We compare a Bayesian model that explicitly handles censored and truncated canopy gap size observations (canopy-gaps-2021) to one that naively ignores the incompleteness of these data canopy-gaps-naive-2021.
As with the other examples, the proper Bayesian estimates are compared to the Horvitz–Thompson (HT) estimates CARE_HorvitzThomsponFiles.
Both Bayesian models include trend terms. All models were applied to data from a single stratum only for two reasons. First, park-wide estimates of canopy gap sizes were not desired; and, second, subsetting the data allows us to run and interact with the results of the model in a reasonable amount of time (tens of minutes, as opposed to hours of compute).
The goals for this analysis are to:
- demonstrate the inability of HT to provide inference on trend; and
- show that ignoring the censoring and truncation leads to biases, and that these biases may have non-trivial implications for management.
The plotting script ("plotting.R") creates the figure presented in the manuscript after results from each of the models have been compiled.