Challenge 3. Gerrymander Fix

My gerrymander fix consists of two completely independent components. That is, either component stands alone and represents a considerable improvement over where we are today. But both components taken together really slay this monster once and for all.

The first component is a rule-based scheme for drawing congressional districts without human decision-making. The second component is multi-seat congressional districts combined with Ranked Choice Voting.

Redistricting without Gerrymanders

An old saying is most relevant here: “Perfection is the enemy of the good.” I do not claim that this solution is perfect, that it meets every conceivable objective of redistricting reform, or that it is the very best way to draw congressional districts. I do claim that it is far preferable to the system we have now. Further, it is achievable on a state-by-state basis or, preferably, through national legislation.

Redistricting Principles

We need a completely mechanical, replicable, transparent, uniform, and (especially) non-partisan process for redistricting.

Seven states have only one seat in the House, so this discussion pertains to the 43 states that have more than one House seat. However, the problem of gerrymandering may be even more pronounced with respect to drawing election districts for state legislatures. The solution recommended here for Congress can be applied to legislative districts in all 50 states.

Most observers would agree that the solution to this problem is to remove redistricting from the control of politicians and to use computer software to make the redistricting process easy and automatic. The solution proposed in this paper satisfies both of those criteria. This section explains the proposed redistricting solution as a manual process, which lends itself to automation.  (As a professor who taught computer programming for 30+ years, I’m both qualified and happy to volunteer my services to help us create the computer software to implement this solution.)

PRELIMINARILY, we should all agree on the basic objectives for any non-partisan redistricting scheme:

  • Make the population per House seat within a state as equal as possible;
  • Afford some consideration to geography, keeping neighbors together in the same congressional district and creating compact election districts consisting of a single contiguous land mass as far as possible;
  • Ignore all other considerations.

Equalize population per House seat: The first of these objectives, equalizing the population per House seat, reflects basic fairness, the time-honored notion of “one person – one vote”.

Keep neighbors together: The second objective is intuitively obvious to Americans, but geography is not necessarily the only criterion that could be used. In Florence in the Middle Ages, representatives to the city Senate were chosen by occupation: construction trades elected one rep, farmers another, merchants another, and so on. We could use religion: Catholics elect their representatives, Protestants elect theirs, Jews elect theirs, Muslims elect theirs, non-believers elect a rep also.  We could use age: everyone 18 to 30 elects their reps, those 30 to 55 elect theirs, and so on. We could divide ourselves by social class, as France did before the French Revolution: churchmen in the First Estate, noblemen and aristocrats in the Second Estate, and the bourgeoise in the Third Estate. We could do it by income or by net worth: Starting at the top, the wealthiest 710,000 residents of a state elect one congressperson, then the next wealthiest 710,000 elect the second congressperson, on down to the poorest 710,000 residents who elect the last congressperson. We could also dispense with these distinctions altogether, and just let a computer randomly assign each voter in a given state to a congressional district. But having said all that, we in this country still have some affinity for geographic proximity as a basis for deciding which folks belong in the same election district. Hence an automated solution to redistricting should at least take geography into consideration. With rare exceptions, each congressional district should (1) keep together everyone who lives in the same zip code and (2) occupy a contiguous and compact land mass.

Ignore all other considerations: Many state statutes, some federal statues, and countless court cases have addressed the allowable criteria in drawing election districts. Laws and courts have mandated approximately equal populations in each district, as well as geographic compactness. Some redistricting plans consider race or other demographic factors to ensure that the redistricting scheme does not unfairly disadvantage any protected class. Those responsible for drawing district boundaries have also considered (whether openly or secretly) party registrations, previous voting patterns, city and county boundaries, and wealth, among other criteria. Since the 1960’s, the major debates have centered on race, socio-economic status, and political affiliation. It appears that every redistricting plan from every state has been subject to litigation; and courts have had a devil of a time trying to sort out these competing interests in order to decide what is legal, fair, and reasonable. After the last round of redistricting based on the 2010 Census, the redistricting plans in 38 of the 43 states who must redistrict faced legal challenges.

The multiple permissible criteria for setting election district boundaries and the few mandated criteria all seem to relate to the perceived misuse of redistricting as a method to achieve partisan political objectives rather than the need to ensure fair and free elections. But if we remove politicians from the process and focus only on the twin objectives of equalizing population and geographic proximity, we will eliminate the need for all other criteria. This proposed plan does exactly that.

NEXT, we should agree that a non-partisan redistricting solution should satisfy these guidelines:

  • Avoid human decisions concerning any particular election district boundary;
  • Adopt a completely rule-governed process;
  • Follow the same process in every state;
  • Make the process transparent to and replicable by anyone so interested

Avoid human decisions: Most published proposals for redistricting have included some kind of state-level commission to oversee the process in each state. Some plans specify a completely non-partisan body. Other plans specify a commission balanced between the major political parties, perhaps with a third group of non-partisan participants. The problem with all of these proposals is that any such commission can be manipulated or controlled by politicians or lobbyists or other influence peddlers, and the task of guarding against such corruptions becomes a permanent feture of the political landscape. Furthermore, in all of these schemes the state commissions, legislatures, and courts still spend a considerable amount of time and effort drawing election district boundaries; a better system would make the whole thing automatic so we could spend our time discussing substantive issues rather than focusing on process.

Let’s get back to basics here: A bedrock principle of democracy is that the people get to choose their representatives; the representatives do not get to choose their constituents. Anytime politicians draw the election districts, we tend toward a system where the representatives choose the people rather than the reverse.

Therefore, a truly non-partisan redistricting system should remove humans from making any election district boundary decisions. Full Stop.

Rule-Governed Process: Any proposed process should be governed by specific rules that spell out precisely how to draw the election districts, leaving no role whatsoever for human decisions. The process must work for any jurisdiction in the country subdivided into any number of election districts, again maintaining allegiance only to the two principles of equalization of population and geographic proximity. As a corollary, such a mechanical process could be automated. As a computer programmer myself, I know that the process proposed here can be automated. The Federal Elections Commission should publish the resulting computer program and make it freely available to everyone.

