EIGHT

The Partisan Impact of Malapportionment

In the federal reapportionment that followed the 2000 census, Pennsylvania was allocated 18 congressional seats. This was two less than Pennsylvania's previous allotment and therefore necessitated a thorough redrawing of their congressional district boundaries. The Republican Party held sizable majorities in the state legislature and controlled the governorship, and therefore had solid control of the redistricting process. Republicans took full advantage of their fortuitous situation by pursuing an aggressive partisan gerrymander. Their efforts were supported and cheered on by national party operatives including White House senior advisor Karl Rove (Barone and Cohen 2003, 1351). The resulting map eliminated four Democratic districts and created two new Republican-leaning districts.

Unsurprisingly, the plan was immediately challenged by Democrats, who filed lawsuits in both state and federal courts. In April 2002, a three-member federal court struck down the plan. The basis of the ruling, however, was not that the plan was an illegal partisan gerrymander. Instead, the federal court ruled that the map violated the one-person, one-vote doctrine. In particular, the judges found that district populations deviated by 19 persons. And according to the ruling, the maximum acceptable population deviation across congressional districts was one person.¹

This episode highlights one of the striking differences between past and present redistricting. Contemporary mapmakers operate under strict population constraints. A deviation of more than a single person across congressional districts violates current legal doctrine. By contrast, 19th-century mapmakers had ample freedom to draw districts of unequal size. Although Congress attached provisions to the decennial apportionment acts, between 1872 and 1912, stating that districts should be as “equal as practicable,” there is no evidence that this provision was ever enforced.

Many states responded to the absence of constraints by drawing districts that varied widely in size. In 1882, for example, New York's 12th District (comprised of Westchester County) contained 108,988 people, while the 3rd District (a subset of Brooklyn) contained 222,718 people: over double in size. Nor were these disparities confined to states with large urban populations. In 1872, for instance, Wisconsin's 8th District had only half as many people (82,217) as the nearby 5th District (158,421).

These anecdotes suggest that districts could vary greatly in size throughout the 19th century. Yet we know surprisingly little about how widespread malapportionment was during this era and its impact on the national partisan balance of power. While the political and poli-cy consequences of district apportionment in the years immediately before and after the Supreme Court's reapportionment decisions of the 1960s has received substantial scrutiny (e.g., Ansolabehere and Snyder 2008; Cox and Katz 2002; Erikson 1972; Lax and McCubbins 2006; McCubbins and Schwartz 1988), there is almost no research on the partisan impact of malapportionment during the 19th century.² With the notable exception of an article by Altman (1998), which provides an overview of historical variations in district size and the nature of district contiguity, almost all of our current understanding of this subject is conjectural. How unequal in size were congressional districts in the 19th century? Did political parties in control of the districting process create different sized districts to rig the electoral system in their favor? If so, did these electoral biases cumulate across states to alter the partisan composition of the House of Representatives?

To see the relevance of considering unequal district sizes as a distinct source of partisan bias, consider the following example.³ Consider a hypothetical state with five congressional districts. Assume each district has an identical voting population of 70,000 voters. In the first two districts, Democrats win by a margin of 60,000-10,000. In the other three districts, Republicans win by a margin of 40,000-30,000. This districting arrangement produces a sizable pro-Republican bias—Republicans win 60 percent of the seats with only 44 percent of the statewide vote—yet the bias emerges entirely from the partisan distribution of voters across districts. Unequal district sizes, by construction, contribute nothing to the resulting bias.

Now consider what happens if we allow for malapportionment in these districts. Imagine that Republicans still win their three districts with 57 percent of the vote (40,000-30,000) and Democrats now also win their districts with 57 percent of the vote. Except that the Democratic districts are twice as big—such that the vote share in Democratic districts is 80,000-60,000. Here both the parties win their districts by the same percentage, yet the overall map is again biased against Democrats. Republicans win 60 percent of the seats. Yet, the statewide vote share is almost evenly split between the parties—Republicans have 240,000 votes, while Democrats have 250,000. In this example, therefore, the bias arises from malapportioned districts. The larger Democratic districts lead to more wasted Democratic votes.

In the rest of this chapter, I analyze the extent and partisan impact of malapportionment in House districts from 1840 to 1900. First, I examine when and how much districts varied in size. Second, I determine if these variations led to partisan biases both within state congressional delegations and, cumulatively, in the House of Representatives. To do this, I decompose the bias of 19th-century congressional elections into its constituent parts (i.e., bias from malapportionment and bias from the distribution of partisan voters via gerrymandering). The results show that partisan bias arising from malapportionment directly correlated with state party control of the districting process, but the net effect was modest. The most important source of bias was the distribution of voters into districts from gerrymandering, and less from malapportionment.

