Utts’s statistical literacy publications span from a 1997 review of her own introductory textbook Seeing Through Statistics⁷ through a 2022 paper integrating ethics into the GAISE (Guidelines for Assessment and Instruction in Statistics Education) framework.⁵ The 2003 special section she edited in The American Statistician on statistical literacy⁸ and her companion paper on what educated citizens should know¹ are the most-cited anchors of her literacy agenda. Her 2016 ASA Presidential Address, published in JASA, synthesized these themes in the context of the profession’s rapid growth.⁴

Utts’s 2003 paper in The American Statistician identified six categories of statistical understanding that educated citizens should possess:¹ (1) When coincidences and surprising events are not surprising, understanding that rare events are common in large samples; (2) the difference between statistical significance and practical importance, that a p-value below 0.05 says nothing about effect magnitude; (3) the distinction between relative and absolute risk, that a “50% increase in risk” from a baseline of 1 in 10,000 is very different from the same relative increase from a baseline of 1 in 10; (4) the role of confounding, that observational studies cannot establish causation without ruling out alternative explanations; (5) the limits of personal experience and anecdote as evidence; and (6) the distinction between a study finding and a media headline about that finding. These six categories map directly onto the most common statistical errors Utts documented in media coverage and public discourse. The paper was published as part of a special section on statistical literacy that Utts edited for the same journal.⁸

Baldi and Utts (2015), published in The American Statistician, surveyed the content of introductory undergraduate biostatistics courses and compared it against the statistical competencies that practicing physicians actually need.³ Their analysis identified systematic gaps: courses over-emphasized classical hypothesis testing mechanics while under-emphasizing Bayesian reasoning, the interpretation of screening test results (sensitivity, specificity, positive and negative predictive value), the distinction between relative and absolute risk reduction in clinical trials, and the communication of uncertainty to patients. The paper argued that these omissions are not incidental, they reflect a curriculum designed around mathematical tractability rather than clinical relevance. No sample size or power analysis was reported for the curriculum survey itself; the paper’s methodology was a structured content review rather than an empirical study, making the provenance exploratory/analytical rather than experimental.

Utts’s 2010 paper² and her 2002 review of News and Numbers⁹ both return to the relative-versus-absolute risk distinction as the single most consequential statistical concept that the public systematically misunderstands. A drug that reduces the risk of a disease from 2% to 1% can be reported as “cutting risk in half” (relative risk reduction of 50%) or as “reducing risk by 1 percentage point” (absolute risk reduction of 1%). Both are arithmetically correct; they produce radically different impressions of clinical significance. Utts argued that statisticians bear professional responsibility for this confusion because introductory courses rarely teach the distinction explicitly, and because the statistics community has not consistently demanded that journals and press releases report both measures. The 2010 paper was a contributed piece rather than an empirical study; its methodology was argumentative and illustrative, drawing on documented examples of media misreporting.

Utts (2021) in Harvard Data Science Review⁶ argued that the rapid expansion of data science as a field has created a new version of an old problem: practitioners who apply statistical tools without understanding their assumptions, and researchers who report statistical results without engaging with their real-world implications. The essay was a commentary/opinion piece rather than an empirical study. Its central claim, that statistical practice requires active, critical engagement rather than passive method-application, connects directly to Utts’s earlier literacy agenda: a statistically literate citizen and a statistically responsible practitioner are both defined by the same capacity for critical interpretation, not by computational skill alone.

Utts (2015) in The American Statistician¹⁰ reviewed the history of statistics education from its origins through the early 21st century, identifying persistent tensions that have never been fully resolved: (1) whether introductory courses should prioritize mathematical foundations or applied reasoning; (2) whether real data or constructed examples better serve learning goals; (3) how much computing should be integrated into the curriculum and at what level; and (4) whether Bayesian methods should be introduced alongside or instead of classical frequentist approaches. The paper was a historical and analytical review, not an empirical study. Its value is in establishing that the debates Utts engaged in her own curriculum work, including the push for Bayesian literacy and the emphasis on real-data examples in Seeing Through Statistics⁷, are not novel but are the latest iteration of longstanding disciplinary tensions.

Utts and Johnson (2008) in The American Statistician¹¹ traced the evolution of approaches to teaching Bayesian statistics to students who will not become professional statisticians, including scientists, social scientists, and medical researchers. The paper identified the core pedagogical challenge: Bayesian reasoning requires students to think about probability as a degree of belief rather than a long-run frequency, a conceptual shift that many find counterintuitive but that better matches how probabilistic reasoning actually operates in clinical and scientific decision-making. The paper was a pedagogical review and analysis rather than an empirical study of student outcomes. It connects directly to Utts’s broader literacy agenda: the relative-versus-absolute risk confusion that dominates media misreporting is, at its root, a failure of Bayesian reasoning about base rates and conditional probabilities.

Raman, Utts, Cohen, and Hayat (2022) in The American Statistician⁵ proposed specific mechanisms for integrating ethical reasoning into the GAISE (Guidelines for Assessment and Instruction in Statistics Education) college report. The paper identified four domains where ethical issues arise in statistical practice: data collection and consent; algorithmic decision-making and fairness; the communication of uncertainty and risk to non-expert audiences; and the responsible use of statistical significance in research reporting. The paper was a position/proposal paper rather than an empirical study of curriculum outcomes. Its significance is in formalizing what Utts had argued informally across her career, that statistical literacy without ethical reasoning is incomplete, and that the statistics education community has a professional obligation to address both.

Utts (2016) in JASA⁴ documented the scale of the profession’s growth: the AP Statistics exam, which Utts had been involved with since its first year in 1997, grew from fewer than 8,000 takers to approximately 208,000 by 2016. Bureau of Labor Statistics projections placed statistician as the 9th fastest-growing occupation overall and 3rd among occupations requiring at least a college degree, with a projected 34% growth rate against a 7% baseline for all occupations. Bachelor’s degrees in statistics and biostatistics had increased by more than 300% since the 1990s; Master’s and PhD degrees had increased by 260% and 132% respectively between 2000 and 2014. The address used these figures to argue that the statistics community faces a choice: it can allow this growth to proceed without deliberate attention to what is being taught and how, or it can use the moment of expansion to embed statistical literacy, including the conceptual tools Utts had been advocating since 2003, into the foundations of the discipline’s growth. The address was a presidential speech subsequently published as a JASA article; it was not an empirical study.

Utts and Goretsky (2021) in Leadership in Statistics and Data Science¹² argued that the statistics community’s influence on public discourse, policy, and science is limited not by the quality of its methods but by the scarcity of statisticians who actively engage with non-specialist audiences. The chapter proposed a framework for developing thought leadership, including writing for general audiences, engaging with journalists, participating in policy processes, and teaching in ways that foreground real-world application. The chapter was a contributed book chapter in an edited volume; it was not an empirical study. Its argument connects directly to Utts’s 2010 paper on media misrepresentation²: the same professional failure that produces statistical misreporting, statisticians who do not engage with journalists and policymakers, is what thought leadership is designed to address. What would settle the question of whether thought-leadership training actually improves public statistical literacy is a preregistered, longitudinal study tracking the downstream effects of statistician-journalist engagement programs on the accuracy of statistical reporting, a study that, to Utts’s knowledge, has not yet been conducted.