Uniform in Every State: The U.S. Constitution, Article I, Section 4 begins with this paragraph:

The Times, Places and Manner of holding Elections for Senators and Representatives, shall be prescribed in each State by the Legislature thereof; but the Congress may at any time by Law make or alter such Regulations, except as to the Place of Chusing [sic] Senators.

Based on that paragraph, Congress established a uniform day for holding the national elections on the first Tuesday after the first Monday of November in even-numbered years. Before that law was passed in 1845, each state chose when to conduct its election. Today we take it for granted that all our national elections take place on the same day, but it was not always so. Based on the same Constitutional authority, in 1965 Congress mandated single-seat election districts. Similarly, Congress has the authority to determine how congressional districts are drawn, since this is very much a part of the “Manner” of holding elections.

The reason for Congressional action with regard to gerrymandering is obvious. If Party A dominates one state and has gerrymandered its congressional districts, while Party B dominates another state and has likewise gerrymandered its districts, why would either state unilaterally disarm? The only viable and long-lasting solution is one that applies equally to every state.

Transparent and Replicable: Drawing election districts should not be rocket science. This is a matter of grouping towns and neighborhoods into relatively compact districts; one should not need an engineering or math degree to do it. The process only becomes complicated when we introduce objectives and criteria which ought to be irrelevant, such as race, wealth, or political affiliation. Therefore, a process that eliminates those irrelevant factors should be something any citizen can comprehend and any citizen with the desire to do so can replicate.

The actual physical process of drawing boundaries for congressional districts should be straight-forward, uncomplicated, easy to understand, completely non-political, and fully automated. Since the computer software to do this has not yet been created, I will explain the process in narrative form.

Redistricting Process

I propose to end all the debate and all the litigation. This solution is not perfect, but it’s far better than any other I’ve encountered to date. I’m not wedded to this particular solution just because I came up with it. Any solution that satisfies the aforementioned principles should be considered. But let me put this one on the table, and then we can all have a discussion about this proposal and any others that may also fit the bill.

The solution proposed here starts with maps and population data for each county in a state and for each Zip Code Tabulation Area (ZCTA).

Since the notion of “counties” (“parishes” in Louisiana and “boroughs” in New York City) is familiar to everyone, I need only explain ZCTA’s. The U.S. Census Bureau developed Zip Code Tabulation Areas (ZCTA’s) from the US Postal Service’s 5-digit zip codes. Think of a ZCTA as a zip code with geographic boundaries. Everyone resides within a ZCTA, even if your mailing address is a P.O. Box and even if your house is not on any mail delivery route. Further, since some zip codes include addresses in two counties or two states, the Census Bureau draws a ZCTA in each jurisdiction and tabulates the population in each. Therefore, we can begin with a map of a given state’s counties and ZCTA’s, along with the population of each county and each ZCTA.

We will draw congressional districts by scanning a state according to a prescribed set of scanning rules, adding counties to a congressional district one by one. When the addition of one more county would make the district’s population too large, we scan the ZCTA’s within that county, adding ZCTA’s to the district until the district contains the requisite population.

Zip Codes and ZCTA’s

Zip Codes, developed by the United States Postal Service, represent mail delivery routes, not populations or geographic boundaries. Mail delivery routes sometimes cross county or state boundaries.

The Census Bureau extended the postal zip codes to represent geographical areas, called Zip Code Tabulation Areas, or ZCTA’s, and then counted the number of people in each ZCTA. ZCTA’s, unlike zip codes, are geographic and do cover the entire country.

Every residence in the country, even Uncle Joe’s cabin on the mountaintop, which is not on any mail delivery route, is inside a ZCTA.

When the area served by a particular zip code crosses a county or state boundary, the Census Bureau creates a ZCTA in each jurisdiction and tabulates the population in each jurisdiction.

 

For example, zip code 20601 in Maryland has mail delivery routes that extend from Charles County across the border into Prince George’s County, so a ZCTA 20601 exists in each county.

For most people, your ZCTA is identified by the name of the county you live in plus your 5-digit zip code.

Scanning Counties and ZCTA’s – The Basic Idea

With a map of counties for a given state in hand, we pass a scan line across the state. When the scan line first touches a county, we add that county to the current congressional district.  We stop adding counties to the current CD when the addition of one more county would cause the target population for that congressional district to be exceeded. At that point, we must decide whether to add, not add, or partition that county, according to this rule: If the CD population is within 5% of the target either with or without this LAST County, then we either add or skip LAST County; otherwise, using the same scan line, we scan the ZCTA’s in that county, adding each ZCTA to the CD until the CD reaches the target population. At that point, the congressional district is complete, and a new district begins..

Because entire counties and, in the last county, entire ZCTA’s belong to one congressional district, neighborhoods vote together. Since the scan line passes incrementally over an entire geographic area, congressional districts are usually both compact and contiguous.

What remains is to determine how to establish the scan lines.

One method that can work well is to draw a Reference Line between the two most widely separated points in the state and then add a Scan Line perpendicular to the Reference Line. Scanning will take place along the Reference Line. Redrawing the Reference Line and Scan Line after completing each CD changes the slope of the Scan Line, and avoids excessively long and narrow CD’s, which would otherwise ensue in large states. Without this rule, Texas, for example, might end up with a number of CD’s 600 to 800 miles long and only one county wide.

The following images show the placement of the Reference Lines, Scan Lines, and election district boundaries for an imaginary state with a population of 2.8 M and four CD’s.

Imaginary State with county boundaries
Initial Reference Line and starting Point

To begin building Congressional District 1, start by drawing a Reference Line between the two most distant points.

