Census Estimates for 2007 -> 2010 Apportionment

[muon2]

Just about any way you crack it up, this should mean one more republican Congressman for Oregon.  Though looking at the present state of politics in Oregon combined with an utterly incompetent country club state GOP, odds are they could lose the new "sure thing" congressional district as well.

A tenth WA district would probably be Republican, too. It would ruin the perfect East-West balance in districts we've had for a while, meaning the tenth district would have to have large portions of both Western and Eastern Washington. The most logical place for a district to crossover would be in the southernmost part of the state, containing parts of suburban Portland (Vancouver area) and into South Central Washington to the Yakima area. I would expect the GOP to have the advantage here, but, as you point out, the GOP in the NW is in a pretty sorry state as of late, so who knows (WA-3 is a lean R district that voted for Bush in SW Washington that is held by a Democrat who has been easily winning re-election). I doubt the GOP will be able to keep WA-8 too much longer, especially if a tenth district is added (changing the rest of the districts significantly). With 10 districts in Washington, 7 held by Democrats and 3 by Republicans seems the most likely.

If there is a 10th district then I would predict 2.2 seats east of the Cascades, whereas with nine seats it should be an almost perfect 7-2 split. I have a question to the WA experts: if 2/10 of a seat come from the east to attach to the west will it be 1) attach Okanagan, Chelas, and Kittigas to NW WA; 2) attach Richland, Kennewick or part of Yakima to suburban Portland; or 3) attach Okanagan and Chelas to NW WA, and only attach Klickitat to Clark Co.

[bgwah]

They'll do it in the southern area of the state, connecting suburban Portland to the Yakima/Klickitat area. That's where they always seem to crossover for legislative districts.

[Brittain33]

I'm curious where the knife is going to fall in New York. 2010 will mark the first census in modern times in which Nassau and Suffolk will not have sufficient population for a full four districts, although it will be very close.

I don't see how the 5th, 7th, and 9th districts all survive. Anthony Weiner said he felt like the Thanksgiving turkey before the 2002 redistricting until the numbers came out and New York City was found to have grown more than predicted. The geography has only gotten worse for his district since then.

[muon2]

Just about any way you crack it up, this should mean one more republican Congressman for Oregon.  Though looking at the present state of politics in Oregon combined with an utterly incompetent country club state GOP, odds are they could lose the new "sure thing" congressional district as well.

A tenth WA district would probably be Republican, too. It would ruin the perfect East-West balance in districts we've had for a while, meaning the tenth district would have to have large portions of both Western and Eastern Washington. The most logical place for a district to crossover would be in the southernmost part of the state, containing parts of suburban Portland (Vancouver area) and into South Central Washington to the Yakima area. I would expect the GOP to have the advantage here, but, as you point out, the GOP in the NW is in a pretty sorry state as of late, so who knows (WA-3 is a lean R district that voted for Bush in SW Washington that is held by a Democrat who has been easily winning re-election). I doubt the GOP will be able to keep WA-8 too much longer, especially if a tenth district is added (changing the rest of the districts significantly). With 10 districts in Washington, 7 held by Democrats and 3 by Republicans seems the most likely.

Based on county projections from July 2006 estimates, here's how a 10th district might be placed in WA. District 1 moves up into Snohomish county and takes up most of the county. District 2 stays west of the cascades, but now links Bellingham to Bremerton across the islands. District 3 has to move west along the Columbia and would stretch from Vancouver up to Kennewick. District 4 links Yakima and Walla Walla, which district 5 gives up to keep to the new smaller sized districts. District 6 remains in Tacoma but only extends to the near suburbs like Lakewood and Puyallup. District 7 and 8 are much the same being Seattle and Bellevue/eastern King/eastern Pierce respectively. District 9 could be entirely in King stretching from Renton to Federal Way. And then the new district 10 is formed from Olympia and the Olympic peninsula all the way down the Pacific coast.

