Why the meal it preselects depends on where you are

The evidence against, and what this is not

This says nothing about whether the time you eat is good for you. It is a default in a form. Whether meal timing affects health is a real and disputed field of research. We have not read enough of it to have a position, and a meal logged at 03:00 is filed as a snack, nothing more.

Several diaries are old. Latvia, Lithuania and Slovenia are from the 2000 round; Belgium, Spain, France, Italy, Greece, Luxembourg, Romania, Turkey and the United Kingdom are from 2010. Habits change. Where a country appears in more than one round we use the most recent, which is why the 2020 round is used for eleven of them.

“Eating” in a diary includes drinking, so a country that spends its morning with coffee has a long, low morning rather than a peak. That is exactly why the window is cut at 65% of the peak rather than at half of it. At half, the Dutch morning ran into lunch and the Finnish evening covered six hours. The 65%, the 90-minute limit and the rounding to half hours are engineering choices we made by looking at the curves, not findings. The rounding moves an edge by up to a quarter of an hour, which is less than the difference between two survey rounds of the same country.

A national average is not a person. Shift workers, students, anybody fasting, and anybody who simply eats late are all inside these numbers and none of them is described by them. Living in Madrid does not make your habits Spanish either. All of this is one preselected chip you change with one tap. Nothing overrides your choice afterwards, not even the analyser. A plate of pasta at 13:00 once came back as “dinner” and overwrote a correct lunch.

The hours are read against the photo’s clock, not yours. A picture carries the moment it was taken, so a plate photographed at 09:14 and logged at ten at night is still breakfast, and it is filed at 09:14 on the day it was eaten rather than at the moment you got around to typing it. The only thing read out of a photograph is that timestamp: never the location, which sits in the same metadata and is not looked at. A picture that carries no timestamp — a screenshot, anything out of a chat app — falls back to the clock, exactly as before.

Eight countries are derived rather than measured. Table X publishes peaks for employees on a weekday, who eat earlier than everybody else, so the four rows taken from it carry a correction: the median gap between employees and everybody else in the four countries the paper reports both for, 38 minutes at breakfast and roughly nothing at lunch and dinner. A median of four pairs is thin evidence, and it is the weakest number on this page.

And every derived row borrows a shape it did not measure. A published peak is one number; a window needs a width, so these rows take the average width of the measured European ones and wrap it around their own peak. For Canada that is a small claim. For Indonesia it is a larger one, and in Taiwan's case it is visibly wrong in a known direction: its own survey reports lunch as extremely concentrated and breakfast as spread out, and the borrowed shape draws them the other way round. It is still closer than the European peak it replaces, which is what these rows are competing against.

And most of the world is still not in here. There is no row for Japan, Australia, Nigeria, or most of Asia, Africa and Latin America. Everyone there gets the average of the 23 measured European countries, which is a schedule nobody actually keeps and may poorly fit habits elsewhere.

Four countries are missing on purpose, which is worse. We went looking for China, India, Bangladesh and Mexico and decided not to add them. Not because there is nothing published, but because what is published is not a sample of a country: the Chinese clock-time studies we located sampled airline shift workers and sports students, India's from a 94-person study in Delhi and two online questionnaires, Bangladesh's from sixteen adolescent girls whose evening meals were never actually observed, and Mexico's from 379 women in Mexico City surveyed by telephone during the pandemic. Mexico's national time use survey cannot help either: it asks people to total up their week rather than keep a diary, so it never records what time anything happened. For India in particular we are fairly confident the current default is wrong — three unrelated studies all put Indian dinner later than our fallback window closes — and we would still rather be visibly wrong than invent a number that looks sourced. These searches do not establish that no suitable national data exists.

We would rather add real rows than invent them, so if you know of a national time use survey or dietary survey with an eating-by-time-of-day breakdown we have missed, send it to us. That is exactly how Portugal, Indonesia, Taiwan and Brazil got here.