README.md

In AndrMenezes/unitModalReg: Unit Modal Parametric Regression Models

unitModalReg: A Collection of Parametric Modal Regressions Models for Bounded Data

The goal of unitModalReg is to provide tools for fitting parametric modal regression models for bounded response variables.

Installation

You can install the development version from GitHub with:

# install.packages("remotes")
remotes::install_github("AndrMenezes/unitModalReg")

Example

The packages follows the structure of glm objects. The main function is unitModalReg.

library(unitModalReg)

# Simulate data
n <- 500
betas <- c(beta0 = 2, beta1 = -1, beta2 = 0.6)
x1  <- runif(n)
x2  <- rbinom(n, size = 1, prob = 0.65)
eta <- betas[1] + betas[2] * x1 + betas[3] * x2
mu  <- exp(eta) / exp(1 + eta)
phi <- 4
y   <- rugamma(n = n, mu = mu, phi = phi)
data_sim <- data.frame(y = y, x1 = x1, x2 = factor(x2))

# Families of distributions
distr <- list(
  Be = "betaMode",
  UGz = "unitGompertz",
  UGa = "unitGamma",
  Kum = "Kumaraswamy"
)

# Fit the models
fits <- lapply(distr, function(m) {
  unitModalReg(y ~ x1 + x2, data = data_sim, family = m, link = "logit")
})

# Get the coefficients
vapply(fits, coef, numeric(4))
#>                      Be         UGz         UGa        Kum
#> (Intercept) -0.72578488 -1.54060142 -0.76066727 -0.6221908
#> x1           0.19229626  0.37335286  0.21208397  0.1398764
#> x21          0.06202745  0.04657786  0.04251816  0.0850151
#> phi          1.52605812  0.98750372  3.74129743  2.1089950

We also implemented the gof function to compare the models based on the information criterion (AIC, BIC and HQIC) and Voung test for non-nested models.

# Compare the models fitting
gof(lt = fits)
#> 
#> Goodness-of-Fit for modal regression models 
#> 
#> Call:  gof(lt = fits)
#> 
#> Voung tests for all model combinations: 
#>                                 statistic p.value
#> betaMode versus unitGompertz    3.508     0.000  
#> betaMode versus unitGamma       -0.357    0.360  
#> betaMode versus Kumaraswamy     1.385     0.083  
#> unitGompertz versus unitGamma   -6.267    0.000  
#> unitGompertz versus Kumaraswamy -5.313    0.000  
#> unitGamma versus Kumaraswamy    1.103     0.135  
#> 
#> 
#> Information criterions of modal regression models: 
#>      betaMode unitGompertz unitGamma Kumaraswamy
#> aic  -328.821 -203.590     -329.241  -324.864   
#> bic  -311.963 -186.732     -312.382  -308.006   
#> hqic -322.206 -196.975     -322.625  -318.249

The currently methods implemented are

methods(class = "unitModalReg")
#> [1] coef      confint   fitted    hnp       logLik    print     residuals
#> [8] summary   vcov     
#> see '?methods' for accessing help and source code

AndrMenezes/unitModalReg documentation built on March 12, 2021, 5:24 a.m.