statistics

Can algorithms replace expert knowledge for causal inference? A case study on novice use of causal discovery

  • Objective: To test whether causal discovery algorithms used by novices could replace expert knowledge in selecting covariates for estimating the known null effect of placebo adherence on mortality in the CDP trial.
  • Key Finding: Adjustment sets derived from causal discovery algorithms resulted in more residual bias in complex time-varying analyses compared to expert-selected sets.
  • Conclusion: Due to high subjectivity in selecting algorithms/parameters and resolving complex graph outputs, the authors do not recommend novice use of causal discovery without an expert’s guidance.
23 January 2026

The Abuse of Power: The Pervasive Fallacy of Power Calculations for Data Analysis

  • Core Argument: This paper argues that retrospective power calculations (or observed power), performed after a study yields a non-significant result, are fundamentally flawed and should not be used to aid in interpretation.
  • The Flaw (Circularity): Retrospective power is a simple monotonic transformation of the p-value; it provides no new information. A non-significant test will always have low retrospective power, as both measures are driven by the large variance or small observed effect.
  • Recommended Alternative: To interpret a non-significant finding, researchers should focus on the confidence interval, which reveals the range of plausible true effect sizes and indicates whether biologically or clinically important effects have been reasonably ruled out.
23 January 2026

Statistical tests, P values, confidence intervals, and power: a guide to misinterpretations

  • Core Principle: This influential essay details the pervasive misinterpretations and abuse of fundamental statistics—P-values, Confidence Intervals (CIs), and statistical power—in scientific literature.
  • P-value Clarification: The authors correct 12 common misconceptions, asserting that the P-value is not the probability of the null hypothesis (\(H_0\)) being true, but rather the probability of the observed data (or more extreme) given that \(H_0\) is true.
  • Misuse of Significance: The essay strongly condemns the practice of drawing dichotomous conclusions based on an arbitrary threshold (e.g., P<0.05), which falsely implies certainty and hinders scientific progress.
  • Recommendation: Researchers are urged to shift focus from the P-value to the effect magnitude and its precision, which are better conveyed through Confidence Intervals, and to integrate statistical results with contextual, external evidence.
23 January 2026

Triangulation in aetiological epidemiology: Approaches to causal inference

  • Core Principle: This article formally advocates for Triangulation in aetiological epidemiology, defined as the integration of results from multiple research approaches where each approach possesses uncorrelated sources of bias.
  • Causal Inference: If different methods (e.g., observational studies, Mendelian Randomization, and clinical trials) all converge on the same causal conclusion, confidence in that finding is significantly strengthened, especially when method-specific biases would predict opposing results.
  • Addressing Inconsistency: When results are inconsistent, the framework helps identify which source of bias (e.g., confounding, reverse causation) is most likely distorting a given approach, thereby guiding the direction of future, more focused research.
23 January 2026

Causal Inference Is Not Just a Statistics Problem

  • Core Argument: This article argues that causal inference relies primarily on study design and domain knowledge to establish untestable assumptions (like exchangeability and no unmeasured confounding), with statistical methods serving only to quantify the effect under those assumptions.
  • Design Trumps Analysis: The authors emphasize that a causal estimate is invalid if the study design fails to account for critical (especially unmeasured) confounders, regardless of the sophistication of the statistical modeling used.
  • The Role of Causal Diagrams: Identifying which variables to control is a causal question, not a statistical one, necessitating the use of Directed Acyclic Graphs (DAGs) and scientific expertise to avoid introducing bias by controlling for mediators or colliders.
  • Conclusion: Teaching and practicing causal inference requires prioritizing the formulation of a causal question and the design of the study before the application of statistical tools.
23 January 2026

Spurious interaction as a result of categorization

  • Core Argument: Categorizing continuous exposure variables in regression models, a common practice in epidemiology, can introduce spurious interaction effects that do not exist in the original continuous scale.
  • Mechanism: This phenomenon is demonstrated analytically and is interpreted as a form of differential measurement error or residual confounding caused by the categorization process.
  • Conclusion: Spurious interaction is induced unless the correlated variables are dichotomized at the median or are uncorrelated. The paper strongly recommends avoiding categorization to ensure valid results, suggesting non-parametric models as a preferred alternative.
23 January 2026

