Slashdot Mirror

← Back to Stories (view on slashdot.org)

Flawed Map Says L.A.'s Crime Highest Next to Police HQ

Posted by timothy on from the statistics-are-coming-from-inside-the-station dept.

CNET briefly describes how a poorly chosen default behavior has led to an online crime map of Los Angeles (on a site designed at a cost of $362,000) that shows that "a location just a block from the department's new headquarters is the most crime-ridden place in the city." I wonder how often this sort of error would completely skew things like real-estate maps that attempt to show whether houses in a certain neighborhood are worth more than those in the one next door.

3 of 123 comments (clear)

  1. Perhaps criminals are getting more brazen by superyanthrax · · Score: 5, Insightful

    More seriously, they should probably have had the program throw an error in case they could not find a certain location rather than putting the crime report at an arbitrary location. That would have caused the problem to be discovered earlier.

    1. Re:Perhaps criminals are getting more brazen by Trepidity · · Score: 5, Insightful

      There's ways to balance that, depending on what your needs and visualization methods are. For example, if you know that a significant proportion of your crime reporting gives only district-level precision, not pinpointing to specific addresses, then it'd be more honest data presentation to just produce a colored-in map on a district-by-district level, and not attempt to give more detailed maps. If you do still want to give the more detailed maps, then at least average the un-localized things across the district instead of putting them all in one place.

      To use an actual (fairly simple) example that came up in my work recently: say you have some date figures, most of them with years but some only with decades. The wrong thing to do is to put the "1960s" datapoint at 1965, because then you get spurious spikes in the middle of every decade. Several more correct options are: just produce decade-by-decade visualizations, or else produce year-by-year visualizations, but assign a "1960s" datapoint as a 1/10-weight datapoint in each of 1960 through 1969.

      --
      10 PRINT CHR$(205.5+RND(1)); : GOTO 10
  2. Criminal activity detection... by RyanFenton · · Score: 5, Insightful

    It's not a legally recorded crime unless someone is caught and convicted. It's not surprising that these crime maps would show this result - the places that police officers are most likely to be, are the places where the most crime is "found".

    This is akin to saying that the places where the most vehicular crime occurs are where speed traps and automated traffic cameras are located.

    If you had a world with absolute and omnipresent law enforcement, and that society could somehow actually function, my guess is that the map would match a map of the average human traffic in a given location.

    Ryan Fenton