Searcharxiv⌕ Search

arXiv subjects

Sander Greenland

Publications and source records attributed to Sander Greenland.

13 recordsLinked to original sources
arXiv2025

Statistical methods: Basic concepts, interpretations, and cautions

The study of associations and their causal explanations is a central research activity whose methodology varies tremendously across fields. Even within specialized subfields, comparisons across textbooks and journals reveals that the basics are subject to considerable variation and controversy. This variation is often obscured by the singular viewpoints presented within textbooks and journal guidelines, which may be deceptively written as if the norms they adopt are unchallenged. Furthermore, human limitations and the vastness within fields imply that no one can have expertise across all subfields and that interpretations will be severely constrained by the limitations of studies of human populations. The present chapter outlines an approach to statistical methods that attempts to recognize these problems from the start, rather than assume they are absent as in the claims of 'statistical significance' and 'confidence' ordinarily attached to statistical tests and interval estimates. It does so by grounding models and statistics in data description, and treating inferences from them as speculations based on assumptions that cannot be fully validated or checked using the analysis data.

stat.ME↗
arXiv2023

Divergence vs. Decision P-values: A Distinction Worth Making in Theory and Keeping in Practice

There are two distinct definitions of 'P-value' for evaluating a proposed hypothesis or model for the process generating an observed dataset. The original definition starts with a measure of the divergence of the dataset from what was expected under the model, such as a sum of squares or a deviance statistic. A P-value is then the ordinal location of the measure in a reference distribution computed from the model and the data, and is treated as a unit-scaled index of compatibility between the data and the model. In the other definition, a P-value is a random variable on the unit interval whose realizations can be compared to a cutoff alpha to generate a decision rule with known error rates under the model and specific alternatives. It is commonly assumed that realizations of such decision P-values always correspond to divergence P-values. But this need not be so: Decision P-values can violate intuitive single-sample coherence criteria where divergence P-values do not. It is thus argued that divergence and decision P-values should be carefully distinguished in teaching, and that divergence P-values are the relevant choice when the analysis goal is to summarize evidence rather than implement a decision rule.

stat.OT↗
arXiv2023

Connecting Simple and Precise P-values to Complex and Ambiguous Realities

Mathematics is a limited component of solutions to real-world problems, as it expresses only what is expected to be true if all our assumptions are correct, including implicit assumptions that are omnipresent and often incorrect. Statistical methods are rife with implicit assumptions whose violation can be life-threatening when results from them are used to set policy. Among them are that there is human equipoise or unbiasedness in data generation, management, analysis, and reporting. These assumptions correspond to levels of cooperation, competence, neutrality, and integrity that are absent more often than we would like to believe. Given this harsh reality, we should ask what meaning, if any, we can assign to the P-values, 'statistical significance' declarations, 'confidence' intervals, and posterior probabilities that are used to decide what and how to present (or spin) discussions of analyzed data. By themselves, P-values and CI do not test any hypothesis, nor do they measure the significance of results or the confidence we should have in them. The sense otherwise is an ongoing cultural error perpetuated by large segments of the statistical and research community via misleading terminology. So-called 'inferential' statistics can only become contextually interpretable when derived explicitly from causal stories about the real data generator (such as randomization), and can only become reliable when those stories are based on valid and public documentation of the physical mechanisms that generated the data. Absent these assurances, traditional interpretations of statistical results become pernicious fictions that need to be replaced by far more circumspect descriptions of data and model relations.

stat.ME↗
arXiv2022

To Aid Scientific Inference, Emphasize Unconditional Compatibility Descriptions of Statistics

