Credit Where it’s Due? Valence Politics, Attributions of Responsibility, and Multi-Level Elections

All this assumes that voters have turned out at all, which for obvious reasons they are less inclined to do in second-order elections. Since this paper is about how rather than whether people vote, I will not dwell on the turnout issue.

Similarly, studies have shown that many Canadians see the federal and provincial arenas as separate ‘political worlds’, the relative importance of which varies between citizens (Wilson and Hoffman 1970; Blake 1985).

There is also the added complication that, as Iyengar (1989) notes, responsibility is a multi-faceted concept. On a ‘causal’ interpretation, it concerns which level of government is deemed answerable for what has or has not been achieved in a particular policy area; on a ‘treatment’ interpretation, the key question is which level has the power to affect future outcomes in that area. The survey measures used here tend to guide respondents towards the latter reading. Nonetheless, especially in a case of relatively recent devolution, some may adopt the former view of responsibility instead.

Cutler’s (2004) study compared two Canadian provinces (British Columbia and Alberta).

With this in mind, a closer comparison would be between Scotland and Quebec (e.g. McLean 2001; Henderson and McEwen 2005). Ontario was chosen instead precisely because it allows us to explore how voters attribute responsibility in a less politically charged atmosphere.

An exception is the series of studies of voting in US state-level elections (e.g. Kone and Winters 1993; Atkeson and Partin 1995; Carsey and Wright 1998).

There was an additional SES wave in December 2007, but no data from that survey are used in this paper.

One reason to avoid such speculation is that both the OES and SES will overstate the awareness of their respective publics. In both samples the well educated are overrepresented and abstainers are underrepresented, and both turnout and education are known positive correlates of political engagement and knowledge. The demographic weights applied here are likely to reduce but not to eliminate such bias.

In fact, those who deemed the province responsible for the economy, and the federal government responsible for health, were slightly more confident than the average.

All analyses are run using Stata, and all significance tests are based on robust standard errors.

To save space, I do not set out precise wordings and codings of all the measures used in the empirical analyses. In Appendix A, the variables used are listed along with their OES/SES variable names. Readers can check these details, and if desired replicate the analyses, by accessing the questionnaires and datasets at the studies’ websites.

The biggest difference was in the measures of political knowledge, which took the form in the SES of a true/false quiz on parties and issues in the 2007 election, and in the OES of open-ended questions asking respondents to name the leaders of the provincial parties.

In this and subsequent tables, statistical significance is indicated as follows: * p < 0.10; ** p < 0.05; *** p < 0.01.

Older respondents were noticeably more likely to ascribe responsibility for the economy to Westminster. However, since there was no parallel effect with health, this result cannot really be explained via the earlier speculation about belated reaction to the impact of devolution.

Since this latter value is partly a function of sample size, listwise deletion was used such that only cases available for Model 2 were used in calculating the statistics for the slimmer models.

Mode differences between the surveys suggest an alternative explanation. Internet surveys tend to attract more politically aware and engaged respondents and so seem likely, other things remaining the same, to elicit more structured and predictable attitude responses. This possibility cannot be tested directly given the impossibility of weighting by latent concepts like political interest and knowledge. However, it is possible to weight by reported turnout. And, in line with the suggestion about mode effects, the gap between survey-reported turnout and the actual figure was wider in the Scottish internet survey. However, when the analyses in Tables 3, 4, 5 were rerun applying a turnout weight (in addition to the standard weighting), the results—both the coefficients and the model fits, and hence the differences between Scotland and Ontario—were more or less unaltered.

The SES question was open-ended, while the OES version asked respondents to choose between five issues: the economy, health, education, electricity supply, and taxes.

In Scotland’s mixed electoral system, voters have two ballots. For these analyses I use the constituency rather than the regional list votes, because with fewer minor parties competing these constituency contests are more clearly fought between the main contenders.

The issue opinion variables were not available for two of the Ontarian issues—public transport and social security—from Table 1. For the two new variables, taxes and electricity supply, the attribution of responsibility variable is not the simple provincial-or-federal question as in that earlier table. Instead, respondents were asked “how much responsibility does the Ontario government bear for ______?” and then an equivalent question but about the federal government. These questions involved an 11-point response scale from ‘not at all responsible’ to ‘fully responsible’. To create a parallel dichotomy from these variables, respondents were coded as attributing primary responsibility to that level of government which they rated higher on this scale. Those who gave the same rating to both the provincial and federal governments are omitted from these analyses, as are those who volunteered ‘both’ in response to the simpler question. Where both types of attribution question were available, as with health and the economy, I reran the analysis using each in turn. The substantive conclusions were the same. Those analyses are available from the author on request. Data were pooled across conditions in the question order experiment conducted with these questions (with the order of provincial and federal responsibility switched). Again, there is no evidence that this affected the results of the analysis here).

Although Labour was in coalition (with the Liberal Democrats) in the Scottish executive between 2003 and 2007, it was clearly the dominant coalition partner, and viewed as such by the voters (see Johns and Carman 2008).

Correct estimation of these coefficients requires that the main effect of the attribution of responsibility is also included in the specification (see Appendix B).

Those non-significant coefficients for taxes and electricity give reassurance that endogeneity bias is not an overwhelming problem here. If it was, then perceived performance on any issue would appear to be a strong influence on party choice.

They should also perhaps be treated with some caution given the low Ns, especially in the OES data. This is due to the fact that the attribution of responsibility questions were asked of only around one-third of the total sample, many of whom were not included in the post-election survey (and of course some of those who were included had not voted or had chosen other parties).

The task is still more difficult in cases where there is horizontal as well as vertical separation of powers, and in particular where presidential systems create the possibility of divided government.  There is therefore good reason to suppose that clear attribution effects, of the kind that proved elusive in these analyses, would be still rarer in such contexts.

A couple of caveats are necessary here. First, voters’ tendency to rely on general impressions of parties, blurred across levels of government, is probably exacerbated when—as in Scotland in 2007 but not Ontario in 2003—the same party is in power at both levels. Different governing parties tend to sharpen the distinction between levels, thus facilitating the maintenance of separate evaluations. Second, separate evaluations are also more likely where the party system diverges across levels of government (Hough and Jeffery 2005), and where—as in Canada—many voters even maintain separate party identifications at the different levels.