Wednesday, June 26, 2013

Commuting Time Correlation and the Commuting (In)Efficiency Index

I came across some new data today that I didn't currently have in my Metropolitan Statistical Area (MSA) infrastructure - density - highways - mode share database mashup.  It was average commute time, which I had begun, but didn't have a full array of data to begin making any charts.  Here are those charts.

What I wanted to find was whether there was any correlation between a variety of statistics in relation to commute time (that are actually quantifiable and quantified), such as what role do things like density, congestion, highway infrastructure, etc. have on commute times, if any.

My hypothesis was that the correlation would likely be very low, as I expected that commute times had more to do with job to housing balance by location.  Knowing cities, you know that jobs and housing tend to co-locate in relation to each other, as (somewhat) evidenced by Prof Peter Newman's hour-wide city theory stating that no matter the transpo technology of the day, a city is always generally an hour wide from end to end and 30 minutes from end to center.  The market simply doesn't put up with much more than that.

This jives with surveys searching for the ideal commute time, which tend to be about 22 minutes (though that would be below the average commute time in every city).  The reality is there is an equilibrium where people like to be a certain amount of time from their work, enough time to prepare for the day mentally on the way to work and enough time to decompress on the way home.  Too close or too far, could disrupt this seemingly necessary transitional mood or phase shift.

Onto the charts.  Let's see if I can structure them in an order that makes sense:

Let's start with the simplest.  Distance traveled daily to commute time (which will show up as one axis in every chart):

Nope. That's all over the map.  Distance has little to do with time.  Which makes a good bit of sense, since distance here isn't factoring mode.  Way too many variables still in action.

How about Density to Commute Time:

Also all over the place.  Again, this doesn't factor mode.  The availability of a 20-minute walk is no different than a 20-minute drive of a lower density metro here.

How about to total driving, ie Vehicle Miles Traveled per capita to commute time:

Any correlation here is extreeeeeemely minimal.  Though a slight, very slight, negative correlation.  The more people drive there is a slight tendency towards shorter commute times.  But that's a stretch.

How about highway capacity to commute times:

Slightly more of a pronounced pattern, perhaps building on the prior.  The more highway lane miles, ie capacity, the shorter the commute.  The theory being that it's a clear highway for you, the faster you move.  This is TTI's ideal scenario (I had a less appropriate analogy here, redacted).  However, you'll see the R-squared factor is quite low.  Also, the shortest commute time (Kansas City) also happens to have the most highway capacity.  Remove that dot way up in the top left corner, the correlation starts to disappear.  Can we really count on a relative outlier as justification of correlation?  I don't think so.

How about Congestion to Commute Times?  Theoretically the more congested a place, the slower traffic would move compared to ideal speeds and travel conditions.  In this case, I'm using daily highway traffic per lane mile per capita, ie traffic over capacity.

Again, some verrrrry slight correlation that increased congestion equates to increased commute times.  But this must also be taken in context of the idea that increased congestion correlates (more strongly than this) with increased economic activity.  Trade-offs is the theme at work here.

How about travel mode:

Oooohhh.  Now we're starting to see some patterns emerge.  The increase non-vehicular commuting mode correlates with increased travel time.  This makes some sense with the potential waits between headways of various forms of transit.

But what if we take NYC out of the equation?

The correlation drops a good bit, R2 dropping from .46 to .30 though that is still stronger than anything else I've compared (and many didn't even make this post).  However, there is still a potential lesson here and again, that's in trade-offs.  People are willing to take alternative modes of travel if the time of commute is slightly longer.  Why?

My guess is because of the potential benefits of alternative modes of travel, whether that means living in a more dense, vitally active place, the amenity of increased choice availability, increased productivity -- the ability to do other things while on a train for example like read or work, or improved health benefits from walking, cycling, or not dying on the many rush hour crashes each day.

On the surface at least, this quick study seems to confirm the theory that no single variable has a signficant effect on commute times more than the natural equalizing force towards that standard 25-30 minute commute.  The better, or more interesting question might be, is a shorter average commute preferable or even worth pursuing, if that isn't the market's preference (provided it doesn't create infrastructural and energy inefficiencies in relation to the diseconomy of decentralization).

While playing around with this data I started thinking about new mash-ups.  Here's one that I call the Commuting Efficiency Index.  Or in the case of some cities, the Commuting Inefficiency Index.  It adds Average VMTs (as proxy for extra energy expended by driving over increased distance) to Average Minutes of Commute.  The thinking here is trade-offs.  Some cities could be better at less driving/distance/energy consumption while others could succeed at quick trips, like Kansas City (though this doesn't measure KC's big weakness which is infrastructure to tax base deficiency):

3 Most Efficient Commutes:

1.  Portland  43.6
2.  Sacramento  44.6
3.  San Diego  46.7
All West Coast.  Interesting.

3 Least Efficient Commutes:

1.  Houston  61.0
2.  Atlanta  58.2
3.  Orlando  57.2
All Sun Belt.  Interesting.

Just about every other city is pretty tightly packed between 49-53, but those six stood out quite a bit beyond just being top three on each side of the spectrum.