I'll leave it to locals to determine the likely partisan makeup of those districts.

[Brittain33]

Based on county projections from July 2006 estimates, here's how a 10th district might be placed in WA. District 1 moves up into Snohomish county and takes up most of the county. District 2 stays west of the cascades, but now links Bellingham to Bremerton across the islands. District 3 has to move west along the Columbia and would stretch from Vancouver up to Kennewick. District 4 links Yakima and Walla Walla, which district 5 gives up to keep to the new smaller sized districts. District 6 remains in Tacoma but only extends to the near suburbs like Lakewood and Puyallup. District 7 and 8 are much the same being Seattle and Bellevue/eastern King/eastern Pierce respectively. District 9 could be entirely in King stretching from Renton to Federal Way. And then the new district 10 is formed from Olympia and the Olympic peninsula all the way down the Pacific coast.

I'll leave it to locals to determine the likely partisan makeup of those districts.

If Washington gains a 10th district, the Republicans can thank illegal immigration. Those two trans-Cascadian districts have enormous non-citizen Latino farm worker populations that bulk up their numbers but don't directly affect the vote totals, except insofar as there's a conservative backlash to their presence.

[minionofmidas]

Only the western one, actually. Mostly around Yakima, but also Richland-Kennewick.

[jimrtex]

Could you tell me the formulas you enter into your Spreadsheet? I am trying to make one for myself.

A brute force method:

In Column A: State Name.
In Column B: Estimated Population.
In Column C: Priority value for 2nd seat (C.row = B.row/sqrt(2*1))
In Column D: Priority value for 3rd seat (D.row = B.row/sqrt(3*2))
repeatr for at least 54 columns (or however many seats you think California will have).

So you will have a 54 by 50 array of priority values.   Find the 385th largest value (each of the 50 states is guaranteed one seat, and 435-50 is 385).  Determine the number of priority values for each state that is greater than or equal to the threshold value.

Short Cut (simplest calculation but may have an error in extreme cases):

1) Determine average per CD for country (AVE = total_estimated_population /435)
2) Estimate number of CDs for each state (XCD = state_population/AVE )
    eg a State with 7.85/435 of the total population would have a XCD value of 7.85
3) Determine whole number of CDs for each state (NCD = max(int(XCD),1) )
    The state with a XCD value of 7.85 would have an NCD value of 7.
4) Determine total number of whole CDs for US.  This should be around 435 - 50/2 = or 410,
    since the average truncated fraction will be around 1/2.
5) Determine priority value for next CD for each state:
    PV = state_pop/SQRT(NCD*(NCD+1))
6) Determine states with N highest priority values for next CD (where N is 435 minus
    the value from step 4).

This does not produce a complete set of priority rankings, but rather the relative ranking for the next CD among the 50 states (eg Wyoming would be somewhere around 800 for a 2nd CD, but at worst would be 50th in this ranking).

[jimrtex]

I noticed that the Census Bureau adjusts its previous estimates for all previous years in the decade.  That is, the latest estimate gives a different estimate for July 2006, July 2005, etc. than was given in the report made in December 2006.   This is not a new phenomena, it has happened every year.   For example, these are the estimates for Alabama.

Each row represents the estimates made in December of that year.  Each column represents the estimated July 1st population.   So for example, the December 2004 estimate for the July 1, 2004 population was 4,530,182; while the December 2007 estimate for the July 1, 2007 population was 4,508,540.

	2007	2006	2005	2004	2003	2002	2001	2000
2007	4,627,851	4,590,240	4,539,611	4,508,540	4,488,071	4,471,006	4,463,224	4,451,887
2006		4,599,030	4,548,327	4,517,442	4,495,089	4,477,571	4,466,618	4,452,375
2005			4,557,808	4,525,375	4,501,862	4,480,139	4,467,461	4,452,339
2004				4,530,182	4,503,726	4,481,078	4,468,031	4,452,307
2003					4,500,752	4,478,896	4,466,440	4,451,601
2002						4,486,508	4,468,912	4,451,975
2001							4,464,356	4,451,493

There are interstate variations.   For example, the revision in July 1, 2006 population estimate (estimate issued in December 2007 vs. estimate issued in December 2006) ranged from +1.2% for Utah to -1.0% for Louisiana.   For Louisiana, the 2007 estimates show a slightly smaller (-8K) population in 2005; a larger drop for 2006 (252K vs. 220K in the previous estimate); and a recovery of 50K for 2007.

