Coffee Talk, Coffee Walk!
Something we have heard from proponents of the Future Land Use Map, and in particular those who have a preference for dense living, is that they “would like to be able to walk to a coffee shop.” We don’t disagree: walkability is an important amenity for towns and cities. But how far short are we falling of this ideal in Charlottesville right now? The consultant materials do not provide any data on this, at least as far as we have seen. We wondered — how much of the city is truly outside of walking distance from a coffee shop? We therefore compiled a list of twenty coffee shops in Charlottesville or just outside the city limits. We then measured the distance from every parcel in the city to the nearest coffee shop. Now, we admit, this is an analysis that suffers from some limitations due to the data available to us and the mapping technology at our disposal. We are looking at linear distance, not road distance, and we aren’t considering elevation changes. Topographical challenges may make short linear distances difficult for people with limited mobility due to age or disability. All that said, we still think this is a useful exercise. If we found that very few parcels are within a kilometer of a coffee shop, we would take that as strong evidence that “walking to a coffee shop” remains merely aspirational. On the other hand, if we found that much of the city was within a short distance, then we could conclude that a large fraction of our residents already can enjoy this amenity.
Below is our map of the results. Green zones are within 750 meters of one of of the coffee shops on our list. Yellow zone are between 750 meters and 1.25 kilometers. Red zones are further than 1.25 kilometers. Our software measures from the nearest point of a parcel to each coffee store, so some large parcels (like parks) may all show one color, even though not all of that parcel is within the specific range.
Most of the city is within 750m of a coffee shop. Greenbrier appears to be the Sahara of Cville coffee deserts. It is also useful to look at the cumulative density function of “distance-to-coffee shop” by parcel. You can see that there are very few parcels beyond 1.5km. 1.5km is not exactly comfortable walking distance, but is certainly quite feasible on a bike or scooter.
In terms of population, we believe the function would look even better. Many more of the parcels in close proximity to one or more coffee shops are multi-family, whereas fewer parcels beyond 1.25km are single family. For parcels in the <750m category, 17% are multifamily residential (by state use code). For parcels in the >1.25km category, it is 15%.
We performed an analogous exercise with grocery stores, using 18 identified grocery stores in Charlottesville or just over the city line. The results were largely similar. The same caveats apply. The map below uses 750m (green), 1km (orange) and 1.25km (red) cut-offs.
Greenbrier and the Fontaine area once again appear to fare worst on walkability to an amenity. It is interesting to note that on both walkability to coffee shops and grocery stores, Venable (the northern part in particular) and Barracks/Rugby look somewhat isolated. These, along with Greenbrier, are neighborhoods where opposition to the FLUM has been strong. Therefore, it is not always the case that a neighborhood would willingly trade more walkability for greater density and presence of commercial activity. Below, we show the same cumulative density function as before, this time for grocery stores.
Again, relatively few parcels lie more than 1.5km from a grocery store. In this case, though, we see a different pattern of the prevalence of multifamily units in each category. While 13% of parcels in the <750m category are multifamily residential, 19% of the (smaller number) of parcels in the >1.25km category are multifamily residential.
Now, we noted above the limits of the “linear distance model.” As it turns out, we can (with some difficulty) access Google Maps to check walking times between addresses. We did this for grocery stores, though we think there is a likelihood that for some addresses we have plotted a walk that it not actually to the closest grocery store. Moreover, we think that’s Google Maps’ assumed walking speed is too low (anecdotally, your author just walked to lunch — Google Maps pegged the walk at 12 minutes, and your author just did it in 7.5 minutes). We go into more detail in the methodological note below, which also explains the cost limitations that kept us from doing this for more types of amenity and via the most accurate method possible. Nevertheless, the result is revealing. In the map below, areas in Green are within 12 minutes walking of a grocery store, in Yellow within 15 minutes, in Orange within 25 minutes and in Red above 25 minutes. Note, again, that in our anecdotal experience, Google’s walk time estimates are considerably higher than what we achieve in “real life”.
