Expected value is the probability-weighted average of every way a case can end. It answers a narrow but crucial question: if you could run this lawsuit a thousand times, what would the average net outcome be? That number gives you a rational anchor for the settle-or-fight decision, a way to test whether an offer on the table is generous or insulting relative to the risk you are carrying.

But expected value is a starting point, not a verdict. It assumes you are risk-neutral, it hides the spread of outcomes, and it is only as trustworthy as the probabilities you feed it. This guide computes it cleanly and then shows the adjustments a careful litigant makes before acting on it. It is an educational explainer, not legal advice.

In its simplest form, the expected value of pursuing a claim is the probability of winning times the net recovery on a win, plus the probability of losing times the net cost of a loss. Because a loss still incurs legal fees and possibly cost-shifting, the losing branch is usually a negative number, not zero. Netting costs into both branches is what makes the figure honest.

Each term hides a judgment call. The win probability compresses every element and procedural gate into one number. The recovery term assumes you have modeled damages realistically. Writing the formula out forces those assumptions into the open, where they can be argued and revised instead of silently driving the decision.

Consider a hypothetical plaintiff weighing a claim. Suppose a win yields a net recovery after costs of a given amount, the odds of winning are about forty percent, and a loss costs the plaintiff their own fees. The expected value is forty percent of the net win figure, minus sixty percent of the loss cost. If that result is positive and exceeds the current settlement offer, continuing has higher expected value than settling; if it is lower, the offer is the better bet.

The numbers here are illustrative, not data. The point is the shape of the reasoning: two branches, each weighted by its probability, each net of cost, summed to a single comparison figure. Change any input and the recommendation can flip, which is exactly why you never treat the first calculation as final.

Two cases can share an identical expected value and still call for opposite decisions. A case with a positive expected value built on a small chance of a huge award is very different from one built on a large chance of a modest award. A plaintiff who cannot survive the downside should weight the variance, not just the average, and may rationally accept a below-EV offer to eliminate ruin risk.

This is the difference between a risk-neutral actor, who follows expected value mechanically, and a risk-averse one, who pays a premium for certainty. Corporate defendants with deep pockets often behave closer to risk-neutral; an individual plaintiff rarely can. Neither is wrong; they are optimizing different things, and a good analysis surfaces the spread so the client can decide with eyes open.

The weakest link in any expected-value calculation is the win probability. Build it from the ground up: estimate the chance of surviving each dispositive motion and carrying each element, then combine them. Ground each estimate in admissible evidence and the applicable burden of proof, not optimism. A candid thirty-five percent beats a comfortable sixty percent that cannot be defended.

Then treat the estimate as provisional. Bayesian updating is the formal name for revising a probability as new evidence arrives: a strong deposition nudges it up, an adverse document nudges it down. Re-running the expected-value calculation after each material development keeps the settle-or-fight decision anchored to what you actually know now, not what you assumed at intake.

A single expected-value number averages away all the texture that actually drives risk. The richer move is to simulate the full distribution of outcomes: sample across plausible ranges for damages and each probability, run many trials, and read off not just the mean but the tails. This shows the chance of a catastrophic loss and the chance of a windfall, which the point estimate silently buries.

Juricratic runs exactly this as a seeded Monte Carlo simulation over a model of the matter. You set your reads as dials, the engine samples thousands of reproducible trajectories, and you see how the expected value and its spread respond as you turn each dial. The same inputs always produce the same distribution, so the number is auditable rather than a black box.