Abi: From the campus of Harvard Medical School, this is ThinkResearch, a podcast devoted to the stories behind clinical research. I'm Abi.

Brendan: And I'm Brendan, and we are your hosts. ThinkResearch is brought to you by Harvard Catalyst, Harvard University's Clinical and Translational Science Center.

Abi: And by NCATS, the National Center for Advancing Translational Sciences.

Brendan: Data visualization is a powerful tool that allows investigators to communicate their data quickly and concisely. Today, Jess Cohen-Tanugi, visualization specialist with the Harvard Libraries, walks us through what makes good data visualization and common mistakes clinical and public health researchers make when incorporating data visualization into their work. So, Jess Cohen-Tanugi, thank you for joining us, and welcome to ThinkResearch.

Jess Cohen-tanugi: Thanks for having me.

Brendan: So you're a visualization specialist at Harvard Libraries. Tell us about your job. What does that entail?

Jess Cohen-tanugi: Well, part of my everyday work is helping anyone in the Harvard community who wants to make some sort of data visualization from even the notional phase. Like you have an idea for a project, but you're not sure what it will entail. To all the way to the other end of the spectrum where you're struggling with a particular thing you need to get your data visualization polished, and finished, and ready. I also put on a series of training workshops during the semester and sometimes during the summer And. Do one-on-one consultations as needed.

Brendan: Who are some of the people that you work with?

Jess Cohen-tanugi: I work with folks from all across the Harvard suite of school, I suppose. So not just at FAS, though certainly a bunch of folks at FAS, particularly in the humanities. Like the history and even literature departments. I also work with people at the Kennedy School, the School of Public Health, even the Medical School. So I see a good number of people from all across Harvard.

Brendan: And maybe a weird question, but what is data visualization?

Jess Cohen-tanugi: That's a good question and a surprisingly hard question. Data visualization, the way I think of it, is usually when you have some sort of tabular data. Or not tabular, but organized data in some fashion, and you want to make something visual out of it. This could be anything from a well-organized chart with some color coding to indicate higher and lower values to any charts and graphs you might picture that you might see in papers, or presentations, or even online. Data visits everywhere these days.

I would say that there are other forms of presenting information visually that I'm not so much of an expert in, like making good looking posters and reports. That's more graphic design, and while you're visually organizing data, it's less the charts and graphs that I really help with.

Brendan: So talk a little bit more about that, the difference between data visualization and graphic design.

Jess Cohen-tanugi: Yes, they are different, but they often come together. So data visualization, for me, I usually think of it as you're visually encoding data. So you might encode the number of people in each age group as the height of a bar, so that's visually encoding your actual information. Versus graphic design is making all sorts of decisions on how to organize text such that one's eye goes to the thing you intend first, second, third that people don't get confused about the flow of a report.

That things look like they line up, or they don't according to the style you're going for. That your colors all mesh well together. So not just choosing a color palette for your chart or graph, but choosing what colors would mesh well and be harmonious for a whole report. When you're making a chart or graph, you often have to make a bunch of graphic design decisions, so it does become hard to disentangle the two. You will inevitably be doing both. But my focus is really on getting people up and running on charts and graphs.

Brendan: Let's talk about why data visualization's important. Why it's important for in general, but also for science and people who are in public health or clinical research who are-- their work actually has potential impacts on people's health. And so, getting that information out is incredibly important. So maybe if you could just talk a little bit about why data visualization and effective data visualization's important?

Jess Cohen-tanugi: Sure. Well, data visualization is very important and ubiquitous, actually. Hard to get away from it these days. Even if you do a simple Google search, you'll see a histogram of the busiest times of day for the store you're looking for. It's very important for a number of reasons.

One is, if you collect a large amount of data or even a medium amount of data, what do you do with it until you visually encode it in some way? You can't even really parse what you have unless you make at least simple charts and graphs. You, audience of this podcast, are most likely creators of data visualization. You have a responsibility to make effective data visualizations, and make it as clear as possible.

So that people don't misinterpret your findings and come up with the wrong conclusions and the wrong results. Even if you've done your methods correctly, if you haven't created an effective data visualization, and people can't read it correctly, then you're in trouble.

So it really is important to learn the basic principles of how people parse charts and graphs to include annotations when necessary to provide additional clarity, to point out key things you want the people who are looking at your data visualizations to notice. Make it easy for people to interpret your graphs in the way you want them to be interpreted.

Brendan: Right. So, how do you do that? What are some basic things? Because I think there's-- and I asked you about effective data visualizations, but I think you also brought up the idea of accurate and not misleading data visualizations.

Jess Cohen-tanugi: So in order to help people correctly interpret your data visualizations, it helps to be very sure of what message you want to put out there first. Often, people make the mistake of trying to do too much with a single chart or graph, and I get it. It's really hard when you have limited space in an academic paper or a poster. It's really hard not to include everything.

