Looking for better way to visualise distribution i

Question

I'd like to visualise the following data: a hotel observes that each year some of its customers are repeat customers. So, each year about half of all customers are fist-time customers, 20% are 2nd time-customers, and so on. Below is some R code that includes the data and a visualisation. However, I'm not happy with it and I'm looking for improvements:

• R doesn't like color bands with many colours - so maybe group data?
• would a step curve be a better visualisation altogether?

• The number of visits is treated as a factor - is this the right approach?

• Stacking bars makes it easy to compare 1st-time guests but not the other ones. Should I pick a different visualisation?

  #! /usr/bin/env R CMD BATCH
  
  library(ggplot2)
  
  d <- read.table(header=TRUE, text='
      year visit count
      2013 1 1641
      2013 2 604
      2013 3 256
      2013 4 89
      2013 5 32
      2013 6 10
      2013 7 4
      2013 8 3
      2014 1 1365
      2014 2 637
      2014 3 276
      2014 4 154
      2014 5 86
      2014 6 39
      2014 7 19
      2014 8 6
      2014 9 4
      2014 10 2
      2014 11 1
      2014 12 1
      2015 1 1251
      2015 2 608
      2015 3 288
      2015 4 143
      2015 5 88
      2015 6 52
      2015 7 21
      2015 8 8
      2015 9 8
      2015 10 3
      2015 11 2
      2015 12 1')
  
  d$year  <- factor(d$year)
  d$visit <- factor(d$visit)
  
  p <- ggplot(d, aes(year,count))
  p <- p + geom_bar(aes(fill=visit),position="fill",stat="identity")
  p <- p + xlab("Year") + ylab("Distribution")
  # pdf("returners.pdf",9,6)
  print(p)
  # dev.off()
  

Answer 1

It seems that you're trying to compare the contributions to the total number of visits to the hotel by number of prior visits as well as do a year-to-year comparison. The following code puts this together in one chart.

d$year  <- factor(d$year)
# d$visit <- factor(d$visit)
d <- transform(d[order(d$year, d$visit),], cum_count=ave(count, year, FUN=cumsum))

x_max <- max(d$visit)
y_max <- max(d$cum_count)
color_sch <- c("red","tan","blue")

p <- ggplot(data=d, aes(x=visit,  colour=year))
p <- p + geom_bar(aes(y= count, fill=year), position="dodge",stat="identity", width=.7)
p <- p + geom_line(aes(y = cum_count), linetype="dotted", size=1)
p <- p + geom_point(aes(y = cum_count), size=4)
p <- p + scale_y_continuous(breaks = seq(0,y_max, 250))
p <- p + scale_x_continuous(breaks=1:x_max)
p <- p + scale_colour_manual(values=color_sch)
p <- p + scale_fill_manual(values=color_sch)
p <- p + xlab("Visit") + ylab("Count and \nCummulative Count")
p <- p + geom_text(aes(x = 2, y  = count[2], label = "Count by Number of Visits"), hjust=-.5, vjust=-2.0, size=6, color="Black") 
p <- p + geom_text(aes(x = x_max-5, y  = tail(cum_count,1), label = "Cummulative Count"), hjust=0, vjust=2.0, size=6, color="Black")
# pdf("returners.pdf",9,6)
print(p)
# dev.off()

which gives the chart

This representation suggests that the drop in 2015 as compared to previous years is due to fewer first time customers as opposed to a reduction in returning ones.

Answer 2

Why not visualize them like actual distributions?

p <- ggplot(d, aes(visit, count))
p <- p + geom_bar(stat="identity", width=0.75)
p <- p + scale_x_discrete(expand=c(0,0))
p <- p + scale_y_continuous(expand=c(0,0))
p <- p + facet_wrap(~year)
p <- p + labs(x=NULL, y="Visits")
p <- p + ggthemes::theme_tufte(base_family="Helvetica") 
p <- p + theme(legend.position="none")
p <- p + theme(panel.grid=element_line(color="#2b2b2b", size=0.15))
p <- p + theme(panel.grid.minor=element_blank())
p <- p + theme(panel.grid.major.x=element_blank())
p <- p + theme(axis.ticks=element_blank())
p <- p + theme(strip.text=element_text(hjust=0))
p <- p + theme(panel.margin.x=unit(1, "cm"))
p

To see the visit count deltas by year, you can just swap the facets:

d$year  <- factor(d$year)
d$visit <- sprintf("Visit: %d", d$visit)
d$visit <- factor(d$visit, levels=unique(d$visit))

p <- ggplot(d, aes(year, count))
p <- p + geom_segment(aes(xend=year, yend=0), size=0.3)
p <- p + geom_point()
p <- p + scale_x_discrete(expand=c(0, 0.25))
p <- p + scale_y_continuous(label=scales::comma)
p <- p + facet_wrap(~visit, scales="free_y")
p <- p + labs(x="NOTE: Free y-axis scale", y="Count")
p <- p + ggthemes::theme_tufte(base_family="Helvetica") 
p <- p + theme(legend.position="none")
p <- p + theme(panel.grid=element_line(color="#2b2b2b", size=0.15))
p <- p + theme(panel.grid.minor=element_blank())
p <- p + theme(panel.grid.major.x=element_blank())
p <- p + theme(axis.ticks=element_blank())
p <- p + theme(strip.text=element_text(hjust=0))
p <- p + theme(panel.margin=unit(1.5, "cm"))
p

Or, you can look at YoY growth by visit (%):

library(dplyr)

group_by(d, visit) %>% 
  arrange(year) %>% 
  mutate(lag=lag(count),
         chg_pct=(count-lag)/lag,
         chg_pct=ifelse(is.na(chg_pct), 0, chg_pct),
         pos=as.character(sign(chg_pct))) -> d

p <- ggplot(d, aes(year, chg_pct))
p <- p + geom_hline(yintercept=0, color="#2b2b2b", size=0.5)
p <- p + geom_segment(aes(xend=year, yend=0, color=pos), size=0.3)
p <- p + geom_point(aes(color=pos))
p <- p + scale_x_discrete(expand=c(0, 0.25))
p <- p + scale_y_continuous(label=scales::percent)
p <- p + scale_color_manual(values=c("#b2182b", "#878787", "#7fbc41"))
p <- p + facet_wrap(~visit, scales="free_y")
p <- p + labs(x="NOTE: free y-axis", y="YoY % Difference per visit count")
p <- p + ggthemes::theme_tufte(base_family="Helvetica") 
p <- p + theme(legend.position="none")
p <- p + theme(panel.grid=element_line(color="#2b2b2b", size=0.15))
p <- p + theme(panel.grid.minor=element_blank())
p <- p + theme(panel.grid.major.x=element_blank())
p <- p + theme(axis.ticks=element_blank())
p <- p + theme(strip.text=element_text(hjust=0))
p <- p + theme(panel.margin=unit(1.5, "cm"))
p <- p + theme(legend.position="none")
p