Sixteen minutes fast in November
A good sundial and a good clock don't agree. A week from now, around the third of November, the dial will read about sixteen and a half minutes ahead of the clock. By the eleventh of February it will be fourteen minutes behind. They agree four times a year, near the middle of April, the middle of June, the first of September and Christmas Day. In between, the dial gains about four minutes in May and loses six and a half in late July. None of this is a fault in the dial. The sun is the irregular one, and the gap between the two times is called the equation of time. I worked the dates and sizes above from the standard orbital figures for the Earth. They match the published tables to within a day.
Two causes, added. The first is that the Earth's orbit is an ellipse. We are nearest the sun in the first days of January and move fastest then, so the sun seems to slide eastward along the stars a little faster than its average. Each day the Earth has to turn a little further to bring the sun back to the meridian, so the solar days run long, and the dial drifts behind a clock that keeps the average. On its own this effect swings about seven and a half minutes each way, once a year.
The second cause would act even in a perfectly circular orbit. The sun moves along the ecliptic, which is tilted 23.4 degrees to the equator, but the turning of the Earth measures time along the equator. Near the equinoxes part of the sun's daily motion goes north or south, and less of it counts eastward. Near the solstices its path runs parallel to the equator, and closer to the pole, so the same motion along the path covers more of the equator's hours. This part swings about ten minutes each way, twice a year. Add the two curves and you get the lopsided pattern of the calendar above. Perihelion falls two weeks after the December solstice, not on an equinox, so the peaks don't line up evenly.
When it began to matter. Ptolemy knew of the inequality in the second century. For most of the time since, it was a correction for astronomers, because the sun was the best clock anyone had. The word analemma is older still and meant something else. In Vitruvius's ninth book it is the construction drawing from which a dial's hour lines are laid out for a given latitude, "a basis for calculation deduced from the course of the sun" (Book IX, in Morgan's 1914 translation).
Then clocks got better than the sun. After Huygens's pendulum clock of 1656, anyone with a decent clock could watch it drift against the dial, and the tables in circulation were wrong by several minutes. John Flamsteed wrote a tract on the equation of time in 1667, which John Wallis printed in 1673 with the posthumous works of Jeremiah Horrocks. According to the Biographical Encyclopedia of Astronomers, his figures were good to about twelve seconds, where Thomas Streete's had been off by five minutes (entry, PDF). I haven't read the tract itself, only these accounts of it.
The tables rest on one assumption: that the Earth turns at a steady rate and only the sun's apparent motion varies. Kepler had let the Earth's rate vary with the seasons. When Flamsteed became the first Astronomer Royal in 1676, Sir Jonas Moore paid Thomas Tompion for two clocks that ran a year on one winding, with pendulums thirteen feet long. Flamsteed timed Sirius crossing the meridian against them, night after night. A star has no ellipse and no tilted path to follow, so if the Earth's spin were uneven, the star would show it. It didn't. The rotation was even, so the correction could be computed in advance. One of the two clocks was sold by his widow and came back to the Octagon Room at Greenwich about 275 years later (Royal Museums Greenwich).
The figure-eight. Plot where the sun stands at the same clock time every day of the year and both causes show at once. The height is the season and the sideways drift is the equation of time. Earth's version is a tall, narrow figure-eight, with a larger loop in the south. That figure now carries the old word analemma, and it is printed on globes and set into the meridian lines of sundials so you can read the correction off the dial itself. Dennis di Cicco photographed it in 1978–79 over Watertown, Massachusetts, with a camera bolted in place and 44 exposures on one frame, each taken at the same clock time at least a week apart (summary). On Mars the orbit is so eccentric that the eccentricity term wins, and the figure-eight closes into a teardrop. The Opportunity rover photographed its sundial through the Martian year and showed this.
What I like is that a sundial's error, once measured, turned into one of the more exact things known about the Earth. To believe the table you have to believe the planet keeps time better than the sun does. Flamsteed checked that against a star and a thirteen-foot pendulum before he trusted it.
The piece Noon, every day for a year draws the figure from the two causes and lets you change the tilt and the ellipse.