So the changes in projected population for 2010 (and consequent apportionment changes) reflect not only varying rates of growth throughout the decade, but also revisions of earlier estimates.  In general, Muon's method of annualizing the 2000-2007 change and projecting it forward to 2010 should still be valid - though as time passes it will be less sensitive to changes in growth rates.  For example, the estimated rate of change from 2004-2007 might be better for projecting the rate of change for 2007-2010 than the estimated rate of change from 2000-2007 while still benefitting from the use of a multi-year average.

[Alcon]

Awesome

Only the western one, actually. Mostly around Yakima, but also Richland-Kennewick.

The two biggest concentrations of illegal Hispanics are around Pasco (Kennewick to a lesser extent, but not so much Richland; it's pretty much lily-white) and Yakima.

I assume that the district that links Walla Walla and Yakima would do so via Franklin County (Pasco), which would mean it has plenty of illegals too.  In fact, I'd wager that the eastern would have just as many, if not more (Clark County doesn't have an exceptionally high Hispanic population).

I'd be able to tell you more if I could see a mapped version of muon2's districts, because I'm not sure I totally understand from the text descriptions.

There are some problems, though.  I assume that you're lumping Bainbridge Island into the 1st with Snohomish County?  That would look a little awkward, but otherwise it would put Jay Inslee in Rick Larsen's district.  Adam Smith would also be transferred to what was mostly Norm Dicks's district, and Norm Dicks would have a district (the new one) containing no military bases - he'd complain.

Those districts would be:

1st - Democratic
2nd - Toss-up (no real lean, maybe slightly Democratic)
3rd - Lean Republican
4th - Strong Republican
5th - Strong Republican
6th - Slight Democratic lean
7th - Strong Democratic
8th - Toss-up (slight Democratic lean)
9th - Democratic
10th - Democratic

[CultureKing]

The most likely placing for a 10th district in Washington I would guess to include the Olympic penninsula, the Pacific coast counties and some mix of Olympia or Bremerton. It would be a pretty strong Democatic district, though if it incorporated Olympia then the 3rd would go from tossup to lean or likely republican (it went to Bush in 2004 by 2-3%, take out Olympia and it shoots up).

[Left-Wing Blogger]

I'm surprised California might go down.

[Padfoot]

I'm surprised California might go down.

California barely squeaked in its 53rd seat during the 2000 reapportionment.  CA-53 was the 434th seat to be handed out in 2000.  Given California's average growth rate as compared to other states and the fact that nearly 1 million Californians have left the state since 2000 this isn't all that surprising.  Right now California's growth is dependent on new births and foreign immigrants.

[ottermax]

At least a reapportionment in WA isn't gerrymandered. Other states need to follow WA and Arizona.

[AkSaber]

How long has it been since a census didn't give California any Congress seats?

[Joe Republic]

How long has it been since a census didn't give California any Congress seats?

1920 was the last.  California had 11 House seats before and after the census.

[jimrtex]

How long has it been since a census didn't give California any Congress seats?

1920 was the last.  California had 11 House seats before and after the census.

That was because Congress did not reapportion after the 1920 census.  California should have had 14 representatives (it went from 11 to 20 from 1910 to 1930).

California did not gain a seat after the 1850 census (staying at 2), but it did not enter the Union until September 9, 1850, which was probably after the census.

