ggplot2 and binomial GLM

Generalized Linear Models in R

Richard Erickson

Instructor

What can I see in my data?

Does commute distance change the probability of taking the bus?

ggplot(bus, aes(x = MilesOneWay, y = Bus)) + geom_point()

Plot with only points

geom_jitter()

gg_jitter <- ggplot(bus, aes(x = MilesOneWay, y = Bus)) +
             geom_jitter(width = 0, height = 0.05)
print(gg_jitter)

Jittered point plot

geom_smooth()

gg_jitter + geom_smooth()

Previous jittered plot with a smoothed geom.

factor to numeric

str(bus)
bus$Bus2 <- as.numeric(bus$Bus) - 1

geom_smooth()

gg_jitter + geom_smooth()

Change data to be numeric rather than a character

linear models

gg_jitter + geom_smooth(method = 'glm')

Add lm to ggJitter

Logistic regressions

ggJitter + 
geom_smooth(method = 'glm',
            method.args = list(family = "binomial"))

ggJitter with a logistic curve

gg_jitter + 
    geom_smooth(method = 'glm',
                method.args = list(family = binomial(link = 'logit')),
                se = FALSE, color = 'red') +
    geom_smooth(method = 'glm',
                method.args = list(family = binomial(link = 'probit')),
                se = FALSE, color = 'blue')

ggJitter with both a logit and probit plot

Summary of steps

Plot as jitter to avoid overlap
Add a smoothed geom
Specify correct method and family
Polish your figure (not covered in this course)

Let's practice!

Generalized Linear Models in R