I am mortified. In the comments, Michael has identified a methodological error in the way I analyzed proximity to a regional center (downtown is after all, also a regional center). But in the process of re-computing those numbers, I discovered a computation error in my calculations for the system center, so both new correlation coefficients were incorrect, and my conclusion about poly-centrism unsupported…
Mea culpa… corrections below.
As promised, I’m continuing to add data to our transit equity data set. The latest addition is distance based – how far are you from a center.
The concept behind the Region 2040 plan is that we are going to become a poly-centric region. Portland’s central city will remain the center of the region, but we then also have 7 regional centers and a number of town centers. Housing, employment and services should cluster around each of these centers, reducing the need to travel long distances on a daily basis.
So today’s question is how strong is the effect of these centers on transit service? To figure this out, I calculated two distances for each block group:
1) The center of the block group to the ‘transit system center’ (I used SW 5th and Yamhill where all four LRT lines meet as my ‘system center’).
2) The center of the block group to the nearest regional center (I used the transit center in each regional center as that point, pulled from a TriMet data set on the CivicApps site).
Those columns have been added to our correlation data set spreadsheet.
The correlation results (negative because service decreases as you get further from a center):
- Distance to system center:
- Distance to nearest regional center:
By comparison, our correlation coefficient for density is 0.53, so distance from the system center is even more predictive of service level than density (although they certainly vary together to a strong degree – the correlation coefficient between density and distance from the center is 0.53).
The more significant finding is that the correlation for distance to a regional center is weak, our transit system is NOT very poly-centric yet. That shouldn’t surprise anyone, but it’s interesting to put a number on it. Unsupported by the revised data… In fact, the correlation with distance to a regional center rivals density as a reasonably strong correlation.
5 responses to “Updated: Transit Equity: Where Are You?”
More awesome number crunching, Chris.
The weak correlation in the regional center data might be due in part to the fact that the “system center” doesn’t also count as a regional center. Look at the scatter plot — that appendage of high transit scores reaching upward at 10 miles away is the city center, which is halfway between the Beaverton and Gateway regional centers.
Or am I confused?
That’s a fair point and I should recompute that.
Damn! Too bad the truth is much less interesting.
Don’t get mortified, Chris! We depend on having you alive to run the streetcar, plus lots of other stuff.
Thanks for the good work.
A reasonable possible explanation of why distance from the central city predicts the transit service level is the effect of all the commuters going downtown. What happens if you add a jobs factor?
My guess is that if jobs are added as a factor, distance to the central city becomes quite a bit less significant.
Also, adding floor to area ratio might be interesting, on the theory that the higher the FAR, the harder it is to find a parking place and therefore the more attractive transit is relative to cars.
You may want to add the “Analysis ToolPak” (Microsoft spelling) to Excel. It come with Excel but is not part of the normal install. It allows two factor ANOVA regression which usually gives a better result than a correlation test. I have a statistics package at work, so I have never used the Excel function. If I get the time, I may try plugging the data into it and run all the factors at once. This will also return p values for statistical significance. I may be able to get a R squared number for how much of the variation can be explained by the factors given, (Right now I am forgetting which methods give R squared and which methods do not) Without p values, it is impossible to know if the results statistically significant or is the difference too small to rule out the random variation in most real world data sets?