Utah Appellate Court
Analytics, 1997–2025
A statistical analysis of every published opinion from the Utah Court of Appeals and Utah Supreme Court spanning 29 years — examining reversal trends, legal topic risk, and standards of review.
| Standard of review | S1 · Cert, CoA affirmed | S2 · Cert, CoA reversed | S3 · Direct appeal |
|---|
This analysis covers 7,674 published opinions from the Utah Court of Appeals (n = 5,412) and Utah Supreme Court (n = 2,262), spanning January 1997 through December 2025. The dataset was compiled from official Utah court opinion archives and includes every published, precedential opinion issued during the study period. Unpublished memorandum decisions are excluded, as they do not carry precedential weight and are not subject to the same review standards.
Each record includes: case name, citation, court, decision date, outcome, standard of review, holding, legal topics, and practice area. The raw outcome field contained over 80 distinct values; these were normalized into seven macro-categories using rule-based text classification.
Raw outcome strings were mapped to seven canonical categories using the following priority-ordered rules:
| Category | Criteria | n | % |
|---|---|---|---|
| Affirmed | Contains “affirmed” but not mixed with reversal or remand language | 4,617 | 60.2% |
| Reversed | Contains “reversed” without “remanded” or partial-affirmance language | 1,587 | 20.7% |
| Mixed | Contains both “affirmed in part” and “reversed” or “remanded in part” — partial wins and losses | 763 | 9.9% |
| Dismissed | Contains “dismissed” without affirmance or reversal language | 354 | 4.6% |
| Remanded | Contains “remanded” without reversal language (procedural remand) | 165 | 2.2% |
| Reversed & Remanded | Contains both “reversed” and “remanded” as the primary disposition | 82 | 1.1% |
| Other / Vacated | Vacaturs, certified questions, extraordinary writs, and unclassifiable outcomes | 106 | 1.4% |
For reversal rate calculations, the binary reversal indicator is coded 1 for Reversed, Mixed, Reversed & Remanded, and Vacated outcomes. Dismissed and pure Remand cases are excluded from reversal rate denominators, as they do not represent merits determinations on the same terms as affirmed or reversed outcomes.
The reversal rate for any group is defined as:
Confidence intervals are computed using the Wilson score interval, which is preferred over the normal approximation (Wald interval) for proportions, especially when rates approach 0 or 1 or sample sizes are moderate. The Wilson interval for proportion p with sample size n at 95% confidence is:
where z = 1.96 for 95% confidence
To test whether reversal rates exhibit a statistically significant monotonic trend over the 29-year study period, we applied the Mann-Kendall non-parametric trend test. This test is preferred over OLS regression for time-series trend detection because it makes no distributional assumptions about the data and is robust to outliers.
The test statistic τ (Kendall’s tau) measures the correlation between time rank and outcome rank. Values near +1 indicate a consistent upward trend; values near −1 indicate a consistent downward trend. The result for the combined reversal rate series:
Rolling averages (3-year and 5-year windows) are computed as simple arithmetic means of consecutive annual observations, used to smooth year-to-year volatility and reveal underlying structural patterns.
To test whether the Court of Appeals and Supreme Court differ significantly in reversal rates, we applied a two-proportion z-test. The pooled proportion p̄ is used as the null hypothesis estimate:
CoA: p₁ = 0.303, n₁ = 5,058 | SC: p₂ = 0.462, n₂ = 2,026
z = −12.66 | p < 0.0001 | Cohen’s h = 0.32 (medium effect)
Cohen’s h is the appropriate effect size measure for the difference between two proportions: h = 2·arcsin(√p₁) − 2·arcsin(√p₂). Values of 0.2, 0.5, and 0.8 correspond to small, medium, and large effects respectively.
To test whether court identity (CoA vs. SC) and outcome category are statistically independent, we constructed a 2×7 contingency table and applied Pearson’s chi-square test:
Cramér’s V = √(χ²/n·min(r−1,c−1)) provides a normalized effect size. A value of 0.19 indicates a meaningful but not overwhelming association — court identity explains some but not all variation in outcome distributions.
Standards of review were extracted from the structured “standard_of_review” field in each record and normalized into five canonical categories using keyword matching. The hierarchy from least to most deferential is: correctness/de novo → abuse of discretion → plain error → substantial evidence → clearly erroneous. Cases applying multiple standards (common in multi-issue appeals) are classified by the primary standard governing the dispositive issue.
Legal topics were extracted from a structured taxonomy field. Because many opinions involve multiple legal issues, each opinion may be tagged with multiple topics (comma-separated). For topic-level reversal rates, the denominator is the count of opinion-topic pairings (not unique opinions), so an opinion with three tags contributes to three topic counts. Only topics with n ≥ 50 opinion-pairings are reported in the Court of Appeals and Supreme Court filtered views (n ≥ 100 for the combined view) to ensure statistical reliability.
Selection bias: Published opinions represent only a fraction of all appeals filed. Cases resolved by settlement, voluntarily dismissed, or decided in unpublished memoranda are not included. This means reversal rates reflect the population of fully litigated published appeals, not all appellate proceedings.
Ecological fallacy: Group-level reversal rates (by topic, court, or year) should not be used to predict the outcome of any individual case. Many confounding variables — quality of briefing, panel composition, specific facts — are not captured in this dataset.
Topic overlap: Because opinions carry multiple topic tags, topic-level statistics are not independent. High-reversal topics that co-occur frequently with other high-reversal topics may inflate apparent reversal risk in both.
Temporal confounding: Changes in reversal rates over time may reflect changes in docket composition, judicial philosophy, legislative activity, or litigation patterns rather than any single causal factor. The structural break around 2011 coincides with multiple potential causes and cannot be attributed to any single variable with the data available.