Assessing Charlottesville property assessments
Residential property assessments in Charlottesville rose by 11.7% in 2022, and local officials and residents are concerned. Is this an unusual change in assessments? Which properties saw the greatest increases, and why?
We can find the relevant data at the city's Open Data Portal. For simplicity, I'm going to treat parcels as residential if they have an entry in the residential details table. I'm also going to look up at to 20 years of data, so I'm only including homes built after 2002.
# Hack: Ignore shapely/geos version mismatch warning
import warnings
warnings.simplefilter("ignore")
import altair as alt
import branca
import folium
import numpy as np
import seaborn as sns
import statsmodels.formula.api as smf
from google.cloud import bigquery
bq = bigquery.Client()
First we can look at average assessments over time:
raw_assessments_by_year = bq.query(
"""
select
assessments.taxyear,
avg(assessments.totalvalue) as totalvalue,
count(assessments.totalvalue) as count,
from `cvilledata.cville_open_data.real_estate_all_assessments` assessments
join `cvilledata.cville_open_data.real_estate_residential_details` residential on assessments.parcelnumb = residential.parcelnumb
where assessments.taxyear >= 2002
and residential.yearbuilt <= 2002
group by assessments.taxyear
order by assessments.taxyear desc
"""
).result().to_dataframe()
alt.Chart(raw_assessments_by_year).mark_bar().encode(
y=alt.Y("totalvalue:Q"),
x=alt.X("taxyear:N", sort="x"),
tooltip="totalvalue:Q",
).properties(
title="Mean assessment by year",
)
Assessments have grown substantially over the past 20 years, almost tripling since 2002. But this doesn't tell us whether the change in values in 2022 was typical or not. Let's look at proportion change in assessments relative to the previous year over the past 20 years:
assessments_by_year = bq.query(
"""
with assessments as (
select
details.parcelnumb as parcelnumber,
details.geoparceli as gpin,
assessments.taxyear,
assessments.totalvalue,
lag(assessments.totalvalue, 1) over (partition by details.parcelnumb order by assessments.taxyear) as totalvalueprev,
safe_divide(
assessments.totalvalue - lag(assessments.totalvalue, 1) over (partition by details.parcelnumb order by assessments.taxyear),
lag(assessments.totalvalue, 1) over (partition by details.parcelnumb order by assessments.taxyear)
) as totalvaluechange,
from `cvilledata.cville_open_data.parcel_area_details` details
join `cvilledata.cville_open_data.real_estate_all_assessments` assessments on details.parcelnumb = assessments.parcelnumb
)
select
assessments.taxyear,
avg(assessments.totalvalue) as totalvalue,
avg(assessments.totalvaluechange) as totalvaluechange,
count(assessments.totalvalue) as count,
from assessments
join `cvilledata.cville_open_data.real_estate_residential_details` residential on assessments.parcelnumber = residential.parcelnumb
where assessments.taxyear >= 2002
and residential.yearbuilt < 2002
group by assessments.taxyear
order by assessments.taxyear desc
"""
).result().to_dataframe()
alt.Chart(assessments_by_year).mark_bar().encode(
y=alt.Y("totalvaluechange:Q"),
x=alt.X("taxyear:N", sort="x"),
tooltip="totalvaluechange:Q",
).properties(
title="Mean assessment proportion change by year",
)
Property values rose about 12.4% in 2022 relative to 2021. That's a large increase relative to the past decade, but assessments rose faster every year from 2003 to 2007. Assessments also rose relatively slowly in 2020 and 2021; we'll say more about this later.
