Electoral reform and the behavioural personalisation of voters: the impact of system change on the importance of party leader and party evaluations in New Zealand elections

ABSTRACT

This paper investigates the effect of a change in electoral system on the relative importance of party leader versus party evaluations in the case of New Zealand national elections, in which an electoral reform from a majoritarian to a mixed system was instituted in 1996. Prior literature on the personalisation of politics suggests that differences in electoral system can affect the importance of party leader and party evaluations in voting choice whereby majoritarian systems exhibit stronger leader effects and mixed systems weaker effects. At the same time, more proportional systems are theorised to enhance party leader visibility and their consequent importance in voting choice. Drawing data from the New Zealand Election Study, the results provide partial support for the moderating effect of system change on the increasing importance of leader evaluations and decreasing importance of party evaluations. The introduction of MMP visibly reduces the importance of party evaluations, but the increasing importance of leader evaluations is to some extent independent of reform.

KEYWORDS:

• Personalisation
• electoral reform
• party leaders
• political parties
• voting behaviour
• first past the post
• mixed member proportional
• New Zealand politics

Introduction

Formal political institutions have been theorised to be the incipient site of personalisation (Rahat and Sheafer 2007, 75). In parliamentary systems, institutions favouring the collectivisation of power in the cabinet, parliament, the electoral system and political parties are thought to give way to the institutionalisation of individual power in the body of the head of state, individual ministers, members of parliament, party leaders and candidates. Inversely, institutions also have the potential to depersonalise power, that is, bring about the collectivisation of power over individual forms.

One domain of representative democracy in which institutions have particular relevance is the electorate. Ample literature has been generated on the personalisation of voter behaviour whereby the influence of the political party as a collective entity is displaced by its most powerful representative, the party leader. The few studies that have sought to incorporate institutional factors have yielded significant results, but these have: (1) predominantly been comparative cross-country diagnoses of personalisation in vastly different national contexts; and (2) yielded competing logics on the causal relationship between institutions and the impact of evaluations on voting choice. This paper contributes to the body of knowledge on institutional effects by testing for the relative importance of party leader and party evaluations on voting choice across two different electoral systems in a longitudinal single-country study. In doing so, it seeks to address the broad-church question, Do electoral institutions affect the relative importance of party leader versus party evaluations in voting choice?

The case of New Zealand offers just such a setting for analysis. An electoral reform was instituted in 1996 via referendum from First Past the Post (FPP) – a plurality majoritarian single-member district (SMD) system, to Mixed Member Proportional (MMP) – a mixed two-vote ballot that combines an SMD vote (the district vote) with a Proportional Representation (PR) party-list vote. The 1996 reform offers a unique opportunity to observe the effects of a change in system on party leader and party importance in voting choice. The first section to follow surveys the existing literature on personalisation and the logics that govern institutional effects. In the second, these logics are applied to the New Zealand context to generate two competing hypotheses. The third section provides a justification of the data and methodological choices employed in the analysis. The fourth contains a description of the results. The concluding section reflects on the theoretical implications of the findings, limitations of the study and future directions for research.

Electoral institutions and party leader versus party importance

Political personalisation occurs when ‘the political weight of the individual actor in the political process increases over time, while the centrality of the political group declines’ (Rahat and Sheafer 2007, 65). While there has been some exploration of institutional variation and its differential impact on party leader and party evaluations from within the personalisation paradigm, this has largely been an indirect acknowledgement. Rahat and Sheafer (2007, 75–76) suggests institutional personalisation to be the first step in the causal chain of interactivity, defined as the ‘adoption of rules, mechanisms and institutions that place more emphasis on the individual politician and less on political groups’ (Balmas et al. 2014, 38); political institutions are the first to personalise, followed by the media and lastly the voting public. Karvonen (2010, 20), too, stresses institutional personalisation’s germinal role in precipitating the behavioural personalisation of voters. Institutional personalisation is limited in its explanatory power as a conceptually predefined sub-programme of political personalisation, focused more on institutional reform as a manifestation of personalisation than as a causal factor of party leader and party importance (Karvonen 2010, 37; Renwick and Pilet 2016, Rahat and Kenig 2018). As a consequence, most studies are comparative and cross-country in design, intended to diagnose a general trend rather that downstream effects (Carey and Shugart 1995, 420–424; Colomer 2011, 10–11). Moreover, the majority of personalisation research involving institutional variation has focused on the dichotomy between individual candidates and parties at the inter-party level – the decentralised variant of institutional personalisation (Balmas et al. 2014, 38). This scholarship is conceptually, and therefore should be empirically, distinguished from centralised forms of personalisation, concerning the importance of leaders – in this case, party leaders.

