01 / RESEARCH VALIDITY
DATA observations · studiesASSUMPTIONS design · measurementCHECK replication · transparencyCLAIM How reliable is the evidence?
EVIDENCE · MODEL · UNCERTAINTY
Reliable conclusions begin with assumptions that can be inspected.
Robust Inference explores how neuroscience, earthquake engineering, and statistics test models, expose uncertainty, examine dependence, and determine how far evidence can support a conclusion.
Independent educational resource
02 / SEISMIC RISK
DATA ground motions · responseASSUMPTIONS source · site · structureCHECK validation · sensitivityCLAIM How uncertain is the risk?
03 / STATISTICAL INFERENCE
DATA curves · repeated measuresASSUMPTIONS dependence · random effectsCHECK diagnostics · comparisonCLAIM What can the model infer?
Three research environments — connected through questions of validity and uncertainty, not through identical scientific mechanisms.
ROBUSTNESS IS EARNED
A conclusion becomes more trustworthy when assumptions are visible, alternative explanations are tested, uncertainty is quantified, and the model survives meaningful checks.
Data are not conclusions.
Dependence changes inference.
Models require validation.
Uncertainty belongs in the result.
FOUR EVIDENCE DOMAINS
Different questions require different kinds of reliability.
Research Validity
Explore reproducibility, replication, experimental design, reporting, bias, systematic review, internal validity, external validity, transparency, and the credibility of scientific evidence.
Translational Neuroscience
Examine experimental neurology, stroke research, preclinical models, clinical translation, brain imaging, cerebral ischemia, neuroprotection, and research-quality challenges in experimental medicine.
Seismic Risk & Reliability
Explore ground-motion models, simulated seismic input, seismic hazard, structural reliability, fragility, risk analysis, performance-based engineering, and disaster resilience.
Statistical Dependence
Study functional data, stochastic processes, random effects, longitudinal observations, dependent data, biostatistics, statistical inference, and model assumptions.
MODEL CHECKS
The same word “reliable” means different things across research fields.
DESIGN → EVIDENCE → REPLICATION
What makes a biomedical result more credible?
Field: Research validity
Relevant considerations: study design, measurement quality, sample selection, randomization, blinding, analysis plan, reporting, replication.
Possible checks: independent replication, systematic review, sensitivity analysis, transparent methods, data-quality assessment.
Reproducibility does not guarantee that a biological or clinical interpretation is correct; it addresses specific dimensions of research reliability.
GROUND MOTION → MODEL → RISK
How can a probabilistic model support seismic-risk assessment?
Field: Earthquake engineering
Relevant considerations: earthquake source, site characteristics, ground-motion variability, structural response, model uncertainty, fragility, exposure.
Possible checks: comparison with observations, simulation validation, sensitivity studies, alternative models, uncertainty propagation.
A probabilistic seismic-risk estimate does not predict the exact time or exact outcome of a future earthquake.
DEPENDENCE → MODEL → INFERENCE
What changes when observations are not independent?
Field: Statistics
Relevant considerations: repeated measurements, time dependence, spatial or clustered structure, functional observations, random effects, stochastic processes.
Possible checks: model diagnostics, residual dependence, simulation, alternative covariance structures, and cross-validation where appropriate.
A statistical model is a structured representation of uncertainty, not a complete description of the real system that generated the data.
THE ROBUST INFERENCE PROTOCOL
Seven checks between an observation and a defensible conclusion.
Define the question
State exactly what the analysis is trying to learn.
Identify the data-generating process
Ask where observations came from, how they were measured, and what selection process produced them.
Make dependence visible
Determine whether observations are independent, repeated, clustered, longitudinal, spatial, functional, or otherwise related.
State the model
Expose the assumptions connecting observed data to the quantity being estimated or predicted.
Test alternatives
Ask whether bias, different structures, parameters, or explanations could produce similar observations.
Quantify uncertainty
Represent uncertainty explicitly rather than hiding it behind a single estimate or simulation.
Limit the claim
Match strength, scope, generalizability, and causal language to the evidence actually available.
EDUCATIONAL REFERENCE POINTS
Six researchers across research quality, seismic risk, and statistics.
These profiles are presented as educational reference points for exploring public academic work. They are not presented as members, employees, partners, collaborators, representatives, endorsers, or affiliates of Robust Inference.
The first three email addresses are platform contact addresses supplied for this site and are not presented as verified university or institutional email accounts.
