bias_correction(model::GLFixedEffectModel,df::DataFrame;i_symb::Union{Symbol,Nothing}=nothing,j_symb::Union{Symbol,Nothing}=nothing,t_symb::Union{Symbol,Nothing}=nothing,L::Int64=0,panel_structure::Symbol=:classic)

Asymptotic bias correction after fitting binary choice models with a two-/three-way error.

Arguments

Required Arguments

• model::Integer: a GLFixedEffectModel object which can be obtained by using nlreg().
• df::DataFrame: the Data Frame on which you just run nlreg().

Optional Arguments

• L:Int64: choice of binwidth, see Hahn and Kuersteiner (2011). The default value is 0.
• panel_structure: choose from "classic" or "network". The default value is "classic".
• i_symb: the variable name for i index in the data frame df
• j_symb: the variable name for j index in the data frame df
• t_symb: the variable name for t index in the data frame df

Available Model

We only support the following models:

• Binomial regression, Logit link, Two-way, Classic
• Binomial regression, Probit link, Two-way, Classic
• Binomial regression, Logit link, Two-way, Network
• Binomial regression, Probit link, Two-way, Network
• Binomial regression, Logit link, Three-way, Network
• Binomial regression, Probit link, Three-way, Network
• Poisson regression, Log link, Three-way, Network
• Poisson regression, Log link, Two-way, Network

Estimate a generalized linear model with high dimensional categorical variables

Arguments

• df: a Table
• FormulaTerm: A formula created using @formula
• distribution: A Distribution. See the documentation of GLM.jl for valid distributions.
• link: A Link function. See the documentation of GLM.jl for valid link functions.
• CovarianceEstimator: A method to compute the variance-covariance matrix
• save::Vector{Symbol} = Symbol[]: Should residuals/predictions/eta's/estimated fixed effects be saved in the dataframe augmentdf? Can contain any subset of [:residuals,:eta,:mu,:fe].
• method::Symbol: A symbol for the method. Default is :cpu. Alternatively, :gpu requires CuArrays. In this case, use the option double_precision = false to use Float32.
• contrasts::Dict = Dict() An optional Dict of contrast codings for each categorical variable in the formula. Any unspecified variables will have DummyCoding.
• maxiter::Integer = 1000: Maximum number of iterations
• maxiter_center::Integer = 10000: Maximum number of iterations for centering procedure.
• double_precision::Bool: Should the demeaning operation use Float64 rather than Float32? Default to true.
• dev_tol::Real : Tolerance level for the first stopping condition of the maximization routine.
• rho_tol::Real : Tolerance level for the stephalving in the maximization routine.
• step_tol::Real : Tolerance level that accounts for rounding errors inside the stephalving routine
• center_tol::Real : Tolerance level for the stopping condition of the centering algorithm. Default to 1e-8 if double_precision = true, 1e-6 otherwise.
• separation::Symbol = :none : method to detect/deal with separation. Currently supported values are :none, :ignore and :mu. See readme for details.
• separation_mu_lbound::Real = -Inf : Lower bound for the Clarkson-Jennrich separation detection heuristic.
• separation_mu_ubound::Real = Inf : Upper bound for the Clarkson-Jennrich separation detection heuristic.
• separation_ReLU_tol::Real = 1e-4 : Tolerance level for the ReLU algorithm.
• separation_ReLU_maxiter::Integer = 1000 : Maximal number of iterations for the ReLU algorithm.

Examples

using GLM, RDatasets, Distributions, Random, GLFixedEffectModels

rng = MersenneTwister(1234)
df = dataset("datasets", "iris")
df.binary = 0.0
df[df.SepalLength .> 5.0,:binary] .= 1.0
df.SpeciesDummy = categorical(df.Species)

m = @formula binary ~ SepalWidth + fe(SpeciesDummy)
x = nlreg(df, m, Binomial(), GLM.LogitLink() , start = [0.2] )