The most promising scholarship comes from the behavioural personalisation of voters’ sub-programme, which approaches party leader evaluations and voting choice as dependent variables. Again, this literature is primarily based on comparative cross-country work and therefore exposed to a myriad of unaccounted for contextual factors whereby the electoral system is included more often as a control than an independent variable (King 2002, 215–217; Curtice and Holmberg 2005, 251–252; Holmberg and Oscarsson 2011, 50). From this literature, we can derive certain expectations about the impact of institutional variation on party leader and party importance. Curtice and Holmberg (2005, 241–245) finds the contribution of party leader evaluations is greatest in majoritarian systems, less so in mixed systems and of least impact in PR systems based on regression coefficient sizes across the electoral systems of the UK, Germany and the PR systems of Europe. Curtice (2011) likewise finds support for the proposition that party leader evaluations matter more in majoritarian systems which have the largest mean coefficient size, although this is more attributable to the fact that party evaluations are less important in majoritarian systems than in party-centred systems, the logic being that the former places more emphasis on individual candidates and party leaders whereas PR systems are designed to promote parties. The basic intuition behind these studies is that there exists a positive relationship between electoral systems that favour candidate-centeredness and party leader importance. In majoritarian systems where electoral competition more closely resembles a race between two dominant parties, voters are more likely to factor in party leader evaluations (Curtice and Holmberg 2005, 240). Similarly, the effective number of parties has a significant impact on evaluations: party leader evaluations matter more in systems with fewer parties (Bittner 2011, 118).

However, there is good reason to be sceptical about generalising the prevailing logic of candidate – versus party-centredness to MMP systems. Interestingly, Bittner (2011, 124) also finds that proportionality engenders a greater party leader evaluation effect on voting choice. As one of the few studies to seriously consider institutional effects, the finding promises a competing line of enquiry. Proportionality and the number of parties yield counterintuitive implications. Majoritarian systems tend to favour two-party politics but they are the least proportional, raising questions of relative magnitude and interaction effects (Vowles 2005, 301–2). The proportionality thesis proposed by Bittner (2011, 117) is derived from Banducci and Karp (2000, 820), which reasons, based on NZ MMP elections, that party leader evaluations are more important for the party vote because of party leaders greater visibility at the top of the party list whereas candidate evaluations are more important for the district vote. While voter evaluations in the UK and PR systems such as the Netherlands cohere nicely with the two ends of the candidate- versus party-centeredness spectrum, research on MMP systems has yielded competing logics. On the one hand, MMP is theorised to balance the importance of the personal and the party, attributable to its mixed electoral logics (Shugart 2001, 26). Colomer (2011, 10–11)’s personal-party representation scheme of electoral systems places New Zealand and Germany mid-way on the party axis and ‘semi-open’ on the personal axis. On the other hand, Shugart (2001, 41–46), Johnson and Wallack (2012) and Karvonen (2010, 37) code both New Zealand and Germany as slightly party-centred, placing them closer to the PR systems. The most widely researched MMP case of Germany tells a different empirical story. The main parties started to adopt leader-centred electoral strategies two decades ago, in no small part due to the emergence of dominant political figures such as Helmut Kohl and Gerhard Schroder who enjoyed a ‘Chancellor effect’ whereby incumbent leaders matter more in federal election outcomes (King 2002, 215; Brettschneider and Gabriel 2002, 128–129; Garzia 2013, 72). Garzia (2014, 38) finds partisanship to be consistently driven by attitudinal characteristics and leader evaluations in Germany, whereas similarly high levels have only come to be observed in the UK since the 2000s. Germany predominates as the MMP proxy in longitudinal case studies, but the particularities of chancellor-centred voting limits the generalisability of the German case to MMP systems as a whole.

