Incorporating other group size calibration approaches

[amandalbradford]

From JB on 3/11/24:

"Tim Gerrodette has published a 2019 paper on group size calibration using different analysis methods. He no longer bases the calibration on the high and low estimates of group size (only the Best). This is good, I think, because the spread between high and low is different for calibration schools than for typical schools, because more time is spent doing calibration estimates and observers get better looks at them. Also, observer effects are treated an a random variable and the published coefficients include species effects but not individual effects. Also, group sizes of 25 or less appear to be unbiased, but the bias depends on group size (getting larger for larger group sizes).

In my current analysis of ETP data, I'm using a very simple group size calibration approach that ignores species differences:
####create new variables for bias-corrected group size estimates for groups of 25 or greater
sightings$BC_ss_tot= exp((log(sightings$ss_tot)-0.63)/0.796)
sightings$BC_ss_tot[sightings$ss_tot < 25]= sightings$ss_tot[sightings$ss_tot < 25] #no bias correction <25 groups
sightings$BC_lnsstot= log(sightings$BC_ss_tot)
sightings$BC_best= sightings$BC_ss_tot * sightings$ss_percent

As you can see, I keep the original estimates and create new bias-corrected estimates as new variables (so that I can easily use or not use the BC estimates at a later stage of analysis)."

There may also be new drone-based calibration datasets available in coming years. Should we update the group size calibration approach in LTabundR?

[amandalbradford]

More insight from JB: Gerrodette et al. "treat observer ID as a random variable and present only the coefficients for the "average" observer. I think this may be the way to go, especially as we have fewer and fewer of the individually calibrated observers. Due to the configuration of the models used, they saw a much larger difference in bias for small and large schools than did prior analyses."