Next, we can map changes in assessments by parcel across the city:
assessments_by_parcel = bq.query(
"""
with assessments as (
select
details.parcelnumb as parcelnumber,
details.geoparceli as gpin,
assessments.taxyear,
assessments.totalvalue,
lag(assessments.totalvalue, 1) over (partition by details.parcelnumb order by assessments.taxyear) as totalvalueprev,
safe_divide(
assessments.totalvalue - lag(assessments.totalvalue, 1) over (partition by details.parcelnumb order by assessments.taxyear),
lag(assessments.totalvalue, 1) over (partition by details.parcelnumb order by assessments.taxyear)
) as totalvaluechange,
from `cvilledata.cville_open_data.parcel_area_details` details
join `cvilledata.cville_open_data.real_estate_all_assessments` assessments on details.parcelnumb = assessments.parcelnumb
), sales as (
select
*,
count(*) over (partition by bookpage) as parcelcount,
from `cvilledata.cville_open_data.real_estate_sales`
)
select
assessments.parcelnumber,
assessments.taxyear,
assessments.totalvalue,
assessments.totalvaluechange,
concat(details.streetnumb, ' ', details.streetname) as address,
neighborhoods.name as neighborhoodname,
sales.saleamount is not null as soldprevyear,
st_geogfromgeojson(details.geometry) as geometry,
from assessments
join `cvilledata.cville_open_data.parcel_area_details` details on assessments.parcelnumber = details.parcelnumb
join `cvilledata.cville_open_data.real_estate_residential_details` residential on assessments.parcelnumber = residential.parcelnumb
join `cvilledata.cville_open_data_derived.geopin_to_assessment_neighborhood` neighborhoods on assessments.gpin = neighborhoods.gpin
left join sales on assessments.parcelnumber = sales.parcelnumb and sales.saleamount > 0 and sales.parcelcount = 1 and extract(year from sales.saledate) = assessments.taxyear - 1
where assessments.taxyear = 2022
and residential.yearbuilt < 2002
"""
).result().to_geodataframe()
parcel_colormap = branca.colormap.linear.YlOrRd_09
parcel_color_values = assessments_by_parcel["totalvaluechange"][~assessments_by_parcel["totalvaluechange"].isna()]
parcel_colormap = parcel_colormap.scale(*list(np.percentile(parcel_color_values, [5, 95])))
def style(feature):
color_value = feature["properties"]["totalvaluechange"]
if color_value is None or np.isnan(color_value):
fill_color = "white"
else:
fill_color = parcel_colormap.rgb_hex_str(color_value)
return {
"color": "black",
"weight": 0.05,
"fillOpacity": 0.8,
"fillColor": fill_color,
}
parcel_map = folium.Map(
location=[38.04, -78.49],
tiles="cartodbpositron",
zoom_start=13.5,
zoom_delta=0.5,
)
folium.GeoJson(
assessments_by_parcel,
tooltip=folium.GeoJsonTooltip(fields=["parcelnumber", "address", "totalvalue", "totalvaluechange", "neighborhoodname", "soldprevyear"], labels=True),
style_function=style,
).add_to(parcel_map)
parcel_colormap.add_to(parcel_map)
parcel_map
One striking observation from this map is that assessment changes seem consistent within neighborhoods. For example, assessment increases were generally low in North Downtown, moderate in Greenbrier and Locust Grove, and high in Orangedale.
The City Assessor groups parcels by planning neighborhood. To simplify, we can map assessment changes after aggregating by neighborhood:
assessments_by_neighborhood = bq.query(
"""
with assessments as (
select
details.parcelnumb as parcelnumber,
details.geoparceli as gpin,
assessments.taxyear,
assessments.totalvalue,
lag(assessments.totalvalue, 1) over (partition by details.parcelnumb order by assessments.taxyear) as totalvalueprev,
safe_divide(
assessments.totalvalue - lag(assessments.totalvalue, 1) over (partition by details.parcelnumb order by assessments.taxyear),
lag(assessments.totalvalue, 1) over (partition by details.parcelnumb order by assessments.taxyear)
) as totalvaluechange,
from `cvilledata.cville_open_data.parcel_area_details` details
join `cvilledata.cville_open_data.real_estate_all_assessments` assessments on details.parcelnumb = assessments.parcelnumb
), sales as (
select
*,
count(*) over (partition by bookpage) as parcelcount,
from `cvilledata.cville_open_data.real_estate_sales`
