Census population estimates 2011-2019

[jimrtex]

https://www.mercurynews.com/2020/07/08/study-in-a-first-california-poised-to-lose-house-seats/?utm_source=twitter.com&utm_campaign=socialflow&utm_medium=social&utm_content=tw-mercnews

It smells like clickbait, but at the same time California has a large number of 'hard to count' census regions, mostly areas in the valley with Rural Hispanic populations. The article mostly talks about the effects and why it might happen, corona and ensuing response rates, rather than the math behind it - typical of puff pieces. So, could our resident calculators like Jimrtex and Cinyc voice in on the scenario outlined about California losing two seats.

It was questionable whether California would lose one seat, but that was largely confirmed if you took into account the declining rate of growth toward the end of the decade.

Here is the news release that the article is based on.


Southern California Faces Loss of One Congressional District, May narrowly hold on to second

while narrowly holding onto the 52nd district by only 1,324 people over Arizona and 3,248 over Minnesota.

I think what they have done here is used the quotient for the 51st seat. For example, my projections show that California is 222,000 below the number needed to keep a 53rd seat, but the quotient would be 4,228 below that needed. The report would lead you to believe that California would only need to lose 1,324 persons, but it would actually be 51.497 (sqrt(51*52) times that. My estimate assumes continuation of the average 2010-2019 rate for the last 3/4 of a year before the census. This is not likely given the slowdown in California growth/increase in domestic outflow. But it is over a short period. As we approach the census, errors in projection become less significant (and we don't really have the capability to make better estimates than the census bureau). I still, assuming the 2019 estimate is reasonably accurate, believe California will have 52 seats, but this assumes an error of less than 1% in the Census estimate.

Note that like the news article bungles Southern California and California together.

This may be the report underlying the news release. Note if you click on Download you can get the report in PDF format.

Winners and Losers: The 2020 Census and California's 2021 Redistricting

The underlying estimates are based on the 2017 one-year estimates projected forward to 2020. They make a small error by not accounting for estimates being based on July 1, and the census occurring on April 1. The ACS is a year-round collection, with July 1 being midpoint of the year.

A better projection would use the 2019 county estimates dividing by 9.25 and projecting forward by adding 0.75 per year.

They divide Los Angeles into 5 regions. You could make projections based on the 2018 ACS, project forward to 2020, and then control the regional populations to the countywide projection.

Their estimates for congressional representation take into account a loss of one representative for California, but due to rounding to 0.1 districts, the 2010 counts total 52.8, leaving the loss of 0.2 districts unaccounted for.

The only ESRI report suggesting a loss of another seat that I could find was based on a supposition of 4 seats being apportioned to a state of Puerto Rico.

Apportionment for California is kind of weird. It gains/loses a seat for about every 2% gain or loss relative to the US (1/52 = 1.92%). But apportionment is among all 50 states, so while the quotients for California are evenly spaced, that of the collective 50 states is not. So on the priority list California might be 435th, 431st, and 419th, rather than a regular spacing of every 8th or 9th seat.

The story is naive about losing a particular district. Imagine that if in 2010, California had been informed they only had 52 seats, but had already completed their 53-district plan.

Every district would have to gain about 15,000 persons. So you start working south from the Golden Gate bridge, and Oregon and converging around Stockton. The first district nibbles 15,000 from a neighbor, which takes 30,000 from its neighbor and so on. By the time you get to Bakersfield, Palmdale, Apple Valley, you are shifting half a district around, pairing incumbents, and creating open seats.

I reworked the estimates in the report, using the 2019 county estimates, linearly projected forward. The reports districts within Los Angeles county are based on assembly districts. I used the 2012 and 2018 5-year ACS which are used as approximations of the July 2010 and July 2016 populations. which were projected forward to 2020, and then controlled to the Los Angeles county estimate. 6 assembly districts cross county lines, and the report gave a 2010 estimate of their population in Los Angeles County. I prorated the 2020 projections on that basis. The projections of the areas in Los Angeles County are a bit dicey, but don't really effect the overall distribution.

