Michele Kiss, Analytics Demystified - Ten Tips For Presenting Data

November 6, 2017

Ten Tips for Presenting Data

Slide 1:

Thanks for having me this year at ObservePoint’s virtual Analytics Summit. I’m excited to be participating again.

Slide 2:

A little bit about myself, I’m a Senior Partner at Analytics Demystified. Chances are, if you’ve been in the analytics industry for a while, you have heard of some of my partners, Adam Greco, Eric Peterson, etc., have written some of the foundational books in the analytics space.

Slide 3:

A little bit about myself personally, I have three babies, two of the furry variety, one of the human variety, and all of them love the interwebs (sic).

Slide 4:

One thing to keep in mind before we begin…

Slide 5:

The things that I’m going to be talking about are tool agnostic. I won’t be talking about how this is how you should be doing this in Data Studio, this is how you should be doing this in DOMO or Tableau, or whatever you’re using for data visualization. What I’m going to be talking about are the principles that really lie behind it and why you should make some of the decision around data visualization that we’re going to discuss today.

Slide 6:

The second thing you should be aware of is the principles of data visualization are constantly divided by the vendors that we commonly use. So, if you think about the default chart that Excel is going to create or the severe love of pie and doughnut charts that many analytics vendors have, it’s just something to keep in the back of your mind so you can focus on those principles rather than going with what your vendor does as a default.

Slide 7:

On to our tips. Tip one is to recognize that how you present data actually matters. It’s really common for analysts to feel like they’re not being heard by stakeholders. They’re putting a ton of work into their reports and people aren’t necessarily paying attention or aren’t taking their recommendations. The problem is, if you’re not communicating data in a way that works for business users, then it’s really easy for them to tune out in that way.

Slide 8:

As we go through the years, this is really becoming something that is no longer optional for somebody that works in our space. As far as thinking about what skills I’m going to work on for 2018, where is my career taking me? You can see any of these job descriptions that are out there right now, even examples where they’re looking for somebody with a fairly small amount, one to three years of experience, they want people who know how to put together data presentations. They know about the principles of data visualization. This is really important for all of us as we continue down the path in analytics.

Slide 9:

Gone should be the days of a 20-page PDF, “dashboards.”

Slide 10:

Another thing clients have asked me to think about is why should I spend the time to make something look pretty? Most of us are really, really busy with the work of doing analysis, of pulling data, of automating it. It takes a lot of time, so it can be hard to say on top of that, I’m going to find the time to stop and think about how I’m visualizing and preventing that.

Slide 11:

The problem is that it’s not about making things pretty.

Slide 12:

It’s about making your data be understood.

Slide 13:

I’ll give you an example in a text basis. If you look at this paragraph that comes from a book on the subject: “Why should we be interested in data visualization? Because the human visual system is a pattern seeker…” Easy to tune out and nothing is presented in an important way. The thing that we do with data visualization is the same thing I’m doing here…

Slide 14:

Which is, “Why should we be interested in data visualization? Because the human visualization system is a pattern seeker. We can easily see pattern presented in certain ways. And we can present our data in such a way that the most important and informative patterns stand out.” Think of it that way, that you are taking what was a ton of data and presenting it in a way that people are able to detect the patterns that you as an analyst are seeing, and also that you’re able to pull the most important data out of it.

Slide 15:

My boss, when I was first starting in analytics, once said to me that, “As an analyst, you are an information architect.” I think that’s a really good way of thinking about it, that you need to be thinking about how you’re architecting that information so that it’s easy for your stakeholders to understand.

Slide 16:

It’s not actually about art, it’s about neuroscience.

Slide 17:

What you are trying to do when you think about data visualization is to optimize the memory funnel. There are several steps that this goes through.

Slide 18:

The first is Iconic memory. This is where pre-attentive processing takes place. We’re not even thinking about it. If you think about sitting in your office right now, there are things going on all around you. It might be somebody sending something to the printer, there’s a phone call going on down the hallway, maybe the sound of the elevator. You are able to tune out and understand the things you should be paying attention to, and these are the ones that you’re not. This is where all of that is taking place. We group objects together and we hone in on what’s important, but we’re not really thinking about any of this. We’re not deciding that we should overhear this conversation, but tune out the ding of the elevator.

