Maps and geographic data

Why this matters

When data has a location, a map is often the first visual people ask for, and it can be the most revealing: clusters of customers, regional gaps in coverage, the spread of an outbreak. Maps are also easy to get wrong. A map of raw counts usually just shows where people live, a large sparsely populated area dominates the eye, and the choice of boundaries can change the story.

The exam's visuals objective names maps explicitly. This lesson covers the main map types, the location data they depend on, normalising before mapping, what map projections distort, and when a map is the wrong answer to a geographic question.

The lesson

Choropleth, point and heat maps

Three kinds of map cover most analytical uses.

A choropleth map shades each area, such as a country, region or postcode district, by a value. It is the standard way to show a rate or a total by administrative area: unemployment by county, sales per head by region. It is only as meaningful as the areas: large areas draw attention in proportion to their size, not their importance, and the value applies to the whole area even when it varies inside it.

A point map (or dot map) places a marker at each location, such as each store or each customer. Markers can be sized or coloured by a value, as in a proportional symbol map where larger circles mean larger sales. Point maps show exact locations and clusters, and become cluttered when points overlap.

A heat map of geography, sometimes called a density map, shows where points are concentrated as a smooth colour surface, useful when there are too many points to plot individually, such as thousands of deliveries.

Related types include flow maps, with lines showing movement between places, and cartograms, which distort area sizes so each area's size reflects its value, such as population, rather than its land area.

Geocoding and the location fields data actually contains

Maps need locations the mapping tool understands. Business data rarely contains coordinates; it contains addresses, postcodes, city names, region names or country codes.

Geocoding converts an address or place name into coordinates, latitude and longitude, so it can be plotted. It depends on the quality of the input: typos, abbreviations and incomplete addresses fail or, worse, match the wrong place. Ambiguous names are a specific trap: many towns share a name across countries or regions, and a geocoder may choose the wrong one without warning. Reverse geocoding does the opposite, turning coordinates into an address or area.

For choropleth maps, records must be matched to boundary definitions, such as the shapes of regions or postcode areas. Names must match exactly, and region spelling variants, the same problem profiled in the course's raw customer file, leave areas blank on the map. Boundaries also change over time, so data from one year may not match another year's boundaries.

Checking how many records failed to geocode or match an area, and where they are, is as important here as counting unmatched rows in a join.

Normalising by population or area before mapping

The most common mapping error is shading areas by raw counts. A map of total customers, total crimes or total sales by region mostly shows where the most people live, because more people produce more of almost everything.

Normalising divides the count by a relevant base so areas can be compared fairly:

• per capita: customers per 1,000 residents, crimes per 100,000 people;
• per area: stores per square kilometre;
• as a rate or share: conversion rate, percentage of households.

The right base depends on the question. "Where are we under-represented?" needs customers relative to population; "where should the next depot go?" may need raw volumes, because a depot serves orders, not a percentage.

A related caution is the small numbers problem: rates for areas with very few people are unstable, so a tiny area with two customers out of forty residents can show an extreme rate by chance. Suppressing or flagging areas below a minimum count avoids drawing attention to noise.

Map projections and what they distort

The Earth is round and maps are flat, so every flat map uses a projection, and every projection distorts something: area, shape, distance or direction.

The familiar Mercator projection keeps shapes and directions correct locally, which is why it is used for navigation and web maps, and greatly exaggerates the size of areas far from the equator: Greenland appears similar in size to Africa, which is about fourteen times larger. On a choropleth, where colour is spread across area, this makes northern regions look more important than they are.

Equal-area projections keep relative sizes correct, at the cost of distorting shapes, and are the better choice when area is part of the message, as on world choropleths.

For small areas, such as a city or a single country, the distortion is usually negligible. For continental and world maps, the projection is part of the design decision.

When a map is the wrong answer to a geographic question

Data having a location does not mean a map is the best chart. A map is right when spatial pattern matters: whether values cluster, whether neighbouring areas are similar, or where things are in relation to each other.

When the question is a comparison or ranking, "which region sold most?", a sorted bar chart is clearer: lengths are compared more accurately than shades of colour, and every region is equally visible regardless of its size on a map. With only a few areas, such as the four regions in the practice data, a map adds little that a bar chart does not show better.

Maps also struggle with precise values, with change over time (animated or side-by-side maps are hard to compare), and with areas too small to see. Often the best design pairs a map, for spatial pattern, with a bar chart or table, for exact comparison.