Policy and Uncertain Science

Every 10 years, the census officially counts our population, information used to redraw the lines of Congressional districts. What could go wrong? 

In a word, gerrymandering, a practice as old as American democracy itself. Science has been asked to define fairness, identify manipulation, and even guide legal remedies. Yet, as our tools become more sophisticated, the interpretations they yield grow more contested. A new study in the Election Law Journal explores the tension between what the data can show and what democracy ought to demand, confronting the limits of science in adjudicating value-laden disputes.

A Founding Father and Political Creature

Our story begins with Elbridge Gerry, signer of the Declaration of Independence, and one of three individuals refusing to sign the Constitution because it did not contain individual rights. Instrumental in shaping the Bill of Rights as a member of the Massachusetts Congressional delegation, he served two one-year terms as Governor of Massachusetts. While his first term was marked by moderation, as the opposition party (the Federalists) controlled the state senate, his second term, when his party controlled both legislative chambers, was more partisan and controversial, reflecting growing political polarization. Gerry oversaw a dramatic “reform” of the Massachusetts judiciary, increasing the number of judgeships and filling them with members of his party, removing Federalist officeholders and replacing them with party loyalists. In 1812, the Republican-controlled legislature created new district boundaries to enhance its control over state and national office. The oddly contorted shape of one district was mocked in a satirical cartoon that compared it to a salamander, giving rise to “Gerry-mander,” later simplified to gerrymandering.

The Limits of Legal Remedies

Redrawing Congressional lines has ebbed and flowed, from post-Construction rules turning back the victors in the Civil War to improved mapping technology, empowering the rising partisanship of the 1990s. Each side of the political aisle has contested the gerrymandering of the opposition, taking their concerns to Court. Chief Justice Roberts, writing for a 2019 majority, concluded: 

“Excessive partisanship in districting leads to results that reasonably seem unjust. But the fact that such gerrymandering is ‘incompatible with democratic principles,’ …does not mean that the solution lies with the federal judiciary. We conclude that partisan gerrymandering claims present political questions beyond the reach of the federal courts.”

If the courts cannot supply an answer, can science do it? Spoiler alert for the tl:dr, it's complicated. 

Defining “Unfair”: Can Data Deliver?

Black’s Law Dictionary defines gerrymandering as,

“[t]he practice of dividing a geographical area into electoral districts, often of highly irregular shape, to give a political party an unfair advantage by diluting the opposition’s voting strength” 

Of course, unfairness is often in the mind of the beholder. 

The research analyzed district-level redistricting in 37 states after the 2020 census, focusing on federal elections for presidential elections in 2016 and 2020 and Senatorial races in 2016, 2018, and 2020. They utilized a set of 5,000 maps representing “neutral, counter-factual redistricting scenarios for each state,” based on maintaining the status quo of the number of seats controlled by each party. [1] The researchers considered five methods proposed by scholars to assess gerrymandering, defining an unfair “shortfall” as a roughly 7% unaccounted change in districting

Efficiency Gap evaluates fairness by calculating the difference in “wasted votes,” those cast for losing candidates or more than needed to win votes between parties. While providing a numerical signal of value partisan bias, it tends to overstate unfairness when one party’s voters are naturally clustered, e.g., Democrats in urban areas or Republicans in rural areas.

Partisan Symmetry assesses whether each party would receive the same number of seats for their percentage of the statewide vote. While it captures the principle of equal treatment in electoral outcomes, it relies heavily on hypothetical scenarios that may not reflect actual voting behavior. It is less reliable in noncompetitive states where one party dominates. 

Median Minus Mean compares the median vote share of districts to the mean, detecting an unequal weighting or skewing in the vote distribution. It is a simple statistical test aligned with “majority rules,” flagging imbalances caused by population geography rather than deliberate gerrymandering. It is not as meaningful in highly and noncompetitive states.

Observed vs. Expected Wins uses computer-generated, neutral maps to estimate outcomes without partisan interference compared to those enacted. It allows for a robust comparison grounded in realistic alternatives; however, it is sensitive to natural residential patterns, reflecting geographic concentrations rather than intentional manipulation.

