This post is everything you want
it’s everything you need
it’s every viz inside of you that you wish you could see
it’s all the right viz at exactly the right time
but it means nothing to you and you don’t know why

horizon gif

LET US MAKE SOME HORIZON CHARTS.

What is a horizon chart you ask? That’s exactly what I was thinking earlier this weekend. Well, not exactly. I sort of knew what horizon charts were, but I couldn’t say exactly what they were good for. But then, after making some it struck me. Come with me on this journey.

As we usually do in this space, we’ll use R to create our plots.

Horizon Charts

Check out Flowing Data on horizon charts and this document for some other examples of horizon charts.

I was inspired to look into this by this awesome tweet from [@timelyportfolio](https://twitter.com/timelyportfolio):

In this example timelyportfolio showed how to embed horizon charts into a datatable widget. We’ll get to that, but first let’s build our own horizon chart.

Building a horizon chart with ggplot2

In building this chart, I was greatly aided by this post from timelyportfolio, which dates back to 2012 (!). Since then, ggplot2 has evolved, so the code needed minor tweaking. Using that modified code I was able to build up a horizon chart. We’ll go over the steps, but first let’s get some data.

Getting data

For today’s examples I’m going to use employment data from the U.S. Bureau of Labor Statistis (BLS) and house price data from the Freddie Mac. We’ve used these data before, here for employment and here for house prices. The posts linked to describe more about how to the data. For the employment data we’ll load it from the web, and the house price data is available in a csv file hpistate.csv.

Loading, preparing and checking data

Let’s follow the strategies from our earlier post and get the employment data from BLS and the house price data from our .csv.

First, let’s check our house price data:

################################################################################
### Load libraries
################################################################################

library(tidyverse)
## Loading tidyverse: ggplot2
## Loading tidyverse: tibble
## Loading tidyverse: tidyr
## Loading tidyverse: readr
## Loading tidyverse: purrr
## Loading tidyverse: dplyr
## Conflicts with tidy packages ----------------------------------------------
## filter(): dplyr, stats
## lag():    dplyr, stats
library(data.table)
## 
## Attaching package: 'data.table'
## The following objects are masked from 'package:dplyr':
## 
##     between, first, last
## The following object is masked from 'package:purrr':
## 
##     transpose
library(scales)
## 
## Attaching package: 'scales'
## The following object is masked from 'package:purrr':
## 
##     discard
## The following object is masked from 'package:readr':
## 
##     col_factor
################################################################################
### Load libraries
################################################################################

df<-fread("data/hpistate.csv")
df$date<-as.Date(df$date, format="%m/%d/%Y")

# Compute percent change by year
#df=df[,hpa:=(hpi-shift(hpi,12,fill=NA))/shift(hpi,12,fill=NA),by=c("statename")]

# Print table for checking
htmlTable::htmlTable(tail(df))
state geo date hpi year month type hpa12 lat long metro.pop fips statename division region
1 WY WY 2016-04-01 185.7677316 2016 4 State 0.0157535023308515 41.15 -104.79 97121 56 Wyoming Mountain Division West Region
2 WY WY 2016-05-01 186.6263416 2016 5 State 0.0111053021335621 41.15 -104.79 97121 56 Wyoming Mountain Division West Region
3 WY WY 2016-06-01 187.2050436 2016 6 State 0.00508809832570378 41.15 -104.79 97121 56 Wyoming Mountain Division West Region
4 WY WY 2016-07-01 187.6827817 2016 7 State 0.00144925455864708 41.15 -104.79 97121 56 Wyoming Mountain Division West Region
5 WY WY 2016-08-01 188.0179497 2016 8 State 0.00186408591734799 41.15 -104.79 97121 56 Wyoming Mountain Division West Region
6 WY WY 2016-09-01 187.743011 2016 9 State 0.00291143018858708 41.15 -104.79 97121 56 Wyoming Mountain Division West Region

Okay, seems all right. We’ve got a few extra columns from our earlier post, but we’ll drop them.

