// Tabla html con la clase del sitio, no Inputs.table: son siete filas fijas, así
// que no gana nada con selección de renglones ni con scroll interno.
html`<table class="data-table">
<thead><tr>
<th>Benefit</th><th>National</th><th>Median municipality</th><th>Municipalities below</th>
</tr></thead>
<tbody>
${nacional.map(r => html`<tr>
<td>${medidas[r.prestacion].nombre}</td>
<td class="data-table-num">${r.nacional.toFixed(1)}%</td>
<td class="data-table-num">${r.mediana_municipal.toFixed(1)}%</td>
<td class="data-table-num">${r.pct_debajo}%</td>
</tr>`)}
</tbody>
</table>`The average lives nowhere
Curiosity · Encuesta Intercensal 2025
A national average weights every person equally. That makes it a correct number about the country and a poor description of almost any piece of it. Mexico’s new intercensal survey shows the gap at municipal resolution.
Curiosity · INEGI, Encuesta Intercensal 2025 · Public microdata
A national average weights every person equally. That is what makes it the right number for a question about people, and the wrong one for a question about places. If half a country’s population lives in a handful of large cities, the national average resembles those cities. It does not resemble the typical town.
The two are different quantities. A population-weighted average answers “what happens to the average person”; an average across places answers “what happens in the average place.” Almost always the first gets computed and the second gets reported. They can sit thirty points apart.
Mexico’s 2025 intercensal survey, published in September 2026, makes the gap easy to see. It interviewed about seven million dwellings, which is enough to estimate each of the country’s 2,478 municipalities separately.
Before the map: the whole country
The survey asks about seven employment benefits, and it asks only of people who work for an employer (not the self-employed, who were never posed the question). Among those 40.9 million people, this is the country:
Two things stand out before any geography. Profit sharing is last at 42.7% and the year-end bonus is first at 64.2%, and both are legal obligations, not employer favours. Twenty-two points separate two entitlements that the law treats identically, which says less about what the law requires than about what gets complied with and what workers know they are owed.
And the distribution is not a gradient. It is two blocks:
36.4% have all seven and 33.0% have none. Only 2.9% sit at the exact middle of the scale. One in three employees in Mexico (13.5 million people) works for an employer with no year-end bonus, no paid holiday, no health coverage and no retirement saving. The mean, 3.77 of seven, describes almost nobody.
Now the map
Everything above is an average of people. The map changes the unit: each polygon is one municipality, and they all weigh the same whether two thousand people live there or two million.
Hatched municipalities are the ones whose estimate is imprecise. They are shown anyway, in full colour, for a reason the last section explains.
Each dot is one of the 2,471 municipalities with an estimate. The national figure sits above of every 100 of them. For the year-end bonus the median municipality is at 33.8% against a national 64.2%: thirty points of distance, and not a calculation error. The national figure weights people; the map weights places.
Why some municipalities are hatched
A survey estimate carries sampling error, and the usual way to summarise it is the coefficient of variation: the standard error divided by the estimate itself. INEGI publishes a municipal figure when that ratio stays under 15% and withholds it when it does not. It is a sensible rule and here it is a trap.
The coefficient of variation carries the estimate in its denominator, so it grows whenever coverage is low, even if the error does not change. In these data the correlation between coverage and the coefficient of variation is −0.61. Sorting municipalities into five groups by precision, the median coverage of the year-end bonus falls from 65% in the most precise group to 12% in the least, and only 13% of that last group clears the 15% bar.
So hiding the imprecise estimates does not remove noise scattered at random. It removes, almost exactly, the municipalities where the benefit is rarest. The map would look cleaner and say something false. Hiding them would raise the median municipality by between 5.3 and 8.9 points depending on the benefit; for profit sharing the median would go from 10.8% to 19.7%, nearly double, without a single number being miscalculated.
That is why this map draws them, and it is worth separating two readings. Taken one at a time, hatching does mean the estimate is weak: the interval is wide and the number will not bear quoting on its own. Taken together, the hatched set is not a random sample of municipalities, it concentrates where coverage is low. So removing it would not remove noise, it would move the reading of the country upward. The practical rule is to avoid quoting any single hatched municipality, and to avoid reading the hatched areas as territory without information.
How this is built
Source: INEGI, Encuesta Intercensal 2025; microdata published 22 September 2026. Universe: people aged 15 and over, employed and working for an employer. The survey asks these seven questions only of that group, so this does not measure informality and is not comparable to the rate ENOE publishes. Estimated with the survey’s own design (single-stage, stratified, clustered), using the strata and primary sampling units INEGI publishes. Municipal medians on this page are computed over all 2,471 municipalities with an estimate, not only the precise ones, for the reason given above. Geometry: the Marco Geoestadístico of the Encuesta Intercensal 2025 itself, with the 2,478 municipalities in force at the time of the survey; seven are left uncoloured because INEGI did not reach minimum coverage there and publishes only dwellings and population. Descriptive and exploratory, and unrelated to the author’s work at any institution.
The script that turns the validated estimates into the files this page reads is scripts/09-curiosities-prestaciones.R in this site’s public repository.