Skip to content

Latest commit

 

History

History

example-3

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 
 
 
 
 
 
 
 
 
 
 

example-3: censored and truncated data

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.