Traditional GLM

library(tidyverse)
attendance = haven::read_dta("https://stats.idre.ucla.edu/stat/stata/dae/nb_data.dta")
attendance <- attendance %>% 
  mutate(
    prog = factor(prog, levels = 1:3, labels = c("General", "Academic", "Vocational")),
    prog = fct_relevel(prog, c('Vocational', 'General', 'Academic')),
    gender = factor(gender, labels = c('Female', 'Male')),
    id = factor(id)
  )

We’ll use Poisson regression² to model the count of the number of days absent

attendance_glm <- glm(daysabs ~ math + gender + prog,
                      data = attendance, 
                      family = poisson)
## summary(attendance_glm)

term	estimate	std.error	statistic	p.value
(Intercept)	1.489	0.081	18.302	0
math	-0.007	0.001	-7.437	0
genderMale	-0.242	0.047	-5.184	0
progGeneral	1.271	0.078	16.309	0
progAcademic	0.845	0.068	12.450	0

---

2. If not familiar with Poisson regression, we are modeling the log counts as a function of the covariates. Often the exponentiated coefficients are reported. For example, exp(coef(attendance_glm)['genderMale']) is 0.785. Subtracting 1 tells us there is a -21.5% decrease in the incident rate of days absent as we switch from female to male.↩