Highest averages method
The highest averages method is a family of divisor methods used to apportion seats in multi-member constituencies under proportional representation electoral systems, allocating legislative seats to parties based on the highest quotients obtained by dividing their vote totals by a predetermined sequence of divisors.[1] These methods, which include prominent variants such as the D'Hondt and Sainte-Laguë procedures, aim to translate vote shares into seat shares while addressing the indivisibility of seats through iterative assignment to the highest resulting averages.[2] Developed in the late 18th and 19th centuries— with roots in Thomas Jefferson's 1792 proposal for U.S. congressional apportionment and formalized by Victor d'Hondt in the 1880s for Belgian elections—the method has become one of the most widely adopted tools for seat allocation in party-list systems across Europe and beyond.[1] In practice, each party's votes are divided successively by divisors starting from 1 (and increasing, e.g., by 1 for D'Hondt or by 2 for Sainte-Laguë), with seats awarded sequentially to the party producing the largest quotient until the available seats are exhausted.[2] While effective in promoting stable majorities by favoring larger parties—ensuring that a party with an absolute majority of votes secures a majority of seats—the highest averages method exhibits a bias against smaller parties, often resulting in their underrepresentation compared to more quota-based alternatives like the largest remainder method.[1][2] This characteristic has led to its use in over a dozen European Union member states for parliamentary elections, balancing proportionality with incentives for broader electoral coalitions.[1]Historical Development
Origins in Early Apportionment Challenges
The apportionment of seats in the United States House of Representatives presented one of the earliest systematic challenges in fairly distributing indivisible legislative positions based on population data, as mandated by Article I, Section 2 of the U.S. Constitution, which requires representatives and direct taxes to be apportioned among the states according to their respective numbers, determined by an actual enumeration conducted every ten years.[3] Following the first census in 1790, which counted a total population of 3,929,214, Congress faced the task of allocating 105 seats—a number derived from aiming for one representative per approximately 30,000 inhabitants—while grappling with fractional quotas that arose when dividing state populations by this standard divisor.[4] This process highlighted the core tension in apportionment: ensuring integer seat assignments that approximated proportional representation without systematic bias toward small or large states, a problem compounded by the constitutional prohibition on fractional representatives.[5] Initial efforts relied on Alexander Hamilton's proposal, which calculated each state's quota as population divided by the national divisor and assigned seats via the integer part plus largest remainders to reach the total, a method akin to the later Hamilton or largest remainder approach.[6] However, the bill embodying this method, passed by Congress in early 1792, was vetoed by President George Washington on April 5, 1792, primarily because it fixed the total seats at a level that deviated from strict constitutional ratios for taxation and representation, potentially allowing future inconsistencies.[7] This veto, the first in U.S. presidential history, underscored the need for a method that adhered more closely to proportional principles while avoiding paradoxes like assigning more seats to states with proportionally smaller growth.[8] In response, Thomas Jefferson, serving as Secretary of State and responsible for census implementation, advanced an alternative divisor-based approach in 1792, which allocated seats by generating successive quotients for each state—dividing its population by 1, 2, 3, and so on—and selecting the largest such quotients across all states until the total seats were filled.[6] This highest averages method, enacted via the Apportionment Act of April 14, 1792, which set the House at 105 members, prioritized larger quotients to favor states with higher populations, ensuring no state received fewer seats than its integer quota but often granting extra seats to larger entities to fill the total.[5] Adopted out of necessity to resolve the impasse, Jefferson's method addressed the indivisibility issue through iterative division and ranking, marking the inaugural application of a highest averages procedure in legislative apportionment and setting a precedent for handling quota fractions via divisor sequences rather than remainders.[8] It remained in use for apportionments from 1792 until 1832, despite later criticisms for biasing toward larger states, as evidenced by its tendency to allocate additional seats beyond strict quotas to meet fixed totals.[5]Independent Inventions and Key Contributors