Area	2010 (53)	2020 (52)	Change
North	2.514	2.418	-0.096
San Francisco Bay	10.547	10.560	0.013
Central Coast	2.827	2.724	-0.103
Central Valley	8.198	8.171	-0.027
Southern	14.943	14.952	0.009
Los Angeles	13.970	13.174	-0.796
Westside	2.681	2.479	-0.202
Downtown/Gateway	4.626	4.360	-0.266
San Gabriel	2.953	2.758	-0.195
San Fernando/Antelope	3.710	3.577	-0.133

So 8/10 of the loss of the 53rd district is accounted for in Los Angeles county, and the other 2/10 in the northern part of the state (Ventura, Kern, Inyo and other points north).

The 24 northern districts can be retained with a shift of about 100,000 out of western Los Angeles, or northern Los Angeles, or northwestern San Bernardino.

Riverside is the only county in southern California that has increased in district share, but San Diego and Imperial are basically static and on the edge of the state. There would be no reason to change those districts unless the commission decided to do a wholesale reshuffle (it is a legal requirement to not take into consideration existing districts or incumbents, but if the demographic facts are about the same, it is not unreasonable to come to the same conclusions).

I assume the commission will not cross the mountains between Orange and Riverside, particularly to grab a small bit on the other side of the mountains. So the Orange districts will add a bit from Los Angeles, Los Angeles will take some from San Bernardino, and Riverside will move some into San Bernardino. (these adjustment are relative small, around 100K).

The loss of a seat in Los Angeles is distributed throughout the county, which will force the district that disappears to be in the center among (CA-34, 37, 38, 40, 43, 44). In 2011, there were not Section 2 VRA seats drawn for a black district (i.e. black voters participate in greater numbers than non-citizen Hispanic voters). In 2021, there may be an emphasis on creating a districts that would be by a Hispanic candidate, essentially pairing Karen Bass and Maxine Waters.

[jimrtex]

In 2021, there may be an emphasis on creating a districts that would be by a Hispanic candidate, essentially pairing Karen Bass and Maxine Waters.

By which we mean a candidate of the Hispanic community's choice, not necessarily a Hispanic candidate.

Certainly.

The California Constitution says: "communities of interest shall not include relationships with political parties, incumbents, or political candidates" For example, a claim that Hispanic voters in a particular area happen to prefer Maxine Waters is irrelevant. It might not even be possible to use endogenous congressional results to determine the candidate of choice for a particular racial or ethnic group. Instead you might have to use an election such as Feinstein-DeLeon to determine if the group prefers different candidates. You might also use county supervisor or city elections. California Top 2 and non-partisan elections may be particularly useful in this regard.

The first step in a VRA analysis is to determine compact areas where a group constitutes a majority of the CVAP. If the 2020 census shows an expanded footprint for Hispanic voters, which it almost certainly will, it may be necessary to maintain 5 Hispanic districts to the east of one Black plurality district.

The reason that the redistricting commission is being organized earlier is to help define and delineate communities of interest prior to the line drawing process. The delay in the census data should help even more.

In 2011, the lawyers for the commission advised that the commission should take the word of local experts or advocates, since the commission did not have the time to do this. Many of these advocates may have been political plants, who were suggesting communities of interest where they wanted lines drawn or not drawn.

[jimrtex]

Normally the Census would be releasing the actual 2020 Census apportionment data next week, and not the Jul 1, 2020 estimates. Due to the expected delay in the release of the apportionment data, we can look at these estimates that do not include any 2020 Census data. It will be interesting to compare this to the actual Census population when they are released.

So, here's my annual projection from the new estimates. I used these July 2020 estimates and the April 2010 Census base to get an annual growth rate. This accounts for a 10 and a quarter year period between the 2010 Census and the estimate. I then applied the annual growth rate to the 2010 reapportionment population to get the 2020 projection. This accounts for the extra overseas population used in reapportionment but not for redistricting (It doesn't adjust for the new distinction between deployed vs stationed).