Select one end of the Reference Line as the Starting Point – Selection of the Starting Point is like reading a page: If you typed a sentence on the red line, where would you expect the text to begin? We read left to right and top to bottom, so the Starting Point here is on the left.

Reference Line and Scan Line for creating CD 1

Next, draw the Scan Line (blue) perpendicular to the Reference Line and through the Starting Point.

 

Scan Line after scanning for CD 1

Drag the Scan Line along the Reference Line, touching counties as you go. Add each touched county to CD 1. Stop adding counties when the addition of one more county would cause the CD 1 population to be too large. This “one more county” becomes LAST County for the current CD. Partition that county by scanning ZCTA’’s and adding them to the CD; stop when the target population is reached.

CD 1 – Green

This is the completed CD 1. For this fictitious example, the 2.8 M population is fairly evenly distributed across the state. The CD boundaries would be different if, for example, a city of 2 million were concentrated in one corner of the state.

(In this imaginary example, we are not partitioning LAST County – Each CD will consist only of entire counties. We are also not detailing the number of people in each county.)

Reference Line and Scan Line for CD 2 (previously-assigned CD’s appear in black)

For CD 2, redraw the Reference Line, Starting Point, and Scan Line.

CD 2 – Yellow

Voila! CD 2 complete.

Reference Line and Scan Line for CD 3

For CD 3, again redraw the Reference Line, Starting Point, and Scan Line

CD 3 – Red

And CD 3 is complete.

CD 4 (everything not yet assigned) – Blue

CD 4 complete: By definition, after all CD’s save one have been drawn, the last CD perforce contains everything left over. Hence, for the last CD in any state, we do not need to draw the Reference Line, Starting Point, or Scan Line.

Here is a summary of the proposed process for drawing congressional districts in every state:

  1. Create a Reference Line, connecting the two most distant points in the area of the state not yet assigned to a CD.
  2. Determine the Starting Point at the NW end of the Reference Line. (Later in this paper, I provide more detailed rules for determining the Starting Point.)
  3. Create a Scan Line, perpendicular to the Reference Line & thru the Starting Point.
  4. Keeping the Scan Line perpendicular to the Reference Line, drag the Scan Line along the Reference Line, touching counties.
  5. Add each touched county (land and population) to the current district.
  6. Stop when adding one more county (“LAST County”) would put too many people in the CD.
  7. Add or skip LAST County in its entirety, if doing so would result in a total CD population within 5% of target. Otherwise, partition LAST County: using the same Scan Line, scan LAST County, adding ZCTA’s until the CD population target is reached.
  8. Repeat these steps until all CD’s for the state are created

Here are a few sample states, showing the initial Reference Line and Scan Line, used for drawing CD 1 in each state:

Proposed Solution Applied to Maryland

Let’s apply this scheme to Maryland as our test case. I chose Maryland as the sample state for three reasons:

  • With eight congressional districts, Maryland is of average population.
  • Maryland has two major metro areas (Baltimore and the Maryland suburbs of Washington, DC) and two large rural areas (western Maryland and the Eastern Shore, the area east of the Chesapeake Bay), and Maryland is one of the most weirdly shaped states in the nation. Hence, a redistricting scheme that can work for this oddly-shaped and unevenly-populated state might well work for them all.
  • I live here! So I’m partial to my home state. In addition, Maryland is seriously gerrymandered in favor of Dems, and even though I myself am a Democrat, I consider our own gerrymandering shameful, so I’d like to come up with a scheme to fix this.

We begin with a map of Maryland’s counties (Maryland has 23 counties plus Baltimore City) and the population of each:

Maryland Population by County

County

Population

Allegany MD

75087

Anne Arundel MD

537656

Baltimore County MD

805029

Calvert MD

88737

Caroline MD

33066

Carroll MD

167134

Cecil MD

101108

Charles MD

146551

Dorchester MD

32618

Frederick MD

233385

Garrett MD

30097

Harford MD

244826

Howard MD

287085

Kent MD

20197

Montgomery MD

971777

Prince Georges MD

863420

Queen Annes MD

47798

St. Marys MD

105151

Somerset MD

26470

Talbot MD

37782

Washington MD

147430

Wicomico MD

98733

Worcester MD

51454

Baltimore City MD

620961

total

5773552

Maryland Congressional District 1

Next, let us create Maryland’s 1st congressional district. We begin by drawing the Reference Line connecting Maryland’s two most widely separated points. We determine the Starting Point as the northwest corner of the state, and we draw the Scan Line perpendicular to the Reference Line, like this:

You may have noticed two possible human decision points when creating the Reference Line and the Scan Line. Since we wish to avoid human decisions, let’s dispense with these issues. In making these rules, we use the convention of reading a book in English. When you open a page of a book, you start in the northwest corner and proceed to the southeast corner. That is the source of the following rules:

  1. Reference Line rule: If an area has more than one potential Reference Line, meaning two potential longest lines are equally long, then choose as the Reference Line the one whose slope is closest to a diagonal from NW to SE.
  2. Starting Point rule 1: If just one end of the Reference Line is in a county that abuts an area of the state previously assigned to a congressional district, then that end of the Reference Line is the Starting Point.
  3. Starting Point rule 2: If rule 1 does not apply, choose the northwest end as the Starting Point. If the northernmost point is not also the westernmost point, then: if the slope is more vertical than horizontal, choose the northernmost point; else choose the westernmost point
  4. End Point rule: The point at the opposite end of the Reference Line from the Starting Point is the End Point. 

