An honest measure of where the public stands on criminal justice — for every state.
Second Look Alliance is a non-partisan 501(c)(3) dedicated to criminal justice, and built around public-opinion research. We field rigorous, even-handed surveys and apply advanced statistical modeling to measure where voters actually stand on criminal-justice policy, state by state — and we partner with organizations working on these issues to put those findings to use.
How we work
An even read, first
Most issue polling is built to produce a number that looks good in a press release — the opposite of what a decision-maker needs. We start with an even read: questions worded to find where opinion truly sits, not to inflate it. An honest baseline is the only thing you can compare across states, communities, and time. Message testing then builds on that baseline, identifying which arguments are associated with stronger support, and for whom.
Every state, one model
We field in a deliberately chosen mix of states, then use a hierarchical Bayesian model — multilevel regression with poststratification (MrP) — to learn how opinion on criminal justice varies by demographics and local context. Those patterns project onto any state’s Census-measured composition, including states no one has surveyed. Pooling is within-wave only: each state’s model is fit from that wave’s own respondents, never blended with another state’s data.
Built to be acted on
Organizations working on criminal justice use our research to see, across many states at once, where opinion is already favorable, where it is genuinely contested, and where the ground is not yet ready — so limited resources go where they matter most, and so they know which arguments resonate, and with which communities.
Those figures come from leave-one-state-out validation: each surveyed state was removed from training and predicted from the remaining five using only U.S. Census data about its demographic composition — a full rehearsal of the unsurveyed-state configuration. Source: Cross-State Estimation Validation Memo (July 2026) · read the full validation, including its limitations. Every modeled estimate is labeled as modeled, carries its 95% interval, and describes opinion as of its field period — a snapshot, not a forecast. A state’s first direct survey supersedes its modeled estimates.
The discipline behind the numbers
Association, never causation
We report which arguments are associated with stronger support — never that a message moves opinion. Correlation is what the data support, so correlation is what we state.
No claim rests on noise
We report no subgroup finding on fewer than 100 respondents. Estimates that cannot pass the model’s convergence gates are discarded and logged, never patched over.
Every number declares itself
Every estimate reflects opinion as of the period it was fielded — a snapshot for decision-making, never a forecast — and every number we publish declares its source and method. These rules are enforced in our systems, not just promised in a mission statement.
Our board
Preston Robinson
Co-Founder & Executive Director. Preston leads Second Look Alliance and its public-opinion research program, directing the survey and modeling work behind the platform. He previously served as Chief of Staff to United States Senator John Kennedy (R-La) from 2016 to 2020 and as Louisiana’s Deputy State Treasurer for Management and Finance, and holds an MBA from the Kelley School of Business at Indiana University and an MPA from Villanova University.
Edward S. Shihadeh
Professor Shihadeh joined LSU in 1992 with a Ph.D. from Penn State. He began his career as a mathematical demographer before turning to crime and the social problems of American cities, and now directs CAPER, the interdisciplinary Crime and Policy Evaluation Research group at LSU. Through a federal grant he created TIGER, a machine-learning risk-and-needs tool used by Louisiana corrections to inform evidence-based release and treatment decisions.
Hunter Lundy
Full biography forthcoming.