Large-scale composite hypothesis testing procedure for omics data analyses

  • Novel method (qch_copula) for testing composite hypotheses across multiple traits/omics levels using mixture models with copula functions
  • Controls Type I error effectively while achieving superior detection power compared to 8 state-of-the-art methods across diverse correlation scenarios
  • Memory-efficient implementation enables analysis of 105-106 markers with up to 20 traits, validated through psychiatric disorder pleiotropy and plant virus resistance studies
23 January 2026

Comparison of Imputation Strategies for Incomplete Longitudinal Data in Life-Course Epidemiology

  • Topic: A study comparing the performance of three Multiple Imputation (MI) methods for handling incomplete longitudinal data in life-course epidemiology, focusing on the effect of longitudinal depressive symptoms on mortality.
  • Methods Compared: Normal Linear Regression (MICE), Predictive Mean Matching (PMM), and Variable-Tailored Specification were tested under nine scenarios varying missingness rate and mechanism (MCAR, MAR, MNAR).
  • Finding: All methods showed similar levels of bias in estimating the causal effect (Hazard Ratios), but Predictive Mean Matching (PMM) was identified as the most appealing strategy due to its consistently low root mean square error (RMSE) and competitive computation time.
23 January 2026

The cost of dichotomising continuous variables

  • Avoidable Costs: The practice of dichotomizing continuous variables for statistical analysis leads to a substantial loss of statistical power and an inflation of the Type I error rate (false positives) in multivariable models.
  • Flawed Interpretation: Dichotomization uses arbitrary or unstable cut-points, provides misleading effect estimates, and ignores the true shape of the relationship across the continuous scale.
  • Recommendation: Continuous variables should be analyzed on their original scale (or using methods like fractional polynomials/splines) to preserve information and avoid these serious statistical drawbacks.
23 January 2026

TESTING FOR BASELINE BALANCE IN CLINICAL TRIALS

  • Core Principle: This paper argues that performing tests of baseline homogeneity (p-value tests) in randomized controlled trials (RCTs) is philosophically unsound, of no practical value, and potentially misleading, as randomization ensures comparability in expectation.
  • The Flaw: A non-significant baseline difference does not prove balance (it reflects low power), and a significant difference is merely a chance event that does not invalidate the randomization.
  • Recommendation: The authors recommend that researchers must abandon baseline testing and, instead, identify known prognostic variables in the trial-plan and fit them in an Analysis of Covariance (ANCOVA) model regardless of their baseline p-value, which increases the precision and power of the final treatment effect estimate.
23 January 2026

Sifting the evidence—what’s wrong with significance tests?

  • Core Critique: This classic commentary criticizes the widespread misuse and over-reliance on statistical significance tests (p-values) in medical literature, arguing that this practice fundamentally distorts evidence.
  • Central Problem: The authors identify publication bias (accentuating positive results over null results) as the key factor causing the mistaken publication of chance findings as real effects, thereby eroding confidence in science.
  • Recommendation: Researchers should shift their focus from the binary question of whether P < 0.05 to assessing the direction and the magnitude of the effect, specifically asking if the effect is of public health or clinical importance.
23 January 2026

An efficient, not-only-linear correlation coefficient based on clustering

  • Novel Statistic: The paper introduces the Clustering-based Correlation Coefficient (CCC), a computationally efficient measure that detects both linear and non-linear relationships, unlike traditional coefficients like Pearson’s \(r\).
  • Mechanism: CCC operates by assessing how well the marginal information of two variables aligns with the clusters formed by their joint distribution, allowing it to capture complex, non-monotonic associations.
  • Biological Utility: In an application to human gene expression data, CCC successfully identified biologically meaningful non-linear patterns, including those driven by sex differences, that were missed by standard linear methods.
23 January 2026