)
select
avg(assessments.totalvaluechange) as totalvaluechange,
neighborhoods.neighhood as neighborhoodname,
st_union_agg(st_geogfromgeojson(neighborhoods.geometry)) as geometry,
from assessments
join `cvilledata.cville_open_data.parcel_area_details` details on assessments.parcelnumber = details.parcelnumb
join `cvilledata.cville_open_data.real_estate_residential_details` residential on assessments.parcelnumber = residential.parcelnumb
join `cvilledata.cville_open_data_derived.geopin_to_assessment_neighborhood` gpintoneighborhoods on assessments.gpin = gpintoneighborhoods.gpin
join `cvilledata.cville_open_data.assessment_neighborhoods` neighborhoods on gpintoneighborhoods.name = neighborhoods.neighhood
left join sales on assessments.parcelnumber = sales.parcelnumb and sales.saleamount > 0 and sales.parcelcount = 1 and extract(year from sales.saledate) = assessments.taxyear - 1
where assessments.taxyear = 2022
and residential.yearbuilt < 2002
group by neighborhoodname
"""
).result().to_geodataframe()
nhood_colormap = branca.colormap.linear.YlOrRd_09
nhood_color_values = assessments_by_neighborhood["totalvaluechange"][~assessments_by_neighborhood["totalvaluechange"].isna()]
nhood_colormap = nhood_colormap.scale(*list(np.percentile(nhood_color_values, [5, 95])))
def style(feature):
color_value = feature["properties"]["totalvaluechange"]
if color_value is None or np.isnan(color_value):
fill_color = "white"
else:
fill_color = nhood_colormap.rgb_hex_str(color_value)
return {
"color": "black",
"weight": 0.05,
"fillOpacity": 0.8,
"fillColor": fill_color,
}
nhood_map = folium.Map(
location=[38.04, -78.49],
tiles="cartodbpositron",
zoom_start=13.5,
zoom_delta=0.5,
)
folium.GeoJson(
assessments_by_neighborhood,
tooltip=folium.GeoJsonTooltip(fields=["totalvaluechange", "neighborhoodname"], labels=True),
style_function=style,
).add_to(nhood_map)
nhood_colormap.add_to(nhood_map)
nhood_map
Here are the same values ranked by neighborhood, from greatest to least proportion change:
assessments_by_neighborhood_2022 = bq.query(
"""
with assessments as (
select
details.parcelnumb as parcelnumber,
details.geoparceli as gpin,
assessments.taxyear,
assessments.totalvalue,
lag(assessments.totalvalue, 1) over (partition by details.parcelnumb order by assessments.taxyear) as totalvalueprev,
safe_divide(
assessments.totalvalue - lag(assessments.totalvalue, 1) over (partition by details.parcelnumb order by assessments.taxyear),
lag(assessments.totalvalue, 1) over (partition by details.parcelnumb order by assessments.taxyear)
) as totalvaluechange,
from `cvilledata.cville_open_data.parcel_area_details` details
join `cvilledata.cville_open_data.real_estate_all_assessments` assessments on details.parcelnumb = assessments.parcelnumb
)
select
neighborhoods.name as neighborhoodname,
avg(assessments.totalvalue) as totalvalue,
avg(assessments.totalvaluechange) as totalvaluechange,
count(1) as count,
from assessments
join `cvilledata.cville_open_data.real_estate_residential_details` residential on assessments.parcelnumber = residential.parcelnumb
join `cvilledata.cville_open_data_derived.geopin_to_assessment_neighborhood` neighborhoods on assessments.gpin = neighborhoods.gpin
where assessments.taxyear = 2022
and residential.yearbuilt < 2002
group by neighborhoodname
order by neighborhoodname
"""
).result().to_dataframe()
alt.Chart(assessments_by_neighborhood_2022[assessments_by_neighborhood_2022["count"] >= 15]).mark_bar().encode(
x=alt.X("totalvaluechange:Q"),
y=alt.Y("neighborhoodname:N", sort="-x"),
color=alt.Color("neighborhoodname:N", legend=None),
tooltip="totalvaluechange:Q",
).properties(
title="Mean assessment proportion change by neighborhood, tax year 2022",
)
Property assessments in the Orangedale neighborhood rose by 24.7%, about double the rate of the city overall. Why? One partial explanation may be that the Orangedale neighborhood saw near-zero changes in assessments in 2021, 2018, and 2016. This doesn't seem to be the case for the adjacent Forest Hills neighborhood, although the Birdwood neighborhood also appears to have skipped assessments in 2021.
