即使小小的決定，也可以產生大影響。本影片採訪史考特．貝里納托（Scott Berinato），史考特為《哈佛商業評論》英文版資深編輯，著有《好圖表練習本：做出更好的資料圖表的訣竅、工具和練習》（Good Charts Workbook: Tips, Tools, and Exercises for Making Better Data Visualizations, HBR Press, 2019）和《哈佛教你做出好圖表》（Good Charts: The HBR Guide to Making Smarter, More Persuasive Data Visualizations）。
Sarah: Scott, there's a lot of interesting psychology about what happens in our brains, when we look at charts, why we find them so persuasive. Can you just give us a quick overview of some of that research?
Scott: Sure. And there's a lot going on right now, and the field is changing rapidly, but it's really exciting. It's really interesting research. Basically, the most important thing to know about the psychology of looking at charts is a lot goes on very quickly. And there are things you can't control about how your brain reacts to charts.
It's a little different than reading a book, where we sort of a contract with the author, and we're reading sentences and in order that they're presented to us. With charts, our eyes go straight to things, and we can't stop it from happening, so we see what stands out. We can't help but notice color. We can’t help but notice outliers.
As soon as that happens, as soon as our eyes land on these things, it immediately starts trying to form narratives. It starts trying to make meaning out of it. And you can't really stop this from happening, which is why some of these challenges with building charts are so difficult, because once you decide on a visual approach, people going to react to it a certain way, whether you want them to or not.
Sarah: Tell us a little bit about problems with charts that may lie to people or just mislead them.
Scott: Sure, I think sometimes people do use charts maliciously. That they try to convey an idea that doesn't exist. There was a case recently in Congress, where somebody presented a chart about Planned Parenthood that plotted cancer screenings versus abortions in a way that made it look like abortions have risen above cancer screenings, when that's not the case at all. It was just a deliberate attempt to mislead.
Sarah: So can the same thing happen when it's not actually a lie, but just sort of by accident?
Scott: I would say that's actually more often the case, is that people just aren't paying attention to the tools they’re using, or they’re not thinking critically about the way they're showing their data, and they end up sort of stumbling into a misleading chart. If you look at this here, we see revenue going up in this chart, and it looks fine. It's a nice trend. Everything's rising. But if you look at that y-axis label, it says cumulative revenue. And what cumulative means is every year we're counting all the previous year's revenue. So if we change this chart and show the year's revenue by color, you see the first year's revenue is counted five times.
And the second year's revenues is counted four times, and so forth. So, you've seen this growth, when in fact, if we just plot each year's revenue by itself, you see that growth is not the story at all. It's actually precipitous decline. And so maybe somebody was trying to pull one over on their boss here, but more likely, they just weren't thinking through what they were showing, and what it really represented.
Sarah: There's clearly a lot of ways to tell stories with data. Is there such a thing as a truly objective chart?
Scott: I'm going to argue there isn't. You can make charts very honest and as objective as possible, but when you think about it, there are a lot of decisions that go into every chart that are somewhat arbitrary. So if you think of a chart with years in the x-axis, 1981, 1982, to 1983, how much space goes between years? How wide should that chart be? The width of a chart is an arbitrary decision. It's usually a determined by the media you're presenting on, on a screen, on the phone, and it turns out, this has a dramatic effect on what people see in the chart.
You make a chart very wide, it's going to flatten out curves. It's going to make things look mellow or slower change. If you smoosh that onto a phone, very tall narrow screen, you're going to see much more dramatic change. And there's no right answer on how wide a charge be. It’s somewhat of an arbitrary decision. So every chart has decisions in it which affect what people see and affect the story that you're telling.
Sarah: So, we talked a little bit just now about that x-axis. What about the y-axis? What are some common problems there?
Scott: Well, there's a couple. The most famous one is the truncated y-axis, when people don't start their y-axis at zero. And that has the effect of creating, again, a more dramatic story, because you're actually seeing curves that are steeper, change that is more dramatic, because that empty space that you’ve cut out of the chart isn't there anymore. So if we look at this chart here, it looks like a job satisfaction really goes down over the course of your life. As you get older, your current and expected job satisfaction continues to decline, and then there's this crossover, and it looks pretty dramatic.
