The Research paper is the paper that we have to follow same structure but using our own data.
Example R script below
# Probit and Logit Models
# Outline:
# Probit and logit models
# Predicted probabilities
# Marginal effects
# Pseudo R-squared
# Percent correctly predicted
# Data files:
# MROZ.csv
# setup
rm(list = ls())
directory <- “C:/Econometrics/DataR/”
# Install packages
PackageNames <- c(“tidyverse”, “stargazer”, “magrittr”, “margins”, “caret”)
for(i in PackageNames){
if(!require(i, character.only = T)){
install.packages(i, dependencies = T)
require(i, character.only = T)
}
}
# Probit and logit models ————————————————
# Model of being in labor force for women
MROZ <- read.csv(paste0(directory, ‘MROZ.csv’))
MROZ %>% select(inlf, nwifeinc, educ, exper, age, kidslt6) %>% str
MROZ %>% select(inlf, nwifeinc, educ, exper, age, kidslt6) %>% head(10)
MROZ %>% select(inlf, nwifeinc, educ, exper, age, kidslt6) %>% stargazer(type = “text”)
count(MROZ, inlf)
# Linear probability model (LPM)
LPM <- lm(inlf ~ nwifeinc + educ + exper + age + kidslt6, MROZ)
summary(LPM)
# Probit model
Probit <- glm(formula = inlf ~ nwifeinc + educ + exper + age + kidslt6,
family = binomial(link = “probit”),
data = MROZ)
summary(Probit)
# Logit model
Logit <- glm(formula = inlf ~ nwifeinc + educ + exper + age + kidslt6,
family = binomial(link = “logit”),
data = MROZ)
summary(Logit)
# Predicted probabilities ————————————————-
# Predicted probabilities for LPM
MROZ %<>% mutate(inlfhat_lpm = fitted(LPM))
# Predicted probabilities for probit model
MROZ %<>% mutate(inlfhat_probit = fitted(Probit))
# Predicted probabilities for logit model
MROZ %<>% mutate(inlfhat_logit = fitted(Logit))
# Summarize predicted values for inlf
MROZ %>% select(inlf, inlfhat_lpm, inlfhat_probit, inlfhat_logit) %>%
stargazer(type = “text”)
MROZ %>% select(inlf, inlfhat_lpm, inlfhat_probit, inlfhat_logit) %>%
head(5)
MROZ %>% select(inlf, inlfhat_lpm, inlfhat_probit, inlfhat_logit) %>%
.[601:605, ]
# Marginal effects ——————————————————–
# Linear probability model (LPM)
summary(LPM)
coef(LPM)
# Coefficients are marginal effects in a linear model
# Probit model
summary(Probit)
Mean <- model.frame(Probit) %>%
map_df(mean) # mean of independent variables
# Probit – marginal effect at the mean
Probit.atmean <- margins(Probit, at = Mean)
summary(Probit.atmean)
# Probit – average marginal effect
Probit.AME <- margins(Probit)
summary(Probit.AME)
# Logit model
summary(Logit)
# Logit – marginal effect at the mean
Logit.atmean <- margins(Logit, at = Mean)
summary(Logit.atmean)
# Logit – average marginal effect
Logit.AME <- margins(Logit)
summary(Logit.AME)
# Pseudo R-squared ——————————————————–
# Probit model – unrestricted model with all variables
summary(Probit)
# Probit model – unrestricted model with all variables
ur.model <- Probit
summary(ur.model)
# Log-likelihood for unrestricted model
(LLur <- logLik(ur.model))
# Probit restricted model with only constant
r.model <- glm(inlf ~ 1, binomial(link = “probit”), data = MROZ)
summary(r.model)
# Log-likelihood for model with only constant
(LL0 <- logLik(r.model))
# Calculate pseudo R-squared
(pseudo_r2 <- 1 – LLur/LL0)
# Percent correctly predicted ———————————————
# Percent correctly predicted for probit model
(Probit.pred <- (fitted(Probit) > 0.5) %>% as.numeric %>% as.factor) # Prediction
(actual <- MROZ$inlf %>% as.factor) # actual data
caret::confusionMatrix(Probit.pred, actual, positive = “1”)
# Percent correctly predicted for logit model
(Logit.pred <- (fitted(Logit) > 0.5) %>% as.numeric %>% as.factor) # Prediction
confusionMatrix(Logit.pred, actual, positive = “1”)