After creating the initial Reference Line and Scan Line, we drag the Scan Line from the Starting Point towards the End Point, keeping its slope constant. As the Scan Line encounters each county, we add that county to CD 1 – both its land area and its population. This takes us through Garrett, Allegany, Washington, Frederick, and Carroll counties and into Montgomery. But we cannot add all of Montgomery County to Maryland Congressional District 1, because CD 1 would then have too many people. Therefore, Montgomery becomes LAST County in CD 1. Continuing with the Scan Line perpendicular to the Reference Line, we scan ZCTA’s across Montgomery until the CD is complete – assigning ~7% of Montgomery’s population to CD 1. Here is the resulting map (showing the CD boundaries) and spreadsheet (showing the county populations):

CD’s Drawn by Scanning Counties (LAST County by ZCTA’s) – Maryland

Total MD 2010 census population = 5773552. Target population per CD = 721694. MD has 8 CD’s.

 

Master Congressional District Table

CD’s Drawn by Scanning Counties

CD#

County

County Population

Running Aggregates

Available

Assigned to this CD

LAST/ CONT

% Used

Left Over

State Total

This CD

Total Target

Diff

1

Garrett

30097

30097

     

30097

30097

721694

691597

1

Allegany

75087

75087

     

105184

105184

721694

616510

1

Washington

147430

147430

     

252614

252614

721694

469080

1

Frederick

233385

233385

     

485999

485999

721694

235695

1

Carroll

167134

167134

     

653133

653133

721694

68561

1

Montgomery

971777

68561

LAST

7%

903216

721694

721694

721694

0

Subdividing MD CD 1’s LAST County, Montgomery County, by Scanning ZCTA’s

Whenever we need to subdivide a county, we will also need a map of that county’s ZCTA’s. Here is a map of Maryland’s 546 ZCTA’s. Obviously, in order to draw the actual congressional district boundary for each district’s LAST County, we will need a map with far greater resolution than this one. This freely-downloadable map can be enlarged somewhat, including approximate boundaries for ZCTA’s within each county, but we still need more detail to complete this task accurately. Detailed maps are available from the State of Maryland’s iMAP project office.

Special note on subdividing LAST County: In some cases the total CD population may fall within 5% of the average population for one CD either by adding or by excluding LAST County. In such a case (which does not occur in the Maryland example), we can add or exclude the entire county, so that there is no LAST County. 

As noted before, Maryland has an odd shape, dating from Colonial times. The northern border separates Maryland from Pennsylvania and is the famed Mason-Dixon Line. Maryland’s three westernmost counties are barely connected to the rest of the state: At its narrowest point in Hancock, the state is only three miles wide along I-70 and I-68. East of the Chesapeake Bay, Maryland’s Eastern Shore shares the DelMarVa Peninsula with Delaware and a piece of Virginia.

Just glancing at the ZCTA map of the whole state, note the cluster of ZCTA’s in the middle of the state from the Washington, DC suburbs to the Baltimore metro area. Tightly packed zip codes betoken concentrated populations, while larger ZCTA’s identify areas with fewer people.

To complete CD 1, we need the ZCTA map of Montgomery County, MD CD 1’s LAST County. Below is an enlargement of the northwestern section of Montgomery County. In addition to the ZCTA map, we need a list of the population in each ZCTA, which we can download from the Missouri Census Data Center as a .csv or Excel file.

 

To complete the current CD, we scan the LAST County using the map of ZCTA’s for that county along with population data for all ZCTA’s in LAST County. We add ZCTA’s to the current CD until the addition of one more ZCTA would cause the target population to be exceeded. We then add, or do not add, this Last ZCTA to the current CD, depending on whether the addition of this ZCTA would or would not bring the aggregate population closer to target.

LAST, CONT, & BOTH – Definitions

In the tables shown in this demo, we add a county to a CD either in whole or in part when the Scan Line touches that county. If we assign the entire county, then the LAST/CONT, % Used, and Left Over columns are left blank.

However, If we assign only part of a county to the current CD, then we show the percentage of the county’s population assigned to the current CD in the % Used column, and we enter the population not yet assigned to any CD in the Left Over column. The LAST/CONT column gets one of three words, as follows:

  • LAST: if the county becomes LAST County in a CD, and if that county has not previously been LAST County for an earlier CD, then the word LAST appears in the LAST/CONT column. Note that the first time a partial county is assigned, it must be LAST; also note that a county designated as LAST must appear in a subsequent CD as CONTinued. Note also that some of LAST County’s population will be Left Over for subsequent assignment to another CD.
  • CONT: If a county was already LAST County for an earlier CD and if it is not LAST County for the current CD, then the word CONT (meaning “CoNTinued from an earlier CD”) is entered in the LAST/CONT column. Note that (1) we can only assign CONT to a county that was previously LAST in an earlier CD; (2) a county can be CONT only once; and (3) we must assign all remaining population of a county designated CONTinued to the current CD, so that Left Over is 0 by definition.
  • BOTH: If a county was already LAST County for an earlier CD and if it is also LAST County for the current CD, then this county is both CONTinued from an earlier CD and LAST county for the current CD, and the word BOTH is entered in the LAST/CONT column. Note that a county designated as BOTH must appear in a subsequent CD as CONT. It occasionally occurs that a county appears as BOTH more than once, but such a county will then have appeared at least four times, as LAST, BOTH, BOTH, and CONT.
Could the Last ZCTA in LAST County Be Too Large?

At this point, before deciding that the current CD is complete, we must check the CD’s population to insure that it falls within 5% of the target population for a CD in this state. Though technically possible, exceeding this ±5% criterion is most unlikely, based on the following logic:

  1. Based on the 2010 Census figures, there were 709,747 residents per House member (that is, total population divided by 435). Hence that is the average Target per House seat across the country. 5% of this target is 35,487. We either add or do not add Last ZCTA to the congressional district, depending on whether or not the addition of Last ZCTA brings the total congressional district population closer to Target or not. Hence to exceed the 5% threshold, Last ZCTA would have to have more than 70,974 residents, and the breakpoint would have to be such that the smaller portion would have to exceed 35,487.
  2. Of the 33,092 ZCTA’s in the United States, only 126 have a population over 70,974. (The average population per ZCTA is 9330.)
  3. The greatest population ZCTA in the country (ZCTA 60629, near Chicago’s Midway Airport) had 113,916 residents in 2010. If that ZCTA happens to be the last ZCTA added or not added to complete a congressional district, and if the target was exactly in the middle of that ZCTA, then that CD would miss Target by a maximum of that ZCTA’s population divided by 2, or 56,958, which is about 9% of the per-seat target for Illinois. Technically possible, but not likely.

