Each one starts with a scope call and a written statement of work, which fixes the price before any work begins.
01
Methods review
We check an analysis for errors and bias before it is published.
From $1,5001 to 2 weeks
Who it's for
Reporters and editors with a data story near publication, and nonprofits or researchers about to release a report.
What you get
A written review of the data, code and claims: what holds and what to fix
Checks for wrong population counts, breaks in the records, miscoded values and groups too small to compare
A read of the writing for claims the data can't support
Key figures rerun from the source data
A call to go through the review
02
City data analysis
Rates per resident on your city or question, with the plain explanations tested.
From $3,0003 to 5 weeks
Who it's for
Local newsrooms, advocacy groups and community organizations with a question about their city: who is stopped, who is harmed, how long people wait for help.
What you get
An audit of the source data: what it covers and where it has gaps
Rates per 100,000 residents for each group, from Census counts
Tests of the plain explanations, such as age, neighborhood, offense type and time period
Charts ready to publish, each with the numbers behind it
A methods note and the code, so the work can be checked
03
Litigation and oversight support
Enforcement and disparity analysis of stops, arrests or charges, built to hold up under scrutiny.
Quoted per matterSet by the case schedule
Who it's for
Public defenders and civil rights lawyers, and oversight boards, inspectors general and auditors at the city, county or state level.
What you get
Stops, searches, arrests or charges by group, measured against residents and against reported crime
Each comparison choice set out and tested, since the answer often depends on it
Rival explanations tested, with every decision documented
A written report with code that reproduces every number
Confidential handling under your agreement or protective order
Timelines run from when we have the data. Project work runs on a fixed statement of work: half at the start, half on delivery.
Selected work
Published analyses
These reports appear in Residents Count, Specta's research publication. Each one links to its full analysis, code and data.
167 → 94homicides, from the year before to the first Guard year
Question
Did crime in DC fall after the National Guard arrived in August 2025, and did it fall more where troops were posted?
What we did
Compared the first Guard year with the 10 years before, using police incident data and gunshot sensors. Published a test plan first, then compared about 800 half-kilometer squares near and away from Guard posts.
What we found
Homicides fell from 167 to 94, and gunshots detected fell 44%. Areas near troop posts did no better than comparable areas on any of six measures.
Limits
The data can't say what drove the citywide drop. There is no official map of Guard posts, so locations were rebuilt from news reports, Guard releases and court records.
2,552,921assault victims in FBI records, one method for every city
Question
Is the gap in reported assault rates between Black women and other women found across large US cities, and does anything measured explain it?
What we did
Fixed a rule for which cities qualify before seeing any results, then applied one method to every city. Tested age, intimate partner assault and how officers record race.
What we found
In all 46 cities with complete data, Black women's reported assault rate is higher than White women's. In the typical city it is about 3.8 times White women's and 2.4 times Hispanic women's. Nothing measured explains it.
Limits
Police records count only assaults reported to police, and willingness to report differs by group and place. The national victimization survey shows a much smaller gap, and a companion report tests why.
5analyses of one police department's public records
Question
How do the Metropolitan Police Department's stops, searches, arrests and crime records land across Washington?
What we did
Stops and arrests against residents by district, search outcomes, enforcement against reported crime, a daylight test of stops, and a test for downgraded crime reports.
What we found
In every district, the Black share of people stopped is larger than the Black share of residents: 1.82 times citywide. Black people who are stopped are patted down 1.7 to 4.5 times as often as others, depending on the district, while searches of both groups find something at about the same rate. One result favors the department and is reported as prominently: at the same time of day, daylight doesn't change who is stopped.
Limits
Records count stops, not people, and race is as recorded by officers. Comparisons with residents can't account for people who work in or visit a district.
13.8M911 calls from the public, across 197 neighborhoods
Question
Do 911 calls wait longer for police in some New York neighborhoods than in others?
What we did
Grouped neighborhoods by income and makeup and compared like calls at like hours. Tests were set in a plan committed before they ran, and the first half of 2026 was held back as a check.
What we found
Critical calls wait under a minute longer in lower-income, majority-Black and majority-Hispanic neighborhoods. Less urgent calls wait 28% to 79% longer. The gaps sit between precincts: inside the same precinct, they disappear.
Limits
The records show how long calls waited, not why. About 40% of calls not about a crime in progress have no arrival time.
Agree on the question and the method, then follow the data
Scope call
A short call on the question, the data and the deadline. If the data can't answer it, we say so.
Statement of work
One page with the scope, deliverables, timeline and a fixed price. Work starts when it is signed.
Analysis
The data is audited first, then the rates and the tests of rival explanations. Every decision is written down.
Review
Checked for errors and unsupported claims. You see the draft findings before they are final.
Delivery with methods
Findings, charts and a methods note, with the code to check every number.
What we hold to
On every engagement, whoever the client is.
Results are reported whichever way they point.
A finding that cuts against what a client hoped to show stays in the report.
Data is handled under agreement.
Client data is used only for the project, on the terms you set, and returned or deleted at the end.
No outcome is promised in advance.
We agree on the question and the method, not the answer.
Limits are stated.
Every deliverable says what the data cannot show.
About
Small and independent
Specta LLC is an independent data analysis practice. The senior data scientist on your scope call is the one who does the analysis, from the data audit to delivery.
We also publish our own research and build tools for the public, so our methods are open to anyone who wants to check them.