# Markov Chain Ensemble Methods The ensemble approach is the most important methodological development in gerrymandering analysis, and it works by refusing to answer the question everyone wants answered. It does not ask "is this map fair?" It asks: **relative to the enormous population of maps that satisfy every legal criterion in this state, where does this particular map fall?** If the proposed map delivers more seats to one party than 99.9% of legally valid alternatives drawn from the same population, geography, and rules, that is evidence of selection. Not proof of intent — evidence of selection. ## How it works You cannot enumerate the space of valid maps; it is far too large (see [[The Mathematics of Redistricting]]). But you can *sample* from it, which is the classic application of Markov chain Monte Carlo (see Monte Carlo Simulation). 1. **Start** from any valid map. 2. **Propose a small random change** — reassign a precinct at a district boundary, or in the more sophisticated variants (ReCom) merge two adjacent districts and repartition them along a random spanning tree. 3. **Accept or reject** the proposal according to whether the result still satisfies contiguity, population equality, and whatever compactness and community-preservation constraints you have imposed. 4. **Repeat** millions of times, recording a map every so often. The recorded maps form the **ensemble**. Compute the partisan seat outcome for each one and you have a distribution. The proposed map's position in that distribution is the finding. ## What makes it powerful **It handles political geography automatically.** This is the decisive advantage over [[The Efficiency Gap]]. The ensemble is drawn from the *same state* with the *same population distribution* — so if urban clustering naturally disadvantages Democrats in that state, the entire ensemble reflects that, and a map is only flagged when it goes beyond what geography alone produces. The single most persistent defense of gerrymandered maps ("Democrats just live too close together") is neutralized by construction rather than by argument. **It requires no normative baseline.** You never have to assert what share of seats a party deserves. The comparison is map-to-maps, not map-to-ideal — which is exactly the objection that sank the efficiency gap. **It quantifies extremity.** "More partisan than 99.97% of 100,000 sampled valid maps" is a statement with a precise meaning, and it is robust to reasonable variation in the constraint settings. ## The honest limitations **Constraint choices drive the answer.** The ensemble depends on which criteria you enforce and how tightly. Require strict compactness and you get one distribution; prioritize county-boundary preservation and you get another. This is a real degree of freedom, and adversarial experts exploit it. Best practice is to report results across multiple constraint regimes and show that the finding survives. **Convergence is a genuine technical question.** A Markov chain must run long enough to explore the space rather than remain near its starting point. Establishing that a redistricting chain has mixed adequately is hard, and the shift from single-precinct flips to ReCom-style spanning-tree moves was driven largely by mixing problems. **It cannot establish intent.** It shows a map is an outlier. Whether that outlier arose from deliberate manipulation, from a legitimate criterion the analyst failed to encode, or from coincidence is a legal inference the mathematics does not make. ## What happened to it in court Ensemble evidence was central to the *League of Women Voters v. Commonwealth* litigation in Pennsylvania (2018), where the state supreme court struck down the congressional map under the state constitution, and it has appeared in state proceedings since. Its federal path closed with [[Rucho v. Common Cause]] (2019). The dissent argued explicitly that ensemble methods supply the judicially manageable standard the majority claimed did not exist — that the courts had been handed the tool they had been asking for and declined to pick it up. The majority's position was that identifying a map as an outlier still leaves the question of how far an outlier must be before it is unconstitutional, and that this line is a political judgment. **That disagreement is the crux, and it is not really a disagreement about mathematics.** Both sides accept the method describes the map accurately. They disagree about whether a court may set a threshold on a continuous measure — which is a question about judicial role, dressed as a question about manageability. ## Related [[The Mathematics of Redistricting]] · [[The Efficiency Gap]] · [[Packing and Cracking]] · [[District Compactness Measures]] · Monte Carlo Simulation · [[Rucho v. Common Cause]] · [[Mathematics and Power Hub]]