Step away from stepwise

  • Core Argument: This report critiques stepwise regression, arguing it is a flawed statistical method, especially in the context of Big Data, because it uses statistical significance to select variables.
  • Key Flaw: Stepwise procedures may include nuisance variables that are coincidentally significant and exclude true explanatory variables that happen to be non-significant, leading to models that fit data well in-sample but perform poorly out-of-sample.
  • Big Data Effect: The belief that stepwise is more useful with more variables is false; Big Data exacerbates its failings by increasing the chance of selecting spurious correlations, causing a dramatic deterioration in out-of-sample predictive accuracy.
23 January 2026

Covariate Imbalance and Adjustment for Logistic Regression Analysis of Clinical Trial Data

  • Core Problem: This study demonstrates that, for binary outcomes analyzed with logistic regression, the unadjusted Odds Ratio (OR) is generally a biased estimate of the conditional OR due to the non-collapsibility of the OR, even in a perfectly randomized trial.
  • Key Finding: Adjusting for prognostic covariates in logistic regression is critical, as it significantly increases statistical power (by up to 17.5% in simulations) and reduces bias in the treatment effect estimate.
  • Recommendation: While good baseline balance in covariates is desirable, it never fully alleviates the shortcomings of unadjusted analyses in the logistic setting; researchers should pre-specify and use covariate adjustment for prognostic variables to ensure the most precise and efficient estimate.
23 January 2026

Synthesizing Subject-matter Expertise for Variable Selection in Causal Effect Estimation: A Case Study

  • Objective: To empirically assess different strategies for Directed Acyclic Graph (DAG) creation and the resulting minimal adjustment sets for estimating a known causal effect (adherence on mortality in the CDP trial).
  • Key Finding (Efficiency): The results confirm that including nonconfounding prognostic factors (outcome predictors) significantly reduced variance (increased efficiency) of the causal effect estimate, aligning with theoretical statistical advice.
  • Recommendation: Researchers should prioritize the exhaustive identification of all potential outcome prognostic factors as the initial and most crucial step when constructing DAGs for causal effect estimation.
23 January 2026

Moving to a World Beyond \(p < 0.05\)

  • Core Argument: This editorial calls for the abandonment of the term “statistically significant” and the practice of dichotomizing results based on the arbitrary \(p < 0.05\) threshold.
  • Central Problem: Relying on the binary threshold distorts evidence, fuels poor practices like P-hacking and publication bias, and leads to the false equating of statistical significance with scientific importance.
  • Recommendation: Researchers must treat the p-value as a continuous measure of incompatibility between the data and the null hypothesis. They should focus their interpretation on the magnitude of the effect and its precision, primarily communicated via Confidence Intervals (CIs).
23 January 2026

A Structural Description of Biases That Generate Immortal Time

  • Core Argument: The true source of immortal time bias is the underlying selection or misclassification of subjects, not the period of event-free time itself.
  • Two Mechanisms: The paper structurally describes two ways the bias arises: 1) Selection bias when eligibility is applied after assignment; and 2) Misclassification bias when assignment is defined using post-eligibility data for strategies indistinguishable at baseline.
  • Prevention: The definitive solution is to use Target Trial Emulation to achieve synchronization of eligibility and treatment assignment at the start of follow-up, thereby eliminating the generating biases.
23 January 2026

The ASA’s Statement on p-Values: Context, Process, and Purpose

  • Core Principle: This article presents the American Statistical Association’s (ASA) six core principles for the proper use of the P-value, aiming to correct pervasive misinterpretations in scientific research.
  • Key Principles: The ASA states that the P-value does not measure the probability that a hypothesis is true (Principle 2) or the size/importance of an effect (Principle 5). Instead, it measures how incompatible the data are with a statistical model (Principle 1).
  • Crucial Admonitions: The statement strongly advises against drawing dichotomous conclusions based solely on a fixed threshold (e.g., P<0.05, Principle 3) and requires full reporting and transparency of all results to combat selective reporting (Principle 4).
  • Recommendation: The ASA advocates for a move beyond the single P-value to integrate statistical results with effect sizes, Confidence Intervals, context, and external evidence (Principle 6).
23 January 2026
No matching items