- Home
- Methodology
Methodology
Where the numbers come from, how they are calculated, and what they cannot tell you.
The three labels
Every figure on this site carries exactly one label. The database that feeds the site will not accept a figure without one.
Sourced Sourced
A number quoted from an official publication, such as a state court system's annual statistics. We record the page, table, row and column it came from and keep the number exactly as printed. Each time the site is rebuilt, the number is read again from the saved copy of the publication, and the rebuild stops if it does not match.
Computed Computed
A number we calculated from public case records, such as a count of cases or a median. The records, the date we retrieved them and the calculation are all recorded, so the figure can be recalculated by anyone with the same file.
Estimated Estimated
A number we worked out because no direct figure exists. An estimate is always shown as a range, always names its method, and is built only from figures that are themselves sourced or computed. Estimates are never made about a named law firm, attorney or person; the database refuses them.
Missing means missing
Where a figure does not exist or rests on too few cases, the site says "Not available". We do not calculate a share from fewer than 20 cases or a median from fewer than 10, because a percentage built on a handful of cases misleads more than it informs.
Federal court figures
Federal figures are calculated from the Federal Judicial Center's Integrated Database, which lists every civil case in the federal district courts. We use the cases the courts classify under "Personal Injury".
- Federal cases filed
- Cases filed in the year, excluding cases flagged as multidistrict litigation.
- Federal cases closed
- Cases closed in the year, excluding cases flagged as multidistrict litigation.
- Federal cases resolved in the court
- Cases closed in the year, excluding multidistrict litigation and excluding cases transferred or sent back to another court.
- Federal cases decided at trial
- Resolved cases whose recorded disposition is a jury verdict, a directed verdict or a court (bench) trial.
- Share of federal cases decided at trial
- Cases decided at trial divided by cases resolved in the court.
- Plaintiff win share at federal trials
- Among cases decided at trial where the court recorded who won, the share recorded as judgment for the plaintiff or for both sides.
- Share of federal cases recorded as settled
- Cases whose recorded disposition is 'settled' divided by cases resolved in the court. Courts record this code inconsistently, so it understates settlements; many voluntary dismissals follow a settlement.
- Median days from filing to closing, federal
- Median number of days between filing and closing for cases resolved in the court.
- Federal multidistrict litigation cases closed
- Cases closed in the year that are flagged as multidistrict litigation. Shown separately because a single mass litigation can add tens of thousands of cases.
Years are calendar years by the actual filing or closing date, so our counts can differ slightly from the courts' own published tables, which use reporting years.
We do not use the database's dollar field. The agency that publishes it advises against using it for analysis, and independent audits found it wrong in about four in ten tort trials. No dollar figure on this site comes from it.
Who "won" is as the court recorded it. A judgment "for plaintiff" says nothing about the amount, and many trial outcomes are later changed by the judge or on appeal.
State court figures
State court figures are quoted from each court system's own published statistics. States define case categories differently. California's categories, for example, combine personal injury with property damage and wrongful death. For that reason we do not compare states with one another or rank them.
Claims against New York City
New York City's Law Department publishes a record of each lawsuit it handles, including the amount the City paid. We use the matters assigned to its Tort division. These are real payments, but only by one defendant, the City, so they do not describe injury cases in general. The City does not separate personal injury from property damage claims.
- Matters closed
- Litigation matters closed in the year.
- Matters closed with a payment
- Matters closed in the year in which the government paid more than zero.
- Share of closed matters with a payment
- Matters closed with a payment divided by matters closed.
- Median payment
- Median amount paid, among matters with a payment.
- 25th percentile payment
- A quarter of payments were at or below this amount.
- 75th percentile payment
- Three quarters of payments were at or below this amount.
- 90th percentile payment
- Nine in ten payments were at or below this amount.
- Total paid
- Sum of amounts paid on matters closed in the year.
- Median days from start of litigation to closing, paid matters
- Median number of days between the recorded start of litigation and closing, among matters with a payment.
We do not read or publish the names of the people who brought the claims.