Slide 19:

From iconic memory, information passes into our short-term memory. This is a really limited capacity system and information gets discarded is it’s not something that we deem to be useful. So, we can think about this like when you’re trying to memorize a phone number. It’s in your head, you’ve got it, you’ve got it, you type it in and it basically leaves your brain after that. Our short-term memory can fit—the estimate is about three to nine chunks of information.

Slide 20:

From there, if something makes it through short-term memory, it will then go into long-term memory. The goal with long term memory is that people retain the meaningful information, but not the precise details, so maybe not some of the exact number don’t matter, but the patterns do, or the differences between the groups. They remember that this test variation performed twice as well as the control.

Slide 21:

When we’re talking about short term memory and we’re talking about chunking data, that doesn’t mean that you can remember three to nine numbers, or that you can remember three to nine letters. It could be three to nine groups of numbers. It can be three to nine words. It can be three to nine song lyrics. The way that you chunk information ultimately allows you to remember more because if you’re thinking about the number of letters—trying to remember nine letters versus nine song lyrics, you’re going to remember a lot more information in the latter. We do this all the time. We do this with phone numbers, the way we make them into patterns. So, 650, 690, 2805. If you think about presenting information, what you want to be doing is chunking the information in such a way that your stakeholders can remember more.

Slide 22:

If you’re looking at something that is a bunch of rows and columns of numbers, somebody is really unlikely to remember this. If you’re lucky, maybe they’ll remember in January, Group A was 2 percent and Group B was 20 percent. But apart from that, they’re really not going to remember the details.

Whereas, once we visualize the data, it’s easy for them to chunk this information and remember more of it. If they look at this chart, they might remember Group B was going down, Group A was going up, and there was a weird spike for Group A.

Slide 23:

Your goal as an analyst is to have people think about what they’re going to do, so what actions they’re going to take based on the data, and not to really have to stop and think what all this data mean.

Slide 24:

Infographics are often abuse this. This is an infographic of top 10 salaries at Google from several years ago and the thing is, apart from the complete train wreck of a—I’ll call it a pie chart, sort of—in the middle, this is not really a visual representation of data. You might as well have prose describing the data because you have to read all of this to understand the information.

Going back to informative being more important than beautiful, this is not as pretty and colorful, but at a glance, you can see the exact same information and you can look and see which job category has the highest salary, which one has the lowest, which one has the widest range of value, and which three—Engineering Director, Human Resources, Senior Partner Technology Manager—are actually fairly similar. You can tell all of that quickly versus in the infographic style chart, you would have to read all of the information and kind of piece this together in your head.

Slide 25:

Informative is way more important than something being beautiful.

Slide 26:

Tip two is to not scare people with numbers. There are little things you can do to make numbers more user friendly.

Slide 27:

I like to remind people that we like numbers. Not everybody does. You may have plenty of stakeholders that you present to, to whom the idea of reading a report of a lot of information is kind of scary for them.

Slide 28:

To make numbers more friendly, you can think of things like making sure you’re including necessary commas. Don’t make people count the number of zeros in something. Again, by default Excel or whatever you’re using may not be doing this for you.

Slide 29:

Skipping unnecessary decimals. If your range of values is from like 2 to 90 percent, you don’t need two decimals places. There is enough difference between those to just be able to show the 2 and it’s easier for people to read.

Slide 30:

But on the flip side, if you have number that are really close, it’s important the you’re including necessary decimal places. That’s where you are looking at it and really understanding what makes sense based on the data is critical.

Slide 31:

Other examples where I’ve seen far too many decimals used is an analyst conducting an estimate of the revenue impact that comes down to 29 dollars and 13 cents.

Slide 32:

It’s an estimate, call it 1.4 million here is an estimate and that’s not really going to make a big difference.

Slide 33:

Unfortunately, what this comes from is a confusion that because you’re being more precise, that your data is somehow more accurate and the two don’t necessarily go together.

Slide 34:

Other quick tips to make numbers more friendly. Right aligning numbers. That allows you data to for a quasi-little bar chart where people can see which the biggest number is by which number is the longest. Little things that you can do in any system that you’re using just to make things look a little bit less scary.

