MerQuR+Statistics Glossary

Statistics Glossary → Modification indices

Modification indicesconcept

In structural equation models, statistics showing roughly how much the model chi-square would drop if a parameter currently fixed at zero were freed.

In confirmatory factor analysis and structural equation modelling, the researcher specifies in advance which paths and covariances exist; the rest are fixed at zero. The modification index (MI) computed for each fixed parameter rests on the Lagrange multiplier (score) test and gives the expected drop in chi-square if that parameter were freed; it is judged against a chi-square distribution with 1 degree of freedom (e.g. MI > 3.84 is significant at α = 0.05). The expected parameter change (EPC) and the standardized EPC show the approximate value the freed parameter would take.

Modification indices help uncover possible errors in model specification, but changing the model step by step on their basis becomes a data-driven search and lowers the chance that the results will replicate in other samples (capitalization on chance). A change should be made only if it can be defended theoretically, every change should be reported openly, and the modified model should ideally be validated in a new sample. Freeing error covariances purely on the strength of MIs is a particularly criticized practice.

An everyday example

An architect draws the structural plan of a building and assumes there is no connection between certain walls. The engineer's calculations show that adding a beam between two walls would make the structure markedly sturdier. Yet the beam should not be added just because the calculation says so; it must also make architectural sense. Modification indices likewise tell you which added path would most improve the model's fit, but theory has to make the decision.

In MerQur

In Structural Equation Modeling the 'Modification indices' output and in Path Analysis the 'Modification indices (Extra output)' list MI, EPC and standardized EPC. Confirmatory Factor Analysis reports the largest modification indices, and a model change based on them needs a theoretical justification.

Analyses where this concept appears

Confirmatory Factor Analysis (CFA)Structural Equation Modeling (SEM)Path Analysis

Related concepts

Model fit indices (χ², CFI, TLI, RMSEA, SRMR)Score (Lagrange multiplier) test