Brazil presidential elections, one voting machine at a time

About the data

Brazil votes on electronic voting machines (urnas). Each polling section (seção) has one machine. When polls close, the machine prints and transmits a tally bulletin (boletim de urna) with its section's results. Sections are grouped into electoral zones (zonas). Here, one bulletin is one "batch" and one circle.

• Circle area is proportional to the combined + votes in the section.
• Election: the menu at the top switches between rounds. Each round compares its own two leading candidates.
• Color by state gives each state a shade of its region's color, so regions stay recognizable while neighboring states contrast. Hover over a state's code in the legend for its totals.
• Color by batch size colors each section by its + votes, spread by percentile so that each color covers the same number of sections. The legend still filters which sections are shown.
• share is 's votes divided by + votes. Blank and null votes are excluded.
• Vote ratio is 's votes divided by 's, on a log scale. Half a vote is added to each side so that sections where one candidate got zero votes stay on the chart. Ratios beyond 64:1 are drawn at the edge.
• Counting time is when the TSE received the bulletin, in Brasília time (BRT, UTC-3). Polls closed nationwide at 5:00 PM BRT. Sections abroad close earlier in their local time zones, so many were received before that. They are off-screen to the left until you zoom out.
• By region orders sections by region, then state, then municipality, electoral zone and section number.
• Overlay lines follow the legend. The running total (by counting time) and the average (by batch size) get one solid line per shown region, or per shown state when coloring by state. A dashed white line shows everything shown combined. By counting time, the menu next to the overlay switches between three statistics. The running total is the share of everything received so far. The moving average is the share of bulletins received in the trailing 5–60 minutes. The spot share is the share within each fixed 5–60 minute interval, so the line jumps when the mix of reporting places changes. Moving-average and spot-share lines fade where their window holds few bulletins (fully opaque from 400 bulletins), and the strip under the chart shows how many bulletins each region sent per interval. Hover over any overlay line to read its values.

Looking for non-random patterns

Most of the variation between sections is geography: some regions lean strongly toward one candidate, and they also report at different speeds. These views remove geography so that anything left over stands out.

• Residual vs polling place is the section's share minus the share of the other sections at the same polling place (usually a school). The section itself is left out of its own reference. Neighbors voting in the same building are very similar, so the residual is usually small. Sections that are alone at their polling place have no reference and are hidden in this view.
• Funnel plot: choose Residual with By batch size. The yellow curves show how far a section would stray from its neighbors by chance alone (95% band and ±3 standard deviations, assuming an evenly split electorate). Small sections naturally scatter more, so the bands narrow as size grows. Real data is expected to spread somewhat wider than pure chance, because voters within one school are not perfectly alike. A modest excess is therefore not evidence of anything by itself; look for structure, such as excess concentrated at one time, region or size.