The next step must then be to control for these contextual differences. Many factors remain unaccounted for between electoral systems that may contribute towards the differential impact of party leader versus party evaluations. Some studies control for within-country factors by way of longitudinal single country designs such as Schmitt and Ohr (2000, 7–8) and Brettschneider and Gabriel (2002, 134–141), but no study to date has incorporated ‘naturally-occurring’ institutional variation as a design feature. Filling this methodological void is an important way of expanding the research programme. The following section applies the literature discussed above to the New Zealand case from which observational implications are derived.

MMP in New Zealand

The effects of institutional change would ideally be investigated under experimental conditions in which a manipulation – electoral reform – is applied to an electoral system in a controlled environment. Reality precludes such a design from social scientific research, but the New Zealand electoral reform of 1996 provides a unique opportunity to observe the effects of system change while minimising some of the contextual variability that is the disadvantage of a cross-sectional comparative design. In-depth case study exploration is of particular necessity to MMP cases because of their complexity; Shugart (2001) cautions against relying wholly on reductionist scoring schemes to evaluate MMP party-centeredness without delving into the cases of interest. As such, the New Zealand case presents its own unique institutional context that requires in-depth examination.

There is considerable theoretical and contextual justification to expect that the 1996 reform will bring about a reduction in the relative importance of party leader evaluations on voting choice. A Westminster system emphasises the importance of party leader characteristics (McAllister 2011, 52). A multitude of cross-country behavioural personalisation studies support this supposition – the impact of party leader evaluations is smaller under MMP than in the majoritarian cases (Curtice and Holmberg 2005, 249; Curtice 2011, 98; Holmberg and Oscarsson 2011, 50). Moreover, the electoral reform was the culmination of several long-simmering factors. Vowles (1995, 96–102) identifies four: a political system that facilitates strong government or ‘elective dictatorship’ (also Mulgan 1992), partisan de-alignment, economic liberalisation and low government accountability. Research on centralised institutional personalisation corroborates the first of these claims – New Zealand prime ministerial power grew between 1984 and 1992 (O’Malley 2007, 17). With respect to the second, intense partisan de-alignment of the late twentieth century saw voting for the two main parties undergo rapid decline. By 1993, more than 30% of the votes was for minority parties and trust in parliament had fallen to a meagre 4% (Vowles 1995, 97–98, Heylen Research Centre [1992] 1993 as quoted in Vowles 1995, 102). With respect to the third and fourth factors, two wrong-winner elections in 1978 and 1981, sustained by the momentum of economic policies pushed through in the face of strong public opposition, have been shown to have placed MMP on the reform agenda (Renwick 2009, 363–5; Denemark 2001, 81–83). The public expressed, via referendum, a desire to check the centralisation of power in parliament and government, and institute a more proportional system. Thus, both theory and the political intentions behind electoral reform support the expectation of a decline in the importance of party leader evaluations relative to party evaluations.

H1: Voting choice is relatively less informed by party leader evaluations than party evaluations under MMP.

At the same time, a case can be made for an alternative hypothesis, that is, the electoral reform will bring about personalisation: party leader evaluations become more important relative to party evaluations under MMP than FPP. First, Bittner (2011, 117) and Banducci and Karp (2000, 820) place greater emphasis on the importance of party leader evaluations in more proportional systems based on their greater visibility at the top of the party list. Disproportionality fell from 18.2 in 1993 to 2.5 by 2002 based on Gallagher’s index (Vowles 2005, 301). Party leader evaluations should therefore become relatively more important after the introduction of a party vote. Second, despite its particularities, the German case demonstrates the greater effect of party leader evaluations on the vote (King 2002, 215; Brettschneider and Gabriel 2002, 128–129; Garzia 2013, 72, 2014, 38). New Zealand, too, evidences a tolerance for capricious leadership, so the same leader effect may be observed. There have been several instances of party leaders transcending their party platforms in recent times. Jim Anderton, formerly of the dominant centre-left Labour Party, by the 2011 election had established and led no less than 3 minor parties over the course of his 40-year political career. Likewise Winston Peters, founder and 20-year party leader of New Zealand First, and Don Brash, who had previously been National’s Leader of the Opposition from 2003 to 2006, had left the dominant centre-right National Party to become party leaders of their own minor parties. These instances of party leader volatility would seem to be symptoms of personalisation already in motion. Third, as Vowles (2005, 310) suggests, the previous majoritarian system may be ‘ingrained’ in New Zealand political culture. Ward (1998, 142–144) demonstrates that despite the similarity in constitutional mandate and candidate experience, list members (MPs) were perceived as ‘second class MPs’ due to New Zealand’s tradition of SMDs and the primacy of district MPs. Additionally, MMP raises the site of electoral competition from the district to the national level (Vowles 2010, 892). Taken together, these structural changes would maintain the stigma of list MPs but raise the profile of the most visible list member, the party leader.

