Missed Detections Demo: Batch data input (ver. 0.2.8)

Observed values ("ground truth")

Actual apprehension numbers (per 100 subjects)


The box above should contain a comma-separated list of K+1 non-negative values, whose sum is 100. The first K values represent the actual apprehension numbers for the K periods per 100 subjects (e.g., if 5% of all subjects were apprehended during the first period, the first value should be 5). The last value is the number of subjects who were not apprehended during these K periods, per 100 subjects. Note that these numbers should add up to 100, representing 100%.

Algorithm parameters

"Scaling the weights" means that the estimator that is closest to the observed proportions will get R times as much weight as the estimator that is farthest from the observed proportions. Closeness is measured with the Jensen-Shannon metric (JSD).

If this option is not selected, the weight is based directly on this metric, i.e. weight=0 for an expert with the maximum possible JSD (= log(2)), and weight=1 for an expert with JSD=0. (This is the same as setting R to be very very large.)

In either case, the weight is a linear function of JSD.

Experts' estimates

There are two ways to upload a collection of estimates:

• You can create a comma separated value (CSV) data file in a text editor, and upload it. The format is illustrated in the text box below. (See format explanation for a detailed explanation of the format.)
• Or you can edit the sample data in the text box, and upload that edited data

Option 2 - edit the sample collection of estimates (to see how the results look you may just upload it as it is):
"An all-knowing oracle", 50, 10,10,10, 70, 0 "Another all-knowing oracle", 100, 5, 5,5, 0, 85 "No one goes", 50, 0, 0, 0, 0, 100 "No one is caught", 0, 20, 20, 20, 40, 0 "Everyone goes and is caught", 100, 33.34, 33.33, 33.33, 0, 0 "Max divergence", 100, 100, 0, 0, 0, 0 "expert A", 20, 10, 10, 10, 10, 60 "expert B", 10, 30, 30, 30, 10, 0 "expert C", 25, 20, 15, 10, 5, 50

The experts' estimates should be described in a comma-separated values (CSV) file (see sample in the box above). Each line should contain information for one expert. If the experts supply predictions for K time periods, each line of the file should contain K+4 fields:

• The first field in the line is a string containing the expert's name. It should be double-quoted if the name contains spaces or commas.
• The second field in the line is the corresponding estimate of the apprehension rate (i.e. what percentage of attempted crossings will result in captures). This is a number value in the range from zero to 100.
• The next K fields are the specific predictions for the number (out of each 100 released persons) who will try to cross again in the corresponding time period. For example, if K=4 and the expert believes that in each of the 4 time periods 15% of the subjects will try to cross the border, these will be four values at 15 each (the % sign is not needed).
• These K numbers are followed by two very important estimates. The first is the number (out of each 100) who will not attempt to cross until later, after all the time periods are passed. If the source or expert estimates this at 12%, the entry is 12 (no % sign).
• Finally, the very last entry in each row is the estimate, for the corresponding source of how many people out of each 100 become completely deterred, and will never try again. Of course, this should complete the 100. In this example it would be 100-4*15-12=100-72=28.

This particular extension of capture-recapture uses short time periods, and assume that people do not cross several times, during these short periods, being only captured on the last of the trips.

Empty lines, or lines starting with the '#' character will be ignored.