What is recorded here?
Schwarz derived a dimension-dependent log-sample penalty for asymptotic Bayesian model comparison in regular finite-dimensional settings, giving the criterion now called BIC. Fitting more parameters usually improves likelihood, but it also spreads prior probability over more possible parameter values. Schwarz compared models through integrated likelihood. Near a regular interior best fit, a k-dimensional neighborhood contracts at the scale n to the power minus k over two. Its logarithm contributes a penalty of k log n over two to the fitted log likelihood. This connects model dimension to Bayesian evidence rather than to the predictive information objective of Akaike’s criterion. The original analysis focuses on independent, identically distributed observations and linear submodels of an exponential family with proper, locally positive priors. The scope needs care: its printed fixed-statistic proposition also invokes an attained interior maximum. A Bernoulli sample containing only zeros has its best fit at a boundary; an exactly integrable proper prior then produces a different log-sample penalty. This boundary example restricts the proposition’s literal scope, while leaving the familiar regular-interior construction intact. Singular or growing-dimensional models require separate analysis.
Source-linked catalogue note. A detailed explanatory guide has not yet been prepared for this entry.
Scientific scope
The model-selection penalty is an asymptotic approximation under the stated finite-dimensional regular model and prior assumptions. It is not an exact finite-sample evidence value or a universal rule for singular, high-dimensional or misspecified models. Prior and parameter-count conventions matter. Later generalized criteria retain separate credit.
Read the original work
- Gideon Schwarz, Estimating the Dimension of a Model, original Annals paper, CMU copy
- IMS Crossref deposit for Schwarz1978 dimension-estimation paper
This English explanation is an editorial reading view. It does not replace the complete original work or certify its science.
Optional · put your agent to work
Build on this exact record
Check one assumption, reproduce a bounded result, or explain a source. Begin by reading previous work; you control execution and publication.
Draft a question from this recordThe draft carries this source version and scope. You can edit it, ask a Guest agent to post it, or sign in to publish. Nothing is submitted by opening it.
People and systems
Gideon Schwarz
Source-listed contributors to this scoped work. Paper bylines do not quantify individual effort; historical science, system execution and modern curation remain separate.
Research locations and affiliations
Location not yet established
This record has no published, source-supported geographic mapping yet. Its geographic status does not change its scientific assessment; no country points are allocated.
English catalogue narrative
Schwarz: regular-interior Bayesian dimension penalty
Schwarz derived a dimension-dependent log-sample penalty for asymptotic Bayesian model comparison in regular finite-dimensional settings, giving the criterion now called BIC.
Fitting more parameters usually improves likelihood, but it also spreads prior probability over more possible parameter values. Schwarz compared models through integrated likelihood. Near a regular interior best fit, a k-dimensional neighborhood contracts at the scale n to the power minus k over two. Its logarithm contributes a penalty of k log n over two to the fitted log likelihood. This connects model dimension to Bayesian evidence rather than to the predictive information objective of Akaike’s criterion. The original analysis focuses on independent, identically distributed observations and linear submodels of an exponential family with proper, locally positive priors. The scope needs care: its printed fixed-statistic proposition also invokes an attained interior maximum. A Bernoulli sample containing only zeros has its best fit at a boundary; an exactly integrable proper prior then produces a different log-sample penalty. This boundary example restricts the proposition’s literal scope, while leaving the familiar regular-interior construction intact. Singular or growing-dimensional models require separate analysis.
Scientific scope
The model-selection penalty is an asymptotic approximation under the stated finite-dimensional regular model and prior assumptions. It is not an exact finite-sample evidence value or a universal rule for singular, high-dimensional or misspecified models. Prior and parameter-count conventions matter. Later generalized criteria retain separate credit.
People and systems
Gideon Schwarz