Introduction
This digital project analyzes artworks in the Tate collection and works jointly held with the National Galleries of Scotland. The goal of this project is to turn a large, messy spreadsheet into clear, accessible information. I have created two visualizations that show the distribution of media across the collection and the cumulative number of works by medium up to each decade.
Sources
This project analyzes the ARTIST ROOMS dataset, which contains metadata for roughly 70,000 artworks owned by Tate or co-owned with the National Galleries of Scotland. I focus on investigating the collection’s medium composition and how the medium varies with acquisition year. Our analysis admits a selection bias because I exclude the missing entries from either of the columns.
Processes
I downloaded the dataset from the provided Github link. Since the dataset is quite large, I relied on Excel and Python for their built-in tools that make data cleaning and organization easier. I created two visualizations, and each needed a different cleaning procedures. Therefore, this section is divided into two parts, with one part explaining the steps for each visualization.
For the pie chart, I started with 69,200 rows and 3,413 distinct medium labels. I calculated the count of each medium, exported those counts to a CSV, and found that the top 12 media account for about 70% of the entire collection. I then created an Excel sheet with a collapsed version of the data that keeps only the 12 largest categories (grouping everything else as “Other”) to derive the pie chart.
For the stacked area chart, I first dropped rows with missing values in the “year created” field. This left 37,145 rows, and I limited the analysis to the top five mediums. I computed the number of works acquired in each year for those media and wrote the results to a spreadsheet. To reduce the noise, I aggregated counts by decade and then calculated cumulative sums to show how the totals build over time.
Presentation
Significance
The Tate dataset has about 70,000 rows and 10+ metadata columns, with plenty of missing entries. In raw form, it’s hard for viewers to learn anything meaningful about the collection. I believe this is one of the best cases in which people can leverage data visualization to organize messy information into clear categories and use intuitive shapes to show patterns and change over time within a structured framework. For example, my visualizations focus on the dataset’s medium field. Before any cleaning, there were 3,413 distinct labels, which is a large number for viewers to make sense of. After consolidating to the 12 largest categories and showing their share of the collection in a pie chart, the medium distribution becomes clear at a glance for the audience. I think this project shows how powerful data visualization can be in helping people quickly and effectively understand what a dataset reveals.
There are also limitations and potential selection bias in this project. First, I focus only on medium trends and don’t examine other variables. Second, I drop records with a missing acquisition year or medium. Third, the pie chart shows only the top 12 media, and the stacked area chart shows only the top 5. These choices keep the presentation focused and readable, but they may underrepresent artworks in less common media and the stories those works could tell.