All scientific interpretations of statistical outputs depend on background (auxiliary) assumptions that are rarely delineated or explicitly interrogated. These include not only the usual modeling assumptions, but also deeper assumptions about the data-generating mechanism that are implicit in conventional statistical interpretations yet are unrealistic in most health, medical and social research. We provide arguments and methods for reinterpreting statistics such as P-values and interval estimates in unconditional terms, which describe compatibility of observations with an entire set of underlying assumptions, rather than with a narrow target hypothesis conditional on the assumptions. Emphasizing unconditional interpretations helps avoid overconfident and misleading inferences in light of uncertainties about the assumptions used to arrive at the statistical results. These include not only mathematical assumptions, but also those about absence of systematic errors, protocol violations, and data corruption. Unconditional descriptions introduce assumption uncertainty directly into the primary statistical interpretations of results, rather than leaving it for the discussion of limitations after presentation of conditional interpretations. The unconditional approach does not entail different methods or calculations, only different interpretation of the usual results. We view use of unconditional description as a vital component of effective statistical training and presentation. By interpreting statistical outputs in unconditional terms, researchers can avoid making overconfident statements based on statistical outputs. Instead, reports should emphasize the compatibility of results with a range of plausible explanations, including assumption violations.

stat.ME↗
arXiv2021

Odds Ratios are far from "portable": A call to use realistic models for effect variation in meta-analysis

Objective: Recently Doi et al. argued that risk ratios should be replaced with odds ratios in clinical research. We disagreed, and empirically documented the lack of portability of odds ratios, while Doi et al. defended their position. In this response we highlight important errors in their position. Study Design and Setting: We counter Doi et al.'s arguments by further examining the correlations of odds ratios, and risk ratios, with baseline risks in 20,198 meta-analyses from the Cochrane Database of Systematic Reviews. Results: Doi et al.'s claim that odds ratios are portable is invalid because 1) their reasoning is circular: they assume a model under which the odds ratio is constant and show that under such a model the odds ratio is portable; 2) the method they advocate to convert odds ratios to risk ratios is biased; 3) their empirical example is readily-refuted by counter-examples of meta-analyses in which the risk ratio is portable but the odds ratio isn't; and 4) they fail to consider the causal determinants of meta-analytic inclusion criteria: Doi et al. mistakenly claim that variation in odds ratios with different baseline risks in meta-analyses is due to collider bias. Empirical comparison between the correlations of odds ratios, and risk ratios, with baseline risks show that the portability of odds ratios and risk ratios varies across settings. Conclusion: The suggestion to replace risk ratios with odds ratios is based on circular reasoning and a confusion of mathematical and empirical results. It is especially misleading for meta-analyses and clinical guidance. Neither the odds ratio nor the risk ratio is universally portable. To address this lack of portability, we reinforce our suggestion to report variation in effect measures conditioning on modifying factors such as baseline risk; understanding such variation is essential to patient-centered practice.

stat.AP↗
arXiv2021

There are natural scores: Full comment on Shafer, "Testing by betting: A strategy for statistical and scientific communication"

Shafer (2021) offers a betting perspective on statistical testing which may be useful for foundational debates, given that disputes over such testing continue to be intense. To be helpful for researchers, however, this perspective will need more elaboration using real examples in which (a) the betting score has a justification and interpretation in terms of study goals that distinguishes it from the uncountable mathematical possibilities, and (b) the assumptions in the sampling model are uncertain. On justification, Shafer says 'No one has made a convincing case for any particular choice' of a score derived from a P-value and then states that 'the choice is fundamentally arbitrary'. Yet some (but not most) scores can be motivated by study goals (e.g., information measurement; decision making). The one I have seen repeatedly in information statistics and data mining is the surprisal, logworth or S-value s = -log(p), where the log base determines the scale. The present comment explains the rationale for this choice.

math.ST↗
arXiv2020

The causal foundations of applied probability and statistics

Statistical science (as opposed to mathematical statistics) involves far more than probability theory, for it requires realistic causal models of data generators - even for purely descriptive goals. Statistical decision theory requires more causality: Rational decisions are actions taken to minimize costs while maximizing benefits, and thus require explication of causes of loss and gain. Competent statistical practice thus integrates logic, context, and probability into scientific inference and decision using narratives filled with causality. This reality was seen and accounted for intuitively by the founders of modern statistics, but was not well recognized in the ensuing statistical theory (which focused instead on the causally inert properties of probability measures). Nonetheless, both statistical foundations and basic statistics can and should be taught using formal causal models. The causal view of statistical science fits within a broader information-processing framework which illuminates and unifies frequentist, Bayesian, and related probability-based foundations of statistics. Causality theory can thus be seen as a key component connecting computation to contextual information, not extra-statistical but instead essential for sound statistical training and applications.