AZ +1
CA -1
CO +1
FL +2
IL -1
MI -1
MN -1
MT +1
NY -2
NC +1
OH -1
OR +1
PA -1
RI -1
TX +3
WV -1

AL and NY have switched seats compared to last year, so that AL would have no change and NY would lose 2 seats. I tested the possibility that inclusion of the Covid-19 fatalities in NY between April 1 and July 1 would affect the result. I added the July 1 NY fatality count to my April 1 projection, and it made NY closer to AL but did not change the overall projection.

The bubble seats in this projection are based on the last five awarded and the next five in line.
The last five awarded are IL-17, CA-52, FL-29, TX-39, and AL-7 (#435).
The next five in line are NY-26, OH-16, MN-8, CA-53, and VA-12.

I think the federal overseas population will be about half of what it was in 2010, even without the distinction between deployed and stationed. In 2010, 96% of the federal overseas population was military, with about 60% of that dependents.

The number stationed overseas is now under 200,000. Most are in areas where dependents can be resident (Europe, Japan, Korea). The share of dependents is likely about the same, with long-term trend towards career military (old enough to have acquired dependents), but perhaps balanced by long-term decline in marriages and child-bearing.

I think I will just use 3/4 of the change from 2019 to 2020 for my April 1, 2020 estimate.

What would you use for 2030 projections? I'm thinking perhaps getting a growth rate from 2015 to 2020.

[jimrtex]

if the new numbers aren't out til February or later then doesn't that mean Trump's little scheme to steal congressional apportionment based on citizenship is screwed?  presumably then Biden could make the decision whether or not to include non-citizens as it's been done in the past.

The most likely illegal aliens that can be identified are:

(1) Persons in ICE detention facilities.
(2) Persons in final deportation orders.
(3) DACA enrollees.

I suspect that they may end up with those in ICE detention facilities that might cost Texas its 39th representative.

[jimrtex]

Fresh 2020 Census Count numbers would be out today, by law, usually.

But not this year.

The Bureau said yesterday that the deadline will not be met and while not mentioning a date of release, the new numbers will likely be out between February and April.

Just talking for my state, there's a big problem with that.

State legislature shuts up shop for business in April, meaning if the Census doesn't release data until April, you'd have to wait until the 2022 session for the legislature to pass a bill with the new districts for federal and state. But people are required to file for primaries in January and February.

Under the original Census planning, the apportionment counts were due by the end of December, and that redistricting data by April 1, 2021. Both these deadlines are in statute.

When the Census Bureau could not do field enumeration they came up with an alternative plan that would slip those back about 3 or 4 months. These plans assumed that Congress would adjust the statutory limits.

When Congress failed to act, the Census Bureau came up with the so-called Replan that would attempt to complete processing by the deadlines in statute. The district court in California interfered with this plan until finally stopped by the SCOTUS.

In the past, the Census Bureau has endeavored to release redistricting data earlier for states with earlier deadlines. So that it might come out in February rather than on March 31. I think the Census Bureau has released their 2020 Census geography, so that state legislature staff can be loading up the data other than population.

My guess is that the apportionment numbers will be out around the end of this month, and that the redistricting data will be out by May.

This might result in some states calling a special session for redistricting. Practically anybody would have standing to file an equal protection lawsuit against the state of Indiana for failing to adjust legislature and congressional districts (e.g. Democratic Front Organizations For Fair Redistricting v. Eric Holcomb). The Indiana AG will claim that the suit is not ripe, and the court will withhold acting on the lawsuit until it is absolutely necessary, probably November 2021.

Holcomb will call a special session. It will be easy to do congressional redistricting, you just need to shift some population around maintaining IN-1 and IN-7 as Democratic vote sinks, and taking care to not make IN-2 and IN-5 competitive.

