shrug-l: Scoring residential tax parcels for COVID-19 risk?
rick at clbcm.com
Fri Apr 17 16:27:33 EDT 2020
Thinking out loud here. Really just brainstorming. Sorry this is long
and a bit "wandering" but that's the nature of research.
How might we score residential tax parcels for COVID-19 risk
appropriately, with an eye towards limited surveillance testing resources?
First we need to know - is any resident of that tax parcel a COVID-19
*#1 Elderly individuals.*
*#2 Individuals with serious underlying health condition**s*
On the "Elderly" factor we have Census data, including the 2020 Census
that is related to a Master (postal) Address File and asks the question
specifically: (for each person living in that household) /"What is
Person 1's age and what is Person 1's date of birth?/...." Of course
Census data is for the most part totally unavailable at such a granular
level, this however might be an excellent "exception".
On a secondary basis we might proxy "age of resident" by number of years
the house was owned from when it was last purchased? That obviously
would miss a great number of elderly residents, renters, etc. *
**Qualified medical conditions *(Individuals with serious underlying
"Individuals with serious underlying health conditions, including
* high blood pressure
* chronic lung disease
* asthma, and
* those whose immune system is compromised such as by chemotherapy for
cancer and other conditions requiring such therapy."
Thinking off the top of the head, could prescription drug purchases
identify many of those individuals?
Of course confidentiality of CENSUS and health care information would be
a huge issue.
The above are just the two easiest to understand criteria.
Heroic assumption: the (one or more) "Vulnerable Person(s)" is fully
informed, responsible, sheltering in place and following all CDC
guidance. (later this assumption can be relaxed.) Then the question
becomes, */who else lives in that household/***and how "exposed" are
they during their typical day? Are they possibly asymptomatic or with
very mild symptoms, fully at risk of inadvertently bringing COVID-19
through the front door, exposing the 1+ /vulnerable /people who live there?
* a greater number of residents in the household who are "out and
about" increases the chance /any one /will carry COVID-19 in.
* if a resident is a front line health care worker the risk goes up
(but the awareness/knowledge/compliance might also be better?)
* if a resident works in an "essential business" there is a greater
chance they have been exposed on the job, or will be exposed
* has a resident traveled through / lingered in any known hot spot?
* more youthful/healthful residents are the MOST likely to carry, but
remain asymptomatic (and hence the least tested group so far.)
* if a resident is employed in a "non-essential business" they may
have been sheltering at home in the recent past, BUT are about to
return to work. How will all that play out?
We can easily factor in some great Census data at the Block Group level
to identify residential parcels that score high on demographic data that
is well known to correlate highly with COVID-19 severity, items such as:
* household income
* ethnic origin
* residential density
* urban / rural differentials
* air quality? (as a proxy for lung disease and asthma - e,g, LA, NY,
air quality measures?)
*Local transportation statistics*
I'm not familiar with this area at all so I can only speculate. I'm
aware of CENSUS commuter statistics regarding commuters living in one
county and working in another (commuting across county lines for work
Very large employers commuter maps may have been constructed as large
employment centers were modeled added or subtracted?
License plate capture on toll / traffic cameras?
Again, I have no solid background here but I speculate transportation
experts might have "network models" that could be juiced to contribute
an individual COVID-19 tax parcel risk score?
*Classic Marketing Data**
*Marketing in the USA is very sophisticated. Much is well known about
each postal address, including past purchases by product type, brand
preferences, estimated ages of residents at that address, etc. (Next to
CENSUS data this data set is pretty granular and accurate.)
For instance I'm sure its possible to get postal lists of heavy buyers
of OTC medications for allergy/asthma/lung disorders (or any of the
other conditions above). And that's just a start...
Credit card data could also be useful. Especially in real time.
*Cell phone location tracking**
They know where you are, when, and for how long you stayed. They know
where that phone sleeps. They infer if you are in a car vs. walking vs.
riding a bike vs. in a kayak....
Some of that data is already VERY widely available to marketers. Some
studies might be run by the "data holders" that do not actually reveal
/personal identity/, but may be linked (with relative ease) to postal
address. Obviously there are great sensitivities here. However if the
relationship were "many-to-many" it probably would fly.
Example to make it "many-to-many" - build up a "points system":
* traveled to any of the following locations and remained more than 1
day in the past 90 days: 10 points
* traveled to and spent time at a known drug store 3 standard
deviations higher than average in the past month: 5 points
* owns a cell phone and uses it more than x times per week but
otherwise shows very little movement around, does not use much data.
(proxy for obesity and/or old age). 15 points
* watches more than 4 hours of daytime TV. 15 points
* does not have a cell phone and does not have any form of Internet,
yet lives in a wired/urban location: 20 points
* ...more like above
Data vendor runs the analysis and gives you the top scoring 5% of your
county population, as just a list of post office addresses, given back
to you in random order.
County sends post cards... "/Your household may be at elevated risk -
more info available by phone or on the Internet...FREE CONFIDENTIAL
TESTING. Only requires a mouth swab, available TODAY! Takes only 15
Any other ideas how to create a tax parcel score for COVID-19 risk?
Above of course would be pretty much a hypothesis to begin with. Would
want to close the loop with actual surveillance testing results. Then
"train the model" further with even wider data to try to improve the pin
Richard J. Labs, CFA, CPA
CL&B Capital Management, LLC
E-mail (preferred for efficiency): rick at clbcm.com
3213 Yorktown Dr, Tallahassee, FL 32312-2015
June-August: 408B Holiday Harbour, Canandaigua, NY 14424
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the SHRUG-L