|
Abstract:
|
Trend estimation with time series is challenging when series are short and exhibit relatively high levels of sampling variability, temporal dependence or both. These challenges are compounded when the number of series precludes careful attention to potential generating processes and model fits. We addressed these concerns using Monte Carlo experiments with regression and state-space models, and series lengths of 10 to 25 years. Our findings are congruent with those of other studies, namely that, when working with short series, trend estimators associated with the simplest and even scientifically unrealistic models exhibited better statistical properties. This work addresses interest by a monitoring program in trend estimates from series (maximum length 24 years) associated with approximately 5000 fish abundance species-sampling gear-subpopulation combinations and approximately 600 vegetation prevalence species-subpopulation combinations. Most series are spatially and taxonomically related. Hence, we plan to explore trend estimators associated with joint modeling of related series.
|
