dataisugly

NASA predicted that Voyager 1 would be one light-day from Earth at 10:16:07 UTC on 18 November 2026: the first human-made object that far, about 24 hours away by radio.

Light from the Moon takes about 1.3 seconds to reach you; from the Sun, about 8 minutes 20 seconds on average; from the Andromeda galaxy, about 2.5 million years.

Voyager 1 moves away from the Sun at about 17 km/s, so a hello takes about 5 seconds longer than light needs to cross the distance at the moment you send it.

Only the travel times are to scale: each wave is one hour of radio travel, so the Moon and the Sun sit right at Earth. The pictures of Earth, the Moon, the Sun and Voyager 1 are not to scale. The moving pulse is a time-lapse: about six hours of travel every second.

Voyager 1 launched on 5 September 1977 and is the farthest spacecraft from Earth. Its 3.7 m dish points at Earth, and NASA still talks to it through its Deep Space Network.

The line is a survey, not police records: the Bureau of Justice Statistics asks people aged 12 and older whether they were victims, so it counts crimes never reported to the police. It covers rape or sexual assault, robbery, and aggravated and simple assault; homicide is not in it. In 1993, 1 in 23 people aged 12 and older was a victim of violent crime at least once; in 2024, 1 in 69.

It is a sample, so each year has a margin of error, drawn as the pale band. Even at the band’s edges, the fall from 1993 to 2024 is at least 63.8%. Most of it came by 2010; 2024 was 21% above 2010, higher than 2020 and 2021 and similar to 2022 and 2023. The lowest year in the series, which starts in 1993, is 2020, at 16.4.

BJS says 2006 should not be compared with other years, so it is left blank. From 2025 a redesigned questionnaire counts more: asked both ways, 2024 came to 28.7 instead of 23.3, about 23% higher for the same year. That is a change of method, not a rise, so it is not on this line.

The dots are Gallup’s question “Is there more crime in the U.S. than there was a year ago, or less?”, asked of adults, mostly in October. It asks about one year, and about all crime, not just violence. There was no poll in 1994, 1995, 1999 or 2012. Gallup gives its 2024 and 2025 polls a margin of error of 4 points.

Since 1993, most said “more” in 26 of 29 polls, and “more” beat “less” in 28. The rate did rise in 10 of 29 pairs of years, but BJS’s own tests, when it published each year, found only three of those rises bigger than sampling error: 2011, 2012 and 2022. In the 16 poll years when the survey’s estimate fell, most still said “more” in 14.

Answers swing with the president’s party: from 2024 to 2025, Republicans saying “more” fell from 90% to 54%, and Democrats rose from 29% to 46%. Yet majorities of both said “more” in 23 of 29 polls. In 2025, 49% said “more”, the fewest since 2001.

The police light stands for crime, not police records. Violent crimes reported to the police fell too, from 33.8 to 11.2 per 1,000 people in this survey, and the FBI’s police-recorded violent crime, murder included, roughly halved from 1993 to 2019, from 747.1 to 366.7 per 100,000. The two sources measure different things and are never drawn on one axis.

For the same electricity, coal power causes about 868 times as many deaths as nuclear: 24.6 deaths per terawatt-hour against 0.03 (0.028 before rounding). A terawatt-hour is about a year of electricity for 155,000 people in the EU.

Why 35 years? At nuclear's rate, a town of 155,000 would need about 35 years of electricity to reach one death. Over the same 35 years: brown coal 1,145 deaths, coal 862, oil 645, biomass 162, gas 99, hydropower about 45, and wind, nuclear and solar about 1 each. No order is claimed among those last three: their uncertainties are likely to overlap.

The deaths are from accidents and air pollution; climate harm is not counted for any source. Coal, brown coal, oil, gas and biomass are modelled for power plants in Europe (Markandya and Wilkinson, 2007), and plants elsewhere likely kill more: a study of coal plants in Shandong, China, estimated about 77 deaths per terawatt-hour. Even at the low end of the 95% range for its air-pollution deaths, 6.1, coal is over 200 times nuclear.

