Duc Quang Nguyen

Fails et succès en mi-saison 2017/18 de Super League: comparaison historique

Thu, 01 Feb 2018 00:00:00 +0000

_{This post is part of a dataviz remix series of the Swiss football league’s results. The motivations behind it are explained in this post.}

La Super League reprend ses droits ce samedi après la trêve hivernale. L’occasion de faire le point sur les (contre-)performances à la mi-saison.

Time for sports to get a dataviz makeover

Sat, 11 Nov 2017 00:00:00 +0000

I have to confess, I am not into sports. Nor do I know much about football. But as media and data professional, I am irked by how sport is summarised and visually conveyed.

I am referring here to the ubiquitous league table. Present everywhere, whether being in print or online. All team sports’ results are recapped in such table.

_{^{RTS Sports’s football table for the Swiss Super League}}

What’s wrong with league table?

I am aware I am having a rather a diehard stance here. Of course league tables are useful to show where team currently stand and to cram stats such as points, wins, loss, goals,…

My issue with it is for people like me who do not regularly follow sports. One cannot get from such table how teams performed over time, like has a team been any good/bad lately?

IMHO, league table is a legacy of old day media/journalism where data/news were scarce. And like with the current general media crisis, there is an unmet demand to synthesise sports information. Furthermore, sports generate more and more data. There are hence plenty of ways to summarise its information to be both insightful and visually attractive.

But most importantly, league table fails to achieve its purpose. According to Wikipedia’s definition:

A league table is a chart or list which compares sports teams ….. by ranking them in order of ability or achievement

League table does the job listing achievements, but does not fulfil its primary purpose to rank teams by ability. Summing up different metrics (such as goals, wins, losses) as it is done routinely, is both primitive and fallacious.

I am not a statistician nor a mathematician, but there have been for decades much more insightful metrics in sports (and in football in particular) that are informative to anybody. For instance, metrics such as the expected number of goals scored or conceded per game by a team would be understandable and useful to anybody (much more so than summing up the number of goals).

Content produced by media needs of course to be understandable and to appeal to the largest audience possible, but it should also be able to give to convey the most relevant information at hand. So my aim here is to post more insightful football metrics instead of a race to the bottom.

There are of course plenty of people and media that came to that same conclusion before. 538 and the Financial Times, to mention only two, have done outstanding work visualising, modelling and conveying football analysis. As often, English-speaking media are the forefront.

_{^{The official Swiss Football League’s actually provides an evolution chart. The expression “spaghetti chart”" brought to a new level…}}

It is easy to only criticze, so as a challenge I will regulary post quick data visualisation “remix” of the Raiffeisen Swiss Super League results.. I hope this will be interesting and not appeal only to stathead or dataviz amateurs. In a second step, I hope to look into alternative (and better) metrics to rank and classify teams’ ability.

If you are interested, check the posts tagged as #remix, #football, #swisssuperleague

Interactive mekko charts in R

Thu, 19 Oct 2017 00:00:00 +0000

Mekko what?

Despite its confusing name, Mekko or Marimekko chart is a simple yet effective data visualisation form.

Here is a rundown in R for the interactive Mekko chart under (which is part of this story). I used ggiraph, a great ggplot2 extension that binds d3.js to ggplot2. This allows to easily turn a ggplot2 object into an interactive graphic.

The bar height here is proportional to the number of jobs and shows the large difference of jobs for different occupations.

It can be thought of a regular stacked bar chart with an additional axis. The only alternative to a Mekko chart would be to “facet” as shown under.

_{^{Data and graphic: NCCR}}

FT Chart Doctor provides an excellent explanation of the Mekko chart’s pros & cons. And most importantly, it explains the origins of its enigmatic name (spoiler - a Finnish textile and fashion company’s renowned for its bright repeating patterns).

Mekko chart can boast one on the largest number of synonyms: Mekko, Marimekko, mosaic, matrix or proportional stacked bar chart. I’ll stick here with FT’s designation, proportional stacked bar chart. IMHO, the most meaningful name.

Proportional stacked bar chart chart in R

There are various ways to create such chart in R. There is of course a dedicated R package for it, ggmosaic built on top of ggplot2. And various gists or posts about it.

I use ggiraph a great ggplot2 extension that binds d3.js to ggplot2. This allows to easily render a ggplot2 object as an interactive graphic.

Wrangle the data

Here is the data to plot in a tidy form. They are the job figures in Switzerland, by occupation group (group, an ordered factor) and by country of origin (origin, Swiss or foreigner)

library(magrittr) library(tidyverse) library(ggiraph) jobs <- c("Managers", "Professionals", "Technicians and associate professionals", "Clerical support workers", "Service and sales workers", "Skilled agricultural workers", "Craft and related trades workers", "Machine operators and assemblers", "Elementary occupations") data <- tibble( group = factor(rep(jobs, each = 2), levels = jobs), origin = rep(c( "foreigner", "swiss"), 9), value = c( 128640.53, 297823.49, 219209.61, 615405.86, 168977.09, 606273.69, 77485.11, 298635.30, 210439.96, 412074.04, 8740.28, 83336.17, 151257.24, 323894.43, 68990.99, 93424.47, 126136.14, 86100.60 ) ) data

Compute for each occupation, the proportion of Swiss/foreigner and the total number of jobs.

data %<>% group_by(group) %>% mutate( share = value / sum(value), tot_group = sum(value) ) %>% ungroup()

A proportional stacked bar is composed of rectangles. Rectangles’ coordinates are computed in two steps.

