Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
Episode #67: Wilson Andrews
169 Plays8 years ago
Thanks for tuning in to this week’s episode of The PolicyViz Podcast! This week, I talk with Wilson Andrews, Graphics Editor at the New York Times. In the wake of Donald Trump’s election win, Wilson and I talk about polling,...
The post Episode #67: Wilson Andrews appeared first on PolicyViz.
Recommended
Transcript
Election Impact on Data Visualization and Journalism
00:00:11
Speaker
Welcome back to the Policy Viz podcast. I'm your host, John Schwabisch. Now that the US presidential election is behind us, we can look back and reflect and review some of the great work that happened in the data visualization and data journalism fields. And we could also sort of take a look back at the landscape of what may have gone right and may have gone wrong when it comes to polling and reporting and to cover
00:00:35
Speaker
those and hopefully many other issues.
Post-Election Information Demand
00:00:38
Speaker
I'm very pleased to be joined by Wilson Andrews, graphics editor at the New York Times. Wilson, welcome to the show. Thanks, John. Thanks for joining me. Are things calming down now a little bit there? I'd say we went from 110% to 109%.
00:00:55
Speaker
as is the case across the U.S. and the world, I think we're all trying to figure out what's going to happen next. And our readers are very interested in that information. And so we are working very hard to deliver that in as clear a way as possible. Still so many unknowns, but it's definitely not really slowed down since the election. Yeah. Yeah. I mean, there's a lot of transitions happening now as we move towards the end of the year. Before we get into some questions I want to ask about
00:01:22
Speaker
the data visualizations and the use of data in the election.
Designing Visual Stories for Elections
00:01:25
Speaker
Could you maybe talk a little bit about your role at the New York Times and what sort of things you do on a day-to-day basis? Sure. So we have a team of very diverse skill sets, designers, developers, cartographers, photographers, reporters,
00:01:41
Speaker
visual motion graphics, all sorts of different skill sets. I kind of come down on the sort of like designer-developer editing role and in the election my job was really to think about what visual approaches we could take to help tell the stories of the election and then ultimately sort of guiding our election results, the live results
00:02:06
Speaker
Helping the upshot folks with the forecast pages that they created and sort of like the overall Package of what we offered to readers on election night. So that's kind of what I've been involved with the past year year and a half
00:02:22
Speaker
It seems like it's been going on for a long time.
Innovative Election Visualizations
00:02:25
Speaker
Let's start with maybe talking a little bit about some of your favorite representations of data that have come out over the last, I don't know, 18 months, I guess. It's kind of a long time, but there's been some incredible work done by, of course, your team over there at the Times and the Upshot, but also some interesting work done by FiveThirtyEight and The Guardian did some interesting things. Were there things that stood out to you as being particularly innovative or particularly interesting that you thought
00:02:51
Speaker
Showed the election data in a unique and different way Yeah, I think the first example that struck me was when the Guardian published their primary their live primary results And they actually had little cartoon characters of the candidates coming out and painting their their counties, right? we've always struggled with the right way to show things changing lives and
00:03:14
Speaker
I don't, we still haven't, um, really solved it. Uh, and it's kind of something that gets put on a back burner until we finish all of the other important stuff. Like is the map displaying the current, correct information in a timely fashion. Um, but they sort of, uh, both, it was both fun.
00:03:30
Speaker
And it sort of helped you see that the page was live, and things were changing, and this is what just changed. And that sort of whimsical approach to it was actually really refreshing. And it was something that we watched with a lot. It was a lot of fun.
Understanding Uncertainty in Results
00:03:46
Speaker
I was going to say, you guys had your live gauge with the probability of each candidate winning, which had some jittering going on, sort of moving the dial around. You want to talk a little bit about that? Because I know some people were
00:04:00
Speaker
I'm not upset about it, but some people were sort of wondering what exactly it was showing. Yeah, sure. So obviously the jitter, at least to us, has become sort of a buzzword, I guess, and almost the word in itself is kind of a joke around the department, but just the amount of tweets about things like that certainly rises up.
00:04:25
Speaker
The jitter was part of a gauge that attempted to show the live probability of either candidate winning. Also, there were two gauges, one that showed estimated electoral votes and the estimated popular vote margin. And the sort of inspiration behind the jitter or one of the reasons to do it was to show probability and the sort of margin of error in probability.