Declination measures asymmetry in the distribution of vote shares, identifying packing, consolidating populations into smaller districts to lessen their impact, and cracking, spreading the population across many districts, again reducing their impact. It signals potential partisan advantage but can oversimplify complex voting dynamics or miss context-specific nuance.

The Tools Disagree

Of the 37 States, four consistently demonstrated gerrymandering; twelve did not. Let’s dive into the uncertainty surrounding the other 50%.

Generally, each method offers a partial lens into gerrymandering. Individually, none of the methodologies produced consistent results; combining them improved convergence, but conflicting results continued to reflect uncertainty. Their biggest shared weakness is misinterpreting geographic clustering (especially of Democrats in cities) as intentional bias. The researchers suggest that observed vs. expected offers the greatest focus on identifying fairness.

The clear cases of gerrymandering were in states “where line-drawing authorities have both opportunity and motive to produce them” – three Republican and one Democrat-controlled legislatures. Reformers have been quick to suggest nonpartisan commissions or the Courts enact districting plans, but the researchers found none of those plans consistently avoided gerrymandering by their five assessments. As it turns out, unfairness is not only in the eye of the beholder but the application of “science” does not reduce uncertainty.  

“You have to identify the values you want to preserve, and then a diagnostic will follow. Those values cannot be so vague as ‘fairness.’”  

- Michael McDonald, Professor of Political Science, SUNY-Binghamton

The debate over which values are essential and which can be sidelined remains scientifically unresolved. What is “fair” will not yield a scientific answer – it is inherently political. 

A Toolkit for Thinking Through Uncertainty

Navigating the space between contested scientific evidence and actionable policy demands a framework that acknowledges uncertainty without being paralyzed. Science can inform, not dictate, values-based decisions, especially when those values remain politically charged and context-dependent. Still, in the face of such ambiguity, we are not without tools. Rather than searching for a definitive answer, decision-makers can adopt a set of heuristics that promote clarity, accountability, and responsiveness. These guidelines do not resolve uncertainty but help us reason through it:

1. Name the beast – Define the problem clearly. As the authors argue, there's no guiding star without clarity on which values or outcomes matter most. Terms like healthy, nutritious, or fair are often as ambiguous as the problems they aim to solve.
2. Not all studies are created equal – Prioritize methodological quality, plausibility, replication, and freedom from conflicts of interest. It is critical when uncertain to acknowledge what’s unknown while cautiously weighing emerging evidence.
3. Start with a belief, then let evidence chisel it – Be a Bayesian. Begin with a prior belief based on logic, experience, or analogy and update it as new evidence emerges. Bayesian thinking is humble: asking not “What is the truth?” but “What does the weight of the evidence suggest?”
4. Embrace Expected Value Thinking –Ask, “What’s the impact if it happens?” Beware the Black Swan; low-probability events deserve attention if the consequences are significant.
5. Risk loves context – Uncertainty comes in many forms: statistical, systemic, or incomplete knowledge. Know which kind you're facing. Let the context be your guide. Use caution when harms are irreversible. Act boldly when risks are low and potential gains are high.
6. Wisdom is plural; seek many voices –Good decisions benefit from many perspectives bring all the stakeholders to the table.
7. A good decision is one you’re willing to revisit – Science evolves—and so should your decisions. Favor reversible choices. Clearly state what you know, what you assume, and what you don’t. That makes your decision auditable by others and your future self.
8. Certainty is rare; clarity is earned – The goal isn’t to eliminate doubt but to make the best possible decision in its presence. And to do that time after time.

Ultimately, science can sharpen our tools but cannot settle our values. The allure of objectivity in a problem as politicized as redistricting offers the illusion of fairness, but behind every statistical model lies a set of human choices—what to measure, what to ignore, what to call fair. As the researchers conclude, even our best methodologies can only go so far when the terrain is as much ethical as empirical. The challenge isn’t finding the perfect algorithm; it’s defining the principles we want it to reflect. 

[1] A computer algorithm generates a randomized set of electoral maps, reflecting a wide range of configurations based solely on geography and population without partisan intent. 

Source: Assessing Gerrymandering after the 2020 Census Election Law Journal DOI: 10.1089/elj.2024.0004