Trend Monitoring with Varying Coefficient Models
Some statistical procedures are presented for monitoring time trends. An F-statistic is given for testing the hypothesis that a particular state-space model fits two time intervals equally. This is useful when a specific time is believed to be the beginning of a structural change. The likelihood-ratio criterion can be used to test the hypothesis of no trend-change by testing that parameter values in the second interval stay constant at the end value of the first interval. When a transition time is not known in advance, cusum procedures can be employed (Brown et al. 1975, Harvey and Durbin 1986) as exploratory tools. An example application is made to some slash pine (Pinus elliottii Engelm.) tree-ring data from the southern United States that shows evidence of a 20-year recurring blight. For. Sci. 37(5):1365-1375.
No Supplementary Data
No Article Media
Document Type: Journal Article
Affiliations: Mathematical Statistician, USDA Forest Service, Institute for Quantitative Studies, Southern Forest Experiment Station, 701 Loyola Ave., New Orleans, LA 70113
Publication date: 01 November 1991