Mathew Kiang (.com)Yes, with one "t".

· 8 min read

Analyzing San Francisco’s parking citation data

San Francisco issues about 5,000 parking citations per day, resulting in over 21.4 million parking citations to over 5.2 million unique license plates since January 2008. That’s a pretty astounding number of parking citations when you consider San Francisco is home to less than a million people, less than half a million registered cars, and covers an area that’s just 7 miles by 7 miles. Parking citations are handled by San Francisco MTA while moving violations are handled by the San Francisco Police Department, which has a well–documented habit of not enforcing (even dangerous) vehicle violations — resulting in about a 90% drop in citations since 2019.

Last week, I nearly got killed by a person running a (blatantly) red light. Ironically, it happened in front of where I live — literally one block from a police station. Thankfully, the person had a fairly memorable license plate so I went home and found the person had over two dozen parking citations in 2024 alone.

I decided to take a closer look at the data (helpfully provided by DataSF) and ended up going down a little rabbit hole. Come join me.

The majority of parking citations are for street cleaning (8.1M) or meter violations (6.6M), which account for about 70% of all parking citations from 2008 through 2024. A couple of things to note: (1) the plot (as do all plots in this post) starts in 2009 because data in 2008 does not appear complete and (2) November 2009 is missing (and this is verified to be missing in the raw data as well)).

A stacked area chart of parking citations from 2009 to late 2024, y-axis 0 to 8 thousand per week. Eleven violation types are stacked in a dark-purple-to-yellow scale, with totals in the legend: Street Cleaning (8,148,599), Overtime Parking & Meter Violations (6,614,846), Prohibited Parking / Driving Area (1,225,311), License Plates & Registration (1,151,668), Blocking Vehicle Access (956,250), Improper / Unsafe Parking (759,129), Commercial Loading (671,628), Transit Fare Evasion (534,761), Pedestrian / Cyclist Right Of Way (508,872), Public Transit & Passenger Loading (359,151) and Other (254,877). The top two bands are most of the height throughout. The total drifts down from about 7,000 a week in 2009 to about 5,000 by 2019, then collapses almost to zero in a narrow notch in early 2020 before climbing back to roughly 5,000. A thin blank stripe just after 2009 is the missing month of November.

Note how the COVID-19 pandemic massively dropped enforcement and/or citations. I don’t show it here but the drop begins the week of March 16th, 2020 (the day of the Bay Area shelter-in-place order). On March 16th, 2020, there were about 4,200 citations and by March 19th it dropped to 971, continuing to drop until it hit 98 on March 29th.

Temporal patterns in parking citations

Most parking citations happen between 12pm and 1pm on weekdays and there are substantially fewer citations on weekends. Whether or not this is due to lower enforcement, fewer violations, or a mix of both is unclear from the data, but if you had to pick a time and day to park illegally, it would be Sunday at 5am.

Two panels sharing a time-of-day axis from midnight to midnight. The top panel is a line of total parking citations by minute of the day, 0 to about 78 thousand: near zero overnight, rising in steps from 6am through a series of peaks around 8am, 9am and 11am, a sharp maximum just after noon at about 78 thousand, then a long decline through the afternoon to near zero by midnight. The line is visibly serrated rather than smooth. The bottom panel is a heatmap with one row per weekday, Sun at the bottom to Sat at the top, coloured on a legend from 0.10% in dark purple to 3.00% in yellow. The yellow band sits at noon on Monday through Friday; Saturday and Sunday stay dark all day.

In my opinion, the most interesting thing about this plot is what happens within the hour. It’s less in clear in some hours (e.g., 2am), but very clear in the afternoon hours that enforcement officers like to round the time to the nearest 0 or 5. This is (I assume) the same phenomenon that occurs with age heaping when people in the US like to round their age to a number that ends in 0 or 5. (Side note, interestingly, other countries age heap around other numbers.)

The other interesting thing that occurs is it seems like for most hours, you are much more likely to get a ticket earlier in the hour. There is no clear mechanism I can think of for this to occur (e.g., more possible violations earlier in an hour). If I aggregate across all days and hours (focusing only on the minute of the hour), this trend is much clearer — you’re most likely to get a ticket at the 10 minute mark and then it steadily (and linearly) decreases from there.