assessments_by_neighborhood_by_year = bq.query(
"""
with assessments as (
select
details.parcelnumb as parcelnumber,
details.geoparceli as gpin,
assessments.taxyear,
assessments.totalvalue,
lag(assessments.totalvalue, 1) over (partition by details.parcelnumb order by assessments.taxyear) as totalvalueprev,
safe_divide(
assessments.totalvalue - lag(assessments.totalvalue, 1) over (partition by details.parcelnumb order by assessments.taxyear),
lag(assessments.totalvalue, 1) over (partition by details.parcelnumb order by assessments.taxyear)
) as totalvaluechange,
from `cvilledata.cville_open_data.parcel_area_details` details
join `cvilledata.cville_open_data.real_estate_all_assessments` assessments on details.parcelnumb = assessments.parcelnumb
)
select
neighborhoods.name as neighborhoodname,
assessments.taxyear,
avg(assessments.totalvalue) as totalvalue,
avg(assessments.totalvaluechange) as totalvaluechange,
count(1) as count,
from assessments
join `cvilledata.cville_open_data.real_estate_residential_details` residential on assessments.parcelnumber = residential.parcelnumb
join `cvilledata.cville_open_data_derived.geopin_to_assessment_neighborhood` neighborhoods on assessments.gpin = neighborhoods.gpin
where assessments.taxyear >= 2002
and residential.yearbuilt < 2002
group by neighborhoodname, taxyear
order by neighborhoodname
"""
).result().to_dataframe()
neighborhoods = [
"Birdwood/Hillcrest Road",
"Forest Hills",
"Orangedale",
]
assessments_by_neighborhood_by_year_filtered = assessments_by_neighborhood_by_year[
(assessments_by_neighborhood_by_year.neighborhoodname.isin(neighborhoods))
& (assessments_by_neighborhood_by_year.taxyear >= 2012)
]
alt.Chart(assessments_by_neighborhood_by_year_filtered).mark_bar().encode(
y=alt.Y("totalvaluechange:Q"),
x=alt.X("taxyear:N", sort="x"),
color=alt.Color("neighborhoodname:N", legend=None),
column="neighborhoodname:N",
tooltip="totalvaluechange:Q",
).properties(
title="Mean assessment proportion change by neighborhood by year, selected neighborhoods",
)
Here's a detailed view of a single parcel in the Orangedale neighborhood. Its assessment stayed constant from 2020 to 2021, from 2017 to 2018, and from 2011 to 2016. Did the true value of that parcel remain unchanged for those years, or is this an artifact of the assessor's methodology?
assessments_by_year_example = bq.query(
"""
with assessments as (
select
details.parcelnumb as parcelnumber,
details.geoparceli as gpin,
assessments.taxyear,
assessments.totalvalue,
lag(assessments.totalvalue, 1) over (partition by details.parcelnumb order by assessments.taxyear) as totalvalueprev,
safe_divide(
assessments.totalvalue - lag(assessments.totalvalue, 1) over (partition by details.parcelnumb order by assessments.taxyear),
lag(assessments.totalvalue, 1) over (partition by details.parcelnumb order by assessments.taxyear)
) as totalvaluechange,
from `cvilledata.cville_open_data.parcel_area_details` details
join `cvilledata.cville_open_data.real_estate_all_assessments` assessments on details.parcelnumb = assessments.parcelnumb
)
select
*
from assessments
where assessments.taxyear >= 2002
and assessments.parcelnumber = '240159000'
"""
).result().to_dataframe()
alt.Chart(assessments_by_year_example).mark_bar().encode(
y=alt.Y("totalvalue:Q"),
x=alt.X("taxyear:N", sort="x"),
tooltip="totalvalue:Q",
).properties(
title="Assessment value for an example parcel, Orangedale neighborhood",
)
Another interesting trend from the maps above is that, while changes in assessment were mostly consistent within neighborhoods, many neighborhoods included a handful of properties with unusually high assessment increases. Hover over those parcels and you'll see that many share one factor in common: they were sold in 2021. Let's compare changes in assessed value for all homes sold in the previous year compared to homes that didn't sell over time:
assessments_by_prev_sale = bq.query(
"""
with assessments as (
select
details.parcelnumb as parcelnumber,
details.geoparceli as gpin,
assessments.taxyear,
assessments.totalvalue,
lag(assessments.totalvalue, 1) over (partition by details.parcelnumb order by assessments.taxyear) as totalvalueprev,
safe_divide(
assessments.totalvalue - lag(assessments.totalvalue, 1) over (partition by details.parcelnumb order by assessments.taxyear),
lag(assessments.totalvalue, 1) over (partition by details.parcelnumb order by assessments.taxyear)