[jimrtex]

The first column is the apportionment based on projected 2010 population, if fractional representatives could be apportioned using the following formula:

   sqrt ((population/average)^2 + 0.25)

where population is the state population, and average is the approximately the US population/435.  The average is adjusted slightly upward so that the sum of state's apportionment is 435.   Using this formula, a state with a population of
sqrt (n*(n+1))*average will have an apportionment of n + 0.5 representatives.

The second column is the change in apportionment from 2000 to 2010.  For example, based on its relative share of the population, Texas should gain slightly more than 3 seats.

The third column apportions whole representatives so that there are 435 representatives, using the priority ranking.

The fourth column is the projected percentage increase in population from 2000 to 2010.  The US rate of growth is 10.1%.

The fifth column is the additional percentage increase needed to gain one more representative (that is, to catch Oregon #435) or to lose one more representative (that is, to catch New York #436).

California           52.692    0.450  53   11.1%  -1.4%
Texas                35.259    3.096  36   20.7%  -0.4%
New York             27.201   -2.071  27    2.3%   0.3%
Florida              26.883    2.228  27   20.1%  -2.5%
Illinois             18.241   -0.920  18    4.8%   0.6%
Pennsylvania         17.499   -1.448  18    1.7%  -1.1%
Ohio                 16.127   -1.390  16    1.4%   1.5%
Michigan             14.185   -1.151  14    1.9%   1.4%
Georgia              14.176    1.540  14   23.6%   1.4%
North Carolina       13.283    0.863  13   17.8%   0.8%
New Jersey           12.321   -0.666  12    4.5%   0.6%
Virginia             11.168    0.238  11   12.5%   2.1%
Arizona               9.631    1.702  10   33.9%  -2.4%
Washington            9.397    0.293   9   13.7%   0.3%
Massachusetts         9.100   -0.705   9    2.2%   3.6%
Indiana               9.044   -0.347   9    6.1%   4.2%
Tennessee             8.898    0.109   9   11.5%  -5.5%
Missouri              8.402   -0.244   8    7.0%   0.3%
Maryland              8.061   -0.123   8    8.5%   4.6%
Wisconsin             7.990   -0.297   8    6.2%  -7.2%
Minnesota             7.449   -0.155   8    7.9%  -0.4%
Colorado              7.149    0.495   7   18.4%   4.1%
Alabama               6.598   -0.279   7    5.6%  -2.6%
South Carolina        6.416    0.209   6   13.9%   0.5%
Kentucky              6.070   -0.185   6    6.9%   6.3%
Louisiana             5.942   -0.968   6   -5.4%  -8.5%
Oregon                5.455    0.155   6   13.4%  -0.3%
Oklahoma              5.181   -0.164   5    6.7%   5.3%
Connecticut           4.982   -0.294   5    3.9% -10.8%
Iowa                  4.247   -0.294   4    2.9%   5.2%
Mississippi           4.158   -0.258   4    3.6%   7.5%
Arkansas              4.090   -0.064   4    8.4%   9.3%
Utah                  3.982    0.501   4   26.3% -13.3%
Nevada                3.981    0.859   4   41.1% -13.2%
Kansas                3.966   -0.211   4    4.5% -12.9%
New Mexico            2.887    0.037   3   11.6% -14.8%
West Virginia         2.588   -0.245   3    0.3%  -4.6%
Nebraska              2.569   -0.118   3    5.1%  -3.8%
Idaho                 2.276    0.219   2   22.5%   9.4%
New Hampshire         1.952   -0.018   2    9.0% -25.9%
Maine                 1.933   -0.096   2    4.6% -25.1%
Hawaii                1.904   -0.031   2    8.3% -23.8%
Rhode Island          1.568   -0.124   2    1.3%  -5.9%
Montana               1.460   -0.018   1    8.6%   2.2%
Delaware              1.353    0.045   1   14.6%  11.6%
South Dakota          1.243   -0.024   1    7.6%  23.3%
Alaska                1.108    0.020   1   12.6%  41.8%
North Dakota          1.025   -0.085   1   -0.5%  56.8%
Vermont               1.009   -0.055   1    2.8%  60.0%
Wyoming               0.900   -0.011   1    8.2%  87.5%