Legislative redistricting may be harder. With 100 House districts you will end up shifting districts around. If you can find an incumbent ready to retire, you may be able to divide up his district. But in other cases, you will end up pissing off incumbents who will suspect all kinds of nefarious plots. If the legislature gets into trouble, they will move the primary, and call another special session. There is zero chance the legislature won't get it done.

The DFFOFR will add a bunch of claims based on the VRA which the district court will eventually dismiss and the SCOTUS will not take up.

[jimrtex]

Fresh 2020 Census Count numbers would be out today, by law, usually.

But not this year.

The Bureau said yesterday that the deadline will not be met and while not mentioning a date of release, the new numbers will likely be out between February and April.

Just talking for my state, there's a big problem with that.

State legislature shuts up shop for business in April, meaning if the Census doesn't release data until April, you'd have to wait until the 2022 session for the legislature to pass a bill with the new districts for federal and state. But people are required to file for primaries in January and February.

That’s really not the Census Bureau’s problem .... (They had enough on their own).

It’s Indiana’s problem.

IN and it’s legislature could also work in September, right ? Other people have to work the whole year as well and not stop in April.

Well how it always worked in the past is the Census was released on time and they released new districts in a year that ends in 1, all complete by April.

They could have a special session later in the year to do it, but that means the state has to spend extra money not previously earmarked because the federal government bureaucrats can't meet a date of release. It's okay, we can just send the Census Bureau a bill for the costs, right?

What was the census bureau supposed to do? Do a rushed job and give the states messed up census data?

On December 31st Tender Branson stated in the census thread on page 51 that the Census Bureau said the data quality was high as he posted a press release from them saying as much. If releasing now is a rush job with messed up census data, then the press release was incredibly premature.

Why ?

They can assess data quality on what has been processed so far ...

Processing the data and quality checks take 5 months after completing the Census.

They were only ready in October this time, instead of the summer.

They have faster computers, and are reducing weekend down time. More of the data is digital rather than digitized from paper forms stained with jelly and coffee and unidentified substances. They can also defer some processing, and didn't have to do the resurvey, where they essentially go back and do a census of a sample of households.

[jimrtex]

Why ?

They can assess data quality on what has been processed so far ...

Processing the data and quality checks take 5 months after completing the Census.

They were only ready in October this time, instead of the summer.

They have faster computers, and are reducing weekend down time. More of the data is digital rather than digitized from paper forms stained with jelly and coffee and unidentified substances. They can also defer some processing, and didn't have to do the resurvey, where they essentially go back and do a census of a sample of households.

True.

But they have other problems that were not the case in 2000 or 2010, slowing down the publication of the numbers:

* ongoing sabotage attempts by Trump
* a pandemic, which cuts staff and work hours due to quarantine of staff, protection measures
* Christmas, New Year and other holidays in the fall

Processing census data in 2000 and 2010 started in the summer of those years, there were hardly any holidays.

Because data collection ended in late October this year, we can still expect 4 months of data processing despite better IT systems because of the factors above.

Mid-February to End-March looks most likely for the release.

There was no ongoing sabotage by Trump. The program to identify illegal aliens is by a different part of the Census Bureau than that conducting the Census.

As you may know, the counting of overseas federal affiliated persons and their dependents is not actually an enumeration, and is done under an executive order rather than statute. The Census Bureau simply receives the numbers from federal agencies (the overwhelming majority are DOD, with a minority of State Department employees, and smattering of other agencies). There is probably some processing to avoid duplicates (e.g. The wife of a soldier stationed in Germany, may have remained stateside, and included her husband when she filled out her census form).

Enumeration ended in mid-October. They have added weekend processing which more than makes up for any losses due to holidays. I suspect they were running their computers on Veterans Day, Thanksgiving, Christmas, and New Years Day.

They are way past the need for huge numbers of temporary workers in call centers and paper forms processing that were slowed because of COVID-19 mitigation.

Three-1/2 months would be the end of January.

