Tutorial–assessing precision ✏️

Photo by Michael Shannon from Unsplash

Practical 4 – Poor precision in an abnormal situation

Remember that the “usual” estimator for encounter rate variance is based upon the idea that transects are distributed randomly rather than systematically with a random start. Under rare circumstances, this can over-estimate encounter rate variance and consequently, variance of density and abundance estimates. Revisit the output from Practical 4 and answer these questions.

Questions

Before getting absorbed in the analysis, what difficulties were caused because of the survey design.

• Identify the design flaws of the survey that produced such poor precision:
  too few replicate transects ignorance of animal gradient truncation too narrow leading to too few detections inadequate survey effort

• Why is the estimate of abundance exactly half the estimate of density in this analysis?
  \(\hat{N} = \hat{D} \times A\), where A=1/2 that is characteristic of the half normal detection function inadequate survey effort because of convert_units, square kilometers are twice as large as kilometers

Here is the summary output from the standard analysis of this data set. Alarm bells should ring in your head as you examine this output; specifically the variance components constituting the uncertainty in your abundance or density estimates.

In normal circumstances, approximately 75% of abundance estimate uncertainty in line transects comes from encounter rate variance. What percentage of uncertainty here comes from encounter rate uncertainty?

Does bootstrapping resolve the issue?

• Why does the use of the bootstrap not resolve the poor precision problem?
  bootstrap only improves precision of estimated detection function bootstrap does not perform well with half normal detection function resampling transects does not alter variability in encounter rate between transects bootstrapping is over-rated, never works well