The
efficient
allocation
of
police
personnel
means
establishing
procedures
so
that
police
officers
can
be
at
the
right
place
at
the
right
time.
The
collection
of
activities
known
as
COMPSTAT
is
clearly a
useful
step
in
that
direction.
However,
the
statistical
procedures
underlying
COMPSTAT
are
not
true
forecasting
tools
and
uncertainty
in
the
crime
patterns
is
usually
not
carefully
addressed.
There
are
also
fundamental
questions
about
what
the
most
effective
spatial
and
temporal
units
should
be.
In
this
paper,
we
begin
to
address
such
concerns
capitalizing
in
part
on
recent
developments
in
statistical
learning.
Data
from
Philadelphia
and
Los
Angeles
are
used
as
illustrations.
Audio (MP3 File, Podcast Ready)