I have about 27,000 survey responses from across about 150 Bus Routes, each with
potentially 100 stops. I've recorded the total Ons and Offs for each stop
on each bus run, as well as the stop pair each survey response corresponds to.
I wish to create weights based on the On and Off stop for each line and
direction. This will create a very sparse "half table" (observations
by From/To) of responses to Rake. I'm wondering if there is any good
"mechanical" method for combining Ons and Offs into groups to help
"fill out the table." I wish to be sensitive to "distance
travelled" when combining pairs. That is when grouping Ons and Offs I want
to minimize the range of "stations traversed" (actually the range of
time on the bus) within each group.
One potential approach to avoid greatly aggregating the data is to create
"Pseudo-responses" for the missing interchanges and seed the weight
table with very small values before Raking. When Raking is done I would scale
the weights of actual responses to account for the dropped weights of the
Pseudo-responses.
Is there any prior art available for me to review? There are a huge number of
groupings to be done, so I'm hoping for an algorithm or process that could
automatically group the stops and report when it fails to find sufficient
"near neighbors" to aggregate.
Thanks in advance,
Robert Farley
LACMTA
1 Gateway Plaza
Los Angeles, CA 90012-2952
(213)922-2532
[[alternative HTML version deleted]]