The highest averages method, encompassing various divisor-based apportionment techniques, emerged through independent inventions across different national contexts, primarily in the late 18th and 19th centuries, as solutions to the challenge of allocating indivisible seats proportionally to vote shares or population figures. Thomas Jefferson first formalized a version of the method—now known as the Jefferson or D'Hondt procedure—in his 1791 report to Congress on apportioning U.S. House seats following the 1790 census, advocating division of state populations by successive integers starting from 1 and assigning seats based on the largest quotients to favor larger states and ensure a constitutional minimum representation.[9] This approach was adopted for U.S. apportionments from 1792 through 1842, despite later criticisms of its bias toward populous states.[10] Nearly a century later, Belgian mathematician and lawyer Victor D'Hondt independently reinvented an identical procedure in 1882 for allocating parliamentary seats in proportional representation systems, publishing it as a mathematical formula to distribute seats by repeatedly dividing party vote totals by 1, 2, 3, and so on, then selecting the highest resulting averages until all seats are filled.[11] D'Hondt's formulation, unaware of Jefferson's prior work, gained traction in European electoral practice, notably in Belgium's adoption of proportional representation in 1899, where it addressed fragmentation in multi-party legislatures without reference to American precedents.[12] A parallel independent development occurred with the Webster/Sainte-Laguë variant, which modifies divisors to odd numbers (1, 3, 5, etc.) for greater neutrality toward smaller parties. American statesman Daniel Webster proposed this in 1832 during debates over equitable House apportionment, emphasizing rounded quotients to balance representation without inherent bias to large or small entities; it was implemented in the U.S. from 1842 to 1852 and again from 1901 onward in modified forms.[13] French mathematician André Sainte-Laguë separately derived the same method in 1910, applying it to graph theory-inspired models of seat allocation and advocating its use in French elections for its mathematical fairness in minimizing vote-seat disproportionality.[14] These reinventions highlight the method's appeal as a first-principles solution to indivisibility in fair division, converging on similar algorithms despite isolated origins in U.S. constitutional mechanics and European parliamentary reforms.Initial Adoption and Evolution in Electoral Practice
Belgium adopted the highest averages method, specifically the D'Hondt variant, as part of its pioneering implementation of proportional representation in national elections through the electoral reform of 1899, marking the first such nationwide use globally.[12] This reform addressed the disproportionate outcomes of the prior majoritarian system, which had favored larger parties amid rising demands for fairer representation from emerging socialist and liberal factions. The method was applied in the 1900 Belgian general election, dividing votes by successive divisors (1, 2, 3, etc.) to allocate seats to parties with the highest resulting quotients, thereby promoting greater proportionality while retaining a slight bias toward larger lists to maintain governmental stability.[15] Following its Belgian origins, the highest averages method proliferated across Europe in the early 20th century, becoming a cornerstone of party-list proportional representation systems. Countries such as the Netherlands, Austria, and Finland incorporated D'Hondt or similar divisor approaches by the 1910s and 1920s to facilitate multi-party parliaments post-World War I, often as a response to fragmented electorates and demands for minority inclusion.[16] The Sainte-Laguë variant, proposed by French mathematician André Sainte-Laguë in 1910 with odd-numbered divisors (1, 3, 5, etc.) to reduce bias against smaller parties, saw initial adoption in Norway in 1921 and subsequently in Sweden and New Zealand, reflecting a shift toward enhanced neutrality in seat allocation.[17] Evolution continued through mid-century adaptations, with southern European democracies like Spain and Portugal embedding D'Hondt in their post-1970s constitutions to balance proportionality and effective governance amid transitions from authoritarianism.[18] Germany, which employed D'Hondt until the 1983 Bundestag election, transitioned to a modified Sainte-Laguë system thereafter to mitigate advantages for major parties, incorporating a 5% threshold for additional fragmentation control.[19] By the late 20th century, highest averages methods underpinned over half of European PR systems, with refinements like initial divisor adjustments (e.g., Denmark's 1.4 factor) addressing critiques of large-party bias while preserving mathematical simplicity and resistance to strategic voting manipulation.[20]Mathematical Foundations
Core Principles of Divisor Methods
Divisor methods, collectively known as highest averages methods, allocate seats in multi-member constituencies by computing a series of quotients for each party, derived from dividing the party's total votes by an increasing sequence of positive divisors, and then awarding seats to the parties corresponding to the highest such quotients until the total number of seats is exhausted. This process ensures that each selected quotient represents a marginal claim for an additional seat, where the quotient value approximates the average votes per seat for that allocation.[21] The divisors are chosen as a non-decreasing sequence d_1 \leq d_2 \leq \cdots, often starting with d_1 = 1, such that the k-th quotient for a party with v votes is v / d_k, reflecting the incremental average if the party receives its k-th seat.[22] [float-right] The core principle underlying this approach is to prioritize allocations that maximize the "highest averages," meaning seats are granted iteratively to the party for which adding the next seat yields the highest possible vote-to-seat average at that step, equivalent to selecting the global highest quotients in batch form. This iterative equivalence holds because the quotients decrease monotonically for each party as k increases, allowing the method to simulate a greedy assignment without recomputing averages sequentially.[23] For instance, in the standard d'Hondt variant, divisors are the integers 1, 2, 3, ..., producing quotients that favor parties with larger vote shares by making their higher-order quotients competitive longer than for smaller parties.