Nuclear counts 433 deaths for Chernobyl, Our World in Data's high estimate (confirmed deaths are under 100), and 2,314 for Fukushima, Our World in Data's count: 1 from radiation, the rest deaths from illness and the strain of evacuation that Japan links to the disaster in the prefecture, without separating the nuclear accident from the earthquake and tsunami. With the World Health Organization's "up to 4000" for Chernobyl, nuclear's rate is 0.065 and coal still kills 378 times more. Even counting all 41,000 cancer cases projected in Europe to 2065 as deaths, nuclear comes to about 16 in 35 years, still 6 times fewer than gas. To match gas, Chernobyl would have needed about 271,000 deaths.

Hydropower's toll is almost all one dam failure, Banqiao in China in 1975, whose toll is disputed; without it, hydro's rate is about 0.04. Wind and solar count accidents in their supply chains, 1990 to 2013, as Our World in Data reports them from Sovacool et al. (2016).

Each dot is one death and each full puff 100, so the dots are to scale; the plants are not. Smoke is drawn only for fuels that are burned, whose deaths are almost all from air pollution. The years that run back and forth are a time-lapse of the same rates.

From 1.4 billion person-years of US tax records, 2001 to 2014. Each dot is 1% of US men, or of women, ranked by pretax household earnings against people of the same sex and age in the same year. Earnings are in 2012 dollars.

Why only $356 a year? Earnings here leave out Social Security and disability benefits, and people with no earnings at all are left out. So the bottom 1% are households that earned almost nothing that year, though some may have had Social Security or disability benefits.

Why 40? Each number is the expected age at death for people who are alive at 40, so deaths before 40 are not in it, and 72.7 is not a lifespan from birth. These are estimates: deaths are observed up to age 76, the authors extend each group's death rates with a standard curve to age 90, and after 90 the rates depend on sex but no longer on earnings. The data end in 2014.

A link, not proof that money adds years: illness can cut earnings, so some of the gap may run from health to income, and the authors say money's own effect is likely smaller. Earnings are measured two years earlier (at 61 for people 63 and over). The bottom two ranks drop sharply; above them the climb is nearly straight, about 0.11 years a rank for men.

Not just the extremes: the middle rank of men reaches 81.7, 5.6 years short of the top 1%, and men at the 5th and 95th ranks are about 10 years apart. Women show the same climb, smaller: 10.1 years between the bottom and top 1%. The numbers are adjusted for race; without that adjustment the gaps are larger: 14.8 years for men, 10.4 for women.

The gap grew from 2001 to 2014. On the authors' fitted trends, the top 5% of earners gained 2.34 years (men) and 2.91 (women); the bottom 5%, 0.32 and 0.04.

A 2025 study measured wealth, not earnings, in adults aged 50 to 85 from 2010 to 2022. Its finding: "Survival in the wealthiest U.S. quartile appeared to be similar to that in the poorest quartile in northern and western Europe."

The drawing: the cupcake and the cake mark the two ends, and their sizes mean nothing. Each stands on its group's age, and the cake's stand is exactly the gap tall on the chart's scale. The age scale does not start at zero because each dot marks a position, not a length.

The chart's counts are final, from US death certificates compiled by the CDC's National Center for Health Statistics: US residents whose underlying cause of death was a drug overdose, whether unintentional (91.6% in 2024), suicide (5.6%), undetermined (2.7%) or homicide (under 1%).

Counts fell 24.4% in 2024. The population grew, so Death rate shows deaths per 100,000 people, adjusted to the 2000 US age mix: 31.3 in 2023 and 23.1 in 2024, down 26.2%. Both peaked in 2022, and both fell most in 2024: the biggest one-year fall since the series began in 1999.

The rise was not steady. Deaths rose in 22 of the 23 years from 1999 to 2022, falling only in 2018. The rate climbed fast from 1999 to 2006 (6.1 to 11.5), slowly from 2006 to 2013 (11.5 to 13.8; NCHS's 2021 trend test called those years stable), then steeply to 32.6 in 2022.

