Michael Marsh, "Candidates or Parties?: Objects of Electoral Choice in Ireland," Party Politics, 13 (July 2007), 500-527.

First paragraph:
Studies of electoral behaviour tend to focus on party choice. When it comes to high-profile single-candidate elections, such as those for president in the United States, there is a recognition that the party label is not all that matters and that personal attributes of the candidates have an importance independent of party. Yet there is a significant and growing literature arguing that candidates themselves should be and are important sources of votes in many countries and in much less significant elections. Candidates may attract support for who they are, or what they have done, or what they might do, rather than simply because of the party to which they belong. There are good institutional reasons for this. Under certain electoral systems, individual candidates have a strong incentive to differentiate themselves from others in their party and to develop a personal following. In a widely cited article, Carey and Shugart (1995) explained how this stimulus would be higher where the vote was cast for a candidate and not a party and where that vote had a significant effect not just on which parties won seats but on which candidates did so (see also Katz, 1986; Marsh, 1985b). Many states use multi-member electoral systems that provide particularly strong incentives, including Finland, Switzerland and the Irish Republic,1 while many others, including mixed-member systems such as New Zealand, and singlemember plurality systems, including Britain, the US and Canada, provide some encouragement for candidates to seek personal support

Figures and Tables:
Table 1. Patterns of voting in multi-candidate situations
Table 2. Main and subsidiary reasons offered for selecting first choice candidate
Table 3. Self-reported most important factor in deciding first preference, party or candidate?
Table 4. Would respondent vote for same candidate if candidate stood for different party?
Table 5. Candidate or party index from direct questions
Table 6. Candidate and party ratings for respondents' first preferences
Table 7. Candidate-party differences for lower preferences
Table 8. Candidate- versus party-centred first preference voting: a comparison of distributions obtained using different measures
Table 9. Principal factor analyses of candidate-/party-centred voter measures
Table 10. Multinomial logit estimation of vote choice model
Figure 1. Predicted probability of actual first preference vote.

Second Paragraph of Conclusion:
There are many ways to explore how important party is to voters. We have chosen four here: open-ended questions, a simulated ballot, direct closedended questions and indirect scale measures. All coincide in indicating a very significant degree of candidate-centred voting; the more direct measures suggest most and the more indirect ones suggest least. All concur in identifying some parties' voters as more party-centred (Greens, Sinn Féin, Fianna Fáil) and others (Labour, Progressive Democrats, Fine Gael) as more candidate- centred, at least in 2002. Several measures do suggest that around 40 percent of voters for these parties are significantly candidate-centred.28 In general, the measures based on behaviour indicate lower levels of candidatecentred voting than those based on reported behaviour, while open-ended questions seem to indicate by far the highest levels. It seems likely that although many voters may vote a party ticket they will rationalize this to themselves in terms of candidate qualities. The open-ended measure correlates least well with the other measures, but including all types of measure in a factor analysis still results in an acceptable scale. The cutoff points of this would be arbitrary, so this does not tell us how many voters are either partyor candidate-centred, but it does provide a measure of differences in degree. Behavioural measures are not applicable across many electoral contexts. However, the results here indicate that the sort of closed-ended questions used can produce results that are reasonably equivalent to those obtained using reported behaviour. This should be helpful in comparative work.