Law firms
The City's records name the law firms that represented plaintiffs. For each firm we count the matters in which it is named and, where there are at least 10 paid matters, the median payment. A matter with several plaintiff firms counts once for each. Firm figures are only ever Computed from the records.
Matching names. The records spell firm names inconsistently. Two spellings are treated as one firm only when they are identical once letter case, punctuation and generic words ("Law Office of", "The", "Esq.", "P.C.", "& Associates") are set aside. A difference in any personal name or initial, even one letter, keeps them separate: wrongly joining two firms would credit one with the other's cases, while leaving a firm under two spellings only splits its own count. Of 115 look-alike pairs, 51 were joined and 64 were kept separate.
Firms are listed alphabetically and are not ranked. Only firms with at least 10 closed matters in the period have a page (918 of 4,830). Some entries are individual attorneys practising under their own name.
How estimates are made
Jury-trial share borrowed from states that publish it
For a state whose courts do not publish how injury cases end, the share of cases decided by a jury is estimated from the states that do publish it (California, Texas and Michigan). The range runs from the lowest to the highest of those states' shares and the value shown is the middle one. Limits: states define case categories and count dispositions differently, and the state being estimated may fall outside the range.
Case tree model
Our own model of what happens when a court decides a case, and what happens next. It has three kinds of branch. Who wins: among cases the court decided at trial, and separately on a motion before trial, the share in which it recorded a judgment for the plaintiff. Appeal: for each way a case can end in the district court (a judgment for the plaintiff, a judgment for the defendant, a dismissal by the court, or a recorded settlement or voluntary dismissal), the share of cases that were then appealed, found by matching each case to the federal appeals database by district, office and docket number. Result of the appeal: among those appeals that the court of appeals decided on the merits, the share affirmed and the share reversed or vacated at least in part. Each branch takes five years of resolved cases; the value is the share over the five years together and the range runs from the lowest to the highest single year, widened at each end by half the gap between them. The appeal branches use five years ending one year back, and the result branches five years ending two years back, so that appeals have had time to be filed and decided. A branch is estimated only where each of the five years has at least 20 cases behind it. Tested by rebuilding each branch for past years from the five years before; the results are published with the model. Limits: who appealed is not recorded, so an appeal after a judgment for the plaintiff is not always the defendant's. A reversal sends many cases back for more proceedings; it does not mean the other side won. Winners are as the courts recorded them, and the courts do not record one for every judgment. It covers federal courts only, describes cases as a group, does not predict any single case, and says nothing about amounts or about any named judge, firm, person or litigation.
Filings outlook
Our own estimate of how many cases will have been filed in a year the federal case file does not yet cover in full. For the year under way, it takes the cases already filed in the months the file covers and scales them up by how the full year compared with the same months in each of the last five years. The value uses the five years together; the range runs from the lowest to the highest of those five yearly comparisons, widened at each end by half the gap between them, and never below what has already been filed. For the year after, it carries that estimate forward by how filings changed from one year to the next over the last five years: the value uses the middle change, and the range applies the smallest change to the low end and the largest to the high end. An estimate is made only where each of the last six years had at least 50 filings. Tested by rebuilding it for each past year as if that year were under way; the results are published with the estimates. Limits: it assumes the rest of the year behaves like recent years, so a sudden wave of filings, such as a new mass litigation, a change in the law or a court closure, will not be foreseen. The latest months in the file can be revised as courts report late. The estimate for the year after is much less certain, and its range is wide for that reason. It counts cases filed in federal courts only, leaves out multidistrict litigation for injury cases, and says nothing about any single case, firm, judge or litigation.
Case path model
Our own model of how a kind of case tends to end and how long it takes. For one kind of case in one place, it takes every such case resolved in the federal courts there over the last five complete years and works out the share that ended each way (recorded as settled, dismissed voluntarily, dismissed by the court, decided on a motion before trial, decided at trial) and the share resolved within one, two and three years of filing. The value shown is the share over all five years together. The range starts from the lowest and highest single years and is then widened at each end by half the gap between them, because testing showed the unwidened range was too narrow: we rebuilt the model for each year from 2020 on using only the five years before it, and the share that actually followed fell inside the unwidened range about 64% of the time and inside the widened range about 88% of the time. The test results are published with the model. An estimate is produced only when each of the five years has at least 50 resolved cases. Limits: it describes cases as a group, as the courts recorded their endings. A voluntary dismissal often follows a settlement, and the record does not say which. It covers federal courts only; state courts may differ. It does not predict any single case, and it says nothing about amounts.
