Blog

Open data and the city's performance scorecards

How to find and use San Francisco's open data portal, the Controller's Transit and Street Safety scorecard and the street-cleaning response measure.

San Francisco has put a surprising amount of the evidence on itself online, and it is there for anyone who wonders whether the city is doing what it says. There is a portal with hundreds of datasets, a set of scorecards that turn some of them into measures with goals, and a handful of "data stories" on sf.gov that explain a single measure in plain language. None of it needs a records request. What follows is how to find it and how to read it without being fooled.

The portal: SF OpenData

The city's official open data portal is SF OpenData, at data.sf.gov. It launched in 2009 and is run by DataSF, the city's open data program, which sits in the City Administrator's group of agencies (Office of the City Administrator). The older address, data.sfgov.org, now forwards to data.sf.gov, and the portal's own notice says anyone with an application or script that queries the old address needs to switch to the new one.

The portal says it holds "hundreds of city datasets." You can search by topic, or browse by category, department, data type or tag. Many individual departments publish their own data. The portal also holds geographic files. The city's list of 41 "analysis neighborhoods," which many reports use to group data, is a dataset there, and it is why a table that says "Mission" or "Bayview Hunters Point" can be mapped consistently.

Three features are worth knowing before you start.

  1. Every dataset has a page with a description and a data dictionary. Read it. It tells you what each column means, how often the data is refreshed and which department owns it.
  2. You can filter, sort and download from the page. For most datasets you can export a spreadsheet of what you have filtered, without writing any code.
  3. There is an API for developers. DataSF's developer resources describe the portal's Socrata-based interface, with a query language similar to SQL. Developers use it to power apps and dashboards.

If you cannot find what you want, the DataSF landing page at sf.gov/data has a "Can't find a dataset?" form where you can ask for data the city does not yet publish, as well as "Data in action" case studies and a contact address.

The scorecards

A dataset gives you rows. A scorecard gives you a measure, a goal and a trend. The city's performance scorecards are kept by the Controller's Office (Controller's Office), which also audits city services and publishes the city's performance and survey reports (The Controller's reports and the City Survey).

The Transit and Street Safety scorecard, for example, sorts its measures into three sets.

  • Citywide trends, which include the number of traffic-related deaths in San Francisco.
  • Transit performance, which includes Muni ridership as average weekday boardings on buses and light rail, the percentage of scheduled Muni service hours actually delivered and the percentage of evenly spaced Muni arrivals, also called headway adherence.
  • Transit and street safety, which includes the percentage of traffic citations issued for the top five causes of collisions and crimes on Muni per 100,000 vehicle miles.

Each measure has its own page with a definition and a chart. The underlying numbers sit in a dataset the scorecard page links to, called Scorecard-Measures, on the open data portal. The scorecard page lists an email address for questions to the Controller's performance team.

Two details change how the numbers should be read. The traffic fatality figure on the scorecard is the calendar-year total of the values in its chart, and the data carries a one-month reporting lag, so the most recent month is usually incomplete. The Muni collision measure shows a fiscal-year average, and the page notes that its source is SFMTA's own performance portal and is being updated. When a number on a scorecard looks too good or too bad, the first question is what period it covers.

A worked example: street cleaning

The sf.gov "data story" on street and sidewalk cleaning response is a good model for what these pages do. It reports the percent of street-cleaning requests answered within 48 hours, and it explains the setup.

  • Requests come from calls to the 311 Customer Service Center and from internal reports. Requests received by 311 go to the dispatch system of Public Works (San Francisco Public Works).
  • The city's goal is to respond to 95 percent of street and sidewalk cleaning requests within 48 hours.
  • The number on the scorecard page is a fiscal-year average of the response chart, so one bad month moves it less than the monthly line does.

