padanalyse

Padanalyse, or path analysis, is a statistical technique used to describe directed dependencies among a set of variables. It represents these relationships with a path diagram in which arrows indicate causal directions and the strength of relationships is summarized by path coefficients. Padanalyse is a special case of structural equation modeling (SEM) that uses only observed variables, not latent constructs. It was introduced by Sewall Wright in the early 20th century to decompose correlations into direct and indirect effects along specified causal paths.

In practice, padanalyse involves specifying a causal model with a set of variables ordered so that effects flow in one direction (often from exogenous to endogenous variables). The path coefficients are estimated using regression techniques, and the model’s implied covariances are compared with the observed covariances. Coefficients are typically standardized, allowing interpretation as the expected change in a dependent variable for a one-standard-deviation change in a predictor. In recursive models, the relationships are unidirectional and there are no feedback loops.

Key assumptions include linear relationships, additive effects, correct specification of the causal order, and adequate sample size. The fit of a padanalyse model is assessed with indices such as chi-square, Comparative Fit Index (CFI), Tucker-Lewis Index (TLI), RMSEA, and SRMR, often in comparison with alternative models. Software packages like LISREL, AMOS, and lavaan (R) are commonly used.

Padanalyse is widely applied in psychology, education, sociology, and epidemiology to test theories about how variables influence one another, and to partition correlations into direct and indirect effects. It provides a structured framework for evaluating causal hypotheses but relies on substantive theory and model specification, especially with observational data.