A line chart of total parking citations against the minute of the hour, 0 to 60, y-axis 250 to about 460 thousand. The line dips to a low near minute 1, jumps steeply to about 420 thousand by minute 5, peaks at roughly 460 thousand at minute 10, then declines almost linearly to about 250 thousand by minute 59. Sharp spikes stand above the trend at minutes 5, 10, 15, 20, 25, 30, 35, 40, 45, 50 and 55 — every multiple of five.

Parking citations are unequally distributed

Below, I plot the proportion of vehicles (on the x-axis) that account for the proportion of citations (on the y-axis). For example, in 2024, the top 0.1% (about 500 vehicles) accounted for 3% of all tickets. Similarly, the top 1% accounted for 12% of all tickets, the top 5% accounted for ~30% of all tickets. In fact, just 15% of all cited drivers (about 80,000 of the 510,000 unique vehicles that were cited) accounted for half of all parking citations.

A Lorenz curve. The x-axis is the proportion of all cited vehicles, ranked, from 0 to 0.25; the y-axis is the proportion of all citations, 0 to just above 0.6. The curve rises almost vertically from the origin, reaching 0.2 of citations by 0.025 of vehicles, 0.4 by 0.10, and about 0.62 by 0.25. Its slope flattens steadily but the curve never approaches the diagonal an even distribution would follow.

This is a pretty wild distribution. It’s not surprising from a behavioral standpoint — at some point, getting another parking ticket is not going to matter to you so the type of person that gets 10 parking tickets is probably also the type of person that gets 50 parking tickets. But just from a sheer numbers perspective, it’s pretty shocking.

For example, one license plate in the data set accounted for nearly a quarter million dollars in fines (from about 2,800 citations). Below, I plot the top nine violators in the data along with where their citations were more likely to occur (conditional on location data being available, which is not always the case with recent years).

A line chart above a three-by-three grid of maps, one per vehicle, for the nine most-cited license plates. Each is titled with its plate, total tickets and total fines, from CA 4C77711 (2,836; $245,368) down to CA 5K61658 (1,472; $113,305). In the line chart, cumulative tickets against year from 2009 to 2025, every line climbs steeply from 2009 to about 2015 and then flattens or stops; the top vehicle reaches about 2,800 by 2019 and none of the nine is still climbing after 2020. Each map plots that vehicle's citations as dots on a pale San Francisco street map, coloured by violation type on a legend of eleven categories from Blocking Emergency Access in dark purple to Street Cleaning in yellow. Eight of the nine dot clusters sit tightly in the downtown and northeast waterfront; CA 8G95568 and CA 5K61658 spread further, the last of them into a second cluster south of Market.

Nearly all citations given to the top violators occur in downtown San Francisco, which is not super surprising — there is both greater density and more opportunities for violations to occur. I assume most of these drivers have jobs that require them to work in downtown and they simply can’t be bothered to find parking. Two drivers, 8G95568 and 5K61658, have slightly different citation patterns than the others. Getting cited more often in different areas (north of Market vs south of Market) and for different reasons.

It also seems like for these nine violators, the citations stopped climbing after 2020. So perhaps we should just look at current violators? Below I plot the top 50 violators in 2024 (rows) where each column represents a day. On average, the top 50 violators had just three days between citations and one person managed the incredible feat of getting at least one citation every single day for 58 consecutive days.

A tile chart of the 50 most-cited vehicles in 2024, one row per vehicle and one narrow column per day, running Jan through Dec. A tile is drawn only on days that vehicle was cited, coloured on a legend from 1 citation in salmon through 2, 3 and 4 to 5 in black. Salmon dominates. The rows are dense with only short gaps, and two rows carry unbroken runs of consecutive days lasting well over a month — one across February through April, another from August into November.

Anyways, I don’t think there’s a take home here other than be careful when you’re crossing the street in San Francisco.

(Code is here.)