) as totalvaluechange,
from `cvilledata.cville_open_data.parcel_area_details` details
join `cvilledata.cville_open_data.real_estate_all_assessments` assessments on details.parcelnumb = assessments.parcelnumb
), sales as (
select
*,
count(*) over (partition by bookpage) as parcelcount,
from `cvilledata.cville_open_data.real_estate_sales`
)
select
assessments.taxyear,
sales.saleamount is not null as soldprevyear,
avg(assessments.totalvaluechange) as totalvaluechange,
count(1) as count,
from assessments
join `cvilledata.cville_open_data.real_estate_residential_details` residential on assessments.parcelnumber = residential.parcelnumb
join `cvilledata.cville_open_data_derived.geopin_to_assessment_neighborhood` neighborhoods on assessments.gpin = neighborhoods.gpin
left join sales on assessments.parcelnumber = sales.parcelnumb and sales.saleamount > 0 and sales.parcelcount = 1 and extract(year from sales.saledate) = assessments.taxyear - 1
where assessments.taxyear > 2012
and residential.yearbuilt < 2002
group by assessments.taxyear, soldprevyear
order by assessments.taxyear desc
"""
).result().to_dataframe()
alt.Chart(assessments_by_prev_sale).mark_bar().encode(
y=alt.Y("totalvaluechange:Q"),
x=alt.X("soldprevyear:N", sort="-x", axis=None),
color=alt.Color("soldprevyear:N"),
column=alt.Column("taxyear:N", sort="ascending"),
tooltip="totalvaluechange:Q",
).properties(
title="Mean assessment proportion change by previous sale by year",
).configure_view(
stroke="transparent",
)
For each of the past ten years, parcels that had sold in the previous year saw greater increases in assessments than properties that didn't sell. We can dig in further and compare properties that sold for more than their previous assessment versus properties that sold at or under assessment:
assessments_by_sale_category = bq.query(
"""
with assessments as (
select
details.parcelnumb as parcelnumber,
details.geoparceli as gpin,
assessments.taxyear,
assessments.totalvalue,
lag(assessments.totalvalue, 1) over (partition by details.parcelnumb order by assessments.taxyear) as totalvalueprev,
safe_divide(
assessments.totalvalue - lag(assessments.totalvalue, 1) over (partition by details.parcelnumb order by assessments.taxyear),
lag(assessments.totalvalue, 1) over (partition by details.parcelnumb order by assessments.taxyear)
) as totalvaluechange,
from `cvilledata.cville_open_data.parcel_area_details` details
join `cvilledata.cville_open_data.real_estate_all_assessments` assessments on details.parcelnumb = assessments.parcelnumb
), sales as (
select
*,
count(*) over (partition by bookpage) as parcelcount,
from `cvilledata.cville_open_data.real_estate_sales`
)
select
assessments.taxyear,
case
when sales.saleamount is not null and sales.saleamount > assessments.totalvalueprev then 'soldoverassessment'
when sales.saleamount is not null then 'soldatorunderassessment'
else 'notsold'
end as salecategory,
avg(assessments.totalvaluechange) as totalvaluechange,
count(1) as count,
from assessments
join `cvilledata.cville_open_data.real_estate_residential_details` residential on assessments.parcelnumber = residential.parcelnumb
join `cvilledata.cville_open_data_derived.geopin_to_assessment_neighborhood` neighborhoods on assessments.gpin = neighborhoods.gpin
left join sales on assessments.parcelnumber = sales.parcelnumb and sales.saleamount > 0 and sales.parcelcount = 1 and extract(year from sales.saledate) = assessments.taxyear - 1
where assessments.taxyear > 2012
and residential.yearbuilt < 2002
group by assessments.taxyear, salecategory
order by assessments.taxyear desc
"""
).result().to_dataframe()
alt.Chart(assessments_by_sale_category).mark_bar().encode(
y=alt.Y("totalvaluechange:Q"),
x=alt.X("salecategory:N", sort="-x", axis=None),
color=alt.Color("salecategory:N"),
column=alt.Column("taxyear:N", sort="ascending"),
tooltip="totalvaluechange:Q",
).properties(
title="Mean assessment proportion change by previous sale category by year",
).configure_view(
stroke="transparent",
)
It turns out that assessment increases for recently-sold properties are mostly driven by properties that sold over assessment. These properties saw average assessment increases of over 20% in 2020, 2021, and 2022. Meanwhile, parcels that sold under assessment generally saw increased assessments, but their assessments grew less than properties that didn't sell at all.