In short, the likelihood of creating a congressional district whose population is too small or too large is tiny. But if such an event were to occur, we can split a ZCTA into two parts by census tracts, which contain ~4000 people per tract. Just use the same Scan Line to scan the census tracts within the Last ZCTA, and stop adding census tracts when Target is reached.

Subdividing LAST County for This Demo

For this exercise I did not obtain detailed ZCTA maps of Maryland’s counties. I am simply pointing out that someone can scan the ZCTA map for a county and add ZCTA’s to the congressional district quite readily, if that someone has in hand the requisite ZCTA maps and ZCTA population data.

I took a shortcut for this demonstration: I added to the CD the population from LAST County needed to complete that CD. I calculated the percentage of LAST County’s population assigned to that CD and drew an approximate boundary line based on the same percentage of land mass. This shortcut is less than precise but seems adequate for demo purposes.

With Maryland’s CD 1 complete, we can proceed to the other CD’s. In each case, we draw a new Reference Line connecting the two most distant points in the state not yet assigned to a CD. We select the Starting Point on the Reference Line, and we draw the Scan Line perpendicular to the Reference Line and through the Starting Point. We then drag the Scan Line towards the End Point, adding counties as we go, and adding ZCTA’s as needed in LAST County. Here are the maps of the Reference Line and Scan Line, as well as the maps of the resulting CD’s, for Maryland CD’s 2 through 7:

 

 

Maryland Congressional Districts 2 Through 7.

The procedure for each MD CD 2 through 7 is the same. We present a map showing the Reference Line and Scan Line for the CD, the map of the CD showing the counties and partial counties that make up the CD, and a table listing the populations of each county or partial county assigned to the CCD.

MD CD 2:

For each CD after CD 1, the first data row in the Master Congressional District Table displays the aggregate state population after completing the previous CD, along with the new target and difference for the current CD.

 

Master Congressional District Table

CD’s Drawn by Scanning Counties

CD#

County

County Population

Running Aggregates

Available

Assigned to this CD

LAST/ CONT

% Used

Left Over

State Total

This CD

Total Target

Diff

2

Previous State Total and New Target for This CD

721694

0

1443388

721694

2

Baltimore County

805029

721694

LAST

90%

83335

1443388

721694

1443388

0

 

MD CD 3:

Note the column named “LAST/CONT” in the Master Congressional District Table for CD 3 below. “LAST” is entered to identify LAST County for each CD; “CONT” is entered to identify a county CONTinued from a previous CD in which that county was LAST. “BOTH” is entered for a county which is both CONTinued from a previous CD and LAST County in the current CD. In all three cases, % Used indicates the percentage of that county’s total population assigned to the current CD, and Left Over indicates the population remaining to be assigned to a later CD.

Master Congressional District Table

CD’s Drawn by Scanning Counties

CD#

County

County Population

Running Aggregates

Available

Assigned to this CD

LAST/ CONT

% Used

Left Over

State Total

This CD

Total Target

Diff

3

Previous State Total and New Target for This CD

1443388

0

2165082

721694

3

Montgomery

903216

721694

BOTH

74%

181522

2165082

721694

2165082

0

 

MD CD 4:

Looking at the map of Maryland CD 4, note that this CD is not entirely contiguous. A few miles of Baltimore County separate the portion of the CD in Baltimore City from the portion in Howard and adjoining Montgomery counties. If we apply the proposed scheme consistently, a few such non-contiguous CD’s will occur, so you should know that it can and does occur. Allowing this to occur is far better than artificially fixing it with a narrow corridor connecting the two parts. Contiguousness is a desirable characteristic, but it is not essential to having fair districts.

Master Congressional District Table

CD’s Drawn by Scanning Counties

CD#

County

County Population

Running Aggregates

Available

Assigned to this CD

LAST/ CONT

% Used

Left Over

State Total

This CD

Total Target

Diff

4

Previous State Total and New Target for This CD

2165082

0

2886776

721694

4

Howard

287085

287085

     

2452167

287085

2886776

434609

4

Montgomery

181522

181522

CONT

19%

0

2633689

468607

2886776

253087

4

Baltimore City

620961

253087

LAST

41%

367874

2886776

721694

2886776

0

 

MD  CD 5:

Master Congressional District Table

CD’s Drawn by Scanning Counties

CD#

County

County Population

Running Aggregates

Available

Assigned to this CD

LAST/ CONT

% Used

Left Over

State Total

This CD

Total Target

Diff

5

Previous State Total and New Target for This CD

2886776

0

3608470

721694

5

Harford

244826

244826

     

3131602

244826

3608470

476868

5

Cecil

101108

101108

     

3232710

345934

3608470

375760

5

Baltimore County

83335

83335

CONT

10%

0

3316045

429269

3608470

292425

5

Baltimore City

367874

292425

BOTH

47%

75449

3608470

721694

3608470

0

MD CD 6:

Master Congressional District Table

CD’s Drawn by Scanning Counties

CD#

County

County Population

Running Aggregates

Available

Assigned to this CD

LAST/ CONT

% Used

Left Over

State Total

This CD

Total Target

Diff

6

Previous State Total and New Target for This CD

3608470

0

4330164

721694

6

Prince George’s

863420

721694

LAST

84%

141726

4330164

721694

4330164

0

 

MD CD 7:

Master Congressional District Table

CD’s Drawn by Scanning Counties

CD#

County

County Population

Running Aggregates

Available

Assigned to this CD

LAST/ CONT

% Used

Left Over

State Total

This CD

Total Target

Diff

7

Previous State Total and New Target for This CD

4330164

0

5051858

721694

7

Charles

146551

146551

     

4476715

146551

5051858

575143

7

Prince George’s

141726

141726

CONT

16%

0

4618441

288277

5051858

433417

7

St. Mary’s

105151

105151

     

4723592

393428

5051858

328266

7

Anne Arundel

537656

328266

LAST

61%

209390

5051858

721694

5051858

0

 

MD CD 8:

We do not need a Reference Line and Scan Line for the last CD in a state, because, by definition, any area/population not yet assigned to a CD must be assigned to the last one. So here is CD 8, consisting of the entire Eastern Shore (area east of the Chesapeake Bay) and the remaining bits west of the bay in Baltimore City and Anne Arundel County.

