Duncan (1966) [115] Introduction to path analysis for sociological audience, with many examples. Decomposition of correlations between variables.
Finney (1972) [134] Discusses indirect effects in path analysis. Points out that the use of correlation in the original definition of total effect leads to problematic definitions of indirect effects when model contains more than one exogenous variable or indirect effects of endogenous variables are computed. Regression coefficient in the reduced-form model suggested instead as the total effect.
Lewis-Beck (1974) [239] Suggests decomposed effects from path analysis as a method of assessing relative importance of explanatory variables. Similar comments on direct, indirect and unanalyzed/spurious effects as in [134].
Duncan (1975) [114] Textbook on structural equation modelling, with a elementary but useful discussion of path analysis and different effects.
Alwin and Hauser (1975) [19] Total association vs. total effect. Complete decomposition of total effect into direct effect and indirect effects via different intermediate variables.
Lewis-Beck (1977) [240] Similar to [239], this one for a political science audience.
Fox (1980) [145] Matrix formulas for decomposition of effects, also for nonrecursive models.
Sobel (1982) [345] Confidence intervals for estimated indirect effects using the delta method.
Fox (1985) [146] Matrix computations for specific indirect effects (give same results as path tracing).
Bollen (1987) [41] Review of decomposition of effects for structural equation models. General formulas for the decompositions. Discussion of specific indirect effects. Formulas for standard errors of the effects.
Allison (1995) [18] Derives exact variance of indirect effects in linear models (with least squares estimates) with no latent variables. Shows that consistent estimates of this is equal to Sobel's (1982) [345] formulas obtained using the delta method.
Bollen (1998) [46] Short introduction to path analysis in the Encylopedia of Biostatistics.