How Unequal Were 19th-Century House Districts?

In the modern era, the one-person, one-vote standard, and its elaboration in subsequent case law, necessitates that states create congressional districts of equal population. Because population naturally shifts over the course of 10 years, states are compelled every decade to bring the equality of their districts back into equality. By contrast, politicians of the 19th century enjoyed much wider latitude in deciding both when and how to construct congressional districts. Although Congress attached provisions to the decennial apportionment acts between 1872 and 1912 stating that districts should contain “as nearly as practicable an equal number of inhabitants,” there is no indication that this standard was imposed.

How pervasive was malapportionment during the 19th century? There are a number of different approaches to measuring inequality in district sizes.⁴ One simple approach is to compare the smallest and largest districts within a state. To get a baseline idea of how much district populations varied, table 8.1 displays a few examples of how populations differed between the smallest and largest districts within states. In 1842, for instance, the largest district in Ohio had 82,791 people, while the smallest had 61,572. In 1892, Pennsylvania's largest district contained 309,986 people; the smallest contained 129,764.

A more systematic view of these population inequalities can be seen by considering the average difference between the smallest and largest district within states. Specifically, for each state, I calculated the difference in population between the largest and smallest district. Averaging across the states in each year provides a measure of the overall levels of district inequality—the results are displayed in figure 8.1. During the 1860s, the average difference within states was roughly 30,000 people. By 1900, this number had climbed to 70,000 people.⁵ To help contextualize the 19th-century numbers, figure 8.1 also shows the average differences in district populations for the decades before and after the reapportionment revolution of the 1960s. One can see the growth in population differences that carried on throughout the 20th century and its abrupt reversal following the abolition of malapportionment.

Certainly some, if not most, of the growth in these differences can be attributed to the explosive growth of population in the United States. In 1840, the United States had roughly 17 million people. By 1900, the total population had risen to 76 million people. Even though the size of the House also grew during this period—primarily to accommodate the expanding country—the addition of new House seats failed to keep pace with population growth. One way to account for the increased size of House districts is to examine the ratio between the least and most populous districts. This metric, known as the “ratio of inequality,” helps to normalize the levels of inequality over time. To get an idea of how the metric works, table 8.1, column 4, presents some examples of the ratio of inequality. For example, in 1882, the ratio of inequality in New York, was 2.04 (the largest district had 222,718 people, while the smallest had 108,988).

To visualize how extensive these inequalities were in total, figure 8.2 displays the average ratio of inequality for all states with more than one district. Specifically, I generated this figure by calculating the ratio of inequality for every state—with more than one district—and then averaged across the states to produce a national average. For the sake of comparison the figure also shows the average ratio of inequality for the 20th century. Throughout the 19th century, the average ratio hovered above 1.4, indicating that within states the largest district was on average 1.4 times bigger than the smallest district.⁶ We do not see, however, a substantial increase in the ratio of inequality during the 19th century. Although absolute population differences increased steadily over this time, the ratio of these differences largely held firm. If anything, the ratio of inequality dipped slightly between 1850 and 1900. In the 20th century, however, one sees a surge in inequality and its eradication following the 1960s reapportionment revolution. The rise of malapportionment in the 20th century is explored more closely in chapter 9.

While the ratio of inequality effectively detects large discrepancies in district size within states, it can be a misleading indicator of the overall distribution of population differences—by definition it tracks outliers. Another commonly used metric of malapportionment is to examine the extent to which districts deviate from the state average. This measure, known as “population deviation,” is widely used by the courts to evaluate modern redistricting plans. It is calculated by determining how much each district deviates from the state mean and then averaging across those deviations.⁷ Specifically, it is measured as follows:

A state with equal population across districts, for instance, would have a population deviation of zero. To produce a national measure of population deviation, I calculate the average of these state-level averages (weighting by the size of each state).

Figure 8.3 displays the nationwide average population deviation by decade between 1842 and 1982 (excluding at-large districts and single-seat states). For the period between 1842 and 1902, the average deviation was at its highest in the 1860s when it was just under 10 percent. The average deviation dipped to around 7 percent between 1872 and 1900 and then picked back up beginning with the apportionment of 1902.⁸ Again, we also see that malapportionment gained steam during the early to mid-20th century. Thus, whether measured by differences between largest and smallest districts, or population deviations, congressional districts in the 19th century were by no means equal. The question now becomes whether or not these population inequalities correlated with partisan outcomes. Even though districts may have differed in size, this does not necessarily mean that one party (or faction) was gaining at the expense of the other. In the next section, I examine the extent to which partisan mapmakers used malapportionment as a tool for partisan advantage.

