Up: Issue 15 Next: Paper 2

Voting matters - Issue 15, June 2002

Positional Voting Bias Revisited

Philip Kestelman

Introduction

It is widely supposed that candidates appearing high on ballot-forms enjoy a considerable electoral advantage. In a highly influential paper on the 1973 General Election to the Irish Dáil, by multi-member Single Transferable Voting (STV), Robson and Walsh (1974) observed that Deputies (TDs) over-represented candidates with A-C surnames. Compared to randomly sampled Irish electors, 'The under-representation of M-O names among politicians is very striking'.

Proportionality conventionally measures the relationship between numbers of Party votes and seats (regardless of candidates). Despite a probable age bias, we are hardly concerned that seats considerably over-represent first preferences for incumbent candidates; let alone that incumbents are far more likely to be elected than 'excumbents' (non-incumbents).

On the other hand, we are concerned not only that seats should proportionally represent votes for women candidates, but also that seats should be proportional to women candidates, in the interests of Parliament representing society. In respect of ballot-form position, we are primarily concerned with the relationship between numbers of candidates and seats (regardless of votes), by surname initial, when candidates are listed surname-alphabetically on ballot-forms.

Electability

This article mainly evaluates positional voting bias in the last 12 general elections in the Irish Republic (1961-97). Electability is quantified in terms of an Electability Index (S%/C%): the ratio of a seat-fraction (S%) to a candidate-fraction (C%); and of a Relative Electability Ratio: the ratio between specified Electability Indices.

Aggregating all 12 elections (Total S/C = 1,875/4,594), Upper/Lower half surname A-J/K-Z Electability Indices were 1.11/0.88, with a statistically highly significant Relative Electability Ratio of 1.26 (P<0.001). By comparison, alphabetically Upper/Lower half forename A-L/M-Z Electability Indices were 1.01/0.99, with an insignificant Relative Electability Ratio of 1.01 (P > 0.05).

Cumbency

In 1961-97, most incumbent candidates (S/C = 1,404/1,687 = 83 percent) were re-elected; whereas few excumbents (471/2,907 = 16 percent) were elected, rendering them more susceptible to alphabetic disproportionality. Surname A-J/K-Z Electability Indices (S%/C%) were 1.01/0.98 for incumbents, and 1.15/0.86 for excumbents, with Relative Electability Ratios of 1.03 (P>0.05) and 1.34 (P<0.05), respectively.

The last 12 Irish general elections have consistently over-represented excumbent candidates with A-C surnames; under-representing those with K-M surnames (overall S%/C%, 1.27 and 0.81: Table A). Even combining the 12 elections into three quartets leaves considerable variability in both forename and surname Electability Indices.


Table A: Excumbent Electability Index, by Elections and Forename/Surname initial letter: Irish Republic, 1961-97 (12 general elections: Dáil Éireann, 1962-98).

* P < 0.05


In 1961-73, excumbent forename and surname alphabetic biases were equally convincing (P<0.05); but insignificant subsequently. Ironically in 1973, the Relative Electability Ratio for A-L / M-Z forenames (2.76) exceeded that for A-J / K-Z surnames (1.57)! The pitfalls of generalising from a single election are manifest.

District Magnitude

Surname disproportionality was virtually confined to four- and five-member STV constituencies: only three-member constituencies returned TDs more-or-less faithfully reflecting excumbent surnames (Table B). Magnitude-specific surname A-J/K-Z Relative Electability Ratios proved statistically insignificant, but much closer to unity in three-member constituencies (1.25, 0.89 and 0.72) than in four-member constituencies (1.62, 1.36 and 1.51), or in five-member constituencies (2.05, 1.77 and 1.42), in 1961-73, 1977-82 and 1987-97, respectively.
Table B: Excumbent Electability Index, by District Magnitude (seats per constituency) and Surname initial letter: Irish Republic, 1961-97 (12 general elections: Dáil Éireann, 1962-98).

*P < 0.05 +Including a few two-member constituencies.


District Canditude and Position

Interestingly, the 1961-97 aggregate, excumbent Relative Electability Ratio by surname (A-J/K-Z) proved identical with that by ballot-form position (Upper/Lower = 1.34: P<0.05). Like the surname A-J/K-Z Relative Electability Ratio with district magnitude (the number of seats per constituency), the positional Upper/Lower Relative Electability Ratio increased with district 'canditude' (the number of candidates per constituency: Table C).
Table C: Excumbent Electability Index, by District Canditude (candidates per constituency) and Ballot-form Position: Irish Republic, 1961-97 (12 general elections: Dáil Éireann, 1962-98).

* P < 0.05 +Excluding odd-Canditude mid-candidates.


Party Policy

Both main political parties in the Irish Republic (Fianna Fáil and Fine Gael) have staunchly denied over-nominating candidates appearing high on ballot-forms[9]. Table D analyses the surname-alphabetic distribution of FF and FG excumbent candidates, compared with other (non-FF + FG) excumbents, in terms of a Relative Nomination Index, over time.


Table D: Two Main Party Excumbent Relative Nomination Index, by Elections and Surname initial letter: Irish Republic, 1961-97 (12 general elections: Dáil Éireann, 1962-98).

*** P < 0.001


Evidently since 1977, both main parties have greatly over-nominated A-C surname candidates (and/or other parties have under-nominated them); with the honours evenly divided between Fianna Fáil and Fine Gael. Relative to the publication of Robson and Walsh (1974), the timing may not have been entirely coincidental!