H1: Voting choice is relatively more informed by party leader evaluations than party evaluations under MMP.

To conclude, the existing literature provides for two competing hypotheses on the transition from FPP to MMP. On the one hand, the party-centeredness and greater number of parties under MMP is expected to result in reduced party leader importance, contrary to the personalisation thesis. On the other hand, the introduction of a party vote increases the visibility of the party leader and their evaluations in voting choice, corroborating the personalisation thesis. The next section outlines the choice of data and methodology used in the analysis.

Data and method

The data used originate from the New Zealand Election Study (n.d.) spanning nine post-election surveys between 1990 and 2014. The unit of analysis is leader-party rating dyads such that each unit consists of a party rating and a corresponding party leader rating. Party leader ratings are derived from survey questions where respondents were asked to score party leaders on an 11-point scale ranging from ‘strongly like’ to ‘strongly dislike’. Party ratings are derived from survey questions in which respondents were asked to score parties on an 11-point scale ranging from ‘strongly support’ to ‘strongly oppose’. As the pre-reform 1990 and 1993 election surveys scored ratings on a 1–5 scale, they are converted to the revised post-reform 0–10 scale in order to standardise across elections.

Party age and party size have been shown to have a marked impact on ratings whereby larger and older parties are more likely to produce influential leaders (Schmitt and Ohr 2000, 22–23; Curtice and Holmberg 2005, 240; Aardal and Binder 2011, 122). Similarly, the number of parties may affect voting choice. The effective number of parliamentary parties in New Zealand rose from less than 2 between 1946 and 1993 to 3.76 parties after 1996 (Barker et al. 2001, 302). The more parties there are to choose from, the lower the probability of any party receiving a vote. Smaller parties in particular are affected by more competition; they are less likely to be voted for and are less likely to be rated. Including them would suppress leader and party effects on the post-reform vote. Another potential source of bias are the parties that tend to run on social structural identity platforms, such as the Maori and Mana parties. Leader evaluations matter less for parties where social structure is the basis of partisanship (Garzia 2014, 36). In order to control for these effects, analysis is limited to the oldest and largest parties – the centre-right National Party and the centre-left Labour Party. More practically, they are the only two parties that ran consistently in every election throughout the period of analysis. This may cause leader effects to be greater than averaged across all parties, as predicted by a two-party, party leader inflation effect which states that the party leaders of the two biggest parties tend to be more influential than those of smaller parties (Curtice and Holmberg 2005, 240). However, only relative trends are of substantive interest. So long as the effect is controlled for across the electoral reform, results will not be biased.

The aim of this paper is to ascertain the aggregate effect of system change on the relative importance of leader ratings versus party ratings. As such, the presented analysis is performed on a stacked sample that pools all National and Labour dyads following the methodological conventions adopted by Holmberg and Oscarsson (2011, 38) and Garzia (2014, 29), both of which pool dyads into a combined data set for analysis. The total sample size after cleaning and controls is 41,778 dyads with analysis divided into 2 steps. First, individual logistic regressions are run by election year and electoral system with leader and party as predictors and a vote dummy as the outcome variable (0 = no vote for the party, 1 = vote). The logistic coefficients are then graphed for a visual indication of change over time. Next, logistic regression is run on the total stacked sample with the same leader, party and vote predictors as in the previous step, but with an additional reform dummy (0 = FPP, 1 = MMP) and a control for election size (number of parties) in each election survey. Interaction terms are added to test for moderating effects. All analyses are run on the district vote, which is consistent across both electoral systems. The individual regressions are also run on the party vote from 1996 onwards to contrast the change between pre-reform effects on the district vote and post-reform effects on the party vote.