But you'll have a better impact if you can manage to narrow down your story to just what needs to be told in a particular chart or graph. And then, if you've done that, if you've decided on the message, it should be a lot easier to choose the chart or graph type you need in order to convey your information visually.

And often, that's quite obvious. In the medical fields, there are certain charts and graphs that you know you're going to use for certain situations, and that's good. I always say, go with the field standards first. But after you've done that, you've done whatever standard in your field, there are opportunities to make your charts and graphs easier to read.

There's been a recent trend in data vis where people do what I call an annotations layer. So instead of just straight-up having the chart and graph with the labeled axes and any additional information in the caption, people have been putting notes directly on data visualizations themselves. A good example of this would be COVID cases over time.

So we've all seen those charts and graphs, and if you've been looking at them, you'll notice they'll often mark holidays because those are times when people gather. So they'll mark up Christmas Day, Memorial Day, 4th of July, and you'll see the corresponding spikes after them. There's been a nice trend in simplifying axes.

So instead of having text rotated, so you have to tilt your head to the side to read it, people have started labeling axes at the top with the text reader-friendly. So just take a look at recent trends. If you look through the news reports, you'll see there are a bunch of ways people have made these charts and graphs easier to read.

Instead of having a key off to the side labeling colors or what marks mean what, they'll label that directly on the graph. So the person looking at it doesn't have to bring their eyes to the side and back over and over again in order to read your chart. So these are some trends that help bring in additional context and also make charts and graphs easier to read and quicker to read.

Brendan: Where are some places that you look to and think are doing data visualization well?

Jess Cohen-tanugi: Well, I get the news every day, and I think the New York Times data vis team is phenomenal. They have so many cool projects. They do ways they collect and then break down data visually. They've got a whole team, and they win lots of awards for it. I also get the Washington Post and find a lot of good examples there.

There are some data vis gurus that I find just extremely inspiring. Alberto Cairo. He is just a wonderful person who does a lot of really cool data of his projects, a lot of more complicated graphic design plus charts and graphs things. And he has several books on data vis out there, and there are lots of good examples of effective charts and graphs in his books.

I also like the work of Nathan Yao, who has a website full of good examples of data vis.

Brendan: If people don't have experience-- maybe these are most of the people that come to you. What can they do besides emailing you and asking for help? What can they do to explore or try and improve their data visualizations?

Jess Cohen-tanugi: That's a great question. What I recommend as people are trying to ramp up and understand how to make better and more beautiful or more effective charts and graphs, I always recommend they start a Google Slide deck. And when they see good examples, they put them in there just so they have a place to put-- when they see a nice data visualization that they admire for some reason, they just put it in one spot with hopefully a link to where they got it from.

And then, when you look through them, you can start to realize, OK, I see a pattern. I like these data visualizations because they do this, and I don't like these others because they don't do that. So you start to be able to put names behind your preferences, your own preferences. And the same with graphic design.

If you're trying to learn graphic design, it's really helpful to look at a bunch of designs and imitate what you like and try to verbalize what you don't like and try to apply that to your own work. So I always recommend people do that. There are a bunch of really fun, flexible tools that allow you to ramp up and make things very quickly.

So it's always fun to play around with tools like Flourish.studio or Datawrapper You'll see even as part of newspapers, people use these tools to make data visualizations all the time. They're really powerful, and you can get up and running quickly. So if you're a little intimidated about learning Stata or a programming language, or even something like Tableau, you don't--

It doesn't have to be that hard. You can dabble and make it something pretty beautiful pretty quickly.

Brendan: Yeah, I liked what you said earlier about even just the idea of data visualization as a step in understanding the data. I've definitely experienced that. When you have a bunch of whatever data it is, it's so helpful to put it in a visual system of some kind to tease out trends or patterns.

And I think that what you said about it being-- it can be part of the research process. It's not the end stage. It's part of you understanding your data.

Jess Cohen-tanugi: It's iterative, yea. The data vis process and data gathering or data manipulation, it's all iterative. So you'll look at your data, which is so exciting. That's always the best part. You'll see what you have there for the first time and see something interesting.

This happened to me yesterday. I was working with someone, and we finally got her data into a form where we could see an uptick in a certain-- I guess I could say the specifics. She saw a really exciting uptick in amount of works translated from German into Polish right before World War Two. And it was totally unexpected, and no one had gathered this data and looked at it before.

And just by doing a simple line chart and seeing whoa, what's that spike there? We were able to say, oh, cool. That's interesting. We have some follow-up questions. We're going to gather some more data. We're going to look into that. We're going to see what were the topics in those works? And who is doing these translations? And dig into it more.