How the case path model did when tested
The case path model is our own. Before relying on it we tested it against the past: for each year from 2020 to 2025 we rebuilt the model using only the five years before, then compared it with what happened. The table shows how often the share that actually followed fell inside the model's range, and the typical gap between the model's value and the actual share. The first version of the range held the actual result 64% of the time, so the range was widened; it now holds it 88% of the time. That still means roughly one result in 8 falls outside the range.
How the model did when tested against past yearsComputed
| What was estimated | Tests | Actual share fell inside the range | Typical gap |
|---|---|---|---|
| Dismissed by the court | 763 | 87% | 3.5 points |
| Dismissed voluntarily | 763 | 85% | 3.5 points |
| Decided on a motion before trial | 763 | 91% | 1.5 points |
| Recorded as settled | 763 | 89% | 5.1 points |
| Decided at trial | 763 | 92% | 0.6 points |
| Resolved within 12 months | 763 | 87% | 5.6 points |
| Resolved within 24 months | 763 | 87% | 3.7 points |
| Resolved within 36 months | 763 | 86% | 1.7 points |
| All measures | 6,104 | 88% | 2.5 points |
A test is one place, one kind of case and one year. The typical gap is the median difference between the model's value and the actual share, in percentage points.
The model's estimates appear on state and case type pages and in the three-year review.
How the case tree model did when tested
The case tree model estimates who wins when a court decides a case, how often each kind of ending is appealed, and how appeals end. Each branch was rebuilt for past years from the five years before and compared with what happened. Branches with few tests rest on few years and should be read with more caution.
How the case tree did when tested against past yearsComputed
| Branch | Tests | Actual share fell inside the range | Typical gap |
|---|---|---|---|
| Affirmed, after a dismissal by the court | 9 | 89% | 4.8 points |
| Affirmed, after a judgment for the defendant | 8 | 100% | 2.0 points |
| Affirmed, after a judgment for the plaintiff | 4 | 75% | 7.7 points |
| Appealed, after a dismissal by the court | 295 | 93% | 2.5 points |
| Appealed, after a judgment for the defendant | 128 | 93% | 4.0 points |
| Appealed, after a judgment for the plaintiff | 36 | 94% | 3.1 points |
| Appealed, after a recorded settlement or a voluntary dismissal | 722 | 91% | 0.8 points |
| Plaintiff wins on a motion | 130 | 86% | 2.6 points |
| Plaintiff wins at trial | 18 | 83% | 5.7 points |
| All branches | 1,371 | 91% | 1.4 points |
A test is one place, one kind of case and one year. Branches with few tests rest on few years and should be read with more caution.
The estimates are on the case tree page.
How the filings outlook did when tested
The filings outlook is our own estimate of cases filed in a year the federal case file does not yet cover in full. We tested it by rebuilding it for each past year as if that year were under way, then comparing it with what was filed. The estimate for the year after is much less certain than the one for the year under way, and its ranges are wide for that reason.
How the filings outlook did when tested against past yearsComputed
| Estimate for | Tests | Actual filings fell inside the range | Typical gap |
|---|---|---|---|
| The year under way | 862 | 90% | 7% of actual filings |
| The year after | 694 | 87% | 18% of actual filings |
A test is one place, one kind of case and one year.
The estimates are on the filings outlook page.
What this site cannot tell you
- Most injury cases end in a private settlement. Private settlement amounts are not public anywhere, and nothing here measures them.
- Trials are rare and unusual. Figures about trials describe the small share of cases that reach one.
- Nothing here predicts the outcome or the value of any particular case.
- Coverage is uneven between states. The sources and coverage page lists what we have for each.
Updates and corrections
Each data page shows the date its source files were last retrieved. If you find an error, see corrections and removal requests.