Blog Post – Tableau

Craigslist Furniture Visualization

I used Tableau to create a visualization comparing Craigslist furniture listings in the three largest cities in the United States: New York, Los Angeles, and Chicago.

Craigslist is fairly easy to scrape because the format is consistent across pages. I used the BeautifulSoup and pandas packages in Python to pull the listing title, price, and listing location (i.e., neighborhood, town, borough, zip code) of furniture listed in for sale in each city into a dataframe, which I exported as a csv file. I did not “clean” the listing titles or locations, and I’m glad I didn’t, because retaining those two fields in their “raw” state raised some of the questions with data visualization that Johanna Drucker and Lev Manovich discussed in their respective articles.

Once I had the data, I went to Tableau. The public platform is robust and allows users to upload tabular or text data for analysis and visualization, and to combine those visualizations into a dashboard. Users can upload more than one file and combine individual tables as you would with relational databases, but you don’t need to know SQL in order to do so.

Initially, I found myself struggling to actualize graphical representations I had in my head as a result of the crowded interface and levels of abstraction. I am used to more explicitly transforming data with functions or formulas through code, with a more minimal visualization functionality in Periscope/Sisense, which is a different business intelligence platform that connects directly into your database and generates charts off code. It took time to figure out where I needed to drag-and-drop a field in order to add it to the x vs the y axis. For example, the first time I dragged the City field into the workspace, it generated coordinates and a map popped up. Its ability to infer what format to present your data in is both helpful and frustrating.

The platform also felt grounded in traditional info visualization the favored bar charts, pie charts, and line graphs, regardless of the data uploaded, which makes sense given that it is largely an enterprise software. I’d also wager this is why Tableau offers little documentation, favoring few example videos and recommending user join a community of others to learn.

With that being said, I was able to explore the dataset and create a final dashboard that made use of statistical charts, while also finding the edges of it’s utility in my current exploration, with my current knowledge of the tool.

In trying to reconcile Neighborhoods, Boroughs, and towns outside from each city, I was reminded of what Johanna Drucker said in “Humanities Approaches to Graphical Display”: that no data pre-exists its parameterization. They are already interpreted values. Any patterns observed or conclusions drawn from location with such fluid boundaries would be subject to criticism given the contested history of neighborhoods in major US cities. In addition, this data was user generated, which might make it easier to understand that the labels and classifications are subjective and influenced by the person creating the listing. I wondered about ways I could use Tableau in ways that were more temporal or flexible, as Drucker described.

I also found myself wanting to explore direct visualization, since I took interest in the listing metadata. Listing titles were long strings of keywords, and the locations were far from standard. I attempted to combine similar or duplicate locations that followed a slightly different format, but debated whether or not that was my role, for reasons mentioned above. The uncleaned listing titles, locations, and images feel critical to understanding cultural, social, and economic ideas underpinning craigslist, yet it was difficult to find ways to incorporate the media objects themselves, other than adding in screenshots and returning the fields themselves.

Ultimately, the platform does enable laymen to create well-crafted visualizations, and I am please with the initial visualizations I put together.