To extend my land value-based analysis of gentrification in Seattle, it was vital to obtain some historic taxlot information. Though King County doesn’t maintain such information in its main spatial database, I was able to find CSV files of tax lot appraised value going back to 1997 from a pretty deeply buried database on the City of Seattle’s website. To join this data with the taxlot shapefile, I had to concatenate a new taxlot id from the “major” and “minor” fields for both historic and present data files. Unfortunately, out of the 600,000+ 2015 taxlots for King County, only about 400,000 had an associated 2000 land value. I was able to attain a somewhat more complete sampling (about 530,000) by including the values from the 1999 assessment file, and I took the maximum of the two assessments as the 2000 value. To further aid in the comparability of 1999/2000 land value to present value, I joined this data to the shapefile of census blocks, obtaining a median land value for each block in the county in 2000 and 2015.
Land value increases since 2000 largely reflect current land value in Seattle, with the same spatial patterns observable. To quantitatively assess this pattern, I then ran a linear regression on land value in Seattle and within the urban growth boundary of King County as a whole, testing for the effects of transit proximity while controlling for distance to downtown (selected as the block with the highest median land value per square foot, at 3rd and Seneca), distance to the center of the nearest designated urban village/center, shortest distance to water, the distance to the nearest road with a speed limit 40mph or higher, the percent of potential sidewalks which are improved (for the Seattle analysis), and the distance to the SeaTac airport (for the county analysis). I ran this regression as a linear-linear, log-linear, and log-log relationship, testing for possible non-linearity in the relationships. For the regressions involving the logging of land value, I had to discard the few blocks which registered a nominal decrease in land value during the period, which were predominately publicly-owned land for which land assessment is a dubious practice to begin with. Nevertheless, to test if this was altering any relationships, I ran a parallel linear regression excluding these same data points; for blocks inside Seattle city limits and within the UGB in King County, the R-squared, p-values and coefficients didn’t change for a singular variable when comparing the excluded data set with the non-altered data set.
First for the King County linear-linear regression:
[table]
Adjusted R Square=.25,B,Std. Error,p-value,
Distance to Transit,5.46,.21,0,
Distance to Downtown,-4.08,.06,0,
Distance to Centers,-1.87,.04,0,
Distance to Roads,-5.16,.29,0,
Distance to Airport,-.32,.04,0,
Distance to Water,-1.66,.07,0,
[/table]
This indicates that a one mile increase in distance from frequent transit, holding the other variables constant, will have been associated with a $5.46 increase in land value per square foot from 2000-2015.
Now for log-linear:
[table]
Adjusted R Square=.58,B,Std. Error,p-value,
Distance to Transit,.08,.03,0,
Distance to Downtown,-.16,0,0,
Distance to Centers,-.04,0,0,
Distance to Roads,-.14,.01,0,
Distance to Airport,.04,0,0,
Distance to Water,-.03,0,0,
[/table]
By logging land value, we see that a 1 mile increase in distance from transit would be associated with a 0.08% increase in land value from 2000-2015.
And log-log:
[table]
Adjusted R Square=.60,B,Std. Error,p-value,
Distance to Transit,-.21,.01,0,
Distance to Downtown,-.90,.01,0,
Distance to Centers,-.09,0,0,
Distance to Roads,.09,.01,0,
Distance to Airport,.50,.01,0,
Distance to Water,.04,.01,.002,
[/table]
But if we log the distance to transit, we then get the result that a 1% increase in distance from transit will have meant a 0.21% decrease in land value in the time period studied.
And now the linear-linear regression within Seattle:
[table]
Adjusted R Square=.29,B,Std. Error,p-value,
Distance to Transit,-14.71,1.32,0,
Distance to Downtown,-15.6,.24,0,
Distance to Centers,-13.59,1.59,0,
Distance to Roads,-3.67,.78,0,
Distance to Water,-3.28,.17,0,
Percent Improved Sidewalks,3.98,.61,0,
[/table]
Restricting the analysis to just the city proper, a one mile increase in distance from transit will decrease land value appreciation by $14.71.
And log-linear:
[table]
Adjusted R Square=.58,B,Std. Error,p-value,
Distance to Transit,-.33,.01,0,
Distance to Downtown,-.25,0,0,
Distance to Centers,-.11,.02,0,
Distance to Roads,.05,.01,0,
Distance to Water,-.1,0,0,
Percent Improved Sidewalks,.37,.01,0,
[/table]
This suggests a 39% (100*((e^-.33)-1)) decrease in land value appreciation for an additional mile in distance.
And log-log:
[table]
Adjusted R Square=.59,B,Std. Error,p-value,
Distance to Transit,-.09,.01,0,
Distance to Downtown,-.82,.01,0,
Distance to Centers,-.04,.01,0,
Distance to Roads,.04,.01,0,
Distance to Water,-.61,.01,0,
Percent Improved Sidewalks,.55,.01,0,
[/table]
And here, we get the data that a 1% increase in distance from transit equated to a 0.09% decrease in land value appreciation, 2000-2015.
So, what do we make of all this? First, statistics are pretty darn subjective. I just found a meta-analysis of studies on transit’s relationship to land value (Higgins and Kanaroglou, 2016), and these three basic models are each well-represented in the literature. Despite the plausibility of all of them, they turn out dramatically different results when examining land value change in Seattle. There are theoretical reasons to log variables related to land value; the bid rent curve does seem to rapidly slope away from downtown, causing this model’s overall explanatory power to be greatly enhanced by logging land value (around 60% of variance explained, rather than 25-30%). One would log distance to the other variables if you expect the benefits of proximity to decline even more rapidly, as it measures the percent change as a function of percent change. That high degree of non-linearity is precisely why the log-log regression flips the directionality of the relationship between land appreciation and transit for King County considered as a whole.
Overall, this investigation, like my independent study, points towards a consideration of transit’s highly spatially disaggregated effects in terms of land appreciation and gentrification. In Seattle, the areas best served by public transit are also those which have high upward shifts in class and land value. At this point, I don’t exactly feel comfortable putting a number to that relationship, considering how contingent the results are on the model chosen. It’s also worth noting that the Eastside, like Seattle, has seen land appreciation and a rise in socioeconomic status (if not gentrification, as it’s never been “low-status”), despite its bare-bones frequent transit network.
Methodologically, using blocks was a great compromise between granularity and usability, providing me with tens of thousands of data points to work with, while not becoming too much of a pain to edit. The slight aggregation has the added benefit of smoothing over a lot of the anomalous cases arising from using poorly-maintained taxlot data. Though there’s an abundance of studies looking at rail investments and land appreciation, the bulk of this literature fundamentally considers that appreciation an unambiguously positive thing. Fleshing out data on the negative impacts of such appreciation is an important next step for me. Additionally, for my thesis, I want to look into exploring some more sophisticated models to determine transit proximity and accessibility, using the actual street paths available instead of distance as the crow flies.
Leave a Reply