With three sources, the data set needed for the capture-recapture application becomes more complicated. The necessary information for the capture-recapture analysis includes:
The numbers of cases from different source combination can also be arranged in a 23 contingency table with one missing cell:
|
Source 1 |
Yes |
Yes |
No |
No |
|
|
Source 2 |
Yes |
No |
Yes |
No |
|
|
Source 3 |
Yes |
g |
e |
f |
c |
|
No |
d |
a |
b |
? |
|
If we use "0" to represent "identified" and "1" to represent "not identified", then the data can also be arranged as :
Source1 Source2 Source3 No. of Cases
============================================================
0 0 0 g
1 0 0 f
0 1 0 e
1 1 0 c
0 0 1 d
1 0 1 b
0 1 1 a
1 1 1 ?
With the above information available, log-linear model could be applied to estimate the number of cases not identified by all three sources.
Pittsburgh Dog-Bite Incidence Data
Three sources were employed to identify the incidence of dog bite injury in Pittsburgh in 1993. They were reports from animal control, reports from hospitals and reports from police/victims. The numbers of cases identified by different source combinations were listed below.
Animal Hospital Police/ # Cases
Control Victim
(p1) (p2) (p3) (r)
============================================
0 0 0 1
1 0 0 7
0 1 0 15
1 1 0 326
0 0 1 27
1 0 1 323
0 1 1 91
1 1 1 ?
In the following, we demonstrated how to use GLIM (Generalized Linear Interactive Model) program to conduct log-linear modelling for the capture-recapture analysis.
GLIM without macro
GLIM using macro created by Dr. Richard Cormack