Slide 35:

Tip three is to maximize the data-pixel ratio. Edward Tufte talked about the data-ink ratios, Stephen Few, a little more in the digital realm, talked about the data-pixel ratio. The technical definition is: it’s the number of non-white pixels that are devoted to data display, versus being devoted to things like borders and gridlines, things that are not necessarily required. If you think about it, this is actually a violation of the data-pixel ratio because it’s adding a bunch of fluff. The visual at the back has nothing to do with the information I’m trying to convey.

Slide 36:

Another way I like to think about it, is just pretend you’re really stingy with the printer ink and that you don’t want to waste ink on certain things.

Slide 37:

This is an example of a gross violation of the data-pixel ratio. Rather than the black or colored pixels being used for data display, they’re being used for backgrounds. They’re being used for big, heavy borders between charts. They’re being used for grid lines. That’s the kind of thing that you’re thinking about.

Slide 38:

I can show you a couple of quick transformations. On the left, ways that you can move to something that maximizes that data-pixel ratio. The word that doesn’t need to be repeated over and over again, once we know from the header that that’s “Region,” simplify. Even the dollar sign doesn’t need to be repeated. If I know that column is dollars, then I’m good, I don’t need to see it every time. We don’t need the heavy borders, we don’t need the heavy title bar. Really kind of light grid lines in between.

Slide 39:

On a bar or a column chart, removing borders—Excel loves to put these in there still. Google Sheets doesn’t give you the option to remove borders, which frustrates me sometimes. Making things easier for people to read; horizontal, condensing axis so we don’t need to repeat over and over, 300,000—three zeros—we can just make it clear that this is in the hundred thousands.

Slide 40:

Looking at a line chart, removing unnecessary legends. If you only have one series on a chart, you don’t need a title that tells you this is visits and a legend that tells you that. It’s the same information. And grid lines, sometimes they can be useful to have them there if you want people to be able to track along rather than just get the idea of an overall trend, but they don’t need to stand out as much and they don’t need to be as visually important as the actual data.

Slide 41:

Tip four is to save 3D for the movies.

Slide 42:

These two charts have the same information. In the top left one, I can see at a glance that the bar is slightly above 150,000. In the bottom one, I can slightly tell it’s above 150,000, but I kind of have to work for that. I have to follow it along that gridline, now I turn the corner, now I find the 150,000. So what you’re doing is adding this cognitive step where somebody has to think about what they’re looking at.

Slide 43:

Don’t even start me on this one. Not only the 3D, but the overlapping, etcetera.

Slide 44:

Now rule, I’m going to call it Rule 4B, there is an exception to every rule. There are certain times where 3D may make sense.

Slide 45:

For example, the recent election. But it’s not necessarily taking into account the population or the size of these different areas.

Slide 46:

3D gives us a very, very different picture. So, if it is accurately telling the story, then you can consider it an exception to the rule.

Slide 47:

Tip five: friends don’t let friends use pie charts. It is really easy to hate on pie charts, but I’m going to give some examples of where they can, and maybe can’t, be useful.

Slide 48:

A typical why pie charts can fail is that they’re not actually representing data that is part of a whole.

Slide 49:

Another reason they fail is that brains aren’t always great at judging the size difference in areas or in circles. So while we’re able to look and say this number is about twice this number while we’re looking at a bar, a column, a line, we’re not great at doing that with areas or circles. So it makes it harder for people to judge the relative differences.

Slide 50:

Often times, way too many pieces or exploding pieces are added to a pie chart.

Slide 51:

There are some acceptable pie charts. Sarcastic pies.

Slide 52:

Artistic pies

Slide 53:

And there are also bad, non-pie charts. If we look at this. We’re looking at gun deaths in Florida. Looking from early 1990s and we see 873, and then we see it trending up and we see this was when a legislation enacted. Then look at the axis, in this case it’s been flipped so that something looks like a decrease, when in fact it was an increase. If you torture the data, it will tell you anything you want.

Slide 54:

Use of area. Like I said, infographics often abuse the principles of data visualization. There’s really no reason for that.

Slide 55:

But there are times when a map might make sense. So if we’re looking at different sports ball teams in the United States.

Slide 56:

This on the other hand, just no, no, just stop.

Slide 57:

But, remember Rule 4B, there is an exception to every rule.

