How To Lie With Maps

Why Your Dashboards Look Like Spam

I spent seven years doing infrastructure reporting for a regional water authority. Every quarter I handed a package to a county commissioner who had about four minutes before his next call. If a single map on page three didn't scream "we are failing" or "we are succeeding," it didn't matter what the other twenty-nine pages said. That is the environment in which How To Lie With Maps became less of a book and more of a reference manual for the opposition. Edward Tufte's 1983 volume is still the shortest useful distance between someone who knows cartography and someone who doesn't. The core argument is that every map is already a lie because projection distorts area, shape, or distance. The book shows you how that lie goes from technical necessity to deliberate manipulation. The useful part isn't the theory. It is the taxonomy of tricks.

How To Lie With Maps in Practice

The taxonomy breaks into seven categories. I will walk through the ones that actually show up in commercial work. The rest are worth reading about but rare in production. Trivial XY charts with a fake zero — a line chart where the y-axis starts at 94 when the data runs from 95 to 102. The trend looks heroic. The absolute numbers are boring. Tufte calls this the "lie ratio" and gives you the formula: the visual exaggeration divided by the actual range. I used to catch this in Excel by selecting the axis and checking "Value axis crosses." If it wasn't at zero without an explicit label, I flagged it. Five seconds per chart. Saved me from producing about twelve misleading slides a month. Area distortion in choropleths — this is the big one. When you color-code US states by density instead of absolute count, Alaska dominates the legend visually even though it has roughly the same population as two mid-sized counties. The fix is either an equal-area projection or a unit like people per square mile with a clear disclaimer. I learned the hard way that most GIS software defaults to Web Mercator for quick exports. Web Mercator preserves angles, not areas. It makes Norway look enormous. It makes Congo look small. If your story is about population or disease incidence, switch to Albers Equal Area before you open the legend. The workflow takes about six minutes and eliminates the single most common accusation I ever received from stakeholders.

Cherry-picked time windows — show a stock price from the day before a crash to the day after the V-shaped recovery. The trend is real. The implication is false. I worked on a climate report where the internal team wanted to highlight a nineteen-year warming slice. The previous administration had picked a twenty-two-year slice ending three years earlier that showed a plateau. Both were defensible. Neither was honest. The workaround I settled on: always anchor the x-axis at a biologically or economically meaningful boundary (start of record, policy change, baseline year). If you must truncate, put the truncation point in the caption with the omitted range listed immediately after. One extra line of text killed about forty percent of the angry emails we used to get. The border shift problem — changing administrative boundaries to gain or lose a district. This is redistricting by cartography. Tufte uses a congressional example from the 1970s that is now old news, but the technique is alive in municipal annexation disputes. The practical tip: always list the source boundary layer and the last modification date in the metadata block. If someone challenges your map and you can't produce a timestamped shapefile within ten minutes, you have already lost credibility regardless of whether you lied. False precision on small numbers — reporting water contamination at 0.003 mg/L when the instrument detection limit is 0.005 mg/L. The extra digit creates authority. The number is noise. I started appending "(approx.)" or the detection limit in parentheses next to any value within a factor of three of the LOD. Editors hated it. regulators stopped pushing back. The truth stayed the same.

What Tufte Misses and What You Should Add

The book is narrow by design. It focuses on print cartography and political graphics. It does not cover interactive dashboards, which are where most lying happens today. Hover states, drill-down filters, and default view selections can distort perception without any axis manipulation. I once saw a dashboard where the default map view centered on a district that happened to align with the ruling party's stronghold. The zoom level made suburban areas look sparse. Rural areas looked empty. No single trick from Tufte's taxonomy was present. The effect was identical. The workaround for interactive media is different. You document the default state explicitly: what center point, what zoom, what filter is applied on load. That single sentence in the footer reduces the manipulation surface more than any chart correction. A second gap is color semantics. Tufle discusses color sparingly. Modern work runs on sequential, diverging, and qualitative palettes chosen in tools like ColorBrewer. The trap is using a qualitative palette for ordered data or a diverging palette without a meaningful midpoint. I keep a one-page cheat sheet on my monitor. Sequential for magnitude, diverging for deviation from a norm, qualitative for categories. If a stakeholder asks me to use red-green for a binary map, I explain that about eight percent of male readers cannot distinguish the classes and offer a blue-orange alternative. The conversation lasts two minutes and prevents a compliance complaint that would last two months.

One Edge Case That Cost Me Sleep

Early in my career I mapped flood risk zones for a county using FEMA floodplain shapes overlaid on a choropleth of property values. The resulting map showed that high-value neighborhoods sat outside the 100-year floodplain while lower-value areas were inside it. The narrative was obvious and politically toxic. I had done everything technically correct. The shapes were public data. The classification was equal interval. The projection was NAD83 / US National Atlas Equal Area. What I had missed was that the floodplain layer was compiled from older topographic surveys that predated recent channelization projects. The high-value neighborhoods had been artificially protected by realigning the creek two miles west of their property lines. The map was accurate to the data. It was misleading about the risk. Tufte would classify this under "omission of relevant data" rather than active lying. The fix was to add a textual caveat noting the survey date and any known geomorphic changes, and to overlay a second layer showing historical channel positions. The map took an extra forty minutes to produce and avoided a lawsuit that would have taken four years to resolve.

Download and Next Steps

Tufte's book is still in print and widely available through university presses and major retailers. The 1997 expanded edition adds the color plates that make the examples clearer. I recommend the expanded edition over the original because the z-score choropleth example in color is impossible to parse in grayscale. If you are doing any serious cartographic work and the publisher is unavailable, the PDF scans that circulate on academic servers are readable, though the reproduction quality drops on the color plates. The practical next step after reading is to audit your own recent maps. Pick three you produced in the last quarter. For each one, answer these four questions: What projection is this in and why? Where does the axis actually start? What is omitted from the frame? If I removed the title and caption, would the wrong conclusion still be the default interpretation? You will find that most of your maps survive the first three questions and fail the fourth. That is normal. The goal is to know which ones are failing and fix them before they leave your desk. There is no universal tool that catches all of these automatically. Some software flags a non-zero axis. Very little software flags a misleading default view. The discipline is the audit. The book gives you the vocabulary. The work is doing it before someone else does it for you.