California is increasing slightly faster than the US population.  With its large population, a relative increase of about 1.9% will add one representative.  In 2000, it got a favorable rounding for its 53rd representative.  During this decade it has solidified its hold on the seat.  The change from last year (when a 54th seat was projected) is due to a slowing down in growth plus revisions from previous years.  Despite a raw entitlement less than 53, California is closer to gaining a seat (+0.7%) than losing one (-1.1%).

Texas had a raw increase of just over 3 seats, but may gain a 4th seat based on a favorable rounding.  There are an unusual number of states with a fractional apportionment below 0.5 (32 vs 18 above).  This results in 4 states with such a fraction getting an extra seat (OR, TX, MN, and PA).  In general, this is favorable to larger states since they can distribute their shortfall among more representatives (in 2000, California gained its 53rd seat that way).  Texas could lose the 36th seat with a very small decrease in the projected increase.

New York could hold on to a seat (losing 1 rather than 2) with a slight uptick in population, or downtick by Texas, Minnesota, or Oregon.

Pennsylvania is losing population share faster than Ohio, but Ohio will lose 2 seats because it was so near to losing its 18th seat in 2000.

New Jersey will likely lose a seat (compared to last year's projection) due to a slowing in its population increase from the early part of the decade.

Arizona could be limited to one new seat if the projected increase is slightly high.

Washington as noted elsewhere could gain a 10th seat.

If Missouri doesn't lose a seat in 2010, it will in 2020.

If Minnesota doesn't lose a seat in 2010, it probably will in 2020.

Colorado will likely gain a seat in 2020, while Alabama will probably lose 1.

South Carolina could gain a seat in 2020.  It has had 6 representatives since 1930.  I suspect some of the recent gain is an overflow from Charlotte, NC across the border.

I did not make an adjustment for Louisiana, but it would have lost its 7th seat even without Katrina.

If Oregon doesn't get a 6th representative in 2010, it will in 2020.

Utah could gain a representative in 2020 as well if they get lucky.

West Virginia and Nebraska will be very suceptible to the loss of a representative in 2020, while Idaho could gain one.

Rhode Island could be vulnerable to loss of a representative in 2020.  It was estimated to have lost population from 2006 to 2007.

Montana appears to be stuck just short of regaining its 2nd representative.

From 2000 to 2010, all western states except HI, MT, and WY are estimated to be growing faster than the country as a whole.  This is also true of the South Atlantic states from DE to FL, with the exception of MD.  The other two fast growing states are TX and TN.

[Brittain33]

Thank you for posting the chart and projections. Fascinating stuff.

[muon2]

I'd be able to tell you more if I could see a mapped version of muon2's districts, because I'm not sure I totally understand from the text descriptions.

Here's a version of the map with 9 and with 10 districts based on county projections for 2010. I've used my usual rules to minimize county splits.

[ottermax]

Thanks for making maps. I'm a visual person.

I see a few problems:
First Map (9 districts)
-The yellow district isn't really connected. The biggest issue is Clallam and Kitsap. This also makes the dark green district disconnected as well because SW King County and Jefferson County do not go together (culturally).
-Other than that there are only a few minor changes you could make, but it looks like you tried really hard. The biggest problem is that some counties just have to be split up, and placed with similar counties otherwise there are geographical dissimilarities.

Second Map (10)
-I really like parts of this map, and I really dislike other parts of it.
-The Olympic Peninsula district looks really well balanced.
-The rest of Western WA looks fine, except my dislike for the Kitsap county with the yellow. I really like the light green district (despite its appearance as a Republican district that I would live in), because it matches the people well.
-The biggest issue is with the 3 E. WA districts. The Tri-Cities are split (Pasco is in a different district) and East Wenatchee (in Douglas County) is split from Wenatchee (Chelan County).