Disentangling the highly correlated relationship between leader and party has always been a challenge (Curtice and Holmberg 2005, 241). Establishing independent effects remains problematic, but the extent of interrelationship can be highly informative with respect to electoral system change. Larger correlations after reform would suggest that MMP enables convergence between leader and party, whereas smaller correlations the opposite effect. This would cohere with research that aims to quantify the endogeneity of leader effects in party effects, and therefore imply a general underestimation of their impact on voting choice (Garzia 2014, 79). Likewise, external factors external to the direct evaluations of a leader or party such as the impact of other leaders and parties are likely to influence voting choice. It is therefore worth stressing the interpretive quality of the data and methods used in this paper and the limitations of taking an aggregate approach. Public opinion surveys are an excellent source for the analysis of leader and party evaluations because they impose a top-down approach whereby structural effects can be interpreted from an aggregation of individual behaviour (Garzia 2014, 4). The limitations of this approach need to be addressed by way of diagnostic indicators including correlation values and model fit, and acknowledged in the interpretation of the results. The following section lays out the key findings.

Results

Logistic regression by election year

Tables 1 and 2 present the results of the base model (vote = constant + party + leader) for the regressions by election year on the district and party votes, respectively. With the exception of 2014, leader and party are significant predictors of the vote (p < .001) in all election years. Party effects are clearly greater than leader effects on voting choice, regardless of electoral system. For the district vote, the difference between electoral systems for party constitutes a 30% drop in contribution to the model (0.95–0.66), which is larger than the 25% increase for leader (0.14–0.19). These results suggest that a modest increase in leader effect occurred after reform, complemented by a substantial decline in party effect. However, the increase in leader effect does not wholly capture the decrease in party effect and the fit of the model is worse under MMP, suggesting that factors external to party and leader evaluations affect voting choice.

Table 1. Logistic regression of district vote choice on party and party leader evaluations with bootstrapped confidence intervals, Nagelkerke’s R² and Pearson’s r correlation values.

	Constant				Party				Leader
Year	b	SE	Lower CI	Upper CI	b	SE	Lower CI	Upper CI	b	SE	Lower CI	Upper CI	Nagelkerke’s R^2	r (Party × leader)
All	−5.12	0.05			0.7	0.01			0.18	0.01			0.64	0.82
FPP	−6.24	0.18			0.95	0.03			0.14	0.02			0.75	0.66
MMP	−4.97	0.05			0.66	0.01			0.19	0.01			0.63	0.82
1990	−6.93	0.28	−7.57	−6.3	0.95	0.04	0.87	1.04	0.2	0.03	0.13	0.27	0.76	0.63
1993	−5.9	0.25	−6.44	−5.35	0.96	0.04	0.87	1.04	0.13	0.03	0.07	0.19	0.74	0.69
1996	−4.38	0.12	−0.47	−0.41	0.58	0.02	0.54	0.63	0.17	0.02	0.14	0.21	0.53	0.77
1999	−5.5	0.14	−5.79	−5.22	0.71	0.02	0.67	0.77	0.19	0.01	0.16	0.23	0.61	0.79
2002	−4.75	0.11	−4.99	−4.51	0.56	0.02	0.52	0.61	0.26	0.02	0.22	0.3	0.6	0.77
2005	−5.42	0.16	−5.78	−5.07	0.75	0.03	0.69	0.82	0.18	0.03	0.13	0.24	0.73	0.89
2008	−5.59	0.18	−5.97	−5.22	0.59	0.03	0.51	0.66	0.32	0.03	0.26	0.38	0.67	0.89
2011	−5.13	0.17	−5.51	−4.76	0.67	0.03	0.6	0.74	0.21	0.02	0.16	0.27	0.67	0.78
2014	−4.77	0.16	−5.11	−4.41	0.79	0.03	0.73	0.87	0.03^ns	0.02	−0.02	0.07	0.66	0.82

ns indicates p > .05. All other coefficients (beta values) are significant at p < .001.

Table 2. Logistic regression of party vote choice on party and party leader evaluations with bootstrapped confidence intervals, Nagelkerke’s R² and Pearson’s r correlation values.

