Wow - this is fantastic! Just quickly skimming the map, something I noticed is that a lot of small town centres perceive themselves as ‘suburban’, even though the word technically denotes an area surrounding a city.
Here is a description of the methodology.
The Urbanization Perceptions Small Area Index: An Application of Machine Learning and Small Area Estimation to Household Survey Data (.docx)There was a bonus question to the 2017 American Household Survey that was conducted by the Census Bureau for the the Department of Housing and Urban Development (HUD) which is a moderately large survey (55,000 housing units). The AHS is a biennial survey, which oversamples large metropolitan areas (e.g. part of the intent is to be able to characterize housing conditions in Detroit vs. Dallas, etc.).
The bonus question was:
Which one of the following best describes the area of your(this)home?
1.Urban
2.Suburban
3.Rural
If you had replied "small town center", you would be classified as non-responding.
To someone in New York City or Nassau County, NY the answer is obvious. To someone in Andrews, TX the answer is not obvious. They clearly are not rural. Almost everyone (75%) of the county residents live in the city. Rural is the empty space with nothing but windmills, oil wells, and some cattle (no water for crops). Urban might mean something like Midland or Odessa, but more likely Dallas or Houston. The people in Andrews would be perfectly capable of describing their place, but that particular question would not reveal an answer.
They then took the responses and developed a predicted response that would match actual response most of the time based on 21 variables. The strongest predictors were population and housing density. Moderate were size of incorporated place, and percentage of single family detached housing. Demographic data such as age, race, and ethnicity were weaker (these are aggregate values, e.g. percentage of population under 18, 18-24, etc.).
Suburbs probably have more under 18, less 18-24, and then more for older ages - kids go off to college, then get a job in the big city. If they marry and have children, they move to the suburbs (affordable housing, safer, better schools).
Census tracts have an ideal population of 4000 persons or 1500 housing units. Those with populations larger than 8000 should be split, those with less than 1500(?) should be merged. But the Census Bureau would prefer that they be stable over the decades since they are intended to be used for statistical purposes for small areas.
They were originally intended for large cities, in particular NYC. Rather than get one huge number for the entire city, or very big numbers for each borough, you would get 100s of numbers for small areas of the city.
Since they are used for statistical purposes, you want at least some coherence. You don't want an area where half is farmland, and the other half Stalinist apartment blocks. You want one to be 90% Stalinist apartment blocks, and the other 90% soybeans - at least as far as statistical characterization.
So in more densely populated areas, census tracts are from 1/2 to 3 square miles. If they contain 4000 persons, this is a density of 1300 to 8000 persons per square mile. Depending on how much of the area is residential and not shopping centers, schools, churches, offices, factories, parks, undeveloped, freeway easements, etc. the residential density will be somewhat higher (i.e. residential lots of less than an acre). Clearly not farmlands, but not so clearly suburban.
In Andrews County there are four census tracts. Two are in the city, and the other two are outside, though they come right up to the city limits. The two in the city are around two square miles each.
The density of Andrews, TX is 2263 ppm.
Census Tracts in Andrews County, TX (PDF)Some densities from Hennepin County, MN
Minneapolis, MN is 7088.
Brooklyn Center 3784
Robbinsdale 4999
Crystal 3833
New Hope 4038
Golden Valley 1998
St. Louis Park 4254
Edina 3163
Richfield 5131
Brooklyn Park 2906
Plymouth 2159
Minnetonka 1847
Eden Prairie 1873
Bloomington 2390
Champlin 2826
Dayton 200
Rogers 1067
Corcoran 150
Medina 192
Orono 465
The census tracts in Andrews are a bit higher since the city limits lap out into the surrounding county census tracts. The census tracts in Minnesota are what you get, the cities are 36 square miles, and undeveloped spaces are included in the cities and census tracts.
So Andrews, from a density standpoint, looks like a second-tier Minneapolis suburb.
So imagine they asked someone in Plymouth
Which one of the following best describes the area of your(this)home?
1.Urban
2.Suburban
3.Rural
98% said suburban (I made this number up, but it might be true).
From this they conclude that persons in census tracts with a population density of 2159 ppm perceive themselves as suburban. They then look at the census tracts in Andrews, Texas and since they have a population density of 2263 they too will perceive themselves as suburban.
If you look at west Texas you will see all kinds of small cities where the census tracts have been drawn tightly around the cities, and then the rest of the county in one or two tracts. This is the correct statistical division. The city/town dwellers are different than ranchers or people who live on an acreage.
One thing the paper notes is that their prediction method over-predicts the dominant response. Since about half the persons responded "suburban", this meant
more than half of the projections would be suburban. To correct for this, they classified the tracts that were most predicted to be urban as "urban" and those most predicted to be "rural" as rural.
These are the results for Andrews County. CT 9502 and 9503 are the two for the city of Andrews.
Tract | Housing Units | UPSAI_urban | UPSAI_suburban | UPSAI_rural | UPSAI_Category |
9501 | 760 | 6% | 31% | 63% | 3 |
9502 | 2295 | 30% | 66% | 3% | 2 |
9503 | 1441 | 39% | 57% | 4% | 1 |
9504 | 917 | 7% | 1% | 93% | 3 |
For Andrews, TX CT 9503, the prediction is that 57% would describe the area of their home as suburban, 39% as urban, and 4% rural. Dr. RI's maps show the tract as a 57% shade of green. But the study classified CT 9503 as "urban" because 39% urban was among the higher predictions for urban. Their map would have shown it as solid red - with no shading distinction.