[jimrtex]

Normally the Census would be releasing the actual 2020 Census apportionment data next week, and not the Jul 1, 2020 estimates. Due to the expected delay in the release of the apportionment data, we can look at these estimates that do not include any 2020 Census data. It will be interesting to compare this to the actual Census population when they are released.

So, here's my annual projection from the new estimates. I used these July 2020 estimates and the April 2010 Census base to get an annual growth rate. This accounts for a 10 and a quarter year period between the 2010 Census and the estimate. I then applied the annual growth rate to the 2010 reapportionment population to get the 2020 projection. This accounts for the extra overseas population used in reapportionment but not for redistricting (It doesn't adjust for the new distinction between deployed vs stationed).

AZ +1
CA -1
CO +1
FL +2
IL -1
MI -1
MN -1
MT +1
NY -2
NC +1
OH -1
OR +1
PA -1
RI -1
TX +3
WV -1

AL and NY have switched seats compared to last year, so that AL would have no change and NY would lose 2 seats. I tested the possibility that inclusion of the Covid-19 fatalities in NY between April 1 and July 1 would affect the result. I added the July 1 NY fatality count to my April 1 projection, and it made NY closer to AL but did not change the overall projection.

The bubble seats in this projection are based on the last five awarded and the next five in line.
The last five awarded are IL-17, CA-52, FL-29, TX-39, and AL-7 (#435).
The next five in line are NY-26, OH-16, MN-8, CA-53, and VA-12.

I used 3/4 of the difference between the July 1, 2019 and July 1, 2020 estimates. The growth assumption in your method assumes that the 2020 is reliable for calculating the annual growth rate, but that the 2019 estimate was not accurate and should be ignored.

I think the federal overseas population will be about 1/2 of what it was in 2010, but I did not include it.

432 MT 1.005
433 FL 1.003
434 TX 1.002
435 NY 1.002
-------------
436 AL 1.001
437 MN 0.996
438 OH 0.995
439 RI 0.985

The bolded states can be considered bubble states. NY is at risk of losing a second seat, while FL would be limited to +1, and TX would be limited to +2, while AL would stay at 7.

CA shot off the table making it certain that it will lose a seat, and for 2030 we may be looking at loss of multiple seats. CA gained +300K every year for the first five years, but has gone +245K, +188K, +100K, 0K, -69K for the last 5.

NY started of +100K in 2011, but gradually lost steam and has been (increasingly negative for the last five years). The loss of -127K for 2020 was almost equal to the combined loss of 130K for 2018+2019. The loss for 2020 also put NY negative for the decade.

AL had a "U" shape pattern, with perhaps enough to hold on to the 7th seat. It could be a state that benefits by inclusion the federal overseas population is overwhelmingly military and dependents, compared to NY.

[jimrtex]

Normally the Census would be releasing the actual 2020 Census apportionment data next week, and not the Jul 1, 2020 estimates. Due to the expected delay in the release of the apportionment data, we can look at these estimates that do not include any 2020 Census data. It will be interesting to compare this to the actual Census population when they are released.

So, here's my annual projection from the new estimates. I used these July 2020 estimates and the April 2010 Census base to get an annual growth rate. This accounts for a 10 and a quarter year period between the 2010 Census and the estimate. I then applied the annual growth rate to the 2010 reapportionment population to get the 2020 projection. This accounts for the extra overseas population used in reapportionment but not for redistricting (It doesn't adjust for the new distinction between deployed vs stationed).

AZ +1
CA -1
CO +1
FL +2
IL -1
MI -1
MN -1
MT +1
NY -2
NC +1
OH -1
OR +1
PA -1
RI -1
TX +3
WV -1

AL and NY have switched seats compared to last year, so that AL would have no change and NY would lose 2 seats. I tested the possibility that inclusion of the Covid-19 fatalities in NY between April 1 and July 1 would affect the result. I added the July 1 NY fatality count to my April 1 projection, and it made NY closer to AL but did not change the overall projection.

The bubble seats in this projection are based on the last five awarded and the next five in line.
The last five awarded are IL-17, CA-52, FL-29, TX-39, and AL-7 (#435).
The next five in line are NY-26, OH-16, MN-8, CA-53, and VA-12.