	Constant				Party				Leader
Year	b	SE	Lower CI	Upper CI	b	SE	Lower CI	Upper CI	b	SE	Lower CI	Upper CI	Nagelkerke’s R^2	r (Party × leader)
All	−5.12	0.05			0.7	0.01			0.18	0.01			0.64	0.82
FPP	−6.24	0.18			0.95	0.03			0.14	0.02			0.75	0.66
MMP	−4.97	0.05			0.66	0.01			0.19	0.01			0.63	0.82
1990	−6.93	0.28	−7.57	−6.3	0.95	0.04	0.87	1.04	0.2	0.03	0.13	0.27	0.76	0.63
1993	−5.9	0.25	−6.44	−5.35	0.96	0.04	0.87	1.04	0.13	0.03	0.07	0.19	0.74	0.69
1996	−5.99	0.17	−6.35	−5.62	0.80	0.03	0.74	0.86	0.20	0.02	0.16	0.24	0.65	0.78
1999	−7.18	0.18	−7.56	−6.80	0.88	0.03	0.82	0.94	0.28	0.02	0.24	0.32	0.71	0.78
2002	−6.78	0.18	−7.16	−6.40	0.73	0.03	0.67	0.79	0.40	0.02	0.35	0.45	0.73	0.81
2005	−6.86	0.22	−7.33	−6.39	0.96	0.04	0.88	1.04	0.22	0.03	0.15	0.28	0.81	0.89
2008	−6.64	0.22	−7.09	−6.17	0.68	0.04	0.60	0.76	0.37	0.03	0.31	0.45	0.73	0.89
2011	−7.06	0.25	−7.60	−6.51	0.86	0.04	0.76	0.94	0.32	0.03	0.25	0.39	0.78	0.79
2014	−7.56	0.28	−8.15	−6.99	1.09	0.05	1.00	1.20	0.16	0.03	0.10	0.22	0.81	0.83

All coefficients (beta values) are significant at p < .001.

A similar pattern for the party vote provides corroboration of a systematic phenomenon irrespective of vote type. The contribution to the model across electoral systems constitutes a considerably larger 50% gain (0.14–0.27), compared to only 25% in the case of the district vote. This may be explained by the greater significance of the party vote at the national level, which gives more prominence to leaders than they would receive at the district level. Although still manifest as a downward trend, the drop in party effect after reform is less extreme in the case of the party vote. The drop in model contribution between electoral systems is only 10% (0.95–0.84), suggesting that the party vote partially re-captures the unaccounted for drop in party effect on the district vote.

The Pearson’s r correlation values have additional implications for result interpretation. Leader and party ratings are highly correlated, with r values between 0.63 and 0.89. The correlation values demonstrate a modest but increasing trend overall, and a marked increase after reform, indicating greater convergence between leader and party under MMP. Considered together with larger leader effects, one could say that leaders are personifying their party platforms to a greater degree after reform.

Figures 1 and 2 graph the logistic regression coefficients (beta values) and bootstrapped confidence intervals by election year where the trend lines and differences between the two votes become more apparent. Leader effects are modestly increasing for both votes, with no marked alteration under MMP. The trend lines are somewhat biased by a lack of high leader-low party ratings in the 2014 data set, producing the drop in leader effect and corresponding increase in party effect for that year. If the data points for 2014 are ignored, a different pattern emerges. One can observe a relatively stable and weakly increasing leader trend line for the district vote that peaks in 2008, strongly indicative of personalisation. In contrast, the differences between leader effects before and after reform become more pronounced for the party vote. There appears to be a reform-independent leader effect on the district vote and reform-dependent leader effect on the party vote.

Figure 1. The influence of party leader and party effects on the district vote by election year.

Figure 2. The influence of party leader and party effects on the party vote by election year.

Leader effects are more consistent across both votes than party effects. The unexpected drop in party effect is more extreme for the district vote and doesn’t ‘recover’ under MMP, which is highly suggestive that the effect is a direct result of reform. Party effects for the party vote demonstrate increasing volatility under MMP, reaching levels that surpass their earlier magnitude under FPP in 2005 and (tentatively) 2014. Party effects for the party vote seem to be more susceptible to inter-electoral volatility than for the district vote.

There is considerable variation between elections, so it is possible that these differences are attributable to the peculiarities of the leaders or parties in each election. Bittner (2011, 125–7, 129) finds centre-left parties to be driven more by leader effects than conservative or left parties. The years in which centre-left leaders were voted in should then correlate with increasing leader effects. A Labour Prime Minister, Helen Clark, was voted in at the 1999, 2002 and 2005 elections. While increasing leader effects are observed for the first two elections, a centre-left Prime Minister, John Key, was voted in at the peak of leader effects in 2008 so leader partisanship cannot account for inter-electoral variation. The next step is to estimate the effects of reform and other structural factors on the complete data set.

