| |
Resources | Consignment
Estimation Software
From this page, visitors can download functions developed by David Lucy for
the estimation of illegal items in consignments of seized articles. There's
a four sample size estimation application, and a suite of functions to support
the computational requirements of Aitken & Lucy
(2004) as a solution for the matching problem in reasonably multivariate
space.
Use of the software
Could
people and organisations who use the consignment estimation software
here
please send us an e-mail (dlucy@maths.ed.ac.uk) to let us know
who they are and what they use it for. We need to know this
as we keep being asked who uses this software, and we simply
have no idea. We know people are using this software as they've
been spotted on UK television, but we could do with knowing
officially.
Note: Those who found
this page with a search engine but really want information
about consignment management should go to the BCSS site.
The usage is for making an estimate of how many units it is necessary
for the forensic examiner to actually inspect in detail to
know what proportion of the consignment contained illicit units.
For example: if a bin liner full of 500 packets of white powder
were found in the possession of a suspect, and it was felt that
a proportion of these packets contained an illegal substance,
then the drugs examiner would test them to see how many were
in fact illegal.
About the software
Current sampling strategies tend to be heuristic, being based
on measures such as 1/10th the number of units, or the square
root of the number of units. In this case the examiner would have
to look at 50, or 22 dependent on sample strategy. The problem
with this approach is that the answer doesn't tell you if you are
90% sure that all the consignment is illegal, rather it just
says that you are fairly sure the consignment is illegal.
By contrast,
the approach used here comes up with an answer which says that
one is 95% (or whatever level is selected) that at least so much of the
consignment is illegal. Not only is this estimative approach more precise
than `rule of thumb' based rules for generating sample sizes, it
can get results which require far fewer units in the consignment
to be measured.
In the bin liner case above were 4 units found to be illegal, then
it is the case that you are 95% sure that 55% or more of the total
consignment is illegal. Were one to sample 8 units, then you could
be 95% sure that at least 72% of the consignment
is illegal, or 99% sure that at least 60% was illegal.
Of course, the use of these functions is not restricted to drug
consignment estimation. It could be used to estimate quantities
of illegally copied CD-ROMs, containers of firearms on a ship
or boat, or even in quality control situations involving
totally legal units. Applications could involve the estimation of
defective items in a batch from a manufacturing process, or the
amount of substance produced during production of a chemical.
Installation notes
With
the exception of the Excel spreadsheets, which give the right
answers, but are a bit buggy, all the software here relies upon
having a fully functioning R installation to work. These are
R scripts, and they simply tell R what to do. As they're graphical
in operation they need a library to provide the features of a
GUI. This library is Tcl/Tk (see bottom of page for download
and configuration details).
Once R is up and running, and can
talk to Tcl/Tk then it is a simple matter to make the working
directory the directory with the package files downloaded here, and
run the main .r script from within R. Alternately the .bat files
can be pointed towards the R_Home (ie, where R is installed)
and they can be run from the file manager. If you
wish to be really flash you can make a symlink (Windows calls
them shortcuts) from the desktop to the .bat file, adjust the
properties so the calling window runs minimized, then you
can run the functions just by clicking on the desktop. Eventually I shall probably try to make them into
proper R packages so they will install in the same way as every
other R package. A help file for sorting out getting R to function
with Tcl/Tk is available here.
Help for getting a little further and getting these functions to
run from a desktop icon is available here.
I may try to make next versions of all consignment
estimations functions using Luke Tierney's excellent tkrplot() which
should make the dynamic parts of the graphics windows a bit smoother,
however, the whole lot could do with being rebuilt as a series
of binaries.
A final point is that anything which has
Tcl/Tk as a dependency will not port directly to S+ as
S+ has it's own set of graphical tools not based upon Tcl.
Beta Beta
This tells you the proportion of illegal units in a consignment,
and calculates the beta posterior density function, from a beta
prior function. The input parameters are the number of positives
and negatives observed in a sample, and the alpha and beta
prior parameters. This is recommended for consignment sizes greater
than 50 units.
Beta binomial
This is pretty much the same as the beta-beta given
above, but this time tells you how many of the remaining units
in a consignment are likely to be illegal. It is a more precise
estimate for consignments with fewer than 50 units.
T beta binomial
The t-beta-binomial describes the amount of illicit
substance in a consignment. If for instance there were 100 packets
of suspected illicit material, and the examiner wished to know how
much material was in the consignment, rather than how many
units , then the t-beta-binomial would be used.
This would
give a posterior distribution based on the prior parameters alpha
and beta, the mean and standard deviation of the masses of illegal
substance found in the units examined, the number of units examined,
and the number of units in the consignment.
The t-beta-binomial
is limited to about 190 units in a consignment, for larger consignments
see the t-beta-beta below. There is no Excel worksheet for the
t-beta-binomial as the calculations are a little too complicated
for Excel.
T beta beta
The t-beta-beta is more or less the same
as the t-beta-binomial, but uses a beta approximation to the
binomial. This mean that consignments with large numbers of units
can be handled, so far I have tried it up to 3 million. There
is no Excel worksheet for the t-beta-binomial as the calculations
are a little too complicated for Excel.
MVRA
These functions support the work done in the paper Evaluation
of trace evidence in the form of multivariate data and are
our solution to the matching problem when the data are continuous,
moderately multi-variate and have two levels of variation and
are for experienced R users only. Just download the zip
file to an arbitrary directory for which you have write access.
Unzip it and it will produce a sub-directory called glass-analysis.
Go into this directory, fire up R and source the file evidence-evaluation.r.
The readme file and the script file will give more details of
what's going on.
Supporting software
Tcl/Tk may be downloaded at this site, from this location. R
may be downloaded here. Tcl/Tk is about 13 megabytes, R is
at about 20 megabytes, so take this into account when downloading
from a slow
connection.
|
|
