As the returns from this month’s presidential election began to come in, many of us were glued to the television watching maps unfold like this one from 270towin.com:
Following the upset outcome, a lot of maps have been circulated on the web and many have created confusion or stirred conflict. How is it that so many maps can materialize out of seemingly universal information? It’s important to remember that maps are just statistics displayed over space, so they innately come with caveats. A handful of news outlets have even covered these issues recently (Washington Post, NY Times, The Atlantic).
So what are some of the issues that produce so many contradictory maps? Unfortunately, so many maps suffer from predictable conflicts such as mapping projections (a topic of discussion from an episode of West Wing) or shady mathematical manipulations (Washington Post and The Telegraph recently called out a certain far right conservative news outlet for highly misleading maps following the election).
The root problem starts with how we visualize data in maps and interpret it. In the 270 to Win map above, data is visualized as a dichotomy; states either go Democrat or Republican. It’s a great map for assessing who will be the next President of the United States, but it doesn’t do much else. It’s tempting to use this map to generalize states (and we all do it) as Conservative or Liberal. Compare the map above to these maps generated by Mark Newman from The University of Michigan:
Through the use of alternative projections and data visualization, these maps reflect a different narrative. Map #1 scales the size of states based on electoral votes. Map #2 scales counties based on population size. Map #3 considers voting results based on gradient scale. Map #4 using the projection system of Map #2 and then uses the gradient scale of Map #3. They are not good maps for assessing the winner of the election, but it does give a clearer understanding of how the nation voted (mostly split between Democrat and Republican).
Another common issue that occurs is statistical bias. There are a lot of different statistical assessments at the disposal of demographers, geographers, and cartographers, and they are all designed to help make sense of complex data. To take a simple example, median household income is a far better assessment of the typical American salary than average household income, because it negates the effects of outliers.
Most commonly, data binning is the culprit of big differences in displaying averages, means, or distributions. Binning is the act of placing data into different buckets to simplify for assessment or visualization. Results of a statistical assessment can also be binned. Below are three examples of binned data results for asthma hospitalization rates in children in Chicago. Each map has 5 bins, but uses a different method of binning. Example #1 uses a natural breaks (Jenks) binning system, Example #2 uses an equal interval binning system, and Example #3 uses a quantile binning system. Each of these maps is statistically valid despite each being different. Natural breaks highlight clusters of data, so the model seeks to group outcomes that are similar to one another. The equal interval option takes your maximum outcome and minimum outcome and then equally places breaks in that range, irrespective of distribution (12.18% [max] divided by 5 is 2.43 [range breaks]). The quantile approach splits your data in equal bins along your distribution, irrespective of similarity of outcomes or range. This ensures every bin has an equal number of variables represented.
An often overlooked issue in mapping is the Modifiable Aerial Unit Problem, which is a statistical sampling bias created by arbitrary geospatial boundaries. In non-technical terms, it’s the problem we all recognize in gerrymandering. Think about the last time you crossed state lines. That state line, while existing for good reasons, is more or less an arbitrary line in space. When we summarize data within these arbitrary geographic spaces, we create a sampling bias. In the below example we have voters for the Green Party and the Purple Party. There are 18 purple people and 14 green people, therefore, in Example #1, purple wins. If we break the map down into smaller arbitrary districts like Example #2, we see, Purple still wins, but in Example #3 Green wins due to a different method of drawing districts.
MAUP also devalues the votes of some of these constituents while heightening the value of others. In Example #2 and #3, a representative democracy ensures that some districts have more voting power per person than other districts. This problem becomes more challenging in demography or public health because the majority in a neighborhood or zip code can actually mask problems within minority communities. For more on equitable mapping, check out Mental Floss’ article on Neil Freeman’s map of America with 50 equal states.
So, what does all of this mean for designers? Here’s a quick list of best practices:
- We have an obligation to ensure the integrity of what we create. The maps and visuals we create need to either display strong and accurate conclusions or clearly state any assumptions or biases.
- Always check the results of any assessment and interpret it. Anyone can take the average of a bunch of numbers and create an info-graphic. Ensure what you produce makes sense.
- Do not extrapolate something from nothing. Maps are tools, they can tell us stories, but not every story is true. Think about ‘conservative Texas’ and ‘liberal California’; that’s not the complete picture.
- Consider multiple data interpretations when possible. In the case of the US election this year, many people voted 3rd party and many did not vote straight Democrat or straight Republican down the ballot. The mapping of exit polls gives a better picture of how Americans really align in terms of party affiliation.
When we engage stakeholders and clients, we should ensure that our maps are accurate and paint a complete picture. The integrity of that work has the power to shape people, we should approach the power cautiously.