Table S2 of Ganley et al. (2019): goodness-of-fit p-values and average detection probability for a detection function fitted separately to each year from 1998 to 2017, from aerial line-transect surveys of Cape Cod Bay.
Format
A tibble with 20 rows and 5 columns:
- year
Survey year, 1998 to 2017.
- cvm_p
Cramér-von Mises p-value for the year's detection function.
- ks_p
Kolmogorov-Smirnov p-value.
- p
Average detection probability, \(\hat{P}_a\).
- p_se
Standard error of
p.
Source
Ganley, L.C., Brault, S. and Mayo, C.A. (2019) What we see is not what there is: estimating North Atlantic right whale Eubalaena glacialis local abundance. Endangered Species Research 38:101-113. doi:10.3354/esr00938 , Table S2. Open access under CC-BY.
This is p, not perception
p is the average detection probability of the fitted detection function,
the same quantity selection_table() reports as p (with standard error
p_se). It is not a g(0) component. Ganley et al. state that perception
bias was not addressed in their study, because estimating it needs a
second observer team. Correcting for one while believing you have corrected
for the other biases abundance by a factor. g0() names any standard
component that is missing.
What it is useful for
It is a long, real example that detection probability is not a constant of
a survey. Twenty years of the same programme, aircraft and bay give p
between 0.431 and 0.866, with standard errors from 0.028 to 0.333.
The goodness-of-fit columns show that a model can rank well and still fail
a fit test. Ganley et al. took p > 0.05 as adequate fit. 2003 fails
Kolmogorov-Smirnov (0.048) but passes Cramér-von Mises (0.103), and has the
largest standard error on p. This is why selection_table() reports
cvm_p alongside AIC.
Examples
ganley_detection
#> # A tibble: 20 × 5
#> year cvm_p ks_p p p_se
#> <int> <dbl> <dbl> <dbl> <dbl>
#> 1 1998 0.751 0.564 0.663 0.156
#> 2 1999 0.521 0.426 0.601 0.094
#> 3 2000 0.531 0.248 0.75 0.041
#> 4 2001 0.867 0.802 0.64 0.07
#> 5 2002 0.97 0.965 0.708 0.104
#> 6 2003 0.103 0.048 0.76 0.333
#> 7 2004 0.969 0.965 0.655 0.142
#> 8 2005 0.985 0.995 0.866 0.059
#> 9 2006 0.431 0.392 0.703 0.163
#> 10 2007 0.866 0.777 0.712 0.065
#> 11 2008 0.97 0.977 0.595 0.05
#> 12 2009 0.581 0.507 0.591 0.038
#> 13 2010 0.668 0.785 0.599 0.046
#> 14 2011 0.847 0.737 0.604 0.043
#> 15 2012 0.991 0.982 0.611 0.062
#> 16 2013 0.98 0.993 0.576 0.03
#> 17 2014 0.93 0.858 0.507 0.035
#> 18 2015 0.583 0.815 0.541 0.028
#> 19 2016 0.115 0.065 0.431 0.038
#> 20 2017 0.341 0.129 0.518 0.028
# Detection probability varies from year to year
range(ganley_detection$p)
#> [1] 0.431 0.866
# The year that fails Kolmogorov-Smirnov also has the least precise p
ganley_detection[ganley_detection$ks_p < 0.05, ]
#> # A tibble: 1 × 5
#> year cvm_p ks_p p p_se
#> <int> <dbl> <dbl> <dbl> <dbl>
#> 1 2003 0.103 0.048 0.76 0.333