I used 3/4 of the difference between the July 1, 2019 and July 1, 2020 estimates. The growth assumption in your method assumes that the 2020 is reliable for calculating the annual growth rate, but that the 2019 estimate was not accurate and should be ignored.

I think the federal overseas population will be about 1/2 of what it was in 2010, but I did not include it.

432 MT 1.005
433 FL 1.003
434 TX 1.002
435 NY 1.002
-------------
436 AL 1.001
437 MN 0.996
438 OH 0.995
439 RI 0.985

The bolded states can be considered bubble states. NY is at risk of losing a second seat, while FL would be limited to +1, and TX would be limited to +2, while AL would stay at 7.

CA shot off the table making it certain that it will lose a seat, and for 2030 we may be looking at loss of multiple seats. CA gained +300K every year for the first five years, but has gone +245K, +188K, +100K, 0K, -69K for the last 5.

NY started of +100K in 2011, but gradually lost steam and has been (increasingly negative for the last five years). The loss of -127K for 2020 was almost equal to the combined loss of 130K for 2018+2019. The loss for 2020 also put NY negative for the decade.

AL had a "U" shape pattern, with perhaps enough to hold on to the 7th seat. It could be a state that benefits by inclusion the federal overseas population is overwhelmingly military and dependents, compared to NY.

"I used 3/4 of the difference between the July 1, 2019 and July 1, 2020 estimates. The growth assumption in your method assumes that the 2020 is reliable for calculating the annual growth rate, but that the 2019 estimate was not accurate and should be ignored."

If the July, 2019 estimate is bad, why are you apparently also using it?

I myself was hesitant on projecting forward from a rate that was over a period of close to a decade, when more recent trends may have been quite different. But then covid hit, and what had happened recently before might not be very useful at all as a rate line.

As we get closer and closer to the next census our projected population is not much different than the latest estimate. Think of having a piece of string tied to a peg at the 2010 Census, and the other end at the 2019 estimate 9.25 inches away. We pull the string out another 0.75 inches being careful not to bend at 2019 estimate. That would our projected population based on the 2019 estimate. We are not going to have much error based on not knowing the change for the last 3/4 of a year.

Now repeat using the July 2020 estimate, but back up 0.25 of an inch. The 2020 estimate was 10.25 years after the 2010 census, and 0.25 years after the 2020 Census.

Most of our estimate is not due to change in that last 0.25 years but over the entire period.

So for New York, the 2020 estimate was 19.337M, which was less than the 2010 Census Population of 19.378M, but only by 41K. That is about 4K per year, or 1K for that last quarter between April 2020 and July 2020. The estimated population would be 19.338M.

But the 2019 estimate was 19.463M. The Census Bureau says that New York lost 126K last year (July 2019-July 2020). We can estimate that 126K was evenly divided over the last year, or about 32K. So my estimate was 19.369M, just a bit more than one based on assuming constant change over the decade.

New York's yearly estimated change over the decade was:
+100K, +75K, +52K, +27K, +4K, -21K, -43K, -49K, -81K, -127K.

It was not

-4K, -4K, ..., -4K, -4K even though the net effect was the same.

It happens that my adding the extra 31K may have saved New York from losing a second seat. That is sure coincidence. If Alabama has 5K extra people after a decade increase of 138K, or New York has -20K fewer after losing 127K in the last year (and those 20K may have gone missing), then Alabama and New York flip. It could be even closer when federal overseas population is added in. Most of this is military and dependents (military and families stationed in Germany, Japan, and Korea).

In 2010, this increased Alabama's population by 0.49%; while New York's only increased by 0.22%.

COVID will be barely reflected in the estimates. Estimates for interstate migration are based on lagging indicators such as income tax records. They only had access to 2019 taxes when they made the estimate for 2020. That gives a measure of migration between 2018 and 2019 which must be projected forward. Birth and death records are more current.