Research

Research and data

Which drafting language actually fails in court.

Most drafting advice is a matter of habit and taste. These studies measure it instead, over the published opinions themselves. The data and the code are open, and so are the counts of everything each study threw away.

I litigate deeds and trusts after they fail, which is where the questions here come from.

A clause that draws a lawsuit has already cost the family money, whether or not the court agrees it was unclear. So each study reports three numbers rather than one. The first is how often the language was litigated at all. The second is how often the challenge succeeded. The third is how often the language was attacked and held up, which is the most useful result a drafter can have, because language that has survived a fight carries precedent that it is clear.

The studies

286,846 term and decision pairs

Disputed terms in state appellate courts

Courts quote the words the parties fought over. This dataset collects those quoted words from every state appellate decision that contains an ambiguity holding, in all 50 states and the District of Columbia, and records whether the court held the language ambiguous or clear.

What it shows. Terms that fix a time and terms that set scope are held ambiguous about 41 percent of the time. Words of succession, meaning issue, descendants, heirs and per stirpes, are held ambiguous 27 percent of the time, which is the lowest rate of any category measured.

Method and build script

932 Florida appellate decisions

Reformation of Florida instruments, 1853 to 2026

A reformation case is a drafting error somebody paid a lawyer to fix. This dataset covers every Florida appellate decision that litigates reformation of a will, trust, deed, contract or insurance policy, with the instrument, the error alleged, and which statute was in force.

What it shows. Florida allowed trust reformation in 2007 and will reformation in 2011. Before 2011 a Florida will could not be reformed at all. Thirty-six decisions state that older rule and six decisions have applied the statute since.

Method and build script

19,085 decision and statute pairs

Florida holdings resting on statutes that changed

A citator reports whether a later case overruled an earlier one. It reports much less about the other way a holding stops being good law, which is that the legislature amended the statute and no court has had occasion to say so. This dataset measures that gap by joining Florida opinions to the Florida Statutes edition by edition.

What it shows. Of the Florida decisions that construe a statute, 77 percent rely on a section that has been amended since the decision, and the median gap is 20 years. Where both editions of the statute are held, about seven in ten of those amendments changed the operative text.

Method and build script

How the work is done

The corpus is 10.8 million published US judicial opinions held locally, joined to the Florida Statutes one annual edition at a time. Retrieval and extraction are ordinary code, so anyone can read exactly what was selected. Classification uses rules that were written against a hand coded sample and ship with their measured accuracy attached, which means the error rate is reported rather than assumed.

Every study states its exclusion funnel with counts, because quiet filtering is the most common defect in research on opinions and it is invisible in the result. Every study also says what would make it wrong. Two examples of that from the current work. Reformation grant rates are not reported, because only 51 decisions state the outcome in a sentence that names the remedy, and a percentage built on that would look more solid than it is. The statutory staleness figure counts sections that were amended, which is not the same as sections whose amendment touched the words the court construed.

These are exploratory studies. The coded samples behind them were coded once, so they support a described pattern and not a measurement, and they say so.

Data and code

The code, the classifiers, the methodology and the executed analyses are on GitHub at stepuplaw/legal-empirics. Author identity is ORCID 0009-0002-1385-8498.

The datasets publish as CSV with a machine readable schema, a checksum and a row count on every file. The opinion studies are not posted here yet. They go up with the first archived release, and a download link that points at a file which does not exist is worse than no link.

One reference dataset is already published, because it does not come from the opinion corpus and did not have to wait. Florida local codes of ordinances records where each of the state's 67 counties and 411 municipalities publishes its code, who publishes it, and when it was last codified, including the jurisdictions that publish nothing online at all.

Text and derived data are licensed CC BY 4.0. This is empirical research about published court opinions. It is not legal advice and reading it creates no attorney client relationship.