Biomathematics & Statistics Scotland
Phenology is the study of the timing of natural events, such as flowering or migrations, and in our work how timing is affected by weather. Stepwise regression is commonly used to investigate such relationships. However this requires aggregation of daily temperatures to avoid multicollinearity problems. Penalised regression offers an intuitively attractive alternative, producing a smooth profile of regression coefficients. We have been using simulations from biologically-based models to evaluate the performance of alternative regression methods. These demonstrated the advantage that penalised regression has over stepwise, particularly if the dimensionality is controlled, e.g. by use of P-spline regression with a basis of not more than 30 B-splines.
Regression coefficients to predict rowan
flowering dates (based on 30 years
of Last family records) estimated by
P-spline regression using a squared
first-difference penalty. The dashed
lines indicate a 95% confidence interval.
The negative coefficients indicate that
the date of flowering is decreased
(ie brought forward) by higher
temperatures between January and
flowering in May.
Further details from: Adrian Roberts