But if we chart the same information with the full y-axis, it almost looks unremarkable, flat, and that's actually an accurate portrayal. Some people say you should never truncate the y-axis, but I think those kinds of draconian rules are not actually useful. There are cases where it makes sense. Scientists, sometimes, are dealing with data sets that are in very limited ranges, and it would just not be useful for them to look at these flat lines all the time. But it certainly is something to think about every time you're making a chart. Are you exaggerating trend? Are you making things look more dramatic than they actually are, by truncating your axis?
Sarah: What about the dual Y-axis? Because this is one where, I confess, I have a hard time even reading these charts.
Scott: Yeah, the dual y-axis is actually, in some ways, more interesting and more complicated. To me it's something that's used more often and without question, but it should be question as much as the truncated y-axis. And the dual y-axis, as you can see here, is when you're measuring two different things in the same visual space. So, it's like playing football soccer on the same field, at the same time. So, you have on the left here, we're measuring percentage share for Tesla from 0% to 10%. And on the right, we have millions of total vehicles sold. Those are apples and oranges.
They’re not really comparable in the same space. But what's more, if you look at that green line, the 2025 projection, it reaches about 25% or so up the height of the bar. Does that mean Tesla is going to reach 25% of the total vehicle sales? Well, if you look at the label, no, it’s about 2%. So we're way off here in our estimation. But because they're in the same visual space, we're making the comparison. We can't help it. That’s just how our brains work. We look a chart; we try to form a narrative out of it. If we change this chart to how it should be plotted, a better way to plot it, where we're using a single measure, total vehicle sales, you see the story’s entirely different. Tesla’s share is much smaller.
Sarah: Can you just give us some examples of what some of those other common decision points are?
Scott: Sure, one common decision point would be what data you choose to show. So say we want to talk about record sales, and we have a chart that shows trend in vinyl sales. Everybody knows vinyl’s him again. Everybody loves records. And sure, this chart shows that vinyl sales are spiking. I think they've gone up since even since this chart was made. But let's take that same data, and let's extend it back to the 1970s, when there were no other options for music. And now, if we just focus on the 2014 bar, in that first chart, it's the tallest bar. It's the spike.
When we go back to 1973, that little nub there at the end is the same data represent entirely differently. And then if we, again, choose to change our reference points, and compare vinyl LP sales not to past and future vinyl LP sales, but rather to all music sales, you get another view of the 2014 records sales. So that's three different ways to show the exact same data.
Sarah: So what's a good way to know if you've crossed the line from trying to be persuasive with your case to actually misleading people?
Scott: Well, the rule I always fall back on is the Golden Rule, which is if you're building a chart and you're looking at what you've done, and you think, what I feel duped, if somebody showed this to me, knowing what I know about the full data? Knowing what I know about other versions of the chart I've tried? I often used the metaphor, are you zooming in on the idea or are you distorting it? Is it like using a microscope or a fun house mirror? Would I feel fooled or tricked? Or would it feel manipulative? That's probably a good sign, you've crossed that line.
Sarah: So, Scott, what's interesting here to me is we've been talking about how to persuade people with charts without lying to them. But as a consumer of charts, you also want to be a savvy chart consumer. What are some things to think about if you are being shown charts; how do you know that they're accurate?
Scott: It's a good question and it's hard to do. But I think we're getting better at it all the time as we see more and more charts. And I think the first thing you can really do, as somebody who's looking at a chart, is evaluate some of these common things that people do.
So, if you do see a y-axis that doesn't start at zero, just understand in your mind that that they've heightened the drama. And you have to ask yourself, what would that look like if it actually did start at zero? You may not be able to actually execute that on the spot, but understand that you're seeing a more dramatic story than may exist.
If it's 2 y-axes, and it's a lot of work, I understand that, but you kind of want to pull those two charts apart, and try to evaluate them individually. Don't compare the values, the visual values, look at one and understand what it says, then look at the other and understand what it says, and then try to compare. Don’t let the visual comparision sort of override the actual values that you’re evaluating.
Sarah: Scoot, thank you so much for talking with me today.
Scott: Thanks Sarah.