DataTerminology.comNEW
Explicit premises underpinning models, analyses, or experiments that must hold for results to be valid.
DESCRIPTION
Assumptions define conditions under which analytical methods and models produce valid conclusions. Examples include linearity, independence, normality, and stationarity. Making assumptions explicit enables appropriate tests, diagnostics, and robustness checks. Violations can bias estimates, inflate error, or invalidate inferences. Sensitivity analysis and out-of-sample validation help assess fragility to assumption breaches. Clear documentation supports governance, reproducibility, and informed decision-making.
ALTERNATIVE DEFINITIONS
Wikipedia
In statistics, a statistical assumption is a condition that is imposed on the data-generating process for a model or procedure to be valid.
Frames assumptions as conditions on the data-generating process tied to validity.
NIST/SEMATECH e-Handbook of Statistical Methods
Most statistical methods rely on certain assumptions about the data; when these are not met, the validity of conclusions can be compromised.
Emphasizes the consequences of unmet assumptions on inference validity.
RELATED TERMS
Hypothesis Systematic Error Ceteris Paribus Confounding Stationarity Independent and Identically Distributed Normality Linearity Model Validation Robustness
COMPETING TERMS
KNOWLEDGE AREAS
Data Science Mathematics Testing & Validation AI Governance Data Governance
TARGET AUDIENCE
Data Scientist Data Analyst Data Management Consultant Data Governance Manager IT Auditor Chief Data Officer
EXTERNAL SOURCES
- Wikipedia - Statistical assumption
- NIST - Comprehensive statistical methods guidance
- CRAN - R task view on inference and assumptions
TOP QUESTIONS
Why are assumptions critical in data science?
They define when models produce valid results. Ignoring them risks biased estimates, misleading significance, and poor generalization.
Category: Relevancy
How can I test model assumptions?
Use diagnostics like residual plots, normality tests, variance checks, autocorrelation tests, and cross-validation to assess assumption fit.
Category: Requirements
What is a common misunderstanding about assumptions?
That they must hold perfectly; in practice, mild violations can be acceptable if robustness checks and sensitivity analyses support conclusions.
Category: Misunderstanding
What are typical assumptions in linear regression?
Linearity, independence, homoscedasticity, normality of errors, and correct model specification.
Category: Example
How do assumptions affect causal inference?
Assumptions like no unmeasured confounding, SUTVA, and positivity are essential; violations undermine causal claims.
Category: Concept
How should assumptions be documented?
State them explicitly, provide diagnostics, and include sensitivity analyses in model cards or technical documentation.
Category: BestPractice
What if assumptions are violated?
Consider transformations, robust methods, respecification, nonparametric techniques, or different models better suited to the data.
Category: Guidance
Are machine learning models assumption-free?
No. While often less parametric, ML methods still have assumptions about data distribution, independence, and stationarity.
Category: Misunderstanding
What is the risk of hidden assumptions?
Undisclosed premises can erode trust, hinder reproducibility, and lead to systemic errors in downstream decisions.
Category: Risks