Most of the city’s residential parcels are in reasonable walking distance, by time, to groceries, confirming that our “linear distance” model was not misleading.
Below, we show the density function for walk-time to groceries. Over 80% of parcels are within 30 minutes walk of a grocery store. The median is 16 minutes. Again, if we consider residential units, this would probably look even better, since the “Green” zones are more heavily weighted toward MFH buildings and the “Red” zones are SFH or non-residential.
Below we perform the same analysis for coffee. An interactive version is available at the link.
The picture for walk time to coffee shops is a little worse than for groceries. Of course, while everybody needs groceries, not everybody needs coffee! The median walk-time to coffee is 17 minutes, by Google Maps’ reckoning, which, again, we figure is realistically more like 12-13 minutes. The spatial pattern is not dissimilar to that we saw for groceries.
Now, as much as we value groceries and coffee, we realize that looking at two specific amenities paints an incomplete picture. Luckily, walkability has become an important variable in the real estate world. Walkscore is a service that has developed an algorithm to generate a single score to summarize the “walkability” of the vast majority of addresses in America. You can read more about the Walkscore methodology here. A Walkscore ranges from 0 (least walkable) to 100 (most walkable). Below 25 is considered “highly car-dependent”, 25-49 “partly care dependent”, 50-69 “somewhat walkable”, and or above 70 is “highly walkable). Walkscore provides a ranking of cities above 200,000 in population. The average Walkscore for these cities is 48. Charlottesville is 59. Cities within 3 points of Charlottesville include Toronto, Denver, New Orleans and Anaheim. Charlottesville in the 4th most walkable (and incidentally 5th most bikeable) city in Virginia.
Below, you can see a map of Charlottesville colored by Walkscore. Red is below 25, Orange between 25 and 49, Yellow between 50 and 69, and Green 70 or above.
We see a similar pattern to our coarser analyses above. Most of the city is highly walkable. At the neighborhood level, some of the neighborhoods that have evinced the greatest opposition to the FLUM and which have been characterized as exclusionary are in fact the least walkable. Nearly all the “Sensitive Areas” in the FLUM plan are in the most walkable parts of the city. Yet again we see the inherent tension between the goal of protecting sensitive areas and the goal of promoting development in areas with easy access to amenities. And finally, again, we see a few areas, notably the southwest corner of the city, where lack of walkability may be a sign of exclusion.
The map shows the spatial distribution of walkability, but does not account for how many residential units are in each walkability “band.” We created a unit count by parcel and tabulated total units by Walkscore band. That picture also makes Charlottesville look relatively good on walkability.
That still raises the question of whether walkability is a feature that is reserved only for the most expensive properties. We took a look at this a few ways. First, we added WalkScore into our “hedonic housing model” regression for SFH. That regression explains 80% of variance of a property’s assessment. WalkScore (entered as a moderator of the living space and land area variables) did not explain much additional variance, but did come out as statistically significant. For an otherwise average single-family house, moving the Walkscore from 25 to 70 added 6.5% to the predicted assessed value. The raw correlation of WalkScore and assessment, however, was negative, explained by the fact that some of the lower-income neighborhoods in Charlottesville (student areas among them), with smaller and lower-graded houses, are the most walkable. We then performed the same analysis with our model for condominiums (which explains about 90% of variance). Here, Walkscore was a more powerful variable. For an otherwise median condo, moving the Walkscore from 25 to 70 added 22% to the predicted assessed value. Unlike with SFH, raw correlation was positive. We also looked at the median assessed value for single family homes and multifamily owner-occupied homes in each WalkScore “band.”
