Normalize Data for Meaningful Maps
Choropleth maps are best when they represent relative, not absolute values. Consider two maps shown in Figure 13.4. They both are about Covid-19 cases in the US states (excluding Alaska and Hawaii) as of June 26, 2020. Figure 13.4a shows total number of recorded cases per state, and Figure 13.4b shows Covid-19 cases adjusted by the state’s population. Darker colors represent higher values. Do you notice any differences in spatial patterns?
Both maps show Covid-19 data collected by the New York Times and published on GitHub. In the map in Figure 13.4b, we normalized (divided) values by population in each state, according to the 2018 US Census American Community Survey, the most recent data available on the day of writing. We didn’t add legends and other important cartographic elements so that you can better focus on interpreting spatial patterns. In both cases, we used Jenks natural breaks for classification.
What are the worst-hit states according to the map showing total Covid-19 counts (shown in Figure 13.4a)? If you are familiar with the US geography, you can quickly tell that these are New York, New Jersey, Massachusetts, Florida, Illinois, Texas, and California. But five of these happen to be some of the most populous states in the US, so it makes sense that they will also have higher Covid-19 cases.
Now, how about the map in Figure 13.4b? You can see that New York and its neighbors, including New Jersey and Massachusetts, have by far the highest rates per capita (per person), which we saw in the first map. But you can also see that in fact California, Texas, and Florida were impacted to a lesser extent than the map on the left had suggested. So the map with per-capita values is a much better illustration to the story about New York being the first epicenter of the Covid-19 crisis in the United States.
Different ways to normalize data
You can normalize data in many ways, and there is not necessarily one acceptable way of doing it.
One of the most common ways of normalization is deriving “per capita”, or “per person” values. If values are small, such as rare disease cases or lottery winners, they can be presented as “per 1,000” or “per 100,000” people. Divide your quantity by population in that area to derive per capita values.
Choropleth maps work well with percentages. The good news is, humans like percentages too. It is quite natural for us to understand that a 9% unemployment rate means that of 100 people who were willing to work, nine were unable to find a job. To derive a percentage for unemployment, divide the number of unemployed people by labor force size (adult population who are willing and able to work), and multiply by 100.
Unlike counts, most measured variables do not need normalization because they belong to a scale. For example, median age (the age of the “middle” person in a population, when sorted from youngest to oldest) can be directly compared among populations. We know that humans live anywhere between 0 and 120 years or so, and we wouldn’t expect median ages to be vastly different from one country to another (maybe twice, but not tenfold). Median incomes, if measured in the same currency, also belong to the same scale and can be compared directly.
How not to normalize values
Absolute values are very important for context. Saying that “20% of blond men living in in town X won the lottery” may sound like a catchy headline, but in reality the town has 450 residents, of those 200 are men, and of those only 5 have light hair color. One of those five (and here comes the 20%) was lucky to win the lottery, so technically the headline didn’t lie.
This is, of course, an extreme and comic example, but exaggerations in this spirit are not uncommon. If you want readers to trust you, make sure you are open about total counts when reporting normalized values (such as percentages or per capita values).
Absolute values are important for another reason: behind numbers there are often people, and smaller, normalized values may hide the scale of the problem. Saying that “the unemployment rate is only 5%” is valid, but the 5% of, say, Indian labor force (around 522 million) is about 26 million, which is pretty much the total population of Australia.
Exercise your best judgement when you normalize values. Make sure you don’t blow numbers out of proportion by normalizing values in smaller populations. But also don’t hide large counts behind smaller percentages for larger populations.
At this point, you should have enough geocoding and spreadsheet skills to aid you with map making. In the following section, we will talk about geographical data in general and will introduce different geospatial file formats to ensure you are ready to create, use, and share map data.