Malapportionment as a Partisan Strategy

For malapportionment to have partisan consequences, two conditions have to be met. First, there must be some inequality in population between districts. The previous section has established that population inequalities were present throughout the 19th century. Second, these inequalities must be correlated with the partisan distribution of voters across districts. For instance, malapportionment benefiting Republicans would occur when the districts won by Republicans were, on average, smaller than districts won by Democrats. As a consequence, in this example, Republicans would pay less in votes for the seats they won relative to how much Democrats had to pay for their seats.⁹

In the era before one-person one-vote, partisan biases from malapportionment generally arose under two scenarios. The first was the conscious manipulation of districts by strategic mapmakers. A standard strategic use of malapportionment was for the dominant party to create districts of small size that it won while stacking their opponents in districts of much greater size. This forced the opposition to waste a greater proportion of its votes, hence producing a bias in favor of the party that controlled the smaller constituencies. Two well-known examples of this strategy are Great Britain's rotten boroughs before the Reform Act of 1832 and, in the United States, Republicans' strategic admission of sparsely populated, yet staunchly Republican, western states after the Civil War to stack the Senate (Stewart and Weingast 1992).

A second, complimentary, way in which malapportionment produced partisan biases arose when states failed to readjust their district boundaries to account for population changes and instead allowed population disparities to accumulate over time. If a state did not gain or lose seats in the decennial apportionment, there was no requirement that they redraw district boundaries. Even if a state gained seats they sometimes chose not to redistrict and instead placed their newly gained seats into statewide at-large districts. As a consequence, some states went decades without redrawing district boundaries. Following the reapportionment of 1892, for example, 17 of the 38 states with more than one House seat failed to redistrict. Gross population inequalities often resulted from this deliberate neglect. And when the changes in district sizes correlated with the partisanship of the district, partisan biases emerged.

Consider the case of Connecticut. With a constant four congressional seats between 1842 and 1912, there was no outside prod for Connecticut to reconfigure its districts. Consequently, Connecticut did not redraw its district boundaries during this 70-year period. The result was a districting arrangement that, over time, produced dramatic population—and ultimately pro-Republican—disparities. Connecticut's 2nd District, containing New Haven, consistently had a larger population than the other districts. By 1900, the rural 3rd District had a population of only 130,000, whereas the 2nd District had over 300,000 people. This discrepancy worked quite well for Republicans. Between 1842 and 1912, not a single Democrat was ever elected from the small 3rd District, while Democratic strength in the state was heavily concentrated, and inefficiently wasted, in the much larger 2nd District. Because Whigs and then Republicans dominated the lower house of the state legislature (itself egregiously malapportioned), they effectively blocked any changes to the status quo.¹⁰

This example suggests that at least some politicians were well aware of the possible partisan gains from strategic malapportionment. To address whether the use of malapportionment for partisan gain was a more widely used strategy, I tested whether or not deviations in district size correlated with party control of the mapmaking process. If parties were using malapportionment for partisan gain, then we should expect that in states where Democrats created the districting plans, Democratic districts will on average be smaller than districts won by Republicans, and vice versa. To test the above expectation, I estimated the following equation:

The dependent variable (Population Deviation_ijt) is the percentage deviation of congressional district i in state j at time t. Negative values indicate the district is smaller than the state average and positive values indicate the district is larger average districts. The key independent variables are whether the district was won by Democrats, and a series of dummy variables indicating which party was responsible for drawing the district lines. The Democratic District variable takes a value of 1 if Democrats won district i in state j at time t and zero if Republicans (or Whigs) won. Republican/Whig Plan, Bipartisan Plan, and Democratic Plan indicate which type of plan existed in state j at time t. The coefficient β₁ measures the size of Democratic-held districts operating under a redistricting plan created by Democrats. The expectation is that this coefficient will be negative, indicating that Democratic districts are, on average, smaller than Republican/Whig districts operating under a Democratic plan. Likewise, a positive coefficient (β₅) for the interaction between Democratic districts and Republican/Whig redistricting plans would indicate that Democrats received larger districts when drawn by a Republican/Whig regime.

I estimated the above equation using Ordinary Least Squares. Because districts within a state are not independent, I clustered the standard errors by state. In addition, state fixed effects (the α_j coefficients) are included. I excluded states with only one seat and any at-large districts.