Deaths involving synthetic opioids other than methadone (drugs such as fentanyl, its analogs and tramadol) fell from 72,776 to 47,735. Overdose deaths not recorded as involving them changed little, from 32,231 to 31,649 (about 4% of deaths named no specific drug). One death can involve several drugs. The pill bottle stands for drugs in general, not for prescriptions.

NCHS names no cause for the fall. Changes in the drug supply, naloxone (which reverses opioid overdoses), treatment, and the loss of many of the people most at risk are all argued over; none is shown here.

Even after the fall, 2024 had the fewest deaths since 2019, yet more than 2019's 70,630, and 4.7 times 1999's (the rate: 3.8 times). The series starts in 1999 because today's cause-of-death codes (ICD-10) start then; earlier counts are not directly comparable.

Provisional counts point the same way: 70,735 predicted deaths in 2025 against 81,313 for 2024, down 13.0% (NCHS, file updated 14 October 2026). They count deaths by where they happened, include non-residents and adjust for late reports, so they are not drawn beside the final counts.

The Living Planet Index follows 35,803 monitored populations of 5,790 species of mammals, birds, fish, reptiles and amphibians, native species only. A population is one species in one place, from a few animals to many thousands. For each population the index measures the change relative to its own size, then averages those changes, giving species-rich groups more weight. Set to 100 in 1970, it stood at 27 in 2022: an average fall of 73.2% (likely range 67% to 78%), about 2.5% a year.

So it tracks an average change, not the number of animals: changes in small populations are averaged with changes in large ones. WWF says so itself: "This doesn't mean that there are 73% fewer wild animals in total today than in 1970." Nor does it mean that 73% of populations are declining.

Over each population's own monitored years, 50.0% fell, 6.9% held steady (a change of less than 5%) and 43.1% grew. The average still fell because the losses were bigger than the gains, and many came in species-rich groups that weigh more, such as amphibians and reptiles of the American tropics and fish of Africa south of the Sahara. The index also carries past declines forward: it is cumulative.

The made-up example is the technical report's Figure 4: 50 animals become 20 (−60%), 33 stay 33 and 400 become 420 (+5%). The average of the three changes is −18%, yet the animals go from 483 to 473, 2% fewer. The figure takes a simple average. The index itself averages changes as ratios, where a halving and a doubling cancel out, so a fall weighs even more: these three would give −25%. It also chains smoothed yearly changes and weights them by species richness. Which animal stands for which population is our drawing.

Not just a few crashes: remove the 10% most extreme rises and falls and the index still falls 64% (61% in the same test in 2024). A 2020 study found that fewer than 3% of populations drove the decline, but it used the 2016 data, and four replies in Nature in 2022 contested its reading.

By habitat, freshwater populations fell 85% on average, land 69% and sea 59%; the global index weighs the three equally. By region, on land and in fresh water, Latin America and the Caribbean fell 95% and Europe and Central Asia 23%.

Limits: vertebrates only, so no insects or plants, and monitored populations only; it does not count extinctions. Monitoring leans toward Europe, North America and well-known species, which the weights correct only in part, and the index cannot see losses before 1970, where it starts. Editions are not comparable: the 2024 report's 73% ran to 2020, on fewer populations. On wide screens the line is drawn on a linear scale (the report draws it on a log scale, where the last decade looks steeper), and its shading is the 95% confidence range of the average: it reflects how much the underlying trends vary, but it does not mean that 95% of populations fall inside it.

One tagged Cuvier's beaked whale stayed underwater for 222 minutes on a single breath; across 23 tagged whales, the median presumed foraging dive lasted 59 minutes.

The record is an outlier. It came weeks after this whale was exposed to Navy sonar for an hour, during a study of sonar's effects, so the study left it out of its main results: 3,680 dives, the longest 132 minutes. The authors think dives like it may show the species' true limits.

Each whale carried a small satellite tag on or near its dorsal fin that recorded when its dives began and ended. In the picture each dive starts at 0 and ends at its true minute; its depth and shape are only sketched, and all three are drawn to about the same depth, since the study found no link between how long and how deep its longest dives were.

"Typical" is the median: half the feeding dives were shorter, half longer. The study counted every dive of 33 minutes or more as a feeding dive.