Structural equation models of annual federal expenditures for elementary and secondary education and for higher education were estimated using time-series data extending from 1947 to the later 1970s. The pattern of expenditures for elementary and secondary education proved to follow closely that for higher education. Factors affecting federal expenditures for elementary and secondary education either directly or indirectly (by affecting expenditures for higher education) have included which political party controlled the House of Representatives, the size of the federal budget surplus or deficit, the level of defense expenditures, and the number of arrests of persons under 18 years of age. Testing the forecasting accuracy of the models incorporating these factors revealed that forecasts diminished in accuracy when extended beyond 1 year, that multivariate forecasting techniques proved superior to univariate forecasts based on linear regression, and that a three-period, moving average model performed better than multivariate models when absolute accuracy of single-year forecasts was the sole criterion. Multivariate models provided greater insight into factors influencing change in time series, however, and appeared likely to be more accurate when influencing factors undergo sudden change. Appendixes present illustrative figures. (PGD)