[1] These methods inherently satisfy house monotonicity, as increasing a party's votes cannot decrease its seat allocation, due to the non-decreasing nature of divisors ensuring that all its quotients rise uniformly. However, the choice of divisor sequence determines bias: linear divisors (e.g., d_k = k) advantage larger parties by compressing small-party quotients faster, while adjusted sequences like d_k = 2k-1 (Sainte-Laguë) mitigate this by slowing the decline for initial seats.[24] Empirical analysis of European Parliament elections from 1979 to 2014 shows divisor methods like d'Hondt yielding effective thresholds around 5-10% in larger districts, balancing proportionality with stability by under-representing fringe parties.762352_EN.pdf) The mathematical rigor of divisor methods stems from their axiomatic foundations, including anonymity (permutation invariance) and neutrality (vote scaling), though they may violate quota conditions where a party's seats deviate from its ideal vote proportion by more than one.[25]The Highest Averages Allocation Procedure
The highest averages allocation procedure is an iterative algorithm within the family of divisor methods used for proportional seat apportionment in multi-party elections. It assigns seats one at a time to the party whose prospective average—defined as the ratio of votes received to the number of seats it would hold after allocation—is the highest among all parties at each step. This approach, equivalent to selecting the largest quotients from the set of all possible vote divisions by positive integers, ensures that seats are distributed to minimize disparities in average representation while favoring parties with stronger vote shares for marginal allocations.[26][23] The procedure begins with zero seats assigned to each party. For the first seat, each party's quotient is its total votes divided by 1, and the seat goes to the party with the maximum quotient. Subsequent seats follow the same principle: for a party with s_i seats already allocated and v_i votes, the quotient for the next potential seat is v_i / (s_i + 1). The party maximizing this value receives the seat, and its seat count increments. This continues until all seats are distributed. Mathematically, the average for the prospective allocation is given bywhere the post-allocation divisor function is typically \operatorname{post}(k) = k + 1, yielding
for the standard arithmetic progression of divisors starting at 1.[27][28] This iterative process is computationally equivalent to generating the sequence of quotients v_i / k for each party i and integer k = 1, 2, \dots, then selecting the h largest values (where h is the total seats), with the number of selected quotients per party determining its allocation. The method's efficiency allows linear-time implementations for large-scale applications, as optimized algorithms avoid exhaustive quotient generation by maintaining priority queues of current maxima.[26][29] It inherently satisfies lower quota bounds but may violate upper quotas, prioritizing higher averages over strict proportionality in edge cases.[27]
Role of Divisors and Rounding Rules
In highest averages methods, also known as divisor methods, the divisor sequence d(1), d(2), \dots , a strictly increasing and unbounded sequence of positive real numbers, plays a central role in determining seat allocations by computing successive averages for each party or state as v_i / d(k), where v_i is the vote count or population for entity i and k is the seat number. Seats are assigned to the M highest such quotients across all entities, ensuring the total equals the house size M. The specific form of the sequence controls the relative thresholds for awarding additional seats, thereby influencing the method's bias toward larger or smaller entities; for instance, sequences with lower initial divisors, such as d(k) = k, yield higher initial quotients for large-vote entities, favoring them in competitive allocations.[27] The shape of the divisor sequence, often characterized by its asymptotic growth and initial values, dictates the method's proportionality properties and potential violations of criteria like the quota condition, where allocations may deviate from the exact proportional share by more than one seat. Sequences growing linearly like d(k) = k + r for some offset r adjust the bias: r = 0 (Jefferson method) biases toward larger parties, while r = 0.5 (Webster method) aims for neutrality by aligning rounding points closer to standard arithmetic means. Empirical analysis shows that slower-growing sequences initially amplify advantages for vote-rich parties, as the first few quotients remain disproportionately high compared to smaller competitors.[27][5] Rounding rules in these methods arise equivalently from selecting a global divisor D and applying a sequence-dependent rounding function to v_i / D, ensuring the sum of rounded quotas equals M; this D is chosen iteratively such that \sum \lfloor \delta^{-1}(v_i / D) \rfloor + 1 = M, where \delta extends the divisor sequence. Common rounding variants include the floor function for pro-large bias (Jefferson, using divisors around 1 to M), ceiling for pro-small (Adams), and nearest integer for balance (Webster), with the effective rounding boundaries set by midpoints derived from the sequence, such as geometric means in Huntington-Hill where d(k) = \sqrt{k(k+1)}. This equivalence highlights how divisors encode the rounding logic: for example, Jefferson's floor rounding after scaling by D \approx total population / seats systematically under-rounds small entities, leading to quota violations exceeding one seat in cases like a state with quota 40.705 receiving 42 seats under certain divisors around 44,600.[27][5]Specific Methods
Jefferson (D'Hondt) Method