The pattern looks very different for SFH than for MFH. SFH tend to be of higher assessed value in the least walkable areas. MFH, on the other hand, shows much higher values for the highest walkability band. Along with our model results, this supports our hunch, which we have discussed elsewhere, that density is a preference. Those who choose to live in a single-family house often do so because they want privacy and space and they are willing to trade other amenities off against those values. Those who prefer multifamily housing prefer to be “in the action” and are willing to pay up for proximity to commercial amenities. In quantitative terms, the simple correlation of Walkscore and Assessed value is negative for SFH units and positive for MFH. The pattern raises serious questions about the underpinnings of RHI’s plan design. Their plan is predicated on the idea that expensive SFH neighborhoods are unique loci of opportunity and access to amenities, artificially pushed to low density by zoning. We see the opposite — expensive SFH neighborhoods are relatively car-dependent. It does not make a ton of sense, therefore, to site high-density MFH in these neigborhoods. On the other hand, we see the risk inherent in gentrification for walkable MFH neigborhoods. There is clearly a market for expensive units in the walkable core of the city. The right-most bar of the graph should give us pause. It shows the incredibly strong economic incentive to replace modest SFH in walkable areas with luxury MFH. More development of this kind risks displacing some of our more affordable SFH. Finally, we recognize that there is a pattern of some of the least expensive MFH sitting in areas that are not walkable or close to amenities. This is a form of exclusion which we agree, in line with the CP, represents a problem in need of a solution. Since that housing will continue to exist for the foreseeable future, rezoning or other encouragements for the development of more commercial amenities in areas such as the southwest corner of the city would seem appropriate.
Our analysis is limited by data quality and coarseness of distance measurement. But it does suggest that there should have been more analysis revealed (it may have been done, but we have never seen it) to support the idea that walkability to amenities is critically poor in Charlottesville. Likewise for the question of whether SFH neighborhoods are truly sitting on land that would otherwise be appropriate for MFH due to location and access. It is another example of how the Comprehensive Plan and FLUM have been buttressed at many points by just-so stories and conclusory assertions rather than by hard data and high-quality analysis.
We close with an observation on that point. We looked at the mean Walkscore for parcels in each FLUM category. Now, as you would expect (and hope) all the Node and High-Intensity categories had very high Walkscores. General Residential had a mean Walkscore of 47.8. Medium-Intensity Residential, you would expect to be somewhere in between. The reality: MIR had the lowest mean Walkscore, at 46.8. Strange, but, given the process that produced the map, perhaps not surprising.
Technical Note: Google Maps Distance Matrix
We derive our walking time estimates via Google’s Map Distance Matrix API. Unfortunately, the cost per element for calls against this API is not negligible. To get a walking distance from every parcel in Charlottesville vs 18 destinations (18 grocery stores) would represent over 300,000 elements, which would involve a cost of over $1,000. While some FLUM proponents seem to think that critics of the FLUM can just dip into the Scrooge McDuck pool of gold coins each one of us has in the basement of his or her mansion, we were not in a position to spend that kind of money on this analysis. After all, nobody paid us $1,000,000 to “Plan Together”, and sadly, the folks the city did pay that money to seem to have spent it on… something or other, but not this kind of analysis (heck, one PC Commissioner appears to have done more hard-core analysis that the whole consultant team, though not, to our knowledge, on this particular question). Therefore, we had to take a shortcut. We cut down on the number of calls by attempting to pre-select the likely closest grocery store to each plot. We did this by using our linear distance calculation for each plot/grocery pair to find the closest grocery by linear distance. We believe that in most cases, this will be the closest grocery by walking distance, too. However, we recognize that won’t always be the case. Therefore, we think the walking distance/time numbers we got are reasonably good, but possibly exaggerated for certain plots, where the closest grocery by linear distance is actually a longer walk due to dead-ends, rivers/creeks, or other natural barriers that force a roundabout walking route. Moreover, we think that Google Maps’ assumed walk speed , at 3 km per hour, is too slow. We have noticed anecdotally that we are able to walk routes in 20-30% less time than Google Maps predicts. Therefore, all things considered, the walk times in this analysis are highly conservative.