The final three profiles are educational reference points based on public academic work. Their inclusion does not imply participation, collaboration, endorsement, employment, representation, or affiliation with this resource.
UD
VALIDITY · PLATFORM CONTACT
Ulrich Dirnagl
Professor emeritus and Guest Scientist · Germany
Charité – Universitätsmedizin Berlin · Berlin Institute of Health at Charité · QUEST Center for Responsible Research. Founding Director of the QUEST Center for Responsible Research.
Academic work spanning clinical and experimental neuroscience, cerebral ischemia, stroke, cerebral blood flow, endogenous neuroprotection, brain inflammation, brain-body interactions, neuroimaging, translational medicine, systematic reviews, meta-research, research quality, reproducibility, generalizability, and biomedical research validity.
Clinical Neuroscience · Experimental Neurology · Stroke Research · Translational Research · Meta-Research
ORCID 0000-0003-0755-6119
MD
SEISMIC · PLATFORM CONTACT
Mayssa Dabaghi
Associate Professor · Lebanon
American University of Beirut · Department of Civil and Environmental Engineering
Academic research in earthquake engineering including generation, validation, and use of simulated ground motions, stochastic ground-motion models, near-fault motion, seismic hazard and risk, performance-based earthquake engineering, structural reliability, uncertainty, probabilistic modeling, and engineering risk analysis.
Ground-motion models · Seismic hazard · Structural reliability · Seismic risk
ORCID 0000-0003-2017-3462
HS
STATISTICS · PLATFORM CONTACT
Helle Sørensen
Professor · Denmark
University of Copenhagen · Department of Mathematical Sciences
Academic research in mathematical and applied statistics including functional data analysis, inference for dependent data, stochastic processes, stochastic differential equation models, random effects, clustered observations, biostatistics, life-science applications, and non-trivial dependence structures.
Functional data · Statistical dependence · Stochastic processes · Applied statistics
ORCID 0000-0001-5273-6093
JPAI
VALIDITY · EDUCATIONAL REFERENCE POINT
John P. A. Ioannidis
Professor and Co-Director of METRICS · United States
Stanford University · Medicine · Epidemiology and Population Health · Biomedical Data Science · Statistics · Meta-Research Innovation Center at Stanford (METRICS)
Academic research on credibility, replication, reproducibility, bias, validity, and reliability of findings, including meta-research, evidence synthesis, clinical research methodology, biostatistics, epidemiology, and meta-analysis.
Meta-research · Research reliability · Bias · Evidence synthesis
ORCID 0000-0003-3118-6859
JWB
SEISMIC · EDUCATIONAL REFERENCE POINT
Jack W. Baker
Professor of Civil and Environmental Engineering · United States
Stanford University · Department of Civil and Environmental Engineering
Academic research using probabilistic and statistical methods to quantify disaster risk and resilience, including earthquake ground-motion characterization, seismic hazard and risk, spatially distributed systems, structural risk, post-disaster recovery, and performance-based engineering.
Seismic hazard · Risk analysis · Ground motions · Disaster resilience
ORCID 0000-0003-2744-9599
SD
STATISTICS · EDUCATIONAL REFERENCE POINT
Susanne Ditlevsen
Professor · Denmark
University of Copenhagen · Department of Mathematical Sciences
Academic research in statistical inference for stochastic processes, mathematical modeling of physiological systems, nonlinear dynamics, computational neuroscience, biomathematics, probability, diffusion processes, regression, and methods for dynamic biological systems.
Stochastic processes · Statistical inference · Mathematical modeling · Computational neuroscience
ORCID 0000-0002-1998-2783
METHOD NOTES
Open a note and inspect the assumptions behind the conclusion.
Explore concise educational notes across research validity, reproducibility, seismic uncertainty, probabilistic risk, functional data, stochastic processes, and statistical inference.
10 notes
Research ValidityWhat is the difference between reproducibility and replication?Explore two related but distinct ways of testing research reliability.
Computational and methodological reproducibility concern research materials, code, data access, and reporting. Replication uses new observations, often by independent teams. Conceptual replication and variation across studies show why obtaining the same computational result from the same data answers a different question from observing a similar finding in a new study.
Research DesignWhy does internal validity matter before generalization?Explore bias, design, causal interpretation, and validity within and beyond a study.