![{\displaystyle {\text{average}}:={\frac {\text{votes}}{\operatorname {post} ({\text{seats}})}}}] (./assets/e7307f9790652446d36257e42e77791abd986bd5.svg)[float-right] The Jefferson method, also called the D'Hondt method, is a highest averages divisor method for apportioning seats in legislative bodies proportionally to votes received by parties or, historically, populations of states. It operates by computing quotients of each party's vote total divided by successive integers (1, 2, 3, and so on) and allocating seats to the highest such quotients until all seats are assigned.[1][30] This approach ensures that larger parties receive a disproportionate share of seats relative to smaller ones, as their quotients remain competitive longer in the sequence.[1] Thomas Jefferson proposed the method in 1792 as a solution to apportion U.S. House of Representatives seats among states following the 1790 census, after President Washington vetoed Alexander Hamilton's largest remainder method for exceeding the constitutional maximum of one representative per 30,000 persons in some states.[8] Congress adopted Jefferson's approach, which used a common divisor applied to state populations to yield integer quotients, with the divisor adjusted to fit the exact number of seats.[8] It remained in use for apportionments based on the 1790, 1800, and 1810 censuses, until replaced in the 1840s due to its bias toward larger states, which could result in allocations exceeding the quota principle.[8] Independently, Belgian lawyer and mathematician Victor d'Hondt formulated an equivalent procedure in 1882 for distributing seats in multi-member districts under proportional representation, aiming to balance linguistic and political groups in Belgium.[11][1] The method spread across Europe, adopted in countries including Belgium, Denmark, and Finland by the early 20th century, and is currently employed for national parliamentary elections in at least 16 European Union member states, such as Austria, France, and Spain, as well as for allocating European Parliament seats in those jurisdictions.[1] The allocation proceeds iteratively: for each party i with vote total v_i, initially compute v_i / 1; assign the first seat to the party with the highest quotient. For subsequent seats, divide the vote total of each party by one plus its seats already allocated (i.e., v_i / (s_i + 1), where s_i is current seats for party i), and award to the highest resulting value, repeating until all seats are distributed.[1][30]| Step | Party A (10,000 votes) | Party B (6,000 votes) | Party C (1,500 votes) | Seat Awarded To |
|---|---|---|---|---|
| 1 | 10,000 / 1 = 10,000 | 6,000 / 1 = 6,000 | 1,500 / 1 = 1,500 | A |
| 2 | 10,000 / 2 = 5,000 | 6,000 / 1 = 6,000 | 1,500 / 1 = 1,500 | B |
| 3 | 10,000 / 2 = 5,000 | 6,000 / 2 = 3,000 | 1,500 / 1 = 1,500 | A |
| 4 | 10,000 / 3 ≈ 3,333 | 6,000 / 2 = 3,000 | 1,500 / 1 = 1,500 | A |
| 5 | 10,000 / 4 = 2,500 | 6,000 / 2 = 3,000 | 1,500 / 1 = 1,500 | B |
| 6 | 10,000 / 4 = 2,500 | 6,000 / 3 = 2,000 | 1,500 / 1 = 1,500 | A |
| 7 | 10,000 / 5 = 2,000 | 6,000 / 3 = 2,000 | 1,500 / 1 = 1,500 | B |
| 8 | 10,000 / 5 = 2,000 | 6,000 / 4 = 1,500 | 1,500 / 1 = 1,500 | A or B (tie possible, but example yields A:5, B:3) |
Adams Method
The Adams method utilizes the highest averages procedure with the post function defined as \operatorname{post}(h) = h - 1 for a party's h-th seat, resulting in quotients of votes divided by successively smaller initial values compared to other divisor methods. For h=1, the divisor is 0, yielding an infinite quotient that prioritizes allocating at least one seat to every party with positive votes before competing for additional seats via quotients votes/1, votes/2, and so forth. Seats are assigned by selecting the largest total quotients across all parties until the house size is reached. This formulation, equivalent to ceiling rounding with a modified divisor below the standard in the divisor method framework, was proposed by John Quincy Adams in 1832 during debates on U.S. congressional apportionment to address perceived unfairness in prior systems.[31]/09%3A__Apportionment/9.02%3A_Apportionment_-_Jeffersons_Adamss_and_Websters_Methods) In practice, the method begins by computing infinite quotients for each party's first seat, filling initial allocations accordingly if seats suffice, then proceeds with finite quotients ordered by magnitude. For example, with parties A (100 votes) and B (10 votes) apportioning 3 seats, A's quotients are ∞, 100/1=100, 100/2=50; B's are ∞, 10/1=10, 10/2=5. The top three (two ∞ and 100) yield 2 seats to A and 1 to B. This contrasts with d'Hondt's post(h)=h, which would give all 3 to A. The approach guarantees satisfaction of upper quotas—no party exceeds \lceil votes / (total seats / total votes) \rceil—but risks lower quota violations and over-representation of small parties.[32][33] The method's bias stems from minimizing early divisors, amplifying small parties' competitive quotients for additional seats relative to large parties' higher denominators. Empirical analyses show it allocates more seats to minor parties than Jefferson or Webster methods, potentially fragmenting legislatures but enhancing minority representation. Despite theoretical appeal for equity in sparse vote distributions, it has seen no widespread electoral adoption, as its pro-small bias can undermine proportionality for majorities and complicate governability.[34][24]Webster (Sainte-Laguë) Method