So data vis really is just part of the process where you isolate something that you think's interesting or strange. And then you dig into it more, either with the data you've already gathered or by some further data gathering or further experimentation.

Brendan: Yeah, that's really cool. So I know that you work with researchers from across the University. When you do work with researchers from medicine and public health, what are some common issues or things that stand out in your mind when you think about that field?

Jess Cohen-tanugi: So the School of Public Health and the medical fields in general, a very common theme I find is that there's a lot of data there that the folks just want to put in one chart or graph. And I see it like it's a struggle. That it's really hard, and I want to recognize that this is challenging. And that the people who do try to put all the-- or do want to put all the data in, I know why they do that. I know why they want to.

But what I really try to do with them is sort through the things that they must have, and the things that they just want to have. And try to figure out how we can pare down the data to make the message come out more clearly. It genuinely is quite hard because there's a lot of stuff that I can see why they want to cram it in there in the one chart or graph.

A grouped bar chart where the groups are eight bars wide. That's a little too much. And I know it's all very interesting, but can we focus a little bit? Or put the data in a different form?

Brendan: Yeah, and I think-- so you gave a webinar for Catalyst PGE a little while back, and I think some clips of that are going to be available on our Writing Center website that go into a lot more depth about some of the tools that are out there. But in that webinar, you mentioned the chart picker tools and a design company that uses a big poster.

Which I thought was really fascinating that these are design professionals, and they have this big poster in their office. And every time they need to make a new visualization, they go look at the poster. It's like, this information-- maybe you can talk a little bit about that. But I thought it was so interesting even though even these seasoned, experienced people still go back. It's almost like going back to the textbook every time they start a new project.

Jess Cohen-tanugi: Yes. So, what I was talking about in my webinar was the team at the Financial Times. And I've heard this, but I should probably confirm rumors. It's a good story, anyway, that the design team, the visual design team themselves, they actually go in front of this poster, this giant poster that has a whole bunch of chart types, broken down by general.

Like the different buckets if you have geospatial data, et cetera. So they break the chart types down by overall category. And then, they have notes to themselves next to each chart. It's really cool and fascinating. It's like their cheat sheet. It's their notes to self.

This chart's really good when we're trying to do A, B, and C, but pay attention. Because if you have too many dots in it, too many points that you need to put on it, things start to get chaotic, and you should choose another chart take. So it really is their cheat sheet of what they've learned in the past.

And they put it all together, and they keep it updated. And they have it on GitHub openly available to anyone. I find that tool to be really fun, and it helps me a lot when I'm thinking about when I've chosen a message. And I'm thinking, OK, now I want to compose a visualization.

Which one of these would tell me, say, would emphasize the difference in quantity between two different points of time? And then I'll look and see which one might help me do that.

Brendan: You talked about narrowing down the message, and you gave the example of eight-- a multiple bar chart that's eight bars wide. How do you narrow down your message?

Jess Cohen-tanugi: That's the really hard part, and the part that will be hardest for the researchers who have done so much work in gathering so much data and have so much that they want to say in a limited space. But it's the necessary evil to just really make yourself think. Boil it down in essence to what do I have to say in order to make my point here?

Will it be a big deal? Will my message not come through if I combine a couple of these bars together and have not so many subcategories? Maybe I can compare one group to the rest. Is there any other way I can simplify this and still get my message through?

So yeah, that's basically what I try to help talk people through. I try to creatively think of ways where once they describe what they're trying to do and the thing they really need to shine through, I push them to think. OK, well, would it really be a big deal if we didn't include this? Or OK, we should include that, but we can combine it with something else. So stuff like that.

Brendan: And that's something that you do as an expert, but I feel like that's something that somebody could do with a colleague maybe in a different field or even a lab mate that isn't on their project. Just that outside set of eyes and ears to be that. They're the fictional reviewer, or poster attendee, or whatever.

Jess Cohen-tanugi: Yes, I usually encourage people to come to me to share their work with even their spouse, or their colleague, or someone who knows enough to be able to reasonably be an audience member for that work. And ask them, what's the first thing that comes through? What are you getting from looking at this? And if they say, OK, well, the first thing I see is blah, blah, blah.

Then say OK, anything else? So just to see the order in which the person notices things, and whether that is sufficient for their purpose as the author of the visualization and the paper.

Brendan: That's great. All right, so common mistakes. What are the things you see over and over again that make you laugh a little bit?

Jess Cohen-tanugi: Well, the classic mistake is truncating bar charts, which is never allowed no matter what.

Brendan: What is that?

Jess Cohen-tanugi: So the thing about bar charts is that in bar charts in particular and not other types of charts, is that you're visually encoding your data such that people are comparing one length to another. So they're comparing how big is this bar compared to this bar? And if you cut the access and have the access not start at zero, the ratio between the sizes of those two bars is off.