Slide 58:

So, places where pie charts might be okay. If it has very few items.

Slide 59:

And I’m going to be a stickler and saw less than three. The data has to represent parts of a whole.

Slide 60:

You can only use one. As soon as you need to compare, for example comparing Facebook in 2011, 12, and 13, it’s no longer useful.

Slide 61:

If you need to compare, you can use an alternate. You can have, for example, stacked bar charts.

Slide 62:

So really pie charts, like 3D, are okay when they are really the best possible way to visualize the data and visualize the message that you’re trying to convey from the data.

Slide 63:

This is actually the best example I have ever seen. On the left, we have four different things. A might be half. On the right, A is half. If that’s what you were trying to say, then in this case, a pie chart is absolutely the right choice.

Slide 64:

That leads me to tip six, which is choosing the appropriate chart. This is where, as an analyst, you have to think about what message you want to convey. When you look at the data, you’re drawing certain conclusions. What takeaway do you want stakeholders to get from that?

Slide 65:

There are a couple guidelines. Typically, a line chart would be used to emphasize a trend.

Slide 66:

You can highlight specific points if you’re trying to say something about, well these two points that are 18 months apart or that it’s double this time. You can kind of highlight specific points to give a little bit more information.

Slide 67:

If you’re trying to emphasize differences, you might use a bar or column chart where it’s easy to compare, say the countries here.

Slide 68:

If you don’t have much space, you might consider using sparklines. Sparklines can be useful they’re not going to give you an axis that tells you the precise information, but they’re going to give you an idea of what the trend if.

Slide 69:

Even tables of numbers can be graphical. For example, adding in little bar charts in the table itself. People are able to quickly glance and say, “Oh, email has the best conversion rate,” without having to read all the numbers. It makes it sound like laziness in some ways, like pretend that you’re trying to have people read as little as possible.

Slide 70:

You can even do these in Google Sheets. They have options for sparklines. I have this quick little doc created. Bit.ly/googlesparklines embedded into Data Studio. Tableau will do them by default. They are typically an option in most places, so that you have that table and visualization at the same time.

Slide 71:

So, there are some good resources if you’re trying to pick an appropriate chart. Extreme Presentation is a little chart picker, etcetera. There are good places you can go where you can say, “This is what I’m trying to do with the data and this is what I’m trying to convey,” and it will help you decide what the best chart selection is.

Slide 72:

Tip seven. If it matters what kind of chart type you use, then you don’t want to be combining them for no reason. In this case, I mentioned that often a line chart might be used for a trend. In this case, this is not a trend, we’re not trying to display that.

Slide 73:

Really the cardinal rule is that they shouldn’t be mixed for no reason to fit information onto a page. So, if you’re thinking, “I just have one more thing I need to fit on, I’m just going to shove it into the same chart that already has other information.” What’s going to happen is, as soon as you put two pieces of information together, people’s brains are going to assume there’s a connection between them.

Slide 74:

In this case, this is a very similar example to the other one. We have a column chart and we have a line chart, but in this case, we’re mixing them and we’re showing them together because we’re trying to demonstrate that they trend similarly. In this case, it makes sense to mix them.

Slide 75:

Instead of here, where we’re not trying to say much of anything. We’re trying to fit two things onto one chart.

Slide 76:

Another option might be to have two side-by-side bar or column charts. This points somebody to say really quickly, “Hey, this is the months that we saw a high volume and a decently high click-through rate.”

Slide 77:

Tip eight is don’t use axes to mislead people. I’ll warn you in advance there’s a lot of Fox News examples. In this case, starting the axis needs to start from zero because when you’re not starting it from zero, it distorts the data to make differences look more significant than they may actually be. Not to mention, cherry-picking specific months, potentially to mislead.

Slide 78:

In this case, a line chart where apparently the axis itself is meaningless because did you know 8.6 is more than 8.8? when you go and you track this yourself, it’s totally not the same thing.

Slide 79:

There are sometimes, especially when you’re putting together a dashboard where information maybe doesn’t change all the time, where the number, maybe it’s your conversion rate and it doesn’t shift dramatically, the most truthful option is to start at zero. But if you need to show differences on a metric that maybe doesn’t shift that much, my recommendation would be to have a second view of the information so that it’s like the idea that you have one that’s kind of truthful, starts from zero, the next one is a zoomed in version of that.