00:04:52
Speaker
There was never ever really one given number that you can show. If you do that, then you're sort of displaying a sense of false accuracy. And Gregor Eich, who developed those pieces, wrote an interesting blog post that sort of spells all of this out much better than most of the Twitter conversations did. The other was simply to transmit to people that this was a live page, that the data was constantly changing, and
00:05:20
Speaker
We took some inspiration from dashboards like Chartbeat that show these types of things live to help people understand that they don't actually need to refresh. They don't need to go to another page. This is actually everything you need to know. Or we think it's sort of the top level information that you might be interested in on election night.
Challenges in Probability Visualization
00:05:40
Speaker
Is the underlying challenge with representing probabilities uncertainty? Do you believe it to be a visualization challenge or is a more of a numeracy or statistical literacy problem?
00:05:53
Speaker
Um, I think it's a little bit of both. Uh, I should say that we, we could have done probably a little bit better job explaining what we were showing, uh, with the, with the jitter with the, with the gauges. Uh, I think we, we put the actual descriptions of what was happening behind the actual, just like getting the work done and getting it ready for election night. And you know, we wrote, we wrote on those pages and around them on the non sort of forecast centric pages.
00:06:21
Speaker
to try to help people understand, but it probably could have been a little bit more verbose about what was actually happening and what people were seeing, because it was very stark, obviously. It really sort of struck a nerve both positively and negatively for a lot of people. And so if we had been a little bit clearer about what was happening there, I think it would have been a better service for years.
00:06:43
Speaker
And do you think having that more explanation, that more annotation, sort of on the other hand, would have detracted from sort of the clean design and where this is the central thing? Or sort of in retrospect, you still think that adding even some light touch of annotation, which I always ascribe to the times and the upshot as being really good at, that even that light annotation would still maintain that central piece of having that gauge right there?
00:07:10
Speaker
Yeah, I think we were pretty spare as it as it was anyway. So I don't think it would have hurt anything. I'm not endorsing the fact that like we needed a whole article on the page explaining it. But I think a sentence. And if you want to further explain what's going on, I'll link off to a like a methodology page or something like that is like
00:07:31
Speaker
pretty much all that is required in that kind of situation. Right.
Creative Electoral Vote Visuals
00:07:36
Speaker
Okay. So I kind of interrupted you, you talked about favorite, we're talking about favorite things. You mentioned the guardian cartoon for the candidates. Um, were there other ones that, that struck you that, that you really liked? Yeah, we were all taken by the snake diagram at five 38. Um, and I think a lot of people, you know, not just at the times would agree with me that that was a really fun approach. There's been a lot of different attempts to like,
00:07:59
Speaker
quantify electoral votes and how they matter over the years. And like a very typical approach is like a cartogram of electoral votes by state. And so I think I feel like this really emphasized that first pass the post sense of what's happening in a US presidential election. All of these states sort of pile up in a way that gets you there. And this sort of novel approach I thought was really fun because
00:08:28
Speaker
You don't just have a really long bar that is really challenging to do and today is like modern screen sizes that sort of compression was really novel and.
00:08:39
Speaker
brought us a lot of joy in seeing how they put that together. That's interesting. So a little bit more of the more mobile devices and changes in screen sizes sort of drove why that worked really well. Yeah, I mean, I don't know if that was sort of the inherent reason why they did it. I haven't seen anybody there about it. But I think the limitations of how we present our work these days kind of
00:09:04
Speaker
forces you to be very creative. And that was, I thought, an elegant solution for that. Yeah, that's really interesting.
Effectiveness of Election Maps
00:09:11
Speaker
You mentioned cardograms, which of course, as elections always want to do in the data visualization field, spawns a whole conversation about maps. And this year, perhaps more than the last few, there was sort of an explosion, all the different kinds of maps.