Let’s get some employment data from BLS:

################################################################################
### Go get data from BLS.gov
################################################################################
emp.data<-fread("https://download.bls.gov/pub/time.series/sm/sm.data.54.TotalNonFarm.All")
## Warning in fread("https://download.bls.gov/pub/time.series/sm/sm.data.
## 54.TotalNonFarm.All"): Bumped column 4 to type character on data row
## 521667, field contains '-'. Coercing previously read values in this
## column from logical, integer or numeric back to character which may not
## be lossless; e.g., if '00' and '000' occurred before they will now be just
## '0', and there may be inconsistencies with treatment of ',,' and ',NA,' too
## (if they occurred in this column before the bump). If this matters please
## rerun and set 'colClasses' to 'character' for this column. Please note that
## column type detection uses a sample of 1,000 rows (100 rows at 10 points)
## so hopefully this message should be very rare. If reporting to datatable-
## help, please rerun and include the output from verbose=TRUE.
emp.series<-fread("https://download.bls.gov/pub/time.series/sm/sm.series")

emp.list<-emp.series[industry_code==0 # get all employment
                     & data_type_code==1 # get employment in thousands
                     & seasonal=="S",]  # get seasonally adjusted data]

emp.area<-fread("https://download.bls.gov/pub/time.series/sm/sm.area",
                col.names=c("area_code","area_name","drop"))[,c("area_code","area_name"),with=F]
## Warning in fread("https://download.bls.gov/pub/time.series/sm/sm.area", :
## Starting data input on line 2 and discarding line 1 because it has too few
## or too many items to be column names or data: area_code area_name
emp.st<-fread("https://download.bls.gov/pub/time.series/sm/sm.state",
              col.names=c("state_code","state_name","drop"))[,c("state_code","state_name"),with=F]
## Warning in fread("https://download.bls.gov/pub/time.series/sm/sm.state", :
## Starting data input on line 2 and discarding line 1 because it has too few
## or too many items to be column names or data: state_code state_name
# merge data
emp.dt<-merge(emp.data,emp.list,by="series_id",all.y=T)

#create month variable
emp.dt=emp.dt[,month:=as.numeric(substr(emp.dt$period,2,3))]
# (this assignment is to get around knitr/data table printing error)
# see e.g. http://stackoverflow.com/questions/15267018/knitr-gets-tricked-by-data-table-assignment

# M13 = Annual average, drop it:
emp.dt<-emp.dt[month<13,]

#create date variable
emp.dt$date<- as.Date(ISOdate(emp.dt$year,emp.dt$month,1) ) 

# merge on area and state codes
emp.dt<-merge(emp.dt,emp.area,by="area_code")
emp.dt<-merge(emp.dt,emp.st,by="state_code")
emp.dt=emp.dt[,c("state_name","area_name","date","year","month","value"),with=F]
emp.dt=emp.dt[,emp:=as.numeric(value)] #convert value to numeric
# Compute year-over-year change in employment and year-over-year percent change
emp.dt=emp.dt[,emp.yoy:=emp-shift(emp,12,fill=NA),by=c("area_name","state_name")]

# Percent change by year:
emp.dt=emp.dt[,emp.pc:=(emp-shift(emp,12,fill=NA))/shift(emp,12,fill=NA),by=c("area_name","state_name")]

emp.dt=emp.dt[,type:=ifelse(area_name=="Statewide","State","Metro")]

# drop states in c("Puerto Rico","Virgin Islands")
emp.dt=emp.dt[!(state_name %in% c("Puerto Rico","Virgin Islands")),]

# only keep state data
emp.dt.state<-emp.dt[area_name=="Statewide"]

htmlTable::htmlTable(head(emp.dt.state))
state_name area_name date year month value emp emp.yoy emp.pc type
1 Alabama Statewide 1990-01-01 1990 1 1626.70 1626.7 State
2 Alabama Statewide 1990-02-01 1990 2 1625.30 1625.3 State
3 Alabama Statewide 1990-03-01 1990 3 1623.50 1623.5 State
4 Alabama Statewide 1990-04-01 1990 4 1635.90 1635.9 State
5 Alabama Statewide 1990-05-01 1990 5 1639.80 1639.8 State
6 Alabama Statewide 1990-06-01 1990 6 1639.20 1639.2 State

Great, looks good. Now that we have our data, let’s merge them together.

# Rename state_name as statename in emp.dt.state data
emp.dt.state<-rename(emp.dt.state,statename=state_name)
# merge on date & statename
dt<-merge(df,emp.dt.state,by=c("date","statename"))

The key variables in our data are called hpa12 which represents the 12-month percent change in house prices and emp.pc which represents the 12 month-percent change in employment.

Looking to the horizon

Let’s build up a horizon chart. To do so, first let’s look at data from the state of Arizona.