Slide 80:

Tip nine is to never rely solely on color. It’s pretty common that people will use colors as “good” or “bad” or “above, below.”

Slide 81:

The problem is, about 10 percent of the population is colorblind. And it’s not all red/green colorblind. There are a lot of different types.

Slide 82:

All black and white printers are colorblind, so as soon as you have an indicator that’s read and green, they may look exactly the same if printed on a black and white printer.

Slide 83:

And all newspapers are colorblind.

Slide 84:

That doesn’t mean you can’t use any red or green, often that’s a pretty understood color scheme. People will know good versus bad. But you do have to do a check to make sure that if somebody’s colorblind, that they’re going to be able to understand it and that there’s something else to give them the information, and that if somebody prints it in black and white, it’s going to be understood.

Slide 85:

And especially if somebody is a global company, you also need to be aware that there can be cultural differences in how colors are perceived. Red can range from a color of mourning to celebration. Yellow can be something that’s a cheery color or it can mean pornography. Blue can be sadness or safety and protection. Green can be environmental color or it can mean death. These colors don’t always have the same meaning to everyone, so that’s an important consideration as well.

Slide 87:

Tip ten is to use color with intention. In this, the colors are completely meaningless.

Slide 89:

Or as I like to think of it—barfing a rainbow.

Slide 90:

A good way to use color with intention, is to use brand colors. Often those are clearly understood by somebody. Facebook’s blue, YouTube’s red, etcetera. It will help them more quickly without them have to necessarily read the axes or look at the logo, just to quickly and visually understand.

Slide 91:

If you are using multiple charts with the same information, you have to keep the colors consistent. Because otherwise what might happen is that somebody might jump from, “Oh Facebook’s blue on this chart and red on this chart.” Not only are you making them think really hard to do those comparisons, but more likely, they’re going to draw an incorrect conclusion.

Slide 92:

This is something that Data Studio added fairly recently where they will keep values consistent.

Slide 93:

Using color with intention, another option is to link your different types of axes. Let’s say you have information that’s on two different axes. You have one that’s with one piece of information and one that’s with another. Just tying the axis to its series color can help people to immediately understand that this one goes with spend and this one goes with people talking about this metric. Coordinate your series and axes colors.

Slide 94:

One bonus tip.

Slide 95:

Tip 11 is that this actually applies to dashboards too. Dashboards follow all the same basic data visualization rules. Plus, they have one important rule on top of that.

Slide 96:

A dashboard, the best definition, the Stephen Few definition, is a dashboard needs to be a visual display, show the most important information. It has to be entirely on one screen or one page.

Slide 97:

The reason for that, and the reason this is important, is if you are expecting people to look at something on a dashboard, you are relying on their short-term memory. You’re relying that they will look at your spend information of the first page, and maybe your conversion rate that might be on the second page and that they’re going to draw conclusions between those two. As soon as you start taxing their short-term memory, you make it less likely that they’re reach those conclusions. Basically, we don’t what to overestimate people’s short-term memories. We want to work with their natural abilities.

Slide 98:

This is not a dashboard.

Slide 99:

This is not a dashboard.

Slide 100:

Keeping in mind the one page rule, you can have reports, they can be longer, but think about dashboards as one page where somebody can look and compare information at a glance

Slide 101:

Some really good books on this topic are The Wall Street Journal Guide to Information Graphics and also Stephen Few’s book about Information Dashboard Design. I hope that this was useful. I hope these tips were something that will help you as you’re putting together analysis or as you’re putting together dashboards and reports.

Slide 102:

You can always reach me on Twitter, via email, on Measure Slack.

Previous Video
Peter Symuleski, NBC Universal - 10 Organizational Tips to Get the Most Out of Your Data
Peter Symuleski, NBC Universal - 10 Organizational Tips to Get the Most Out of Your Data

Tips focusing on creating a working environment that optimizes output across implementation and analysis.

Next Video
Eric Myers, Keystone Solutions - Driving True Value: Moving Your Data Stream Beyond the Click
Eric Myers, Keystone Solutions - Driving True Value: Moving Your Data Stream Beyond the Click

Learn how to drive value from your data stream by driving economic, revenue and cultural value from your data.