Variable selection in dendroclimatology: an example using simulated tree-ring series
1989
Arbaugh, M.J. | Peterson, D.L.
Regression analysis is often used to select important climatic variables from a group of candidate variables in dendroclimatology. Model R2 and F statistics are commonly used as a measure of the importance of these variables. The reliability of these selections in the presence of substantial unexplained variance has not been previously considered (with the exception of some climatic reconstruction studies). In this study multiple regression analysis with stepwise (forward inclusion and backward elimination), best subsets, and principal components was conducted on simulated tree-ring index series constructed from linear combinations of temperature and precipitation data with varying amounts of white noise. Regression techniques increasingly failed to select correct and/or complete sets of variables with increasing proportions of introduced noise, even with a limited number of candidate variables and a known relationship between predictor and dependent variables. R2 estimates were inflated with increasing noise for all techniques used. The results imply that climatic regression models may be incorrectly specified when large proportions of unexplained variability are present. There are implications for the analysis of data in other areas of forestry in which variable selection is attempted with datasets containing large amounts of unexplained variability.
Show more [+] Less [-]AGROVOC Keywords
Bibliographic information
This bibliographic record has been provided by National Agricultural Library