
Three-list capture-recapture: what is and is not identified
Source:R/dark_figure.R
morie_dark_figure_three_list.RdWith three lists the observed table has seven cells; the count of
units on no list is the eighth. The saturated log-linear model (all
pairwise interactions and the three-way interaction) reproduces any
positive eight-cell table exactly (Research.P1.three_list_saturated_fits),
so the missing cell can take any positive value while the seven
observed cells are matched (Research.P1.missing_cell_unconstrained).
The population size is identified only once the analyst fixes the
three-way interaction; the conventional choice sets it to zero, which
gives the closed form
\(m_{000} = m_{111} m_{100} m_{010} m_{001} / (m_{110} m_{101} m_{011})\).
The function reports that estimate, the logical floor (the number of
distinct units seen), the two-list Petersen and Chapman estimates for
each pair of lists (each proved to respect the floor,
Research.P1.petersen_ge_floor, chapman_ge_floor), and the
three-way interaction that any candidate missing cell would imply.
Value
A list with observed (units seen, the floor),
missing_no_three_way, N_no_three_way, pairwise
(a data frame of Petersen and Chapman estimates per pair of lists with
the floor for that pair), implied_three_way (for
candidate_missing) and theorems.
Examples
morie_dark_figure_three_list(c("100" = 120, "010" = 90, "001" = 70,
"110" = 40, "101" = 30, "011" = 25, "111" = 15))
#> $observed
#> [1] 390
#>
#> $missing_no_three_way
#> [1] 378
#>
#> $N_no_three_way
#> [1] 768
#>
#> $pairwise
#> lists n1 n2 m floor petersen chapman
#> 1 1-2 205 170 55 320 633.6364 628.0357
#> 2 1-3 205 140 45 300 637.7778 630.4348
#> 3 2-3 170 140 40 270 595.0000 587.0732
#>
#> $implied_three_way
#> NULL
#>
#> $theorems
#> [1] "Research.P1.three_list_saturated_fits"
#> [2] "Research.P1.missing_cell_unconstrained"
#> [3] "Research.P1.petersen_ge_floor"
#> [4] "Research.P1.chapman_ge_floor"
#>