Data Vis example: Contraception data

library(mlmRev) ## for contraception data

## Warning: package 'lme4' was built under R version 4.3.2

library(ggplot2)

Adjust some graphical parameters:

theme_set(theme_bw()+
          theme(panel.spacing=grid::unit(0,"lines")))

g1 <- (ggplot(Contraception,aes(x=age,y=use))+
       labs(x="Centered age",y="Contraceptive use")
)
print(g1+geom_point())

Not very useful!

Use a fancy component (stat_sum) that calculates the proportion of the data in a given category that fall at exactly the same (x,y) location, and by default changes the size of the symbol proportionally:

print(g1+stat_sum())

Use transparency – not a mapping (which would be inside an aes() function) but a hard-coded value:

print(g1+stat_sum(alpha=0.25))

Distinguish by number of live children:

g2 <- (ggplot(Contraception,aes(x=age,y=use,colour=livch))
    + labs(x="Centered age",y="Contraceptive use"))
g3 <- g2 + stat_sum(alpha=0.25)
print(g3)

Add smooth lines:

g4 <- (g3
    + geom_smooth(aes(y=as.numeric(use)))
)
print(g4)

Adjust the axis limits.

g5 <- (g4
    + coord_cartesian(ylim=c(0.9,2.1))
)
print(g5)

## `geom_smooth()` using method = 'loess' and formula = 'y ~ x'

Technical details:

• we need to know that the numeric values on the \(y\) axis actually correspond to the underlying numeric factor codes (N=1,Y=2);
• we use coord_cartesian(ylim=...) rather than just saying +ylim(0.9,2.1), which will erase the confidence-interval ribbons that go below the lower limits of the graph (although oob=scales::squish is useful here)

Facet by urban/rural: the syntax is rows~columns, and . here means “nothing”, or “only a single row” (cf. facet_wrap)

g4+facet_grid(.~urban) 

## `geom_smooth()` using method = 'loess' and formula = 'y ~ x'

Separate urban/rural by line type instead (and turn off confidence-interval ribbons):

g3 + geom_smooth(aes(y=as.numeric(use),lty=urban),se=FALSE)

## `geom_smooth()` using method = 'loess' and formula = 'y ~ x'

• Technical detail: we need as.numeric(use) because use is a factor.

Or make the confidence-interval ribbons light (alpha=0.1) instead of turning them off completely:

print(g3 + geom_smooth(aes(y=as.numeric(use),
                           lty=urban),alpha=0.1))

## `geom_smooth()` using method = 'loess' and formula = 'y ~ x'

Advanced topic: smooth by fitting a polynomial logistic regression.

g6 <- ggplot(Contraception,aes(x=age,
                                y=as.numeric(use)-1,
                                colour=livch))+
    stat_sum(alpha=0.25)+labs(x="Centered age",y="Contraceptive use")
print(g6 + geom_smooth(aes(lty=urban),method="glm",
                 method.args=list(family="binomial"),
                 formula=y~poly(x,2),alpha=0.1))

Technical details:

• have to convert \(y\) from (1,2) to (0,1) so that logistic regression makes sense

A more common use would be to use method="lm" to get linear regression lines for each group.

Two-way faceting:

print(g1 +
      facet_grid(livch~urban)+geom_smooth(aes(y=as.numeric(use)))
      )

This is nice, although a plot without data is always a little scary.

Use a generalized additive model (takes some more tweaking, see k parameter!)

print(g3
      + aes(y=as.numeric(use)-1)
      + geom_smooth(aes(lty=urban,y=as.numeric(use)-1),
                    method="gam",
                    method.args=list(family="binomial"),
                    formula=y~s(x,k=5),
                    alpha=0.1)
)