And you can look this up. I highly recommend you Google it because it's actually satisfying to see just how wrong people can get with this, and there are so many examples if you Google image search truncated bar chart. I just went to town, and one of my presentations included a billion of them. So never ever do that.

The problem is, a lot of people do truncate their bar charts because their data shows a slow upward trend or something. So if the bar is when you start at zero, the bars look like they're just barely getting bigger, and it looks like there is no trend. Well, that's a good indication that you shouldn't be using a bar chart because with something like a line chart, you're just comparing how high a point is compared to another point.

The point is not comparing how long is this bar compared to this bar? So you can feel free to zoom in if it's a line chart. Does that make sense?

Brendan: Yeah, so like if it's that slow trend, it's not so much about, say, you have 10 time points. It's not so much about the change from time point two to four. It's about the change from 1 to 10, and you don't need a bunch of bars to show you that. You really only need two points with a line connecting them.

Jess Cohen-tanugi: Exactly. Yeah, I would say that's about right. So if you're trying to show a subtle upward trend, a bar chart will not do that for you. But if you're trying to say, this quantity doubled, a bar chart is perfect for that. It's very easy for us to say, oh yeah, that bar is approximately double the bar next to it.

So showing that kind of difference is great for a bar chart. Versus a line chart is good if you need to show. subtle changes. Changes to the stock market would be terrible as a bar chart unless something terrible happened. I guess it could be a bar chart.

Brendan: But that's a good point. I think going back to what you said about looking at the Washington Post or the New York Times, and just seeing how different types of data are encoded in different types of charts. When you see a stock price graph, it's a line graph. And just seeing what's out there in the world.

And I love your idea about the slide deck and thinking like a designer. Like you have your little scrapbook, or your look book, or inspiration book, whatever that is.

Jess Cohen-tanugi: Yeah, because you always think you're going to be able to remember them and be able to Google them, and then you end up being like, cool line chart Financial Times stocks blue. It just doesn't come up.

Brendan: No. Yeah.

Jess Cohen-tanugi: Another thing I see that people often I won't say do wrong, but that people often do that could be done better. Including by some of the people I most admire, like the New York Times database team, the Washington Post team is make the mistake of putting all their data in one chart, and this is a different point from choose your story.

Sometimes, you do need all the data. Think about COVID when we first started seeing the line charts with COVID cases. The news outlets put all of the countries in one interactive chart. I don't know if you remember. It looked like spaghetti. And they had it, so that you could search for your own country. So that they were like, oh, it's interactive, so it's OK that you can't find any particular country.

And then they remembered this magical thing called small multiples. It's a technical term, but it's just basically saying, you could make a bunch of really tiny charts and put them next to each other. Instead of trying to put all the lines in one chart, and you can allow your users to make the visual comparisons.

If they want to compare France to the US, they can look at France and then look at the US and make that comparison. So people often forget that you can have it. You can have many streams of data, use a simple chart, and repeat it, and have a grid of charts. And that's one thing I find myself suggesting to a lot of folks even in the School of Public Health and the Medical School area.

Brendan: No, that's a good point. That's great. Especially on the web. You have interactivity, so you're like you cram everything in there and say, oh yeah, they can pick and choose. But just because you can doesn't mean you should.

Jess Cohen-tanugi: Right. It's not a good practice.

Brendan: What can researchers who are listening to this who are thinking about their own data vis and maybe have a publication coming up. What are a couple of takeaways that you would have for people who are thinking about data visualization?

Jess Cohen-tanugi: I think for people who are currently authoring data visualizations, keeping in mind that it's part of their responsibility to make sure that what they've created is read correctly. I think remembering that in this day and age, people often screenshot and share without context.

And making sure that no wild news article could come out of a poorly-designed data visualization. Not to scare people, but the more context you can put in the visualization itself, rather than the caption, the better. I would encourage them to ask their colleagues or whatever, anyone who might understand the concept, what message they're getting from the chart and graph that they've composed.

And make sure that lines up with what the researcher intended to convey. Within reason, I know there are conventions with certain fields and subfields to try to think outside the box and think if another chart type, even though it's not conventional, might do the job better. Not automatically reach for the box spot because everyone else does.

But maybe think, do I need it? Or maybe there's a better way to do this. So look at those chart picker tools, and think about whether one of them might serve better than what people typically do.

Brendan: Great. Well, Jess Cohen-Tanugi, the visualization specialist at Harvard Libraries. Thank you very much for joining us. It was a pleasure to have this conversation with you.

Jess Cohen-tanugi: You too. Thanks for having me.

Brendan: Thank you for listening. If you've enjoyed this podcast, please rate us on iTunes, and help us spread the word about the amazing research taking place across the Harvard community.

Abi: To learn more about the guests on this episode, visit our website, Catalyst.Harvard.edu/ThinkResearch.