Analytics Magazine Analytics Magazine, July/August 2014 | Page 86

FIVE- M IN U T E A N A LYST Figure 2: Histogram of raw parking meter data. Note the tri-modal nature of the data. “Overtime,” i.e., flashing parking meters are represented by -1 in the red-shaded oval and constitute the large bar at the origin of the graph. Known paid parking meters are at the right and have a blue oval. from the meters, which is displayed for anyone who wishes to see. What we found was surprising. We expected to see uncorrelated parking lot data. We did not expect to find many over-time parking spots. I hoped that the data would be exponential – which would lead to nice, clean analysis. What we discovered was, well, a mess. Of the 100 parking spots surveyed, 25 percent were “flashing” or over-time (violation). Of the parking spots that were not over-time, six showed times over one hour, implying that the persons parked there had in fact put money in the meter. We are completely discarding the possibility that someone would park in a 86 | A N A LY T I C S - M A G A Z I N E . O R G spot that had been previously occupied but was not vacated, i.e., showing up with 30 minutes remaining on meter and not pressing the button/inserting coins. I had hoped that the sojourn times would be exponentially distributed, but that is a case that is pretty difficult to make with this dataset (see Figure 2). Now, we don’t actually know how many patrons have paid, or how many have simply run over. However, there are 100 parking spots considered, and of these, six currently have clocks over one hour. We can (crudely) estimate [2] the true number of paid parking spots by realizing that we are observing the last hour of what may be a two-hour process. Therefore, we think approximately W W W. I N F O R M S . O R G