# plot house price trens for Ohio:

ggplot(data=dt[state=="AZ"],aes(x=date,y=hpa12))+geom_line()+theme_minimal()+
  labs(x="",y="",title="12-month percent change in Arizona house prices",
       caption="@lenkiefer Source: Freddie Mac House Price Index")+
  theme(plot.caption=element_text(hjust=0))+scale_y_continuous(labels=percent)

Okay, if we are interested in seeing how prices compare to zero we might try an area chart.

# plot house price trens for Arizona:

ggplot(data=dt[state=="AZ"],aes(x=date,y=hpa12))+geom_area(fill="blue",alpha=0.25)+theme_minimal()+
  labs(x="",y="",title="12-month percent change in Arizona house prices",
       caption="@lenkiefer Source: Freddie Mac House Price Index")+
  theme(plot.caption=element_text(hjust=0))+scale_y_continuous(labels=percent)

It might be nice to have gradient shading according to whether or not price changes are positive or negative on a year-over-year basis. We cannot do this directly with ggplot’s geom_area, but we can make it happen with a little tweaking:

# tweak data:
dt2<-copy(dt)
dt2<-dt2[,":="(hpa.up=max(0,hpa12),
               hpa.down=min(0,hpa12))
    ,by=c("state","date")]


ggplot(data=dt2[state=="AZ"],aes(x=date))+
  geom_area(aes(y=hpa.up),fill="blue",alpha=0.25)+
  geom_area(aes(y=hpa.down),fill="red",alpha=0.25)+
  theme_minimal()+
  labs(x="",y="",title="12-month percent change in Arizona house prices",
       caption="@lenkiefer Source: Freddie Mac House Price Index")+
  theme(plot.caption=element_text(hjust=0))+scale_y_continuous(labels=percent)

Okay, but sometimes prices are down a little, and other times down a lot. It would be nice to have shading vary by how much prices are up or down. We could take two approaches.

First, we could build a bar chart and have the fill vary according to house prices like so:

ggplot(data=dt[state=="AZ"],aes(x=date,y=hpa12,fill=hpa12))+
  geom_col()+
  scale_fill_gradient2(low="red",high="blue",name="12-month\n% change")+
  theme_minimal()+
  labs(x="",y="",title="12-month percent change in Arizona house prices",
       caption="@lenkiefer Source: Freddie Mac House Price Index")+
  theme(plot.caption=element_text(hjust=0))+scale_y_continuous(labels=percent)

Alternatively, we could keep subdividing the range of house prices and have shading vary by how much house prices have increased or decreased. Let’s try that with three break points on each side of zero.

df.az<-dt2[state %in% c("AZ")]  #subset data

df.az<-df.az[,c("date","state","hpa12"),with=F]  #only keep relevant columns
colnames(df.az) <- c("date","grouping","y")
origin<-0
# compute max deviation
max.y<-max(abs(df.az$y-origin))
nbands = 3
horizonscale<-max(abs(df.az$y-origin))/nbands
h1<-horizonscale
h2<-horizonscale*2
h3<-horizonscale*3
h1n<- -horizonscale
h2n<- -horizonscale*2
h3n<- -horizonscale*3

df.az <- df.az[ , ":="( ypos1  = ifelse(y>0,min(y,h1),0),
                    ypos2  = ifelse(y>h1,min(y,h2),0),
                    ypos3  = ifelse(y>h2,min(y,h3),0),
                    yneg1 = ifelse(y<0,max(y,h1n),0),
                    yneg2 = ifelse(y<h1n,max(y,h2n),0),
                    yneg3 = ifelse(y<h2n,max(y,h3n),0)) ,by=c("date","grouping")]
df.az<- df.az %>% select(-y) %>% gather(type, value, 3:8)
colnames(df.az) <- c("date","grouping","band","value")
df.lk<-data.frame(band=c("ypos1","ypos2","ypos3","yneg1","yneg2","yneg3"),
                  vmin=c(0,h1,h2,0,h1n,h2n))

df.az<-left_join(df.az,df.lk,by="band")
## Warning: Column `band` joining character vector and factor, coercing into
## character vector
df.az$v2<-ifelse(abs(df.az$value)<abs(df.az$vmin),df.az$vmin,df.az$value)
require(RColorBrewer)
## Loading required package: RColorBrewer
col.brew <- brewer.pal(name="RdBu",n=10)


  ggplot(data=arrange(df.az,value)) +
  geom_ribbon(aes(x = date,ymin=vmin, ymax = v2, fill=band,group=band),alpha=0.75)+
  scale_fill_manual(values=c("ypos1"=col.brew[7],  #assign the colors to each of the bands; colors get darker as values increase
                             "ypos2"=col.brew[8],
                             "ypos3"=col.brew[9],
                             "yneg1"=col.brew[4],
                             "yneg2"=col.brew[3],
                             "yneg3"=col.brew[2]))+
  labs(x="",y="",title="12-month percent change in Arizona house prices",
       caption="@lenkiefer Source: Freddie Mac House Price Index")+
      theme_minimal()+
  theme(plot.caption=element_text(hjust=0))+scale_y_continuous(labels=percent)

Here we have sliced prices into 6 regions, three positive and three negative. The further away from 0, the darker the color.