 

Master Congressional District Table

CD’s Drawn by Scanning Counties

CD#

County

County Population

Running Aggregates

Available

Assigned to this CD

LAST/ CONT

% Used

Left Over

State Total

This CD

Total Target

Diff

8

Previous State Total and New Target for This CD

5051858

0

5773552

721694

8

Anne Arundel

209390

209390

CONT

39%

0

5261248

209390

5773552

512304

8

Baltimore City

75449

75449

CONT

12%

0

5336697

284839

5773552

436855

8

Kent

20197

20197

     

5356894

305036

5773552

416658

8

Calvert

88737

88737

     

5445631

393773

5773552

327921

8

Caroline

33066

33066

     

5478697

426839

5773552

294855

8

Queen Anne’s

47798

47798

     

5526495

474637

5773552

247057

8

Somerset

26470

26470

     

5552965

501107

5773552

220587

8

Talbot

37782

37782

     

5590747

538889

5773552

182805

8

Wicomico

98733

98733

     

5689480

637622

5773552

84072

8

Dorchester

32618

32618

     

5722098

670240

5773552

51454

8

Worcester

51454

51454

LAST

100%

0

5773552

721694

5773552

0

 

All MD CD’s, and Political Implications:

Here is the approximate map of all eight Maryland CD’s according to the proposed scheme:

Political Results

Of course, everyone reading this wants to know how this might affect ME. This is especially true if you are a Maryland politician or Maryland voter. While no one can predict how people will vote in future elections, we can get a hint of expected results by examining the voting patterns of the areas that make up each of the proposed CD’s.

As a starting point, Maryland Democrats outnumber Republicans about two to one. In this heavily gerrymandered state, Maryland elected seven Dems and one Repub to the House in each of the last two elections. None of the third party, independent, or write-in candidates moved the needle in any of these 16 elections, so I decided to tabulate only the votes for Republicans and for Democrats.  

To get a hint at expected political results using the CD boundaries proposed herein, I added up the votes cast in the general elections for the House of Representatives in 2016 and 2018. If a county is wholly within one CD, I added all the votes from that county to the totals for that CD. If only a percentage of a county lies within a CD, I multiplied the total county votes by the percentage of the county population that lies within that CD. Here are the results:

Total Votes Cast in MD in 2016 & 2018

CD

W

Rep

Dem

1

R

348840

57%

267680

43%

2

D

230308

38%

381271

62%

3

D

142097

23%

483578

77%

4

D

149235

24%

466924

76%

5

D

238439

42%

325244

58%

6

D

47104

8%

522075

92%

7

D

229021

37%

389790

63%

8

R

315036

52%

292769

48%

All

 

1700080

35%

3129331

65%

 

As expected, Democrats in Maryland got 65% of the vote for major party candidates, and Republicans got 35%. Democrats would have won six of Maryland’s eight seats in the House. Only one CD is fairly competitive: CD 8 would have been a Republican win by 4 points. The other Republican victory (CD 1) would have been by 17% (57 to 43). All six Democratic wins would have been by wider margins.

Racial Profile of Proposed CD’s

The chart below, based on census data, shows the percentage of residents in each proposed CD who identify with each race.

Ever since a Boston cartoonist in 1812 introduced the term “gerrymander” to the American political lexicon, Americans have argued about unfair influences on drawing election district boundaries. Editorials and law suits abound concerning the influence of race, religion, party registrations, age, income, education, and rural-versus-urban demographic characteristics of the population.

 

Of all the characteristics that have been the subject of discussion and debate, the one factor that always seems to dominate such discussions is race. Hence this chart might be instructive. But also consider this:

  • Half the population is female, but 0 of 8 House Members are female.
  • 31% of the population is African-American; currently 2 of 8 House Members are African-American.
  • 30% of voters are registered as Republican, but only 1 of 8 House Members is a Republican.
  • The mostly white, mostly Democratic voters of Maryland elected a black Republican Lieutenant Governor in 2002 and voted twice for an African-American for President. So race and party affiliation are important and instructive but not determinative.

Considering only race as a predictor of election results, the chart shows that a black candidate would be expected to win CD 6, and black candidates would have a reasonable chance to win CD 4 and CD 5 and perhaps even CD 7.

Avoiding Future Gerrymanders

One more item is necessary to complete this redistricting scheme: Once adopted, how do we prevent politicians (or anyone else) from playing games with the county and ZCTA boundaries, in an effort to gerrymander? The answer is that we adopt the county and ZCTA boundaries used for the 2010 census in perpetuity, updated by population data from the census every 10 years. (By the way, this does not mean that states cannot change county boundaries or that the Postal Service cannot add or alter zip codes; what it does mean is that, for purposes of allocating seats in the next Congress, the Census Bureau will count the people in each county and each ZCTA using the boundaries from the 2010 Census.)

Technical note: These maps of Maryland congressional districts are approximations. I used county population data for whole counties and for portions of counties as needed to complete each congressional district. The lines are adequate for demo purposes, but they are not exact. I did not attempt to allocate individual ZCTA’s to each district’s LAST County.

