When to Viz
One of the biggest barriers to making great data visualization is the time it takes to do so. But that’s typically because we wait to start visualizing until it’s late in the game. This post is on when to viz.
The Typical Data Visualization Process
Let’s say you want to run a survey to find out how much employees have integrated what they learned from your company’s half-year intensive training on diversity, equity, and inclusion.
Typical projects like this start off with a topic to explore and perhaps some specific questions that need answers.
Then we convene committees to re-word the specific questions and look at the topic through other angles and eat bagels and yammer on (not a big fan of committees over here). This part of the process takes weeks.
Finally, there’s enough agreement that someone with expertise drafts a survey questionnaire. Someone identifies the pool of potential participants. Someone drafts lovely emails gently nudging employees to take the survey.
The data collection period takes a month or more, because low response rates prompted more committee meetings where consensus landed on extending the survey period and offering incentives for survey participation and someone had to run out for that iPad.
Once the committee has decided enough data has been collected, someone spends a week or more cleaning and organizing the dataset. It was only supposed to take a couple days but the dataset was messier than expected.
Then comes analysis. At this point the project enters the exploratory phase. Weeks of statistical tests and disaggregation on every demographic variable. Folks play with the dataset, graphing every question, maybe in testing out multiple chart types, on the lookout for the interesting patterns and informative results.
Then you get into the communication phase, where you prep your results for discussion. Except the committee wanted the meeting two weeks ago. What’s taking so long? So you slap together the graphs you’ve been generating during exploration and rush into the meeting. Where discussion is scattered, heading down rabbit holes, and unproductive. Everyone leaves frustrated because this whole endeavor was time-consuming, expensive, and ultimately inconclusive.
I’ve been here. Sometimes the meeting turns out a lot better. But it seemed we always had that slap-it-together-this-was-due-yesterday rush at the end. Which compromised the quality of our data storytelling, ultimately under-informing the answers we set out to find.
It happened so routinely, you’d think we would have predicted it. But each time it was like “How did we end up out of time again?”
It’s because we think visualizing only happens after we have the data.
The Revised Process
Good data visualization takes time. So does everything else leading up to it.
The trick is to spread out the visualization process so it doesn’t all occur after data analysis is complete. This might seem counter-intuitive – visualize BEFORE you have the data?? – but if you think about your whole process, you have a pretty good clue about what the data will be before it lands.
Back when you first thought through the committee-generated goals of the project, you were articulating the 3 (plus or minus) big questions the project needed to answer. And you know for certain people will want to see those 3 big answers, disaggregated. You can already start fashioning a slideshow focused on answering those top 3 right up front.
While the committee meets (and meets), you can learn some efficiency tricks and develop templates for your data viz.
As soon as you have the survey instrument constructed, you know exactly what the response options will be that you’ll need to graph. Granted, you don’t have the data, but you can easily get dummy data or data from the last time you ran this survey – right?
While you’re in that quiet data collection period, you can do more than send nudging emails and cross your fingers for a high response rate. You can be graphing.
(Matter of fact, I’ve seen some cases where analysts use dummy data to create some possible outcome scenarios and run those by committee members to get them thinking about how they’d respond. Prepping them for possible decisions and setting them up to take actions.)
That way, when the data comes in and you’ve got it scrubbed up, you just have to replace the new numbers in the graph templates you’ve already made.
You don’t need to wait until crunch time to start graphing. In fact, if you disperse the work of visualizing, you can kinda eliminate crunch time altogether.
When to viz? Always Be Vizzing.