00:09:27
Speaker
Were there approaches that or techniques that you particularly like that, that you think could be useful going, you know, going forward, or are we still going to be in the choropleth map world for, you know, the red, blue choropleth map for the foreseeable future? Um, I still think that the general public, uh, does not understand cartograms. Uh, that's probably, you know, that, that puts a lot of the onus on
00:09:57
Speaker
us as designers and graphics creators to do a better job of making them more accessible. But geography still matters quite a bit because it's how people find themselves and find the things that they know. I'm from North Carolina, so I know immediately the look and the small sliver of
00:10:19
Speaker
counties around Charlotte where I grew up. And that sort of is very difficult to do, even for me, someone who looks at maps every day to do on a cartogram. So I think there is space for both. We're not anti-cartogram. But I think
00:10:38
Speaker
that there's still a lot of value in a basic geographic map that shows election data. One way that sort of attempts to do something a little different than a Choropleth map is the sort of change maps that we're now doing which use arrows to show shift between the candidates or between elections. We did a piece the night of basically for the next morning that analyzed the results and showed the sort of massive swing
00:11:07
Speaker
toward Trump in the Midwest, and I thought that those arrow maps, the swing maps, kind of were a really strong depiction of what happened in the election, much more so than probably a cartogram would have given you in that sense. I think that piece was called How Trump Swung the Electoral Map, but I think that's kind of a nice, elegant approach that, you know, it doesn't show all of the results, but it gives you one slice of the story and really was
00:11:37
Speaker
we think the main story and therefore was I think an effective use of mapping election results. Right. I mean, it's sort of interesting because you had mentioned sort of we, whoever we is, need to sort of educate people on how to read some of these alternate map types. And what's interesting is that when you go to the New York Times website or the Washington Post website or the Guardian website, there are
00:12:00
Speaker
tile grid maps and there are tile grams and there's hex maps and, you know, there's all these other map types. And then when I watched the election coverage on the major networks, it's all just, you know, just your standard core plus map colored red and blue. Yeah. And I guess that's because that audience sort of is being shown the data as opposed to being allowed to sort of go in and explore it, which you can do when you're browsing the website.
00:12:22
Speaker
Yeah, I mean, perhaps if CNN, you know, had somebody on there explaining exactly what a core plus was showing, it would be effective. But that's also a case where they're attempting to basically highlight one or two things in a very short amount of time. And if you're having someone to sort of first parse what they're looking at, figure out that it's actually a US map, and then like follow along through this, like,
00:12:49
Speaker
bulbous winding channel of red and blue. I can see why they have not really gone down that path just because it's a very challenging format for anybody, I think. Yeah. I mean, it's certainly true. I was watching a little bit of CNN when they were sort of zooming in and out of counties in the different states.
00:13:10
Speaker
And certainly if you're a resident of North Carolina or Florida, you recognize your county and the neighboring counties in the shape of the state where you don't get that in these other representations. Yeah, that's right. And we found a lot of times the most effective
00:13:26
Speaker
pieces that we have done really resonate with people being able to put themselves in the graphic. Not always a map, sometimes it's like a piece of data that we allow people to find themselves in. And I think we shouldn't lose sight of that fact that people are very attuned to where they fall within our work.
Polling Errors and Data Journalism
00:13:45
Speaker
And that's something that we try to keep top of mind and I think applies a lot more to the geographic style of the map. Right. Let's broaden the scope a little bit and talk about what
00:13:56
Speaker
Happened and not from a political standpoint, but a lot of the polls and the and the predictions were wrong Had you know? Probably of Hillary Clinton winning by a far margin and those all turn turn out to be wrong Does that mean from your perspective that? data science
00:14:16
Speaker
and or data journalism or the polling industry. I mean, is this sort of a day of reckoning now or is this just the outlier election? And, you know, next time we'll return sort of back to back to normal. I don't think that there was something inherently wrong with the probability models. Nobody said that with the exception of, I think, Princeton Consortium, which maybe gave Clinton and a 99 percent chance, nobody said it was 100 percent chance that
00:14:47
Speaker
Clinton would win, which means that there was always a chance that Trump would win. That's not saying that the polls did not miss this. They certainly did. No poll gave, I think it was Wisconsin, gave Trump a lead there and he ended up winning that state. So that was pretty striking. Nate Cohen did a pretty interesting piece that
00:15:10
Speaker
showed where the polling misses were in 2016 and compared them to previous elections. And we did see that the polling misses in the Midwest were particularly striking and they were. And the reason that the models and the probabilities were so far off of the polling is that the entire Midwest polling error was off in one direction. It was all much more Clinton and the error actually ended up swinging it towards Trump's side.
00:15:37
Speaker
So I think that was the underlying piece. The maybe thing that we could improve next time is visualization of probability could be improved. I think people tend to gravitate towards the big number, and it's really easy to do that as a designer, just put a big number up there. And I think that that is maybe what is confusing a lot of people.
00:16:02
Speaker
When I show it to people that don't work here, just everyday people, my family, whatever, they often say, oh, 85%. That seems like a lot higher than what the polls are saying. And then I have to explain that there's a difference between a polling average and forecasts. And that's sort of where we lose people, I think.
