 confidence interval is being used (see DDL, section 4.2), otherwise the CI3 confidence interval will be used (see DDL, appendix C). We recommend using CI3 only if the user suspects the FE logit model may be a severely misspecified model for the data.
nbCores	(default 4) number of cores to be used for parallel computing, to speed up the estimation of the sharp bounds.
varY	(default "Y") for data already in wide format, the name of the binary variable of interest in the data list/environment.
varX	(default NULL) for data already in wide format, the name to use for each of the variables given by the slices of the array data$X along the third dimension. dimnames(data$X) along the third dimension can also be used. If varX is NULL and no name is given in dimnames(data$X), X_1, X_2, etc will be used.

Value

A list containing: - summary: a dataframe containing the estimation results, - n: the number of used individuals, - ndiscard: the number of discarded individuals, - Tmax: the total number of distinct periods of observed, - vardiscard: the label of the discarded variables, - formul: the formula used (implicitly deduced if the input was NULL), - alpha: the level used for the confidence intervals, - Option: the method used, - summary_CMLE : a dataframe containing the estimation results of the CMLE.

Examples

library(pglm)
data("UnionWage", package = "pglm")

UnionWage$union <- UnionWage$union == "yes"
UnionWage$rural <- UnionWage$rural == "yes"
UnionWage$black <- UnionWage$com == "black" # used as test for discarded variable because constant
UnionWage$NorthEast <- UnionWage$region == "NorthEast"
sub <- UnionWage[UnionWage$year < 1986,]

formul <- formula("union ~ exper + married + black")
output <- felogit(data = sub, formul = formul, Option = "quick", compute_T = NULL)
summary_felogit(output)

cgaillac/MarginalFElogit documentation built on Dec. 24, 2024, 3:23 p.m.