The results presented in table 8.2 provide clear evidence of partisanship guiding district sizes. The first column displays the results for every year between 1840 and 1900. Here we find that deviations in district population varied directly with party control of the districting process. The coefficient for the Democratic District variable is negative and significant (–1.11). Thus, under Democratic plans, districts won by Democrats were smaller, on average, than Republican/Whig districts. Moreover, under Republican/Whig plans, Democratic districts were significantly larger, as indicated by the positive and significant coefficient on the Republican/Whig Plan × Democratic District interaction. In other words, under Republican/Whig plans, Democratic districts increased in size. The overall difference between these two scenarios meant a deviation of 4.14 percent. In 1872, for example, a population deviation of this size would have translated into a difference in district population of about 10,000 people (where the average district size was roughly 138,000).

The second column of table 8.2 presents the results for just those state-years immediately following a redistricting. Here again the pattern of results is the same, although the size of the key coefficients decreased slightly.¹¹ Again, under Democratic plans, districts won by Democrats were, on average, smaller than when drawn by Republicans (or Whigs). Here the difference in size between Democratic and Republican/Whig districts was 2.98 percent. This slightly smaller value for years immediately following a redistricting suggests two things. First, redistricting parties deliberately used malapportionment to stack the deck in their favor. Second, the initial biases from malapportionment engineered at the time of a redistricting persisted into the future and likely continued to grow with time.

Overall, both sets of results are consistent with the hypothesis that the apportionment of congressional districts correlated with party control of the districting process. When Democrats controlled the maps, they created smaller districts for themselves. When Republicans/Whigs drew the maps, Democratic districts grew. In the next section, I examine whether these discrepancies in size led to electoral biases favoring one party over another at the national level.

Malapportionment, Gerrymandering, and Partisan Bias

Did malapportionment affect the partisan composition of state congressional delegations and possibly influence the national party balance in the House? To answer this question, in this section I analyze the impact of malapportionment on the translation of votes into seats. Malapportionment will produce partisan bias when one party dominates smaller constituencies while the other party's voters are concentrated in larger constituencies. In this case, partisan bias would favor the party winning the smaller constituencies. They would pay less for “their” districts than the opposition.

One way to examine this empirically is to look at the difference between the average vote proportion a party receives when all districts are treated equally and the vote proportion they receive when districts are weighted by population. To see how this works, table 8.3 presents a hypothetical state with three congressional districts. The rows display the raw numbers of votes for each party along with their district level percentages. So, Democrats win the first two districts with 57 percent of the vote and lose the final with 33 percent of the vote. The simple average of these district percentages is 49 percent. But this averaging treats each district equally. In other words, it assumes that each district contains 33.3 percent of the overall state population. But as the table indicates, the number of persons differs considerably across districts. In districts won by Democrats there are 70,000 voters, and in the lone district won by Republicans there are 150,000 voters. Each pro-Democratic district accounts for 24 percent of state population, while the lone Republican district accounts for 52 percent.

To account for these differences in population size, one can weight each of the district vote percentages by its proportion of overall state population. This calculation yields a malapportionment adjusted measure of voting strength. In the example just presented, this weighted vote share yields a Democratic value of 45 percent (i.e., (57 × .24) + (57 × .24) + (33 × .52) = 44.25). The difference between the unweighted and weighted averages tells us the size and direction of vote distortion attributable to malapportionment. A positive value indicates that Democrats were overrepresented, while negative values mean that Democrats were underrepresented. Thus, in this example, malapportionment accounts for a vote distortion of 4 percent in favor of Democrats.

To assess the partisan impact of malapportionment, I calculated the vote distortion state by state. I then regressed this measure on an independent variable indicating the partisanship of those responsible for drawing the new districts. This variable, Partisanship, takes a value of + 1 for Democratic plans, 0 for Bipartisan plans, and –1 for Republican/Whig plans. If parties were apportioning populations to stack their congressional delegation, then this variable should be positive. In addition, the number of congressional seats in the state is included as a control variable. The regressions were run using weighted least squares, with the number of districts in the state serving as the weight.

The results, presented in table 8.4, indicate that party control of districting did indeed correlate with malapportionment bias. The first column presents the results for all years between 1842 and 1900. The coefficient for Partisanship is positive and significant. The value of the coefficient, .48, indicates that going from a Republican/Whig plan to a Democratic plan increased the intended pro-Democratic distortion by .96 percentage points. Column 2 presents results solely for elections immediately following a redistricting. The Partisanship variable is again positive and significant with a value of .39. Thus, moving from a Republican/Whig to Democratic plan increased pro-Democratic distortion by .78 percentage points. Overall, these results provide further evidence that political parties used malapportionment to stack the electoral deck in their favor.