The number of jobs by occupation (y dimension) and the proportion of Swiss/foreigner (x) by occupation. It relies on cumsum() to express all values in a 0 to 1 coordinates.

data %<>% group_by(origin) %>% arrange(desc(group)) %>% mutate( ymax = cumsum(tot_group) / sum(tot_group), ymin = (ymax - (tot_group/sum(tot_group))) ) %>% ungroup() %>% group_by(group) %>% arrange(desc(origin)) %>% mutate(xmax = cumsum(share), xmin = xmax - share) %>% ungroup() %>% arrange(group) data %>% select(group, origin, ymin, ymax, xmin, xmax) %>% arrange(desc(group))

## # A tibble: 18 x 6 ## group origin ymin ymax ## <fctr> <chr> <dbl> <dbl> ## 1 Elementary occupations swiss 0.00000000 0.05336812 ## 2 Elementary occupations foreigner 0.00000000 0.05336812 ## 3 Machine operators and assemblers swiss 0.05336812 0.09420840 ## 4 Machine operators and assemblers foreigner 0.05336812 0.09420840 ## 5 Craft and related trades workers swiss 0.09420840 0.21368795 ## 6 Craft and related trades workers foreigner 0.09420840 0.21368795 ## 7 Skilled agricultural workers swiss 0.21368795 0.23684109 ## 8 Skilled agricultural workers foreigner 0.21368795 0.23684109 ## 9 Service and sales workers swiss 0.23684109 0.39337573 ## 10 Service and sales workers foreigner 0.23684109 0.39337573 ## 11 Clerical support workers swiss 0.39337573 0.48795332 ## 12 Clerical support workers foreigner 0.39337573 0.48795332 ## 13 Technicians and associate professionals swiss 0.48795332 0.68289448 ## 14 Technicians and associate professionals foreigner 0.48795332 0.68289448 ## 15 Professionals swiss 0.68289448 0.89276323 ## 16 Professionals foreigner 0.68289448 0.89276323 ## 17 Managers swiss 0.89276323 1.00000000 ## 18 Managers foreigner 0.89276323 1.00000000 ## # ... with 2 more variables: xmin <dbl>, xmax <dbl>

This is enough to plot a basic proportional stacked bar chart chart

gp <- ggplot(data) + geom_rect(aes(ymin = ymin, ymax = ymax, xmin = xmin, xmax = xmax, fill = origin), colour = "white", size = 0.2) gp

Make it interactive

For an interactive version, two additional optional aesthetic can be provided to ggiraph:

data_id an aesthetic to identify elements on hovering

tooltip the HTML tooltip text

data %<>% mutate( data_id = paste0(origin, group), tooltip = paste0( "<em>", as.character(group), "</em><br>", origin, " ", round(share * 100, 1), "%<br>", "(", prettyNum(round(value), big.mark = " ", mode = "character"), " jobs)" ) ) # hack to escape single quote data %<>% mutate(tooltip = gsub("'", "`", tooltip))

Same ggplot2 call as previously, replacing geom_rect() with ggiraph::geom_rect_interactive(), and add the two freshly created aesthetics.

ggiraph() renders a ggplot2 object as an interactive chart in RStudio’s Viewer or for the web browser.

gp <- ggplot(data) + geom_rect_interactive( aes(ymin = ymin, ymax = ymax, xmin = xmin, xmax = xmax, fill = origin, data_id = data_id, tooltip = tooltip), colour = "white", size = 0.2) ggiraph({print(gp)})

Et voilà, a plain looking but functional interactive proportional stacked bar chart

Prettify

Direct labelling

I labelled the jobs directly in the chart’s bars. In this case, because some bars are narrow I scaled the labels according the bars’ height. I added also the % values.

# job labels tibble labels <- data %>% filter(origin == "swiss") %>% mutate(y = ymax - 0.01, yRange = (ymax - ymin)* 100) %>% select(group, xmin, y, yRange) %>% ungroup() value_labels <- data %>% select(group, origin, xmin, xmax, ymax, share) %>% mutate( x = ifelse(origin == "swiss", xmax, xmin), y = ymax - 0.005, label = paste0(round(share * 100), "%"), hjust = ifelse(origin == "swiss", 1.05, -0.25) ) gp2 <- gp + geom_text( data = labels, aes(x = xmin + 0.008, y = y, label = as.character(group), size = yRange), hjust = 0, vjust = 1, colour = "white" ) + geom_text( data = value_labels, aes(x = x, y = y, label = label, hjust = hjust), vjust = 1, size = 2.7, alpha = 0.7, colour = "white" ) + scale_size_continuous(range = c(2.2, 4.5), guide = F) ggiraph({print(gp2)})

Theming

gp3 <- gp2 + theme_minimal() + scale_x_continuous( position = "top", expand = c(0.01, 0.01), labels = scales::percent, breaks = scales::pretty_breaks(n = 4)) + scale_y_continuous( expand = c(0,0), limits = c(0, 1.02) ) + scale_fill_manual( values = c('#4c6570', '#b15953') ) + theme( axis.line.x = element_blank(), axis.text.y = element_blank(), axis.title = element_blank(), legend.position = "none" ) ggiraph({print(gp3)}, width = 0.9)