The Webster (Sainte-Laguë) method is a highest averages divisor method for allocating seats in proportional representation systems or apportioning representatives among entities such as states, using successive odd integers as divisors: 1, 3, 5, 7, and so forth.[35][36] To apply it, each party's (or state's) vote total or population figure is divided by these divisors to generate a series of quotients, and seats are assigned iteratively to the highest quotients until the total number of seats is reached, equivalent to rounding each initial quota to the nearest integer after selecting an appropriate common divisor such that the sum matches the house size.[35][14] This rounding incorporates geometric means implicitly, with decision points at half-integers (e.g., a quotient above 0.5 earns the first seat, above 1.5 the second), distinguishing it from methods like D'Hondt that favor larger parties through even-integer divisors starting at 1, 2, 3.[36] Proposed by U.S. Senator Daniel Webster in 1832 as a refinement to earlier divisor methods, it addressed biases in quota rounding by advocating nearest-integer allocation via a modified divisor adjusted iteratively until the total seats apportion correctly.[35] Congress adopted it for House of Representatives apportionment following the 1840 census, fixing the House at 223 members with a ratio of one per 70,680 residents, though it was repealed in 1852 amid disputes over representation and reinstated briefly in 1901 before further shifts.[35] Independently, French mathematician André Sainte-Laguë described the identical procedure in 1910, framing it as an arithmetic progression to minimize least-squares deviations in seat-vote proportionality and counteract the large-party bias of prevailing methods like D'Hondt.[14] Sainte-Laguë's formulation emphasized its application to party-list systems, influencing its adoption in parliamentary elections. The method exhibits house-monotonicity and satisfies the quota condition more reliably than alternatives for certain house sizes, as it is the unique divisor method meeting the quota criterion for three seats.[37] It is pairwise unbiased, meaning for any pair of parties or states, the probability of favoring the larger over the smaller equals that of the reverse under uniform random vote distributions, avoiding systematic bias toward either large or small entities.[38][36] In practice, this neutrality supports broader representation without excessive fragmentation, though variants like Schepers (starting divisors at 0.5 or adjusted values such as 1.4) have been implemented to further deter dominance by the largest party, as in Germany's federal elections.[36] Countries including Sweden, Norway, and formerly New Zealand have employed it or close variants for multi-member district allocations, valuing its balance over D'Hondt's stability-favoring bias.[14]| Party | Votes | Quotient for 1st Seat (÷1) | Quotient for 2nd Seat (÷3) | Quotient for 3rd Seat (÷5) |
|---|---|---|---|---|
| A | 10000 | 10000 | 3333.33 | 2000 |
| B | 6000 | 6000 | 2000 | 1200 |
| C | 4000 | 4000 | 1333.33 | 800 |
Huntington-Hill Method
The Huntington-Hill method, also termed the method of equal proportions, is a divisor-based highest averages procedure for apportioning legislative seats proportionally to population shares, currently applied to allocate the 435 seats in the United States House of Representatives following each decennial census.[39] Developed independently by mathematician Edward V. Huntington and statistician Joseph A. Hill in the early 20th century, it gained prominence after Congress, stalled on apportionment post-1920 census due to methodological disputes, consulted a National Academy of Sciences committee of experts including Huntington, who endorsed it in 1929 as superior for balancing proportionality and avoiding paradoxes like the Alabama paradox seen in earlier Hamilton and Jefferson approaches.[40] Congress enacted it via the Reapportionment Act of 1929, with refinements confirmed in 1941 legislation that fixed the House size at 435 and made the method permanent, resolving a 1930s impasse where it diverged from Webster's method in seat allocations for states like Arkansas and Minnesota.[39][41] The procedure begins by assigning one seat to each state, reflecting constitutional minimum representation. For remaining seats, it iteratively grants the next seat to the state maximizing the priority value p / \sqrt{k(k+1)}, where p is the state's population and k its current seats; this divisor \sqrt{k(k+1)} represents the geometric mean constituency size at which indifference occurs between awarding the (k+1)-th seat or not.[40] Equivalently, it modifies Webster's arithmetic-mean rounding (at k + 0.5) by shifting thresholds downward via the geometric mean, which lies between arithmetic and harmonic means, yielding initial quotas floored then adjusted upward for states whose fractional part exceeds the geometric threshold.[39] This produces allocations minimizing the maximum relative difference in constituency sizes across states, with the effective quota bounded between 1 and 2 for the marginal seat, though overall quotas may slightly violate exact equality condition.[41] The method exhibits house monotonicity, ensuring that increasing total seats does not reduce any state's allocation, and avoids the population paradox where a state's seat gain causes another's loss despite national growth.[42] It introduces a mild bias toward smaller states relative to pure proportional methods, as the geometric rounding favors rounding up lower quotas more readily than arithmetic means, evident in post-1941 apportionments where states like Montana retained seats longer than under Webster's despite population shifts.[41] Computationally self-executing once census figures are certified—using standard divisor total population divided by seats, then priorities—it has yielded consistent results without legal challenges since adoption, though critics note its equal-proportions criterion prioritizes relative equity over absolute quota adherence.[43] For the 2020 census, it allocated seats effective January 3, 2023, shifting one from California, New York, Illinois, Ohio, Michigan, and Pennsylvania to Texas, Florida, Colorado, Montana, North Carolina, and Oregon.[39]Theoretical Properties
Monotonicity and Related Criteria