Logistic regression with interaction terms

Three models are fitted to the total stacked data set (Table 3). The basic reform predictor in Model 2 suggests that the introduction of MMP reduces voting probability (b = −0.58), supporting the idea of greater electoral competition brought about by an increase in the number of parties. In order to investigate the changing importance of leader and party evaluations on the (district) vote, however, interaction terms are more indicative. A party*leader interaction term was included at first, however the effect size was small with no discernible improvement in model fit (R²), so it was dropped. The reform*leader interaction term is not significant but the reform*party coefficient is significant and negative (b = −0.35), confirming the findings from the regressions by election year. Reform does not seem to have increased the importance of leader evaluations but the introduction of MMP reduces the importance of party evaluations in voting choice.

Table 3. Logistic regression of district vote choice on party and party leader evaluations, institutional control variables and interaction effects for the total stacked data set, inclusive of all elections, 1990–2014.

						Model comparison
		b	SE	Nagelkerke’s R^2	X^2 (df)	X^2 (df)
Model 1	Constant	−5.12	0.05	0.64	27507.08 (2)
	Party	0.7	0.01
	Leader	0.18	0.01
Model 2	Constant	−5.24	0.08	0.64	27574.51 (4)	67.43 (2)
	Party	0.7	0.01
	Leader	0.18	0.01
	Reform	−0.58	0.07
	Size	0.09	0.01
Model 3	Constant	−8.83	0.44	0.65	27747.88 (9)	173.37 (5)
	Party	0.91	0.04
	Leader	0.25	0.03
	Reform	2.86	0.47
	Size	0.69	0.1
	Reform*party	−0.35	0.05
	Reform*leader	0.13^ns	0.04
	Reform*size	−0.57	0.1
	Size*leader	−0.02	0.01
	Size*party	0.01	0.01

n = 41,778. ns All coefficients (beta values) are highly significant (p < .001) except reform*leader (p = .28).

Size*leader has a small effect on voting probability such that, in the presence of a greater number of parties, leader evaluations have a small decreasing effect on voting probability while size*party has a small increasing effect. The direction of the size interaction coefficients contradict the reform*leader and reform*party coefficient values, but the effect sizes are small, and thus of little moderating effect. One could however say that reform has a significant moderating impact on party evaluations independent of any effect brought about as a consequence of the number of parties running for election. This provides some support for the claim that proportionality is a more important structural predicate of leader and party evaluations in voting choice than the number of parties. At the same time, it must be noted that model fit did not improve greatly after the addition of more predictors. There are therefore still factors external to leader, party, electoral reform and size that contribute towards voting choice.

The interaction terms corroborate the results from the regressions by election year. The results lend no support to H1 since leader evaluations become more important and party evaluations less so under MMP. The number of parties has a moderating effect, which confirms Bittner (2011)’s theses about the relationship between electoral competition, proportionality and leader importance, but the analysis presented here goes further to infer that proportionality under MMP counteracts the effect of the number of parties. There is mixed support for H2 as leader effects appear to be increasing, but seemingly as an independent process from system change. Only when the 2014 election is ignored in the analysis of the party vote is there some confirmation of a reform-moderated leader effect. The more tangible conclusion is that leaders are becoming more influential regardless of electoral system. The more surprising finding is how party evaluations are affected under MMP. The intuition behind H2 is that the loss in party effect should come at the expense of leaders, a dynamic that should be more pronounced for the party vote according to Banducci and Karp (2000). Leaders are neither becoming more important as a result of MMP, nor is the phenomenon more pronounced for the party vote. The drop in party effect might be because the introduction of a party vote re-captures party importance, but further research is needed to estimate the magnitude of such an effect.

Conclusion

The impact of institutional factors on the importance of party leader and party evaluations in voting choice has largely been examined from a comparative cross-country perspective. This paper, by adopted a longitudinal approach, controls for more of the contextual vagaries between countries that would otherwise mask such counterintuitive results. The findings are noteworthy because they question prevailing theory on the implications of candidate- versus party-centeredness and electoral competition as predictors, as well as alternative formulations that expect a more proportional system to amplify party leader effects.

