如何可视化成分随时间的“逐步”变化

Question

I have a data.frame containing the distribution of seats in parliament between parties at the year of the election. Eventually, I would like to obtain a graph similar to this one . I want to visualize the composition of the parliament over the years, not only for the election year.

results<-structure(list(party = c("PARTY1", "PARTY1", "PARTY1", "PARTY1", "PARTY2", "PARTY2", 
"PARTY2", "PARTY2", "PARTY2", "PARTY2", "PARTY3", "PARTY3", "PARTY3", "PARTY3", "PARTY3", 
"PARTY3", "PARTY3", "PART4", "PART4", "PART4", "PART4"), year = c(1996, 
1998, 2000, 2010, 1996, 2000, 2002, 2006, 2010, 2014, 1996, 1998, 
2000, 2002, 2006, 2010, 2014, 2002, 2006, 2010, 2014), party.seats = c(8, 
6, 5, 3, 19, 8, 10, 9, 7, 10, 9, 4, 6, 5, 3, 4, 5, 3, 7, 8, 6
)), class = "data.frame", row.names = c(NA, -21L), .Names = c("party", 
"year", "party.seats"))

I am able to produce a bar chart, which however only presents me the data for the election year and misses the years between the elections.

ggplot(data=results,aes(x=as.factor(year), y=party.seats, fill=party, label=party))+geom_bar(stat="identity")

I am able to produce a ggplot chart with geom_area, which is however misleading since it suggests that the distribution of seats is changing during the years following the elections (there is a slop, and not a “step”).

ggplot(as.data.frame(xtabs(party.seats~year+party, results)), aes(x=as.Date(as.character(year), "%Y"), y = Freq, fill = party)) +  geom_area(position = "stack")

Any help? I am particularly wondering whether there is a (time-series related?) command which would take the results of the election year to all subsequent years until new elections were held. So basically, a command which takes the election event at time x as ongoing (= fills years in between) until new elections are held at time y.

Answer 1

I think that geom_step is what you are looking for, though the simplest implementation will not have the bars/areas stack to the total number of seats allotted (though that may be better):

ggplot(data=results
      , aes(x=year
            , y=party.seats
            , col=party)) +
  geom_step()

If you really want you can get the fills, though like in @Haboryme's answer you will need to generate all of the points in between elections. Here, I use dplyr / tidyr to add a new data row for each day between elections (you just need the resolution narrow enough that the "step" appears instantaneous rather than spread over a full year on the final plot) with some added after the most recent election to make those values actually show up. I then fill the party seats from before forward until the next election, and set the missings to 0 for good measure (before the party had any seats).

Note that you could extend this with the exact dates of elections instead of just the years without needing to modify too much

results %>%
  complete(year = full_seq(c(min(year), max(year) + 1), 1/365), party) %>%
  group_by(party) %>%
  fill(party.seats) %>%
  replace_na(replace = list(party.seats = 0)) %>%
  ggplot(
    aes(x=year
        , y=party.seats
        , fill=party)) +
  geom_area(position = "stack")

gives

I still prefer the lines though, as it is easier to compare the parties against each other when they are not stacked on top of each other. For example, from 2010 to 2014, it is difficult to tell from the area versions whether party 2 or 4 has more seats (but it is clear from the lines).

Answer 2

Another option could be to create the complete dataframe with all the missing years:

library(tidyverse)                      
library(zoo)
all_years=seq(min(results$year),max(results$year)) #get the sequence of all the years considered
filled=data.frame(party=rep(unique(results$party),each=length(all_years)), #build a df with the seq of years for each party
                  year=rep(all_years,length(unique(results$party))))

Then merge with your data and fill the NA (with 0 if at the start, with the most recent value else):

df=merge(results,filled,by.y=c("party","year"),all.y=T)%>%
  group_by(party)%>%
  na.locf()%>%
  mutate(party.seats=coalesce(as.numeric(party.seats), 0))  

Plot with geom_bar and width=1 to have something that looks continuous:

ggplot(data=df,aes(x=as.factor(year), y=party.seats, fill=party, label=party))+
  geom_bar(stat="identity",width = 1)

It gives (the x axis needs some tweaking):

Answer 3

You can try fancy streamgraph too (you will get plotly like mouse-hover tooltips too):

library(dplyr)
library(streamgraph)
results %>%
  streamgraph("party", "party.seats", "year") %>%
  sg_axis_x(1, "year", "%Y") %>%
  sg_legend(TRUE, "party")

results %>%
  streamgraph("party", "party.seats", "year", offset="zero", interpolate="step") %>%
  sg_axis_x(1, "year", "%Y") %>%
  sg_fill_brewer("PuOr")