Highest averages methods satisfy party monotonicity, ensuring that if a single party's vote total increases while others remain constant, that party's seat allocation does not decrease. This follows from the iterative selection process, where seats are awarded to the highest current average (votes divided by a divisor sequence); an increase in votes raises all relevant averages for that party proportionally, allowing it to retain prior selections and potentially claim additional ones without displacing its own prior awards.[44][45] These methods also fulfill house monotonicity, meaning that expanding the total number of seats in the assembly cannot reduce any party's allocation. Proofs rely on the monotonicity of the divisor function and rounding rule: additional seats are assigned to the next-highest averages across parties, preserving existing allocations due to the non-decreasing nature of the averages as seats increase.[46] This contrasts with quota-based methods like Hamilton's, which can exhibit the Alabama paradox where enlargement deprives a party of a seat.[47] Related criteria include population monotonicity (analogous to party monotonicity in multi-state or federal apportionment), where an increase in one entity's population share does not cause it to lose seats while another gains. Divisor methods satisfy this via similar average-comparison logic, avoiding paradoxes observed in non-divisor approaches.[22] However, the specific divisor sequence influences bias toward larger parties, potentially amplifying small violations in strict proportionality under extreme vote shifts, though core monotonicity holds universally across the family.[48]Quota Compliance and Inequalities
Highest averages methods, also known as divisor methods, do not satisfy the quota condition, which requires that each party's allocated seats s_i satisfy \lfloor q_i \rfloor \leq s_i \leq \lceil q_i \rceil, where q_i is the party's standard quota defined as votes for the party divided by the Hare quota (total votes divided by total seats).[49] Instead, these methods adjust a common divisor to ensure the sum of rounded modified quotas equals the total seats, which can result in modified quotas that push allocations outside the standard quota bounds.[49] This violation arises because smaller divisors inflate modified quotas (favoring larger parties and risking upper quota breaches), while larger divisors deflate them (favoring smaller parties and risking lower quota breaches).[49] The propensity for quota violations varies by the specific rounding rule in the highest averages procedure. Methods employing downward-biased rounding, such as the Jefferson (D'Hondt) method with divisors $1, 2, 3, \dots and rounding down, tend to produce lower quota violations, under-allocating seats to smaller parties whose quotas fall just below integers.[50] Conversely, upward-biased methods like Adams, using divisors $2, 3, 4, \dots and rounding up, more often cause upper quota violations, over-allocating to larger parties.[49] Neutral methods, including Webster (Sainte-Laguë) with arithmetic mean rounding or Huntington-Hill with geometric mean, exhibit violations in both directions but at lower frequencies; for instance, simulations for Huntington-Hill in U.S. House apportionment indicate lower quota violations occur in approximately 1-2% of cases under historical population distributions, with upper violations rarer. Empirical analyses confirm these patterns persist in multi-party electoral contexts, though exact frequencies depend on vote distributions and house size.[51] Despite quota non-compliance, highest averages methods constrain relative inequalities in representation, measured as disparities in parties' effective votes-per-seat ratios (v_i / s_i). The procedure ensures that the final set of averages (votes divided by the divisor corresponding to allocated seats) forms a threshold where no unallocated next average exceeds the lowest allocated one, bounding the maximum ratio of any party's votes-per-seat to another's at most (k+1)/k for a party receiving k seats, typically yielding ratios under 2 for common implementations like D'Hondt.[22] This property minimizes pairwise relative deviations compared to quota methods, which prioritize absolute quota adherence but can amplify inequalities via remainders.[5] In practice, such bounds promote consistent proportionality, with deviations rarely exceeding 10-20% in seat-vote ratios across European parliamentary elections using variants like D'Hondt.[21]House Monotonicity and Population Constraints
Highest averages methods, also known as divisor methods, satisfy house monotonicity, a criterion requiring that an increase in the total number of seats allocated in the legislature does not result in any party receiving fewer seats than it would have under the previous house size, assuming unchanged vote shares.[26] This property follows from the methods' reliance on iterative division and rounding of vote quotients, ensuring that additional seats are assigned to parties with the highest resulting averages without displacing prior allocations.[22] In contrast to largest remainder methods, which can exhibit the Alabama paradox—where a state or party loses a seat upon house expansion—highest averages methods avoid such violations systematically.[26] These methods also fulfill population monotonicity, stipulating that if the population (or vote total) of one state or party increases while others remain fixed and the house size is constant, that entity receives at least as many seats as before, with no other entity gaining more than one additional seat.[22] Theorem 8.4 in apportionment theory establishes that population monotonicity holds uniquely for divisor methods among common classes, as their uniform divisor sequences preserve relative priorities in seat assignments despite proportional shifts.[22] For instance, under the Jefferson (D'Hondt) or Webster (Sainte-Laguë) variants, an incremental vote gain for a party adjusts its quotients upward without retroactively reducing its rounded allocations below the prior level.[26] Population monotonicity implies house monotonicity, reinforcing the robustness of highest averages methods to expansions in representational capacity.[22] These properties contribute to their adoption in systems prioritizing stability, such as certain European parliaments and the U.S. House of Representatives (via Huntington-Hill), where avoiding paradoxes from demographic or legislative changes is paramount. However, while monotonicity is guaranteed, trade-offs arise with other criteria like strict quota compliance, which divisor methods may violate.[26] Empirical applications, including post-census reapportionments, demonstrate rare boundary cases but no systematic failures of these monotonicity axioms.[22]Biases, Criticisms, and Advantages