Candidate- and party-centeredness typologies of electoral institutions do not inevitably translate to the electorate. Party-centred PR systems do seem to cultivate more electoral competition and result in reduced party leader importance, but the effect is weaker relative to other structural factors. In MMP systems especially, the presence of two votes requires a conceptual separation between candidate importance, which matters more in the district vote and party leader importance, of more significance in the party vote (Banducci and Karp 2000, 820). Expectations of less candidate-driven voting in the district vote don’t foreshadow the same consequence for party leader importance. Likewise, the presence of centralised personalisation doesn’t always imply the same for decentralised personalisation.

There is only limited support for the claim that systems which provide for list votes, and which are more proportional, promote greater leader visibility. Leader evaluations are more important but not as a consequence of the transition from a less to a more proportional system. The surprising implication is that, in the absence of a reform-induced party leader effect, there is in fact reduced party importance when the system is more proportional. The findings echo the dynamics of the German case to a limited extent in that leaders matter a great deal in voting choice, despite the party-centred logics used to categorise MMP systems. The differences arise in accounting for party implications. Leaders are personifying their platforms to a greater extent in Germany (King 2002, 215, Garzia 2013, 72). In contrast, the change to MMP, which was intended to increase proportionality and decentralise power, seems to have eroded the importance of party evaluations in voting choice. In this sense, declining party importance was very much an unintended consequence of electoral reform.

The most logical explanations seem to lie in the particular context of the New Zealand MMP system. For one, the addition of a party vote elevates the site of electoral competition from the district to the national level (Vowles 2010, 892). Party leaders become more visible as the face of the party and more influential as a consequence. The introduction of a second vote under MMP then has the effect of watering down the candidate–party relationship otherwise afforded by a single-district vote. As a result, party evaluations would matter less, not more, under MMP. Second, a two-vote ballot introduces split ticketing and strategic voting incentives, so voting patterns may not correlate with party evaluations; respondents may evaluate parties highly, but vote strategically for potential coalition partners. The increasing volatility of party evaluations after reform certainly speaks to split ticketing as an additional causal factor. Third, New Zealand’s legacy of majoritarianism may offset the intended effects of electoral reform. If voters still perceive of politics as a contest between the two main parties, the introduction of a party vote may compound this perception. Voters would be more, not less, likely to take leader evaluations into account. Fourth, the reform was arguably the most prominent political event of 1996. As the first election to adopt an MMP system, the parties may have received more attention than otherwise with the effect of elevating party leader importance. Whether these explanations are borne out on an empirical level requires further research.

Several methodological issues should also be considered. First, the highly correlated relationship between party leader and party evaluations compounds the difficulty of untangling these two factors and in deriving confident conclusions about the changing dynamic between them. That being said, endogeneity is more likely to result in an underestimation of leader effects than to provide a suitable explanation of declining party importance (Garzia 2014, 79). Second, the change in survey rating scale from a 0 to 5, to 0 to 10 Likert scale coincides with the electoral reform. It is unlikely that the effect is of such a size as to undermine the causal effect of electoral reform or invert trends in party leader and party influence, but a change in scale may have some distorting effects on respondent evaluations. Finally, the incomplete observations of the 2014 survey somewhat limits the interpretability of the overall coefficient trend lines. The inclusion of newly released survey data would enable more decisive conclusions regarding the effect of reform on party leader importance.

Future research should aim to buttress these methodological deficiencies and be taken in the direction of causes: Why do more proportional systems affect a decline in party importance and what moderating factors explain this phenomenon? The influence of mass parties is clearly shaped by the electoral systems in which they operate, but this paper demonstrates that existing theories on institutional variation do not satisfactorily account for their impact on party leader and party evaluations. Party leaders have become more prominent in the voting considerations of the electorate, but this is not attributable to the particular electoral institutions in which they operate.

Acknowledgements

The author would like to thank Rudy Andeweg for his guidance and criticism. Without his support, this paper would never have come to fruition. I am grateful to Petr Kopecky, Maarja Beerkens, Michael Meffert, Jesper Lindqvist and Malcolm McKinnon whose recommendations and advice proved invaluable, as did the overarching support of the Political Science and Public Administration family at Leiden University. Finally, a word of thanks to Jack Vowles, Sam Bigwood, the guest editors at Political Science and my anonymous reviewers for their comments and their time.

Disclosure statement

No potential conflict of interest was reported by the author.

Additional information

Notes on contributors

Imogen T. Liu

Imogen T. Liu is a PhD Candidate at Maastricht University. Her research is focused on questions of institutional change in political and economic systems.