stat.OT↗
arXiv2020

Technical Issues in the Interpretation of S-values and Their Relation to Other Information Measures

An extended technical discussion of $S$-values and unconditional information can be found in Greenland, 2019. Here we briefly cover several technical topics mentioned in our main paper, Rafi & Greenland, 2020: Different units for (scaling of) the $S$-value besides base-2 logs (bits); the importance of uniformity (validity) of the $P$-value for interpretation of the $S$-value; and the relation of the $S$-value to other measures of statistical information about a test hypothesis or model.

stat.ME↗
arXiv2020

Semantic and Cognitive Tools to Aid Statistical Science: Replace Confidence and Significance by Compatibility and Surprise

Researchers often misinterpret and misrepresent statistical outputs. This abuse has led to a large literature on modification or replacement of testing thresholds and $P$-values with confidence intervals, Bayes factors, and other devices. Because the core problems appear cognitive rather than statistical, we review simple aids to statistical interpretations. These aids emphasize logical and information concepts over probability, and thus may be more robust to common misinterpretations than are traditional descriptions. We use the Shannon transform of the $P$-value $p$, also known as the binary surprisal or $S$-value $s=-\log_{2}(p)$, to measure the information supplied by the testing procedure, and to help calibrate intuitions against simple physical experiments like coin tossing. We also use tables or graphs of test statistics for alternative hypotheses, and interval estimates for different percentile levels, to thwart fallacies arising from arbitrary dichotomies. Finally, we reinterpret $P$-values and interval estimates in unconditional terms, which describe compatibility of data with the entire set of analysis assumptions. We illustrate these methods with a reanalysis of data from an existing record-based cohort study. In line with other recent recommendations, we advise that teaching materials and research reports discuss $P$-values as measures of compatibility rather than significance, compute $P$-values for alternative hypotheses whenever they are computed for null hypotheses, and interpret interval estimates as showing values of high compatibility with data, rather than regions of confidence. Our recommendations emphasize cognitive devices for displaying the compatibility of the observed data with various hypotheses of interest, rather than focusing on single hypothesis tests or interval estimates. We believe these simple reforms are well worth the minor effort they require.

stat.ME↗
arXiv2010

Interval Estimation for Messy Observational Data

We review some aspects of Bayesian and frequentist interval estimation, focusing first on their relative strengths and weaknesses when used in "clean" or "textbook" contexts. We then turn attention to observational-data situations which are "messy," where modeling that acknowledges the limitations of study design and data collection leads to nonidentifiability. We argue, via a series of examples, that Bayesian interval estimation is an attractive way to proceed in this context even for frequentists, because it can be supplied with a diagnostic in the form of a calibration-sensitivity simulation analysis. We illustrate the basis for this approach in a series of theoretical considerations, simulations and an application to a study of silica exposure and lung cancer.

stat.ME↗
arXiv2010

Relaxation Penalties and Priors for Plausible Modeling of Nonidentified Bias Sources

In designed experiments and surveys, known laws or design feat ures provide checks on the most relevant aspects of a model and identify the target parameters. In contrast, in most observational studies in the health and social sciences, the primary study data do not identify and may not even bound target parameters. Discrepancies between target and analogous identified parameters (biases) are then of paramount concern, which forces a major shift in modeling strategies. Conventional approaches are based on conditional testing of equality constraints, which correspond to implausible point-mass priors. When these constraints are not identified by available data, however, no such testing is possible. In response, implausible constraints can be relaxed into penalty functions derived from plausible prior distributions. The resulting models can be fit within familiar full or partial likelihood frameworks. The absence of identification renders all analyses part of a sensitivity analysis. In this view, results from single models are merely examples of what might be plausibly inferred. Nonetheless, just one plausible inference may suffice to demonstrate inherent limitations of the data. Points are illustrated with misclassified data from a study of sudden infant death syndrome. Extensions to confounding, selection bias and more complex data structures are outlined.

stat.ME↗