Empirical Biases Toward Larger Parties
Empirical analyses of highest averages methods in proportional representation systems reveal systematic seat biases favoring larger parties, particularly under the Jefferson (D'Hondt) variant with its standard divisor sequence of 1, 2, 3, and downward rounding. In the German state of Bavaria, which employs list proportional representation, examination of 49 apportionments across seven districts from 1966 to 1998 (with district magnitudes ranging from 19 to 65 seats) showed the largest parties receiving consistent excess seats, while smaller parties experienced deficits relative to their vote shares.[52] Similarly, in the Swiss Canton of Solothurn, data from 143 apportionments in 10 districts between 1896 and 1997 (district magnitudes 7 to 29 seats) confirmed this pattern, with larger parties benefiting from the method's mechanics in multi-party contests.[52] Quantitative assessments quantify the magnitude of this bias. For instance, the largest party typically gains about 5 extra seats over the course of 12 elections under D'Hondt, accumulating from district-level rounding effects that disproportionately disadvantage smaller competitors by elevating the effective threshold for additional seat allocations.[52] These findings from historical election data underscore how the method's highest averages priority amplifies advantages for parties with higher initial vote concentrations, as smaller parties must surpass steeper quotient hurdles to compete for seats. In broader empirical contexts, such as European parliamentary allocations using D'Hondt, observed seat-vote disproportionality indices (e.g., Gallagher index) are higher for larger parties' overrepresentation compared to more neutral methods like Sainte-Laguë, with real-world applications in countries like Spain and Belgium reinforcing the pattern through post-election seat distributions that exceed strict proportionality by 2-5% for dominant parties in multi-district systems.[21] This bias persists despite varying district magnitudes, as confirmed by apportionment datasets analyzed for mechanical effects independent of strategic voting.[52]Criticisms Regarding Small Party Representation
The highest averages method, encompassing divisor-based apportionment rules such as the D'Hondt (Jefferson) variant, systematically disadvantages smaller parties by favoring those with higher initial vote concentrations, as seats are iteratively assigned to the highest vote-per-seat quotients. This dynamic enables larger parties to accumulate seats more rapidly after securing initial representation, while small parties often fail to surpass the de facto threshold for even one seat, particularly in multi-member districts with limited seats. Empirical studies confirm this bias, showing that D'Hondt implementations result in small parties absorbing disproportionate residual votes—unrepresented portions of the electorate—thus underrepresenting their support relative to larger competitors.[18][54] Critics argue this structure elevates the effective electoral threshold, excluding minor parties and minority viewpoints from legislatures, which can entrench two-party dominance or coalitions among major actors at the expense of broader pluralism. For instance, in systems employing D'Hondt, parties garnering 5–10% of votes in districts may receive zero seats, amplifying disproportionality compared to largest remainder methods that guarantee quota-based allocations.[55][56] Academic analyses quantify this favoritism, noting that divisor sequences starting at 1 (as in standard D'Hondt) inherently prioritize larger lists, with seat shares deviating more from vote proportions for smaller entities than in Sainte-Laguë variants, which use odd-numbered divisors to mitigate but not eliminate the effect.[57][58] Proponents of alternative systems highlight that this bias persists across highest averages implementations unless modified with higher initial divisors, but such adjustments risk other distortions; unchanged, the method correlates with reduced legislative diversity, as observed in national parliaments where small-party seat bonuses are minimal or negative.[59][60] Consequently, it has drawn scrutiny for potentially undermining the core aim of proportional representation by marginalizing emerging or niche political forces, though defenders counter that the stability gained outweighs representational losses for fringe groups.[18]Advantages for Stability and Governability