Proposed Solution Applied to Pennsylvania and North Carolina

For additional examples, I tried out this scheme on Pennsylvania, which has 18 seats in the House of representatives, and on North Carolina, which has 13 seats. The final results appear below. The details appear in Appendix 2:

Pennsylvania:

North Carolina:

The Rules in Excruciating Detail

For die-hard redistricting enthusiasts among my readers, I prepared the detailed steps for redistricting any state and included them in Appendix 3.

Conclusion and next steps

When it comes to re-drawing legislative boundaries, gerrymandering is a significant impediment to our “Forming a More Perfect Union”. Until now, the technical challenge is that the process itself has been tedious and complicated. The political challenge is that those who stand to benefit from the results of the process have overseen the process.

Gerrymandering can be solved if we simply decide to do it, and it’s not all that hard to accomplish. It only becomes hard when we take into consideration factors other than population and geographic proximity. If we implement redistricting based solely on these two factors, we can pretty well lick the gerrymandering problem in one reapportionment cycle. Furthermore, if we adopt this rule-based solution, we can automate it.

Multi Seat CD’s Combined with Ranked Choice Voting

This solution by itself does not eliminate gerrymandering. Rather, this solution makes gerrymandering less important.

Multi-Seat Congressional Districts

In Part I, we mentioned that from time to time some states have had multi-seat CD’s. States sometimes created multi-seat CD’s as a tool in partisan redistricting and as a method for disenfranchising black voters. This worked as follows: A single-seat CD has a majority black population, let’s say 2/3 black. Two adjacent CD’s have majority white populations, let’s say 2/3 white. All three CD’s together have a majority white population, about 56%. Therefore, in the traditional voting scheme, where voters select three candidates to fill three seats, three whites are likely to win.

A 1965 civil rights era law ended multi-seat CD’s, but that occurred before Ranked Choice Voting was on the table. Let’s apply RCV and multi-seat CD’s to the question of CD boundaries and the problem of gerrymandering. The addition of RCV vote-counting procedures to multi-seat CD’s insures a more equitable distribution of winners.

In this paper, we first describe how to construct the House of Representatives with as many three-seat CD’s as possible. Second, we describe an adaptation of Ranked Choice Voting that will give the minority party in each three-seat CD a clear opportunity to win one seat and even a reasonable chance at winning two.

We should require states with fewer than four seats in the House to elect all their representatives at-large. At present seven states have only one House seat, so these seven are already elected at-large. Eight states currently have two or three House seats. Hence, with this new rule, fifteen states are removed from the Congressional gerrymandering challenge. And eleven additional states have 4, 5, or 6 seats, so these states would have two CD’s and would only need to draw one congressional district boundary between them.

With the mandated three seats per CD, the number of 3-seat CD’s in any state is given by the formula N = INT(Seats/3), where N is the number of 3-seat CD’s and Seats is the number of seats apportioned to a state. If Seats is not evenly divisible by 3, then that state will have one CD with either one or two seats. This will always be the highest numbered seat. So, for example, Maryland’s eight seats will be distributed as follows: CD 1 and CD 2 will each contain 3 seats, and CD 3 will contain 2 seats. Here is a table of the CD’s in all 50 states, based on the apportionment of House seats following the 2010 Census:

 

Seats: # of house seats apportioned to each state following the 2010 census.

3-Seat CD’s: # of CD’s that elect 3 members, viz., INT(Seats/3).

2-Seat CD: x indicates a 2-Seat CD, to wit: If Seats modulo 3 = 2, then this state has a 2-Seat CD.

1-Seat CD: x indicates a 1-Seat CD, to wit: If Seats modulo 3 = 1, then this state has a 1-Seat CD.

All At-Large: x indicates this state has 3 or fewer seats, so all seats are in one CD.

Apportionment Populations 2010 and number of CD’s with 3 seats per CD

State

Population

Pop per Seat

Seats

3-Seat CD’s

2-Seat CD

1-Seat CD

All At-Large

Alabama

4,802,982

686,140

7

2

 

x

 

Alaska

721,523

721,523

1

0

 

x

x

Arizona

6,412,700

712,522

9

3

     

Arkansas

2,926,229

731,557

4

1

 

x

 

California

37,341,989

704,566

53

17

x

   

Colorado

5,044,930

720,704

7

2

 

x

 

Connecticut

3,581,628

716,326

5

1

x

   

Delaware

900,877

900,877

1

0

 

x

x

Florida

18,900,773

700,029

27

9

     

Georgia

9,727,566

694,826

14

4

x

   

Hawaii

1,366,862

683,431

2

0

x

 

x

Idaho

1,573,499

786,750

2

0

x

 

x

Illinois

12,864,380

714,688

18

6

     

Indiana

6,501,582

722,398

9

3

     

Iowa

3,053,787

763,447

4

1

 

x

 

Kansas

2,863,813

715,953

4

1

 

x

 

Kentucky

4,350,606

725,101

6

2

     

Louisiana

4,553,962

758,994

6

2

     

Maine

1,333,074

666,537

2

0

x

 

x

Maryland

5,789,929

723,741

8

2

x

   

Massachusetts

6,559,644

728,849

9

3

     

Michigan

9,911,626

707,973

14

4

x

   

Minnesota

5,314,879

664,360

8

2

x

   

Mississippi

2,978,240

744,560

4

1

 

x

 

Missouri

6,011,478

751,435

8

2

x

   

Montana

994,416

994,416

1

0

 

x

x

Nebraska

1,831,825

610,608

3

1

   

x

Nevada

2,709,432

677,358

4

1

 

x

 

New Hampshire

1,321,445

660,723

2

0

x

 

x

New Jersey

8,807,501

733,958

12

4

     

New Mexico

2,067,273

689,091

3

1

   

x

New York

19,421,055

719,298

27

9

     

North Carolina

9,565,781

735,829

13

4

 

x

 

