How can I generate conditional distributions of data by taking slices of scatterplots?

Answer 1

This page explains it for you http://www.statmethods.net/advgraphs/trellis.html

You basically want to alter the equation for the graphs. They should be more like

csalary ~ bsalary|gender

should break the graphs apart based on different values of gender. There is a bunch of control language for continuous conditional variables.

Answer 2

you can use the cut() function to slice your data into ordinal categories. Then ggplot2's qplot function can then very easily create your desired plots.

library(ggplot2)

#fake data
csalary <- rnorm(100,,100)
bsalary <- csalary +rnorm(100,,10)

#Regular Scatter Plot
qplot(bsalary,csalary)

#Stacked dot plot
qplot(cut(bsalary,10),csalary)

#box-plot
qplot(cut(bsalary,10),csalary,geom="boxplot")

Answer 3

Do you really want to do that? Turning a continuous variable into an ordinal one throws away information since different values of the X variable end up in the same bin. I think your boxplot graphic conveys much less information than your scatterplot.

If you are dissatisfied with the scatterplot because of points overlapping, one way to preserve information would be to add a smooth curve that captures the trend. Look at the documentation for lowess for an example.

In your graph the three observations with salaries higher than $20,000 are pushing the remaining observations into a corner. Dropping those and replotting would give a better graph.

Another approach for skewed data like yours is to plot the logarithms of the variables instead of the variables themselves.

Answer 4

Rather than slice the data by the value of the conditioning variable (turning a continuous variable into a discrete variable), it is more efficient to condition using a kernel function. There is package that does this: hdrcde. Check out the examples in the help files.