Good job on the maps!

[minionofmidas]

Awesome

Only the western one, actually. Mostly around Yakima, but also Richland-Kennewick.

The two biggest concentrations of illegal Hispanics are around Pasco (Kennewick to a lesser extent, but not so much Richland; it's pretty much lily-white) and Yakima.

I assume that the district that links Walla Walla and Yakima would do so via Franklin County (Pasco), which would mean it has plenty of illegals too.  In fact, I'd wager that the eastern would have just as many, if not more (Clark County doesn't have an exceptionally high Hispanic population).

I'd be able to tell you more if I could see a mapped version of muon2's districts, because I'm not sure I totally understand from the text descriptions.

I (and whoever I was replying to as well) was referring to the current map, ie the current Yakima-Richland-Pasco-Kennewick district as opposed to the current Spokane-Walla Walla district.

[muon2]

Thanks for making maps. I'm a visual person.

I see a few problems:
First Map (9 districts)
-The yellow district isn't really connected. The biggest issue is Clallam and Kitsap. This also makes the dark green district disconnected as well because SW King County and Jefferson County do not go together (culturally).
-Other than that there are only a few minor changes you could make, but it looks like you tried really hard. The biggest problem is that some counties just have to be split up, and placed with similar counties otherwise there are geographical dissimilarities.

The yellow (Bellingham) district really is connected according to sources like Mapquest or Rand McNally. San Juan county borders Clallam and Island county borders Kitsap. The dark green (Olympia) district is physically connected through the part of Pierce on the Peninsula and Vashon Island. Cultural connection was not a factor for me, except that I disallowed maps that spanned the Cascades except along the Columbia River. Placing Jefferson with either Bellingham or Vancouver would require an additional county to be split.

You must be logged in to read this quote.

With 10 districts there is either an eastern county like Okanogan or Chelan attached to a western district, or one uses the Cascade barrier rule that I used. That forces the Vancouver district east and it either takes in part of Yakima or it takes in Kennewick without Pasco. The Yakima and Columbia rivers seemed like a natural line compared to any split I could make in Yakima county to move 106 K people out of that district.

Glad you enjoy the maps.

[jimrtex]

Thanks for making maps. I'm a visual person.

I see a few problems:
First Map (9 districts)
-The yellow district isn't really connected. The biggest issue is Clallam and Kitsap. This also makes the dark green district disconnected as well because SW King County and Jefferson County do not go together (culturally).
-Other than that there are only a few minor changes you could make, but it looks like you tried really hard. The biggest problem is that some counties just have to be split up, and placed with similar counties otherwise there are geographical dissimilarities.

The yellow (Bellingham) district really is connected according to sources like Mapquest or Rand McNally. San Juan county borders Clallam and Island county borders Kitsap. The dark green (Olympia) district is physically connected through the part of Pierce on the Peninsula and Vashon Island. Cultural connection was not a factor for me, except that I disallowed maps that spanned the Cascades except along the Columbia River. Placing Jefferson with either Bellingham or Vancouver would require an additional county to be split.

I would side with Ottermax on this.  Across a body of water, I would require either a bridge or ferry link to establish contiguity.  If there were no such links, I would permit nearest point contiguity if necessary to establish contiguity (eg to connect all parts of county, or to connect counties that would otherwise be disconnected.

So, I would reject the Clallam-San Juan and Island-Kitsap links, since the two island counties have connections to the east.  I would also require you to change the Vashon Island link using the ferry from the south end of the island into Tacoma and then across the Narrows Bridge.

If someone were to produce a map with districts crossing the northern Cascades that split fewer counties, I might be inclined to accept that.

[muon2]

Thanks for making maps. I'm a visual person.