The highest averages method, especially the D'Hondt variant, inherently favors larger parties through its increasing divisor sequence, which disadvantages smaller parties relative to more neutral methods like Sainte-Laguë. This bias reduces the effective number of parties in legislatures, mitigating parliamentary fragmentation and thereby promoting government stability. Empirical analyses of electoral systems indicate that lower fragmentation correlates with decreased instability, as fewer parties simplify coalition negotiations and reduce the likelihood of no-confidence votes or governmental collapses. For instance, in Spanish municipalities using D'Hondt allocation, crossing entry thresholds to admit additional small parties increases the probability of destabilizing no-confidence motions by approximately 4 percentage points, particularly in non-majority settings.[61] By allocating disproportionate seats to major parties, the method encourages voter consolidation around larger blocs rather than splintering into ideologically similar small groups, fostering decisive outcomes that enhance governability. This dynamic supports the formation of stable single-party governments or minimal coalitions capable of enacting policy without constant renegotiation. In Turkey, the adoption of D'Hondt with a 10% national threshold since 1995 shifted from fragmented coalitions in the 1990s to single-party rule by the AKP in the 2002, 2007, and 2011 elections, where the party secured 66% of seats in 2002 despite 34.3% of votes, enabling prolonged governance stability.[62] Such advantages are particularly pronounced in multi-party systems prone to instability, where the method's mechanics act as a de facto threshold mechanism, prioritizing effective governance over maximal proportionality. While critics argue this comes at the expense of minority representation, proponents highlight its role in avoiding the gridlock observed in highly fragmented assemblies under purer proportional systems.[63]Comparative Analyses
Versus Largest Remainder Methods
The highest averages methods, also known as divisor methods, allocate seats by repeatedly dividing each party's vote total by a sequence of divisors (such as 1, 2, 3, ... in the d'Hondt variant) and awarding seats to the highest resulting quotients until all seats are filled.[64] In contrast, largest remainder methods first compute a quota (e.g., Hare quota as total votes divided by seats) to assign initial whole seats via integer division of votes by the quota, then distribute remaining seats to parties with the largest fractional remainders.[33] This procedural divergence leads to distinct outcomes in seat proportionality, particularly under varying party fragmentation and district magnitudes. Highest averages methods exhibit a systematic bias favoring larger parties, as the divisor sequence effectively raises an implicit threshold for smaller parties to compete for seats; for instance, in d'Hondt, small parties require disproportionately higher vote shares to secure their first seat compared to larger ones.[65] Largest remainder methods, by prioritizing quota-based initial allocations followed by remainder rankings, produce less bias toward large parties and can yield more seats to smaller ones when remainders favor them, though this may result in greater volatility if vote distributions yield high remainders for minor parties.[33] Empirical analyses, such as those using the Gallagher index of disproportionality, show divisor methods consistently yielding higher disproportionality scores (worse proportionality) than largest remainder under equivalent conditions, with d'Hondt's bias amplifying as district size decreases below 10-15 seats.[33][65] Monotonicity—a criterion requiring that increasing a party's votes does not decrease its seats—is satisfied by most highest averages variants like d'Hondt and Sainte-Laguë, but largest remainder methods (especially with Hare quota) can violate it in multi-party settings where a vote shift alters remainder rankings unfavorably.[33] Conversely, largest remainder avoids the "no-show paradox" more reliably in some quota implementations (e.g., Droop), where abstaining or strategic voting harms the intended beneficiary, though both families are susceptible to other paradoxes like the Alabama paradox in fixed-seat expansions.[33] In practice, highest averages promote coalition stability by concentrating seats among larger parties, as observed in systems like Spain's Congress (d'Hondt since 1986), while largest remainder, used in countries like South Africa (with Droop quota), better accommodates diverse small-party representation but risks fragmented parliaments requiring broader coalitions.[64][33]| Aspect | Highest Averages (e.g., d'Hondt) | Largest Remainder (e.g., Hare) |
|---|---|---|
| Large-Party Bias | High; divisors penalize small parties progressively | Low; quota floors initial seats equally, remainders favor residuals |
| Proportionality (Gallagher Index) | Higher disproportionality, especially in small districts | Lower disproportionality overall |
| Monotonicity | Generally preserved | Can fail due to remainder flips |
| Small-Party Threshold | Implicit (e.g., ~3-5% effective in mid-sized districts) | Explicit only if added; otherwise minimal |
Illustrative Examples Across Methods
The highest averages methods, such as D'Hondt and Sainte-Laguë, allocate seats by repeatedly awarding them to the party with the highest average vote per seat obtained so far, using successive divisors; this contrasts with largest remainder methods, which first assign seats based on a quota (e.g., Hare quota of total votes divided by seats) and then distribute remaining seats to parties with the largest fractional remainders.[66] These approaches can yield divergent outcomes for the same vote distribution, with highest averages generally exhibiting a greater bias toward larger parties than largest remainder variants using the Hare quota.[66][20] Consider an illustrative election with 100,000 total votes and 6 seats to allocate among four parties: A with 42,000 votes (42%), B with 31,000 (31%), C with 15,000 (15%), and D with 12,000 (12%).[66]| Method | Party A | Party B | Party C | Party D |
|---|---|---|---|---|
| D'Hondt (highest averages) | 3 | 2 | 1 | 0 |
| Modified Sainte-Laguë (highest averages) | 2 | 2 | 1 | 1 |
| Hare largest remainder | 2 | 2 | 1 | 1 |