Now, if we just squish everything together, inverting the negative valuet to postiive values and overlaying each of the 6 colors, we’ve got a horizon plot:

df.az<-dt2[state %in% c("AZ")]  #subset data

df.az<-df.az[,c("date","state","hpa12"),with=F]  #only keep relevant columns
colnames(df.az) <- c("date","grouping","y")

df5 <- df.az[ , ":="( ypos1  = ifelse(y>0,min(y,h1),0),
                    ypos2  = ifelse(y>h1,min(y,h2)-h1,0),
                    ypos3  = ifelse(y>h2,min(y,h3)-h2,0),
                    yneg1 = -ifelse(y<0,max(y,h1n),0),
                    yneg2 = -ifelse(y<h1n,max(y,h2n)-h1n,0),
                    yneg3 = -ifelse(y<h2n,max(y,h3n)-h2n,0)),
              by=c("date","grouping")]
df6<- df5 %>% select(-y) %>% gather(type, value, 3:8)
colnames(df6) <- c("date","grouping","band","value")
#df6<-left_join(df6,df.lk,by="band")
df6$vmin<-0
df6$v2<-ifelse(abs(df6$value)<abs(df6$vmin),df6$vmin,df6$value)

  ggplot(data=arrange(df6,value)) +
      theme_minimal()+
  geom_ribbon(aes(x = date,ymin=vmin, ymax = v2, fill=band,group=band),alpha=0.75)+
  scale_fill_manual(values=c("ypos1"=col.brew[7],  #assign the colors to each of the bands; colors get darker as values increase
                             "ypos2"=col.brew[8],
                             "ypos3"=col.brew[9],
                             "yneg1"=col.brew[4],
                             "yneg2"=col.brew[3],
                             "yneg3"=col.brew[2])) 

Yeah, might be hard to see what’s going on here.

How about an animated gif:

horizon gif

What we have done is densify our data. By compressing the vertical area by a factor of 6 (3x >0, 3x <0), we can show the same data in a much smaller space. Of course we have to interpret it too.

You know what’s cool? We can stick these things inside a data table widget!

Adapting code from timelyportfolio, try this:

############# magic! ##############################################################

library(htmltools)
library(DT)
library(d3horizonR)


myf3<-function (s="Ohio"){
  d.out<- filter(dt,statename==s)$emp.pc
  return(d.out)
}


myf4<-function (s="Ohio"){
  d.out<- filter(dt,statename==s)$hpa12
  return(d.out)
}


emp.dt3 <- dt %>% select(statename)
emp.dt3<-unique(emp.dt3)


emp.dt4 <- emp.dt3 %>%
  #mutate(y = lapply(x, function(x) {cumprod(1 + runif(365, -0.05, 0.05))})) %>%
  mutate(x = lapply(statename, myf3 ) )  %>%
  mutate(x = lapply(x, function(dat) {
    d3horizon_chr(
      list(dat),
      options = d3horizonOptions(height=20),
      width = 400
    )
  }) )%>%
    mutate(x2 = lapply(statename, myf4 ) )  %>%
    mutate(x2 = lapply(x2, function(dat) {
      d3horizon_chr(
        list(dat),
        options = d3horizonOptions(height=20),
        width = 400
      )
    })
    )

m<-
datatable(
  emp.dt4,
    caption = 'Annual growth in employment and house prices',
  escape = FALSE,
  colnames=c("State","12 month % change\n in employment ",
             "12 month % change \n in house prices"),
  options = list(
    columnDefs = list(list(width="400px", targets = 2:3)),
    fnDrawCallback = htmlwidgets::JS(
      '
// not the best way but works fairly well
function(){
  HTMLWidgets.staticRender();
}
'
    )
  )
) %>%
  tagList(htmlwidgets::getDependency("d3horizon", "d3horizonR")) %>%
  browsable()

Oh yeah! We’re going to find a lot of uses for this in near future.