phsample.R

library(ggplot2)
library(plyr)
load("phsample.RData")

hhonly <- subset(dhus, TYPE == 1 & NP > 0)
filtered <- subset(hhonly, HINCP > 0)

# Do all variable creation in one step
expense_frame <- summarise(filtered,
  np = NP,                     # household size
  hinc = HINCP,                # household income
  hinc.log10 = log10(HINCP),
  ocpip = OCPIP,               # owner costs as % hinc
  grpip = GRPIP,               # gross rent % hinc
  # A discretized version of household income, for later
  hinc.bin = 10 ^ (round(hinc.log10, 2)),
  # merge the owner and renter expense columns
  living_expenses = ifelse(!is.na(ocpip), ocpip, 
    ifelse(!is.na(grpip), 
      grpip, NA)),
  # cost of living (housing) as % household income
  COL_pct = living_expenses,
  # is the household living beyond its means?
  beyond.means = (living_expenses > 30)
)

# remove households with no rent or living expenses 
expense_frame <- subset(expense_frame, !is.na(living_expenses))

# Use ddply to compute summaries. 
hinc_stats <- ddply(expense_frame, "hinc.bin", summarise,
  n = length(np),
  COL_pct = mean(COL_pct))

# Use scale_size_area() to scale point sizes
# Also add in a little alpha to ameliorate overplotting
ggplot(hinc_stats, aes(hinc.bin, COL_pct)) + 
  geom_point(aes(size = n), alpha = 1 / 3) +
  geom_hline(aes(yintercept = 30), color = "red") + 
  scale_x_log10("Household Income", 
    breaks = 10^(2:6), 
    labels = format(10^(2:6), big.mark = ",", sci = F, trim = T),
    # Due to buglet, need to log transform minor breaks
    minor_breaks = log10(as.vector(outer(1:9, 10 ^ (1:6))))
  ) +
  scale_y_continuous("Living expense as %income") +
  scale_size_area()