Next, we can look at assessment changes as a function of previous-year sale prices. Do homes that sold for more relative to previous assessments also see higher assessment changes the following year?
assessments_by_sale_category_by_parcel = bq.query(
"""
with assessments as (
select
details.parcelnumb as parcelnumber,
details.geoparceli as gpin,
assessments.taxyear,
assessments.totalvalue,
lag(assessments.totalvalue, 1) over (partition by details.parcelnumb order by assessments.taxyear) as totalvalueprev,
safe_divide(
assessments.totalvalue - lag(assessments.totalvalue, 1) over (partition by details.parcelnumb order by assessments.taxyear),
lag(assessments.totalvalue, 1) over (partition by details.parcelnumb order by assessments.taxyear)
) as totalvaluechange,
from `cvilledata.cville_open_data.parcel_area_details` details
join `cvilledata.cville_open_data.real_estate_all_assessments` assessments on details.parcelnumb = assessments.parcelnumb
), sales as (
select
*,
count(*) over (partition by bookpage) as parcelcount,
from `cvilledata.cville_open_data.real_estate_sales`
)
select
assessments.parcelnumber,
neighborhoods.name as neighborhoodname,
concat(details.streetnumb, ' ', details.streetname) as address,
assessments.totalvaluechange as totalvaluechange,
safe_divide(sales.saleamount - assessments.totalvalueprev, assessments.totalvalueprev) as overassessmentchange,
from assessments
join `cvilledata.cville_open_data.parcel_area_details` details on assessments.parcelnumber = details.parcelnumb
join `cvilledata.cville_open_data.real_estate_residential_details` residential on assessments.parcelnumber = residential.parcelnumb
join `cvilledata.cville_open_data_derived.geopin_to_assessment_neighborhood` neighborhoods on assessments.gpin = neighborhoods.gpin
left join sales on assessments.parcelnumber = sales.parcelnumb and sales.saleamount > 0 and sales.parcelcount = 1 and extract(year from sales.saledate) = assessments.taxyear - 1
where assessments.taxyear = 2022
and residential.yearbuilt < 2002
and sales.saleamount is not null
"""
).result().to_dataframe()
sns.lmplot(
data=assessments_by_sale_category_by_parcel,
x="overassessmentchange",
y="totalvaluechange",
).set(
title="Assessment changes by sale prices relative to previous assessments",
)
Unsurprisingly, houses that sold relatively high over their previous assessment also saw relatively large increases in assessed value the following year. However, the slope of the best-fit line is about 0.57, meaning that a parcel that sold for 10% more than its previous assessment saw its assessed value rise by about 0.057%. In other words, even though new homeowners are likely to experience relatively large increases in home assessments, those assessments may still fall short of the price they actually paid.
model = smf.ols(formula="totalvaluechange ~ overassessmentchange", data=assessments_by_sale_category_by_parcel)
res = model.fit()
res.summary()
Our simple model explains future assessment changes fairly well, accounting for about 59% of the variance in assessment changes for parcels that sold the previous year. However, there are some outliers from the best fit line; we can draw an interactive plot (hover over points for details) and color points by neighborhood to look for an explanation, although nothing jumps out at me:
alt.Chart(assessments_by_sale_category_by_parcel).mark_circle(size=60).encode(
y=alt.Y("totalvaluechange:Q"),
x=alt.X("overassessmentchange:Q"),
color=alt.Color("neighborhoodname:N"),
tooltip=["address:N", "overassessmentchange:Q", "totalvaluechange:Q"],
).properties(
title="Assessment changes by sale prices relative to previous assessments",
).interactive()
Well, this was fun! My main takeaway is something like "assessments are confusing", but here are a few more specific observations:
- The increase in assessments in 2022 is high relative to the past few years, but we aren't replaying 2007 yet.
- Changes in assessments are mostly consistent within neighborhoods. For most neighborhoods, the assessor seems to adjust assessments on most properties by roughly the same proportion. Except...
- Properties that sold during the previous year, and especially properties that sold for more than their previous assessments, increased in value much more than nearby properties.
- Some parcels had identical values across two or more years, and sometimes values remained the same for most of a neighborhood.
Did I miss anything? Find a mistake in a query? Let me know @whatthecarp.