Those who know me well know that on any given weekend, I’m probably at the museum. I usually come in as a visitor, but ever since I started working at the Hammer Museum last year, I’ve also become a part-time tour guide and researcher. Art history provides a unique and fascinating way to examine the past and historicize the present, allowing us to ask questions like: what sorts of things matter to us? How have we learned to express ourselves over time? I’ve been itching to find a way to connect my burgeoning interest in data science with my lifelong love of the arts, and art auctions are an interesting intersection of the two. Art matters to us, and we pay a lot of money for it, but why? What sorts of art are valued most? These are some of the questions I asked myself when starting this project, finally narrowing down my queries into a central question: Does art become more expensive with time?
I turned to Wikipedia to begin my investigation. A history of art sales seemed like the natural place to look when thinking about the question of artistic value. I ended up with lists of the most expensive paintings and photographs ever sold. The Wiki list format was great for quick data scraping, though some inconsistent formatting had to be corrected before analyses could begin.
This project couldn’t be possible without the help of dplyr, stringr, rvest and plotly:
library(dplyr)## Warning: package 'dplyr' was built under R version 3.4.1library(stringr)
library(rvest)
library(plotly)The Wiki tables used for this assignment could be easily scraped using the html functions available from the rvest package:
paintings.raw = read_html("https://en.wikipedia.org/wiki/List_of_most_expensive_paintings") %>%
html_node("#mw-content-text > div > table") %>% html_table(trim = TRUE)
paintings = paintings.raw %>% select(adj_price = `Adjusted price\n(in millions)`, orig_price = `Original price
(in millions)`, painting = Painting, artist = Artist, year_painting = Year, date_sale = `Date of sale`, seller = Seller, buyer = Buyer)Wiki tables are unfortunately not very rigorous in consistently recording their data. I had to do some quick data wrangling before I could begin the analysis:
paintings$adj_price = gsub("\\$", "", paintings$adj_price)
paintings$adj_price = gsub("\\~", "", paintings$adj_price)
paintings$adj_price = gsub("\\ ", "", paintings$adj_price)
paintings$adj_price = gsub("\\+", "", paintings$adj_price)
paintings$adj_price = paintings$adj_price %>% as.numeric()
paintings$year_painting = paintings$year_painting %>% substr(1,4) %>% as.numeric()
paintings$artist = gsub( ",.*$", "", paintings$artist ) #fix van gogh
paintings$artist = gsub( " !.*$", "", paintings$artist)
paintings$artist = gsub("Gogh", "van Gogh", paintings$artist)The photograph process had a similar web scraping and data wrangling process:
photos = read_html("https://en.wikipedia.org/wiki/List_of_most_expensive_photographs") %>%
html_node("#mw-content-text > div > table") %>% html_table(trim = TRUE)
photos$year = gsub("(?<=\\()[^()]*(?=\\))(*SKIP)(*F)|.", "", photos$Work, perl=T)
photos$year[5] = photos$year[5] %>% substr(str_length(photos$year[5]) - 3, str_length(photos$year[5]))
photos$year[c(7,16,20)] = photos$year[c(7,16,20)] %>% substr(1,4)
photos$year[8] = photos$year[8] %>% substr(str_length(photos$year[8]) - 3, str_length(photos$year[8]))
photos$year[21] = photos$year[21] %>% substr(str_length(photos$year[21]) - 3, str_length(photos$year[21]))
photos$year = photos$year %>% as.numeric()
photos$Price = gsub("\\$", "", photos$Price)
photos$Price = gsub("\\,", "", photos$Price)
photos$Price = photos$Price %>% as.numeric() / 1000000 # price in millions
photos$Artist = gsub("\\ \\&\\ ", "", photos$Artist)
photos$Artist = sub('.*\\ ', '', photos$Artist)
photos$Artist = gsub("GilbertGeorge", "Gilbert&George", photos$Artist)plotly has great functionality when it comes to zooming in on plots and hovering over points for more information. I’ve included the names of each artist in the hover function. The zoom feature becomes particularly handy once we reach the 20th century and see more records of art being bought and sold:
paint.plot = plot_ly(paintings, x = ~year_painting, y = ~adj_price, type = 'scatter',
mode = 'markers',
hoverinfo = 'text',
text = ~paste(artist),
color = ~adj_price)
x = list(title = "Year Painted"); y = list(title = "Selling Price (in millions)")
paint.plot %>% layout(title = "Paintings", xaxis = x, yaxis = y)photo.plot = plot_ly(photos, x = ~year, y = ~Price, type = 'scatter',
mode = 'markers',
hoverinfo = 'text',
text = ~paste(Artist),
color = ~Price)
x = list(title = "Year Photographed"); y = list(title = "Selling Price (in millions)")
photo.plot %>% layout(title = "Photographs", xaxis = x, yaxis = y)On the surface, it appears our plots tell us a quick and easy answer: that no, these paintings do not become more expensive with age. A naive reading of these results would imply that paintings get cheaper with as they grow older. But an important caveat needs to be made, particularly regarding the data used to make this analysis. Though the most expensive paintings ever sold were not made that long ago, this does not necessarily imply that a painting made by a less well known artist, say, half a century ago, would be worth any more than art made from an artist of the 17th century. One shortcoming of this analysis is in its inability to demonstrate the historical value an object accumulates over time. So this interesting plot is just that: an interesting plot. Perhaps something can be said of the commercial value of artists who gained their fame relatively recently (post-Impressionism and beyond) because their fame burgeoned around the time that many other aspects of western living became commercialized as well (I’m thinking 20th century). These questions and hypotheses only encourage more questions and more research, incorporating more disciplines than data science. This is a start!