Selection, measurement, randomization, blinding, confounding, and design shape internal validity. External validity concerns populations and conditions beyond a study. Replication helps test generalizability; statistical precision alone cannot justify broad causal interpretation.
Translational ResearchWhy can promising preclinical evidence fail to translate?Explore model validity, biological differences, design, and evidence gaps.
Preclinical models, experimental conditions, biological heterogeneity, outcome measurement, publication bias, sample-size limitations, replication, systematic review, and research quality all affect clinical translation. Success in experimental models does not automatically imply clinical benefit. General educational content only.
Earthquake EngineeringWhat does a stochastic ground-motion model represent?Explore how probabilistic models characterize ground-motion variability.
Source characteristics, site conditions, time histories, near-fault effects, directivity, spectral properties, parameters, and uncertainty enter stochastic modeling. Validation compares simulations with observed motions; simulated ground motions are plausible scenarios, not exact future-earthquake predictions.
Seismic RiskHow are seismic hazard and seismic risk different?Separate shaking probability from consequences that shaking may produce.
Hazard concerns ground-motion intensity and recurrence. Risk also involves exposure, vulnerability, structural response, fragility, damage, and consequences. In performance-based engineering, high hazard does not imply the same risk for every structure or community.
UncertaintyHow does uncertainty move through an engineering model?Explore inputs, model form, parameters, simulation, and output variability.
Uncertain inputs, probability distributions, model-form uncertainty, ground-motion variability, structural models, Monte Carlo simulation, sensitivity, conditional results, correlation, and output distributions form an uncertainty-propagation chain. Uncertainty should remain visible throughout risk assessment.
StatisticsWhy does dependence change statistical inference?Explore why repeated or related observations are not always independent.
Repeated measures, clusters, time dependence, within-subject correlation, random effects, and covariance change effective information. Statistical models must represent dependence; ignoring it can produce misleading measures of precision and conceptually incorrect standard errors.
Functional DataWhat makes functional data different from ordinary tables?Explore observations naturally represented as curves or functions.
Curves and trajectories measured over time or location involve smoothing, registration, functional variation, repeated functions, sampling grids, measurement error, and dependence. Functional principal components summarize patterns; treating every curve point as unrelated can lose data structure.
Stochastic ProcessesWhy model a changing system as a stochastic process?Explore random evolution, dynamic uncertainty, and trajectories.
Stochastic processes represent time-dependent observations, states, transitions, random variation, trajectories, diffusion models, and stochastic differential equations at a high level. Repeated paths support parameter estimation; a dynamic process contains information absent from an isolated measurement.
Model ValidationWhen is a model useful even though it is not literally true?Explore approximation, diagnostics, sensitivity, and scientific usefulness.
Models use abstraction, simplified assumptions, parameterization, and calibration. Residuals, predictive performance, sensitivity, competing models, simulation, domain knowledge, and model validation reveal usefulness. Fitness depends on the question, not reproduction of every feature of reality.
No method notes match your search.
ABOUT ROBUST INFERENCE
Good inference makes uncertainty visible instead of treating it as an inconvenience.
Robust Inference is an independent educational prototype connecting research-quality studies, probabilistic earthquake engineering, and mathematical statistics.
It does not suggest that biomedical reproducibility, structural reliability, and statistical dependence are equivalent concepts.
Instead, it examines shared methodological responsibilities: defining questions, understanding how observations were generated, making dependence visible, stating model assumptions, considering alternative explanations, quantifying uncertainty, validating models, and limiting conclusions to what the evidence supports.
Robust Inference is not a university, hospital, research institute, engineering consultancy, statistics company, medical provider, earthquake warning service, or commercial product.
Assumptions should be visible
Readers should distinguish observations from the model used to interpret them.
Dependence matters
Repeated, clustered, functional, spatial, or time-dependent observations contain different information than independent measurements.
Validation needs alternatives
Models become more informative when compared with observations, competing explanations, alternative specifications, or independent evidence.
Uncertainty belongs in the conclusion
A single estimate should not hide variation, model limits, sampling uncertainty, or uncertainty about generalization.
CHECK THE STRUCTURE
Take one conclusion and rebuild the assumptions underneath it.
Browse method notes, compare model checks, and use the Robust Inference Protocol to examine observations, dependence, assumptions, uncertainty, validation, and limits.
OBSERVATION DEPENDENCEASSUMPTION BIASINFERENCEUNCERTAINTY MODELCLAIM VALIDATION · GENERALIZATION