Today I'm going to teach some visualization best practices. Now, this isn't an exhaustive list by any means, but these are a few of my favorites that I think will be most important for you as you start on your quest towards being a viz guru. I'll highlight five tips. First, to know your audience. Second, to know your data. Third, to use color purposefully. Fourth, that less is more and simpler is usually better. Finally, that you should get feedback early and often. First up, know your audience. Who asked you to make this viz and what questions are they trying to answer? For example, executives have very different needs from say, product managers. Executives want to consume data quickly and efficiently. So they might only look at your dashboard for maybe five seconds at a time, meaning you need to treat every pixel as precious. A product manager on the other hand asks for reports so they can understand how their ideas are panning out. How did that AB test go? Did users click the new button? Did the users who were supposed to see the new button actually see the new button? While sea levels tend to care most about the very high-level segments, product managers like lots of detail that they can dig into. They are the ones who will ask you for every slicer and dicer that they or anyone in earshot can possibly imagine. Let me be the first to tell you how is the business doing or how is my tests doing, do not count as valid questions. Don't be afraid to be specific and push back a little when people are being vague. Having a bit of that gumption is part of what makes you a good analyst after all. Next, you should know your data. Now, you might say, "Of course, I'll know my data does it right" But the success of your dashboard depends on you really getting the data behind it. If you don't get it, neither will your end user. Even worse is that they'll be able to tell that you didn't understand it. So to really get it, you should first get familiar with how the data was collected. Was it entered into a system by a person typing it, or by many people typing it, or was it automatically collected by the click of a button? Is it aggregated somewhere between that system and you? What cleaning did it go through, if any, before it got to you? Was it de-duped? Are all text field uniform down the whole column? By that I mean, is there say an instance of a word where in one row it's all uppercase and in another it's all lowercase? You'll have to do that cleaning yourself if so. Next, you should investigate where the holes are, like missing dates and time series data, for example. Know roughly what the totals look like, such as how many rows are there altogether, how many rows per day, and so on. You don't need to memorize the exact numbers, but you do need to be able to see when your results are off by a factor of 10, or two, or some such. The quickest way to lose your credibility is present a dashboard to your stakeholder with dramatically wrong data that shows that you weren't paying close enough attention. Equally important, you really need to know what the grain of the data is. As in, what does one row mean? Is it one click on a button? Is it one customer? Is it one sales transaction? Or is it a whole country? You'll treat all of these data differently from each other, and you'll want to visualize them differently as well. So the grain of the data is very important. Next, a great way to take your viz from good to awesome is to use color purposefully. What I mean by that is that while you might enjoy looking at a dashboard that is all decked out in every color on the color menu, your stakeholders will probably just find it confusing and overwhelming. They might even physically go, whoa, but that is not the whoa reaction you want. When you're designing, I suggest starting with making everything just a basic gray, and then add color as you need it. You'll be surprised first by how much you can convey with a single color, and second, by how much sleeker your dashboard looks. Whatever you do, please don't use more than about five colors. If you do use color as a differentiator, try to avoid using the same color to denote two different things within a single dashboard or report. Now, when it comes to color, you might also need to be mindful of folks with red-green colorblindness. People with this condition can't differentiate between some shades of red and green. They just appear to be the same grayish brown here. This is mitigated by using orange and blue instead of red and green. But if you absolutely must use red and green together, there are things you can do to make sure that everyone can see your viz and at least most of its glory. If you must rely solely on color, above all, make sure that your chosen red and green are at different brightness levels because that differentiates them enough so that colorblind folks can tell them apart. But maybe consider trying to use another visual cue for indicating difference in addition to color. You could use shape like up and down arrows, or Xs and circles, or you could add a label, or you can do all three; use color, shape, and labels. On a related note to color, don't be afraid to use white space. If you cram every last miniscule bit of the page full of information leaving little to no margins and little to no space between elements, you're liable to get that bad whoa reaction again. This would be akin to reading a story that has no spaces between all the words and there's no punctuation to the sentences. No. Just like when you write a sentence and use punctuation and spaces between words to indicate when a reader should take a breath or pause for effect, so you should also use white space. Don't be afraid to pad the outside borders a little more than the defaults, and make sure that your dashboard elements are clearly separate items. Don't just rely on drawing lines and borders between. In fact, I would say don't do that at all. Give your data room to breathe, and your users will be able to breathe more easily as well. Next, remember that less is more and simpler is usually better. Now, I could have lumped this in with talking about using color purposefully, but I feel that it's important enough to be treated on its own. Make your data the star of the show. Minimize the use of numbers unless they are part of the data or unless it's unavoidable, like with days and years. But when you're abbreviating a month on a date axis, try not to use the month number. Instead, use the first three letters of the months name. If you minimize your use of extraneous numbers, it makes your data pop that much more. Also as with color, use decorations like grid lines sparingly. I'd say start off by removing all of the grid lines and only add them back in if you really need them later. Axis should be minimalist and lightweight, not intrusive. If you're labeling any data points, consider whether you'll even need to show the y-axis at all. Because contrary to what your science teacher told you, it's not always necessary. Finally, spare yourself some heartache and get feedback early and often. Don't spend too long working on your viz without making everything pixel perfect, without getting feedback along the way. The single most frustrating thing you'll encounter is finding out after days and days of hard work that you were working on entirely the wrong thing all that time. Also the more time you spend with your data, the more familiar you'll be with it, which is good, but that means it's easier for you to lose sight of what your audience knows or doesn't know. Check in with them often to keep your eye on the prize. So to recap, my five favorite viz best practices: know your audience, know your data, use color purposefully, remember that less is more, and get feedback early and often. Keep these in mind and you're more than halfway there to mastering data visualization.
