The Journal Gazette
 
 
Sunday, September 26, 2021 1:00 am

Simulations show minimal effect on outcomes

Bruce Reidenbach

With the publishing of the most recent census results comes the effort to redraw the congressional districts that will be in place for the next 10 years.

Often, following the redistricting proposals are complaints that the party in power is purposely drawing these districts to skew the voting in such a way to retain its power, i.e., gerrymandering. The most common argument put forth is that with a 60/40 statewide split in party voting, the districts should also represent this same split.

In an effort to try to see what a fair congressional redistricting would look like, I developed a computer algorithm that randomly simulates creation of congressional districts. The algorithm only considers county borders and populations to determine the best case of distributing the overall population into each district.

The decision to use county borders and populations is based on four reasons.

First, counties were created to group people with similar needs and concerns together into a common governmental unit, which is also the reason for creating congressional districts.

Second, the county boundaries are very stable: the last change was in March of 1931 when LaPorte County exchanged a small area with Saint Joseph County.

Third, it helps constrain the magnitude of the possible combinations (even so, there are 1.7×1082 – that's 17 followed by 81 zeros – ways of fitting 92 counties into nine congressional districts).

Finally, there is precedence: Only eight Indiana counties are split between congressional districts in the current proposal put forth by the Republicans. The remaining district boundaries follow county boundaries.

There are obviously several problems with this approach, the primary one being that the overall population will not be evenly distributed within each district. Therefore, the goal is to attempt to find the combination of counties that results in the least population variance within each district. A second problem is that the population of Marion County (977,203) is greater than the ideal district population (753,947).

To overcome this, I split the population of Marion County into the nine townships that make up that county, effectively replacing it with nine “pseudo-counties.”

Besides the goal of minimizing district population variance, the simulations also measure the amount of perceived gerrymandering of the results. Ideally, each district would fit together like squares on a checkerboard.

Unfortunately, with the irregularity of county boundaries and the disparity of their populations, this is not possible.

A method of measuring how well each district fits together is to draw the smallest rectangle around each district that includes each assigned county and measure the total overlap between districts. The smaller the amount of overlap, the better the districts “fit” together and the less likely there is to be perceived gerrymandering.

Trying to exhaustively solve this problem by examining each possible combination is prohibitive. Instead, the algorithm I developed uses what is known as a Monte Carlo simulation, where randomly selected county assignments are made and the resulting district population variance and overlap are computed.

Monte Carlo simulations are a common practice used for problems of this sort. While the ideal combination may not be found, there will be many combinations where the results are likely near-ideal.

Over hundreds of millions of simulation trials, the best result is shown in the accompanying graphic. Interestingly, if one totals up the 2020 congressional party votes for the counties and townships within each district, the results indicate that seven districts would elect a Republican and two districts would elect a Democrat.

This is not an isolated case. In every single case where the population variance and district overlap are minimized, the resultant election results produce the same 7-2 split.

How can this be? Given a 60/40 statewide split in party voting, the complaint is that there should be a near 60/40 split in congressional party assignments. What is missing from this argument is how these votes are distributed geographically.

In the 2020 congressional election, there were only four counties – Lake, Marion, Monroe and Saint Joseph – that had more votes for the Democratic congressional candidate than for the Republican candidate.

In the case of Monroe and Saint Joseph counties, the vote margins are slim enough that when they are combined with the surrounding counties (which vote majority Republican) to form the congressional district, the overall vote totals favor the Republican by quite a large margin (greater than 50,000 votes).

For Lake and Marion counties, the county population and number of votes for the Democratic candidate are enough to keep each of the districts they are assigned to solidly Democratic. The remaining counties vote Republican, so no matter how they are combined, the result is a Republican district. Thus the 7-2 split.

Another common complaint concerns splitting Marion County into two districts, which is seen as a way of diluting that county's vote. However, as described previously, Marion County needs to be split because the population exceeds the target district population by almost 225,000 people.

In the 2020 election, the northernmost six townships voted Democratic while the southernmost three townships voted Republican. The 2020 census population of the six northernmost townships is within 1,734 of the ideal district population (752,213 vs. 753,947), so it makes sense to group these six townships together since the population is nearly ideal and the resulting district reflects the views of its constituents.

Consequently, instead of diluting the vote, the current Republican proposal keeps this Democratic-majority voting bloc intact.

Therefore, the conclusion from this series of simulations is that there is no concrete evidence of gerrymandering based solely on the belief that the party elected to each district should follow the statewide split in party voting. That is not to say that gerrymandering is not taking place; however, the evidence does not match the arguments being put forth.

Bruce Reidenbach is a retired electrical engineer and currently is a lecturer in the electrical and computer engineering department at Purdue Fort Wayne.

Subscribe to our newsletters

* indicates required
Newsletters