North Dakota

675,905

675,905

1

0

 

x

x

Ohio

11,568,495

723,031

16

5

 

x

 

Oklahoma

3,764,882

752,976

5

1

x

   

Oregon

3,848,606

769,721

5

1

x

   

Pennsylvania

12,734,905

707,495

18

6

     

Rhode Island

1,055,247

527,624

2

0

x

 

x

South Carolina

4,645,975

663,711

7

2

 

x

 

South Dakota

819,761

819,761

1

0

 

x

x

Tennessee

6,375,431

708,381

9

3

     

Texas

25,268,418

701,901

36

12

     

Utah

2,770,765

692,691

4

1

 

x

 

Vermont

630,337

630,337

1

0

 

x

x

 

 

 

 

 

 

 

 

Virginia

8,037,736

730,703

11

3

x

   

Washington

6,753,369

675,337

10

3

 

x

 

West Virginia

1,859,815

619,938

3

1

   

x

Wisconsin

5,698,230

712,279

8

2

x

   

Wyoming

568,300

568,300

1

0

 

x

x

TOTAL

309,183,463

710,767

435

145

16

19

15

                 

 

Sample CD’s

Applying the redistricting solution proposed earlier to two sample states, Louisiana and Minnesota, here are the resulting CD’s.

Louisiana (6 seats, hence 2 CD’s of 3 seats each)

Louisiana’s completed CD’s, based on 3-seat CD’s. For the details, see Appendix 4:

Minnesota (8 seats, hence 2 CD’s of 3 seats each plus one CD with 2 seats)

Minnesota’s completed CD’s, based on 3-seat CD’s. For the details, see Appendix 5:

Ranked Choice Voting in Multi-Seat Congressional Districts[1]

The basic idea here is two-fold:

  1. Ranked Choice Voting (RCV) increases the probability that, in any election with multiple winners, opinions held by a significant portion of the electorate will be represented by at least some of the winners.
  2. With RCV, we can expect that multi-seat Congressional Districts will result in a majority of winners who represent the majority view of a CD, but some winners will also represent a widely-held minority view.

The voting and counting procedure proposed for multi-seat Congressional Districts is the same as the procedure for multi-seat Senate races. To wit:

VOTING in all primary and general elections,: Voters select a 1st, a 2nd, and a 3rd choice.

COUNTING of ballots:

  • Elections in single-seat CD’s: Follow the rules for open primaries described above in the section “Counting of ballots in primary elections for Congress”. For the general election, follow the normal RCV rules for an election with one winner.
  • Elections in two-seat and three-seat CD’s: Calculate the weighted vote for each candidate, which equals 3 X the candidate’s 1st choice votes + 2 X the 2nd choice votes + the 3rd choice votes. Rank order the results.

This is the table of successful primary and general election candidates for all House contests:

Number of House seats

Voters select

Counting process

Primary winners

General winners

1

1st, 2nd, and 3rd choices

RCV for one seat

Last 3, or 3 > 25%

1 > 50%

2

1st, 2nd, and 3rd choices

Weighted voting

Top 4

Top 2

3

1st, 2nd, and 3rd choices

Weighted voting

Top 5

Top 3

 

Political Impact

To see how multi-seat CD’s with Ranked Choice Voting might affect the political makeup of Congress, let’s examine the expected results in three-seat and in two-seat CD’s dominated by one political party.

Three-Seat Congressional Districts

The scheme outlined here, based on the 2010 census and apportionment, would result in 145 three-seat Congressional Districts. Based on county voting patterns of the recent past, we can expect that many of these three-seat CD’s would be dominated by either Republicans or Democrats. So what would be the result? Given that RCV systems tend to favor candidates who appeal to all sides and also give an opportunity for independents and third parties, the results surmised here may not obtain for even one election cycle much less for a decade. Nevertheless, we might try to guesstimate the result.

Given two major parties, Party A and Party B, if Party A has 55% of the voters and Party B has 45%, in most elections we can expect Party A to end up with two seats, and Party B to win one seat. Here is the logic, using plausible suppositions. To keep this simple, let’s assume that there are exactly 100 voters.

Each party fields three (or more) candidates for the three available seats. Each party has one very strong candidate, along with two less strong candidates. So the candidates are designated as A1, A2, and A3 from Party A, and B1, B2, and B3 from Party B. Let’s also assume that Party A’s voters select A1 as their 1st choice, A2 as their 2nd choice, and A3 as their 3rd choice. Similarly, Party B’s voters select B1, B2, and B3, in that order.

Applying the RCV procedure for tabulating votes in a three-winner election, we can expect the winners to be candidates A1 (165 weighted total vote), B1 (135), and A2 (110). Here is the table of voting results:

Three-seat CD

Candidate

Votes

Weighted Total

 

1st

2nd

3rd

 

A1

55

   

165

A2

 

55

 

110

A3

   

55

55

B1

45

   

135

B2

 

45

 

90

B3

   

45

45

 

Two-Seat Congressional Districts

Given the same 55% to 45% advantage for Party A over Party B, in a two-seat contest we can expect each party to win one seat. The voting for two positions proceeds exactly the same as the voting for three positions, and the winners are candidates A1 and B1.

Conclusions

We can derive several conclusions from this analysis: 1) We are likely to end up with better representation of voters’ views using RCV and multi-seat CDs, and 2) candidates of both Party A and Party B will increase their chances of winning if they appeal to voters of both parties. This tendency will be even more pronounced if we also adopt single, open primaries for all offices. The result will almost inevitably be a Congress that is less partisan, less extreme at both ends of the political spectrum, and more interested in catering to the needs of all voters in their districts.

The advantages of RCV combined with multi-seat districts also accrue to states that adopt these reforms for their state legislatures.

[1] Appendix 1 summarizes all the RCV procedures and variations recommended in this book.

Leave a Reply

Your email address will not be published. Required fields are marked *