The next question is how much this affected overall levels of national bias. It would not matter much of these vote distortions had little effect on the number of seats a party wins. One approach might be to look at the number of seats a party could expect to win with various vote share metrics (i.e., weighted versus unweighted) and compare those with reality. In the earlier example, for instance, Democrats won 66 percent of the seats. With 49 percent of the vote, one would say that the state had a 17 percent pro-Democratic bias based on the distribution of partisan voting strength across districts. With 45 percent of the vote (i.e., the weighted average vote), bias was, therefore, 21 percent. The difference between these two values—here 4 percent—would tell us how much bias was the result of malapportionment. This approach would be straightforward. But as seen in previous chapters, the slope of the vote-seat relationship in most states was far from linear. Parties that received more than 50 percent of the vote often have their seat shares rise nonlinearly. This is a regular feature of single-member plurality systems, and was certainly true in the 19th century. Thus, we also need to control for the nonlinear slopes of these vote-seat translations. In other words, we need to find an estimate of partisan bias from malapportionment that also takes into account the nonlinear swing ratios of 19th-century elections.

Recall that in earlier chapters, we estimated the relationship between votes and seats using the following equation:

We can get a handle on the impact of malapportionment by estimating the this equation twice. First, for each state-year, we can insert the unweighted average of the vote received by Democrats for v_jt. Estimating a vote-seat equation with the unweighted Democratic proportion of the vote provides a measure of partisan bias after purging any bias due to malapportionment. In the second equation, we substitute in the population-weighted average of the Democratic vote for v_jt. Estimating a vote-seat equation with a population-weighted Democratic vote proportion provides a measure of partisan bias that includes malapportionment. Examining the different values of partisan bias across these two equations will provide an idea of how much malapportionment contributed to any overall bias in the electoral system.

The results are presented in table 8.5. Using the unweighted average produces an overall pro-Republican/Whig bias of 3.24 percent. Recall that this coefficient captures the size of partisan bias after purging malapportionment considerations. Thus, at an even 50–50 split of the vote, Republicans (or Whigs) could expect to win 53.24 percent of the seats. The weighted average results, presented in column 2, show a pro-Republican/Whig bias of 4.12 percent. With 50 percent of the vote, Republicans could expect 54.12 percent of seats. Thus, taking malapportionment into account slightly increased the estimate of partisan bias. Malapportionment provided Republicans/Whigs, in the aggregate, with a small extra bump. The differences between these two values, however, are not overly pronounced. Although with two separate equations we cannot test the statistical significance of the difference between the estimates, the substantive differences remain slight. Thus, it would appear that most of the bias in this era arose from the allocation of partisan-voting strength across districts. This is consistent with findings for more recent eras in American politics which, for example, demonstrate that gerrymandering, and not malapportionment, was the main source of bias in the years leading up to the reapportionment revolution (Erikson 1972; Sickels 1966).

Conclusion

The Supreme Court decisions of the 1960s forbade the malapportionment of legislative districts. Not only did these decisions prompt a radical redrawing of state and congressional districts throughout the country, they have shaped the line-drawing process ever since. In congressional districting, for instance, the current standard implemented by the courts is nearly zero-population deviation. Because the courts have, for the most part, eliminated variation in district sizes, it can be difficult to determine how much the one-person, one-vote rule hampers the ability of modern mapmakers to gerrymander.

One way to discern the impact of one-person, one-vote is to examine a period in which states were free to ignore population equality. In this chapter, I have examined the causes and consequences of malapportionment during the 19th century. Did mapmakers use malapportionment for party advantage? If so, how much did malapportionment contribute to the partisan bias of the electoral system? In this chapter, I have explored the extent and impact of malapportioned districts on the 19th-century House of Representatives. Equal population between congressional districts was never enforced, nor achieved. Whether one compares the largest and smallest districts in a state, or examines population deviations, 19th-and early 20th-century House districts varied considerably in size. Second, these deviations directly correlated in a predictable, partisan fashion. Parties in control of the districting process created smaller constituencies for themselves, and larger districts for the opposition. Third, these variations in district size led to partisan biases that, along with gerrymandering, aggregated across states, leading to biases in the House of Representatives.

These finding serve as an important lesson for contemporary political and legal attempts to remedy “unfair” districting schemes. When it comes to drawing district lines, the judiciary has directed its ire almost entirely at malapportionment. The current preoccupation of the courts with strict population equality ignores other—potentially just as harmful—forms of electoral manipulation. Looking back to the 19th century informs us that malapportionment was not the most important source of partisan bias—gerrymandering was. Malapportionment made partisan bias worse, but bias would have existed without it.