You can test any claim about street cleanliness against it. If a neighborhood group says the city is slow, the scorecard shows whether the citywide response rate sits above or below the goal. If the city says it is on target, the data lets you see the monthly line. The 311 request data is on the portal too, and you can filter by category and neighborhood to see how many requests arrived and how many were closed (Using 311 in San Francisco, by phone, app or web, from a 311 call to a cleaned street).

Here is a hypothetical, labeled as such. Say a dataset lists 1,000 street-cleaning requests for one month in one neighborhood, and 900 of them were closed within 48 hours. The response rate is 900 divided by 1,000, or 90 percent, five points below the city's 95 percent goal. But if the dataset's "closed" field counts a request as closed when a crew arrives, and a request can be reopened, the true rate depends on the definition. That is why you read the data dictionary first.

Open data versus a records request

Open data and public records requests answer different needs. The portal holds information the city has already decided to publish as structured data: tables of requests, permits, budgets and incidents that can be filtered and downloaded. A records request is for documents and communications that are not on the portal, such as an email chain or a report that has not been posted. Under the city's rules, a department must respond to a standard request within 10 calendar days, and in four listed situations it can extend by up to 14 more days. A request headed "Immediate Disclosure Request" must be answered by the next business day if the Sunshine rules for such requests are met. Most departments take requests through the city's NextRequest portal (how to request public records).

A sensible order of operations: search the portal first, because the answer may already be a table. Read the scorecard if the question is about performance. File a records request only for what is left. If the portal has a dataset that nearly answers the question, the department listed on its page can often explain the gap before you file anything.

What the data cannot tell you

Open data is a record of what the city logged, not a complete picture of the city. A few cautions help.

  • Requests are not conditions. A 311 dataset counts what residents and workers reported. A neighborhood with many reports may have more problems, or more residents who know how to report them. Compare rates over time in the same place before comparing places.
  • Definitions move. A measure can be redefined, a category split in two or a data source replaced. The Muni collisions measure on the scorecard says its source is being updated, a flag that earlier values may not line up exactly with later ones.
  • Fiscal years are not calendar years. The city's fiscal year runs from July 1 to June 30. A "2026-27" figure covers July 1, 2026 through June 30, 2027. Mixing a fiscal-year average with a calendar-year total is a common way to get a wrong comparison.
  • Averages hide spread. A fiscal-year average of a monthly line can be the same for two very different years.
  • Lags matter. The traffic fatality figure carries a one-month reporting lag, and other datasets have delays of their own.

How to build a useful question

The portal rewards specific questions. Here is a short method.

  1. Pick a service and a place. "Graffiti reports in the Mission last year" is a question. "Is the city doing a good job?" is not.
  2. Find the dataset and read the description. Note the date range, the refresh cycle and what each status value means.
  3. Filter before you download. Narrow by neighborhood, date or category on the portal so you are not wrestling a huge file.
  4. Check the scorecard for a goal. If a measure has a target, compare your number against it using the same period.
  5. Note the caveats. People enter the data. A lag, a changed definition or a change in how requests are logged can all bend a trend line.
  6. Ask the department if something is unclear. The dataset page lists a contact. If you need records that are not published, a records request can get them (how to request public records).

Other sources on sf.gov

The city's data stories cover several topics besides street cleaning: traffic fatalities, Muni collisions, pothole response and the effort to keep Tenderloin sidewalks and streets clean, among them. The budget pages carry the Controller's six-month and nine-month status reports on the city's finances (The Five-Year Financial Plan and the structural deficit). The Muni service alerts page lists current disruptions, and the SFMTA's performance portal reports on-time and ridership numbers.

Why it matters

Public data lets residents check how well the city keeps its promises, and it lets journalists, researchers, neighborhood groups and the city's own staff work from the same numbers. In a city this large, that is one of the few ways to see how services compare across neighborhoods.

The best habit is a simple one. When you see a statistic about San Francisco in a news story or on a campaign flyer, search the portal for the dataset behind it. Often you can find it in a few minutes, and with it the definition, the date range and the context the flyer left out.

Related city government

Related guides

Latest on this topic