Electorate

Robson and Walsh (1974) observed that the alphabetic distribution of surname initial letters differed insignificantly between randomly sampled Irish electors and excumbent candidates at the 1973 Irish General Election. Presumably nowadays, the surname initials of electors are rather better represented by excumbent, non-FF + FG candidates; and Table E compares overall seat-fractions, by surname initial, with excumbent, non-FF + FG candidate-fractions.

This Surname Concentration Index (Total S%/Excumbent, non-FF + FG C%) highlights Dáil Éireann over-representing A-C surnames in the Irish electorate; while under-representing K-M surnames. Despite the lower surname A-J/K-Z Concentration Ratio since 1987 (1.35), A-C surname electors remain over twice as likely as K-M surname electors to become TDs.


Table E: Surname Concentration Index, by Elections and Surname initial letter: Irish Republic, 1961-97

(12 general elections: Dáil Éireann, 1962-98).

** P < 0.01 *** P < 0.001


Other STV Elections

Compared to the last 12 Irish general elections, with a total surname A-J/K-Z Relative Electability Ratio of 1.26 (P<0.001), the last five European elections in the Irish Republic (1979-99: Total S/C = 75/234) have yielded a higher but statistically insignificant surname A-J/K-Z Relative Electability Ratio of 1.37 (P > 0.05) [5].

On the other hand, the last five Irish Local Elections (1979-99: Total S/C = 4,918/10,250) disclosed a lower surname A-J/K-Z Relative Electability Ratio of 1.12 (P<0.001)[3]. Perhaps better acquainted with local government candidates, voters discriminate more individually; numbering their preferences regardless of alphabetical order.

At the 1973 Assembly Election in Northern Ireland, Robson and Walsh (1974) attributed eight out of 78 Seats to positional voting bias. Yet at the 1998 Northern Ireland Assembly Election (Total S/C = 108/296), the surname A-J/ K-Z Relative Electability Ratio fell below unity (0.87: P>0.05)[4]. Certainly, parties are more sharply differentiated in Northern Ireland than in the Irish Republic.

Discussion

Using forenames as controls, surname-alphabetic electability valuably measures voters' lack of discrimination between candidates within parties. Neither voters nor the Irish electoral system (STV) can be reproached for any positional voting bias.

However, Dáil Éireann remains surname-alphabetocratic, over-representing candidates with A-C surnames, while under-representing excumbents (non-incumbents) with K-M surnames (Table A: compare Table E): especially in constituencies with over three seats (Table B), and/or over eight candidates (Table C).

Perhaps aware of Robson and Walsh (1974), Ireland's two main parties (Fianna Fáil and Fine Gael) have apparently over-nominated A-C excumbents (notably since 1977: Table D). However, thus acting on the belief of increased electability may itself increase A-C surname over-representation: aggregating the last 12 Irish general elections (1961-97), excumbent S%/C% for FF + FG (1.67) was considerably higher than for other parties (0.43).

Reassuringly, aggregating all 12 general elections (1961-1997), the excumbent Surname Relative Electability Ratio (S/C ratio: A-J/K-Z=1.34 overall) proved significantly higher for FF+FG (1.28: P<0.05) than for the other candidates (1.03: P>0.05). however, it remains unclear whether the two main parties have benefited from A-C over-nomination.

Darcy and McAllister (1990) found 'no evidence for position advantage for political parties in any election'. Their review concluded that positional voting bias may be eliminated by removing its causes: notably, compulsory voting; completion of all preferences; and ballot-forms not indicating candidates' Party affiliation (as in Ireland before 1965[7]).

On the strength of the 1973 Irish General Election, Robson and Walsh (1974) advocated randomising the order of candidates on ballot-forms. Citing Robson and Walsh (1974), Sinnott[8] suggested that the problem could 'easily be eliminated by arranging the names in a number of different randomised orders on different sets of ballot papers'.

At the Dublin High Court in 1986, Mr Justice Murphy accepted that candidates with surname initials high in the alphabet were over-represented but, noting that alphabetic order helped voters to find candidates, he found it constitutional[9]. Indeed, the voter's predicament is paramount; and to avoid the palpable frustrations of randomised ballot-forms in locating preferred candidates, a reasonable compromise might be to print half the ballot-forms in surname-alphabetic order, with the other half in the reverse order — if positional voting bias really matters.

Acknowledgement

I am indebted to David Hill for statistical advice. Statistical significance was calculated by combining election-specific, one-tailed exact two-by-two table probabilities[6].

References

  1.  Dáil Éireann (1962-98): Election Results and Transfer of Votes in General Election: October 1961 — June 1997. Stationery Office, Dublin.
  2. Darcy R and McAllister I (1990): Ballot Position Effects: Electoral Studies 9, 5-17.
  3. Department of the Environment and Local Government (1980-2000): Local Elections, 1979-99. Stationery Office, Dublin.
  4. Chief Electoral Officer for Northern Ireland (1998): The New Northern Ireland Assembly Election 25 June 1998. Belfast.
  5. European Parliament Office in Ireland/Department of the Environment and Local Government (1980-2000): European Parliament Election Results, Ireland, 1979-99.
  6. Fisher R A (1970): Statistical Methods for Research Workers, 14th Edition. Hafner Publishing Company, New York (pages 99-101).
  7. Robson C and Walsh B (1974): The Importance of Positional Voting Bias in the Irish General Election of 1973: Political Studies 10, 191-203.
  8. Sinnott R (1999): The Electoral System: in Coakley J and Gallagher M eds: Politics in the Republic of Ireland (Third Edition). Routledge, London (page 124).
  9. Trench B et al (1987): The Alphabet Advantage: in Magill Book of Irish Politics: Election February 87 (page 23).

Up: Issue 15 Next: Paper 2