00:16:22
Speaker
So I think that there are certainly lessons to be learned as far as sort of setting expectations About what probability means doing a better job of visualizing that I think one thing that was really fun that we've done in the past or probability in 2014 we did a Senate model and
00:16:40
Speaker
and we had little spinners. I think Bostock maybe worked on that piece where each, you basically clicked a button that said run the election and each state had little probability spinners that spun and showed you different slices of a pie that it would be Republican or Democrat. And every time you would get a different result based on that simulation. And I think those are the types of like techniques that
00:17:06
Speaker
have a real world analog, but help people to understand what we're actually doing when we say there's 85% chance. And then the Upshot did post-mortem where they kind of explained a lot of these things. And they showed on the histogram of likely outcomes where the Trump electoral vote total fell. And it was
00:17:27
Speaker
somewhere along the 10% range, but it was certainly among the possible outcomes. While I don't think that the underlying math is wrong, I think our way of setting expectations and communicating that could probably be
00:17:46
Speaker
Does that make sense? No, it does. And I wonder if it all ends up coming down to the fact that there is, or is usually a number that's highlighted. It's 85%. And there might be a margin of an error. There's a confidence bound around that. But at the same time, it seems like everybody wants to be able to show their one number.
00:18:07
Speaker
And that number is easy for people to get a handle on and to understand. Oh, the upshot says 85, and 538 says 92, and there's that number. But it comes down, I think, again to margin of error, which I'm not sure a lot of people really fully understand to begin with.
00:18:23
Speaker
Probably the best visualization of these models is the histogram, but that's an incredibly unsexy and sort of not well understood, not common style of database for the general public, you know, so it's often buried on
00:18:39
Speaker
pages you know 538 had one we had one but it was pretty far down and so that that's that's sort of the challenge I think for if and when this we do this next time like what is the right way to both make it understandable for everyone but make sure that they actually understand what they're looking at right
Future of Live Election Visuals
00:18:56
Speaker
Elections are also an interesting and unique time for the country in general because everybody is, not everybody, but many people are on the website at the same time looking at results live. And so you're sort of feeding these live results just like you were talking about with the Jitter and Gage.
00:19:15
Speaker
providing this information in real time, which I would guess offers a different type of opportunity to communicate those data and that uncertainty than a one-off story about something that may have some polling numbers in it.
00:19:29
Speaker
Yeah, I think while people may disagree that the upshot forecast before the election was right or wrong, the live forecasting that they did was absolutely amazing and actually kind of gave you a sense of how the outcome was going to
00:19:48
Speaker
be four or five hours before it actually turned out. I think a lot of people's frustration with how we visualize that was just with the fact that they didn't like how the election turned out. If it had been jittering towards Clinton all night, I don't think we'd be hearing anything right now. So I think that was really fun to watch.
00:20:11
Speaker
the sort of success of live estimating the election. And we want to kind of continue to build on that and sort of the opportunities that that gives us for next time. It's a long way off, so I don't know what that will be yet.
00:20:28
Speaker
Let me just ask one more question. How do these live results and live data visualization stories and dashboards, how does it work in terms of recording that information and sort of saving that information, not so much for posterity, but for learning for next time?
Learning from Live Election Data
00:20:46
Speaker
Do you save or hold on to the results over time so that you can go back and review them? Or is it now I go to the website and it's 100% for Donald Trump since he won?
00:20:55
Speaker
is there a way that you are capturing what happened over the course of the night? Yeah, we snapshot all the data that the AP sends us. And that way we can replay it, you know, when you test out new forms next time, the test data that we get isn't
00:21:11
Speaker
usually very real. We'll often see really strange states going Republican or Democrat, and so having a real election that you can play back, especially for something like a live forecast, is really valuable because then you can test your assumptions about what your database is going to show versus a sort of realistic dataset.
00:21:30
Speaker
So we store all of that. Do you mean with regard to showing it publicly? No, I meant just in terms of your testing and being able to sort of go back and go through it. That's really valuable, I think, to do that. Great. Well, good luck in the next few weeks. Hopefully you'll get a rest, but I'm going to assume you're going to be busy through Inauguration Day and for the next 100 days after that. So Wilson, thanks a lot for coming on the show. It's been really interesting.
00:22:00
Speaker
Thanks, John. Thanks, everyone, for tuning into this week's episode. Until next time, this has been the Policy Vis Podcast. Thanks for listening.