Analytics anyone can use.
Data prep anyone can use.
Analytics for organizations.
Cloud analytics for organizations.
I spend most of my days visualizing data that folks share with me. But sometimes the world has different plans. Sometimes, without any warning, we’re flooded with raw data and need to make sense of it quickly. Or we know there’s a data story out there, but no single data set pulls it all together. How do you approach this as an individual when there are hundreds (or thousands or millions) of data points to collect, before exploration and visualization can begin? By crowdsourcing a data set, of course.
Crowdsourcing a data set means building it with others. Everyone contributes their own data points to create a data set for others to use.
With all the buzz around the Women's March taking place all over the world, I tried my hand at crowdsourcing a data set. I wondered: What are the core issues motivating people to march? To find out, I created a survey, which is one of two ways to crowdsource your data.
If you think about it, a survey is a tool used to collect data from individuals who each offers a piece of a larger story. Scientists and academics have been using survey data for hundreds of years—John Snow being one of the first to give us a clearer picture of cholera epidemics through a field survey in London. He personally interviewed residents in an affected area, and in turn collected the data they shared: how many ill in a home, how many sick, for how long... data, pure and simple.
Rigorous scientific studies require surveys be vetted carefully to eliminate bias, but that doesn’t mean that a quick, informal survey doesn’t provide value. Polling data is a survey, after all. And while surveys are generalized data, those generalizations can provide insight.
Online tools such as Google Forms and SurveyMonkey allow us to build and distribute surveys quickly and cheaply, but there are some gotchas. Last weekend, I built a quick six-question survey in Google Forms. I wanted to build a viz that would update as responses rolled in, so I kept the questions short, the answers concise. I thought I had constructed a survey that would result in a clean, easy-to-viz data set. I was wrong.
The construction of the questions—and the survey options at my disposal—were my downfall. I asked the following six questions to the participants in the Women's March across the world:
When I opened the spreadsheet with the results, I got a bit of a surprise.
My data needed cleaning—and a lot of it. Why? Questions 1 and 3 were the problems, and of course they corresponded to the answers I was most interested in. What had I done wrong?
The solution for multiple choice questions? I had to pivot my data. That means that instead of the data being wide, with each response in a single column, I needed to make the data tall and keep "like" data in the same column.
Instead of this:
I needed this:
Steve Wexler has a terrific post on pivoting your data in Tableau, and it’s worth a read.
Because of the way I’d built the survey, I was trying to collect live data with Google Forms while pivoting and analyzing. Ultimately, I captured the data in a static file to viz the data, which was not ideal.
Much later into the project, I had 456 responses to the survey (it's grown since, alas). I filtered out single-response issues (these were often the "other" items), and I was a little surprised that environmental policy tipped the scale as the issue that was most pressing among the respondents. That might have everything to do with the largest segment of responses coming from the Pacific Northwest, my home.
My vizzes are now completely separated from the survey, which is still collecting data. I’m not happy about this, but I learned an important lesson about how to structure the questions in a survey. Next time, I’ll simplify even more, or I’ll use the next method.
The second method for crowdsourced data sets involves spreadsheet sharing. We’ve done this since the beginning of the internet, but it's only really taken off in the last few years. At the same time as I was collecting data in my survey, Erica Chenoweth (@EricaChenoweth) and Jeremy Pressman (@djpressman) created a shared Google Sheet with some very simple instructions. They asked for crowd estimates for the marches on January 21, 2017. There were simple instructions: include numbers along with city, state (if applicable), country, and verifiable source of the data. They offered up the spreadsheet via social media, and it took off. People from all over the world added data, and in under 24 hours, they had a data set for everyone to use.
Was it perfect for visualizing? No, but it was quick and dirty, and fit the immediate need. In under 24 hours, I’d already seen someone from Tableau Public who had cleaned enough of the data to visualize it with election results underneath. I haven’t seen the media pick data up this fast, but folks playing with data have.
One of the biggest advantages to crowdsourcing data is that you generate interest—everyone becomes a contributor. Working together, subject matter experts, open-data organizations, and data scientists alike come together to help people like me explore data in depth.
What kind of data and projects have you seen come together through crowdsourcing? What are some of the advantages and disadvantages with this type of data? I’m already working on my next crowdsourced data set, so let’s continue the conversation!