I see a few problems:
First Map (9 districts)
-The yellow district isn't really connected. The biggest issue is Clallam and Kitsap. This also makes the dark green district disconnected as well because SW King County and Jefferson County do not go together (culturally).
-Other than that there are only a few minor changes you could make, but it looks like you tried really hard. The biggest problem is that some counties just have to be split up, and placed with similar counties otherwise there are geographical dissimilarities.

The yellow (Bellingham) district really is connected according to sources like Mapquest or Rand McNally. San Juan county borders Clallam and Island county borders Kitsap. The dark green (Olympia) district is physically connected through the part of Pierce on the Peninsula and Vashon Island. Cultural connection was not a factor for me, except that I disallowed maps that spanned the Cascades except along the Columbia River. Placing Jefferson with either Bellingham or Vancouver would require an additional county to be split.

I would side with Ottermax on this.  Across a body of water, I would require either a bridge or ferry link to establish contiguity.  If there were no such links, I would permit nearest point contiguity if necessary to establish contiguity (eg to connect all parts of county, or to connect counties that would otherwise be disconnected.

So, I would reject the Clallam-San Juan and Island-Kitsap links, since the two island counties have connections to the east.  I would also require you to change the Vashon Island link using the ferry from the south end of the island into Tacoma and then across the Narrows Bridge.

If someone were to produce a map with districts crossing the northern Cascades that split fewer counties, I might be inclined to accept that.

I'm not convinced that ferries and bridges are the only cross-water connections to allow. Distance should be given some consideration, as the ferry connection from Edmonds/Snohomish to Kingston/Kitsap is greater than the distance from Kitsap to Island at Hansville. Also I would think that an arbitrary cross-water connection where the counties officially connect is no worse than a cross-mountain connection with no pass or adjacent counties in the plains that share a short but finite border with no road.

[jimrtex]

I'm not convinced that ferries and bridges are the only cross-water connections to allow. Distance should be given some consideration, as the ferry connection from Edmonds/Snohomish to Kingston/Kitsap is greater than the distance from Kitsap to Island at Hansville. Also I would think that an arbitrary cross-water connection where the counties officially connect is no worse than a cross-mountain connection with no pass or adjacent counties in the plains that share a short but finite border with no road.

The Edmonds-Kingston connection is longer only because Kingston has a harbor indentation and Edmonds is slightly to the north of the headland.  There is no connection at Hansville which is completely off the highway system.  The San Juan-Clallam boundary might only exist because of Canada, which makes the Straits of Juan de Fuca an inland waterway.  If Vancouver Island did not exist, the Clallam border would at most only extend 3 miles north, and there wouldn't be the SW corner of San Juan County.

To take another example, I would permit a Richmond-New York link in New York State based on the Staten Island Ferry, while disallowing a Bronx-Nassau link based on the marine boundary.

I would also apply a rule regarding corner connections or near-corner connections: something like the boundary must be 5% of the total boundary of one of the counties, where the boundaries might be idealized:

For the total boundary, use the circumference of a circle with area equal to area of the county, (ie C = 2 sqrt (pi * area) ).  This avoids a penalty for irregularly shaped counties.  And for the shared-boundary, I would apply some sort of algorithm that would reduce kinky borders such as one that follows a meandering stream.

Requiring a highway link might be difficult to apply.  Should Washington 20 be disqualified as link between Whatcom or Skagit and Okanogan counties, but allowed as a link between the two together and Okanagan? 

In other cases, a highway might clip the corner of a 3rd county, but people would be travelling between the two counties.  For example, Colorado 82 NW from Aspen clips the corner of SW Eagle County, but it is a long ways from the Eagle and Vail areas, which would be reached by first going to Glenwood Springs in Garfield County.  The other roads out of Pitkin County are closed in winter.

The county commissioner in southern Hinsdale County, Colorado must travel through 6 other counties (5 county seats) during winter to get to the county seat of Lake City (around 250 miles).  In summer, the trip is reduced to 3 other counties.

I like the idea of requiring a highway link, but I don't know how to practically state a rule.