Coloring by two factors

Bottom line is… this is a hard task and there is no good answer. The three work around I came up with are:

I implemented the last two options here, but the first would be a good choice if it fits your project requirements/goals.

Load packages and functions

Simulate data

I could not load the qlfs.Rdata files, so I made something similar

obs <- 100
qlfs <- data.frame(YoungestChild = sample(x = c(0:20), size = obs, replace = TRUE ),
                   Marital.Status = rbinom(obs, 1, 0.8),
                   AgeGroup = factor(sample(x = c(1:5), size = obs, replace = TRUE )),
                   Sex = sample(c("m","f"), obs, 0.5))

Munging code

This is all straight from your script, unaltered

qlfs$YoungestChild[qlfs$YoungestChild < 0] <- NA
breaks <- c(0, 3, 5, 12, 18)
qlfs$AdHoc.YoungestChild <- group.data(qlfs$YoungestChild, breaks=breaks)
table.all <- tally(AdHoc.YoungestChild ~ AgeGroup, data = qlfs, format="percent", margin="joint")
table.valid <- tally(AdHoc.YoungestChild ~ AgeGroup, data = qlfs, format="percent", margin="joint", na.rm = TRUE)
AgeGroup.Total.All <- table.all[nrow(table.all),]
AgeGroup.Total.Valid <- table.valid[nrow(table.valid),]
old.categories <- levels(qlfs$AgeGroup)
weights <- AgeGroup.Total.All / AgeGroup.Total.Valid
new.categories <- weights[1:length(weights)-1]
qlfs$weights <- recode(qlfs$AgeGroup, "
                       old.categories[1]=new.categories[1];
                       old.categories[2]=new.categories[2];
                       old.categories[3]=new.categories[3];
                       old.categories[4]=new.categories[4]
                       ",
                       levels=c(new.categories))
qlfs$weights <- as.numeric(as.character(qlfs$weights))
qlfs$invalid.cases <- is.na(qlfs$AdHoc.YoungestChild)
qlfs$invalid.cases2 <- is.na(qlfs$YoungestChild)
qlfs$weights[qlfs$invalid.cases==TRUE] <- NA
qlfs$weights[qlfs$invalid.cases2==TRUE] <- NA
new.categories <- c(levels(qlfs$AdHoc.YoungestChild),"No Response")
qlfs$AdHoc.YoungestChild2 <- recode(qlfs$AdHoc.YoungestChild, "  NA='No Response'  ", levels=c(new.categories) )

Fills

These are like yours, but I reduced the number to 4 (because of simulated data) and I created a fill_manual_final that is a combo of male and female

fill_manual <- c("#006d2c", "#2ca25f", "#6bbea3", "#b2e2e2", "#edf8fb")
fill_manual_male <- c("#045a8d", "#2b8cbe", "#74a9cf", "#bdc9e1")
fill_manual_female <- c("#7a0177", "#c51b8a", "#f768a1", "#fbb4b9")
fill_manual_final <- c(fill_manual_female, fill_manual_male)

Original Plot

your plot unaltered

ggplot(na.omit(qlfs)) +
  geom_bar(aes(x=AgeGroup, fill=AdHoc.YoungestChild2, weight=weights)) +
  ylab("Total Respondents") +
  xlab("Age Groups") +
  theme() + 
  facet_wrap(~Sex) + 
  scale_fill_manual(values = fill_manual, 
                    name="Reported Age of \n Youngest Child") + 
  theme(strip.background = element_rect(fill = "light blue"))

Combine groupings

Simply made a new column mf_youngest that combines the two factors of interest. The plot then colors by the new factor and the scale_fill_manual calls the combined scale to cover all of the new factors. I am not a huge fan of this, but it most closely approximates your goal. There would have to be some work done on the legend to make it prettier.

# new column
qlfs$mf_youngest <- paste0(qlfs$Sex, "_", qlfs$AdHoc.YoungestChild2)  

ggplot(na.omit(qlfs)) +
  geom_bar(aes(x=AgeGroup, fill=mf_youngest, weight=weights)) +
  ylab("Total Respondents") +
  xlab("Age Groups") +
  theme() + 
  facet_wrap(~Sex) + 
  scale_fill_manual(values = fill_manual_final, 
                    name="Reported Age of \n Youngest Child") + 
  theme(strip.background = element_rect(fill = "light blue"))

Use Alpha

This was taken from the S.O. post linked below. Essentially, use scale_alpha_manual to give an effect of changing colors by Adhoc.YoungestChile2. I added a theme_bw() and took away the grid lines to gussy it up a tad. http://stackoverflow.com/questions/33221794/separate-palettes-for-facets-in-ggplot-facet-grid

ggplot(na.omit(qlfs)) +
  geom_bar(aes(x=AgeGroup, fill=Sex, weight=weights, alpha = AdHoc.YoungestChild2)) +
  ylab("Total Respondents") +
  xlab("Age Groups") +
  theme_bw() + 
  facet_wrap(~Sex) + 
  scale_fill_manual(values = c("#c51b8a", "#2b8cbe"), 
                    name="Sex of \n Youngest Child") + 
  scale_alpha_manual(values=c(0.35, 0.55, 0.75, 1),
                     name="Reported Age of \n Youngest Child") +
  theme(strip.background = element_rect(fill = "light blue"),
        panel.grid.minor = element_blank(), 
        panel.grid.major = element_blank())