MVA case value calculator.

What moves an MVA case value — at intake and through the file

Motor vehicle accident cases are the most predictable category in plaintiff personal injury. The severity gradient is well-defined, medical histories are well-documented, and jurisdiction-level verdict data is dense. That predictability is the reason the MVA fold of the Predict model holds 88–92% median accuracy (MdAPE) on higher-value cases against held-out cases — and the reason the confidence bands on MVA predictions run tight (±8–14% at the moderate severity tier).

A handful of inputs frame the number at the intake call. Injury severity and the direct damages do most of the work; jurisdiction tunes the rest:

• Injury severity. The single largest driver of the number. Soft-tissue cases (no treatment past 4 weeks) settle at small multiples of medical specials. Moderate cervical or lumbar strain — the most common MVA injury — typically settles at 3–4× multiples. Severe injuries (fracture, surgery, ongoing therapy) move to 6–7× multiples. Catastrophic injuries are jurisdiction-dependent and often bracket-bound by policy limits.
• Jurisdiction. A major swing from one state to the next — after the injury and the damages themselves. The same moderate cervical strain from a low-speed rear-end with the same medical specials can carry a materially different number from one state and county to the next. Same fact pattern, different number — because jury composition and local verdict history vary widely. The model is tuned per state and county, which is why a flat multiplier misses on real cases.
• Gross value, no liability haircut. Predict returns a gross value — what the case is worth on the merits, before any comparative-fault reduction. The model does not apply a comparative-fault discount, so the comparative-fault call stays yours to apply, not the model's.
• Medical specials. The anchor. The "multiple of meds" heuristic is a reasonable first approximation but the multiplier varies by 30–50% across jurisdictions at the same severity tier — which is why a static multiplier-based spreadsheet misses on real cases.
• Property damage. A weak signal on case value but a useful sanity check on severity claims. A $200 PD claim with $40K in chiropractic specials reads as low-credibility to a jury — the model down-weights cases where PD and meds diverge sharply.
• Plaintiff profile and vehicle type. Secondary but real signals — plaintiff age and gender, and whether a commercial vehicle was involved — shift the number at the margin. A commercial-vehicle defendant changes both the exposure and the policy picture.

The full per-state breakdown — comparable verdict density and how the model tunes per jurisdiction — lives on the state-by-state calculator hub.

Why the bands are tight on MVA

Confidence bands on MVA predictions run notably tighter than on premises liability or other case types. Three reasons:

• Verdict density. The MVA fold is the larger, denser of the two folds in the corpus — premises liability is the thinner cohort. More data, tighter folds, narrower bands.
• Standardized case anatomy. MVA cases share a common structural skeleton — injury severity, direct damages, property damage, jurisdiction. The features are well-defined and well-populated. Premises cases vary more in the underlying fact pattern (slip-and-fall versus animal attack versus inadequate security) and the model accordingly carries more uncertainty.
• Comparative-fault stability. MVA liability is binary more often than premises liability — clear or contested, but rarely the long-tail of comparative-fault exposure that drives the bands wide in PL.

The Accurately Predicting MVA Cases post documents the per-jurisdiction MdAPE calibration in detail — the methodology behind the held-out test results that produced the 88–92% median-accuracy figure on higher-value MVA cases.

What the model does not do

The free version on this page runs the demo model — the same architecture as the full Predict model, but with fewer secondary signals. It produces a real number on real inputs, which is the point. What it does not do:

• It does not cite the comparable-verdict cohort for your specific case. The full model surfaces 5–10 cited comps from the training set; the demo model returns the headline number with the density indicator only.
• It does not factor in plaintiff-counsel reputation, judge composition, or per-county settlement medians. Those signals require a case loaded into the platform.
• It does not produce a demand letter. The full Predict subscription drafts a complete demand letter in your firm's own style and formatting, built on the prediction.

Those features run inside the 14-day free trial. $0 today — card on file, cancel anytime; nothing is charged until day 15.