When attorneys evaluate motor vehicle accident (MVA) cases, one question dominates early decision-making:
“What is this case actually worth?”
Historically, the answer has relied on experience, intuition, and incomplete comparisons. With a structured dataset of plaintiff-side outcomes and modern regression methods, the question can now be answered with measurable precision — and the per-tier results are tighter than most attorneys expect.
The Predict model is trained on 198,000 MVA jury verdicts and reported settlements, structured against the case-history attributes that actually drive settlement value: liability clarity, injury severity, medical specials, property damage, plaintiff-counsel reputation, judge composition, and per-county settlement medians.
The metric that matters
To evaluate accuracy, we use Median Absolute Percentage Error (MdAPE) — expressed as Median Accuracy for readability. MdAPE shows how close predictions are in a typical case, without the distortion of a few extreme outliers that inflate or deflate mean-based error metrics.
For a long-tail damages distribution — which is what plaintiff PI is — the median is the right central-tendency statistic. Mean-based metrics over-weight the catastrophic tail and under-represent typical-case performance.
Results across higher-value MVA cases
We focus this post on higher-value MVA cases — $290K and above — where the financial stakes make precision most consequential. The per-tier breakdown:
| Case value range | Median accuracy |
|---|---|
| $290K – $645K | 87.9% |
| $645K – $1.4M | 91.9% |
| $1.4M – $5.0M | 91.3% |
The pattern is clear: median accuracy stays in a tight band between roughly 88% and 92% across the higher-value range.
What this means in practice
Strong accuracy where it matters most
Across higher-value MVA cases, the model delivers ~88% to ~92% accuracy in typical predictions. In practical terms:
- A $300,000 case is typically predicted within about $264K to $336K
- A $1,000,000 case is typically within roughly $920K to $1.08M
At higher case values, small percentage differences become meaningful dollar amounts. The tighter the band, the more defensible the settlement number — particularly in the negotiation against a carrier’s reserve.
Accuracy improves as case value increases
For MVA cases above approximately $600,000, median accuracy exceeds 90%. The reason is data density — higher-value cases tend to involve clearer severity classifications, more documented medical histories, and more frequent appearance of comparable verdicts in the training set.
This is also where accuracy matters most. Settlement negotiation at the $1M+ tier carries materially more financial exposure than at the $50K tier; the model’s tighter performance at higher values is structurally aligned with where precision is load-bearing.
Predictable performance, not best-case performance
By focusing on Median Accuracy, the model performance reported here reflects the typical case — not the ideal case, and not a cherry-picked subset. The same band of accuracy holds across thousands of held-out test cases, not a handpicked few.
What this changes for the practice
Case valuation has historically been more art than science. Calibrated against the held-out test set, the model produces consistent value predictions that:
- Surface stronger cases at intake, before retainer
- Anchor settlement expectations to evidence rather than instinct
- Reduce the cross-attorney variance in case-selection decisions
The objective: bring the same kind of pricing instrument to the plaintiff side that the carriers have used for a decade.
The bottom line
For higher-value MVA cases, Predict delivers:
- ~88% to ~92% median accuracy across the $290K–$5M range
- 90%+ accuracy for cases above ~$600K
- Stable performance across thousands of held-out test cases
For the full methodology — training data, model architecture, held-out test design, confidence-band derivation, and the per-quarter recalibration policy — see the methodology page. To run the model on a real case from your pipeline, the free calculator is open. The full Predict model — case-history citations, demand-letter-ready valuation blocks, full jurisdiction folds — runs inside the 14-day free trial. No credit card at signup.