The article below is reprinted with the kind permission of the Hill Times, from their Monday May 11 edition
. Part I can be found here
Money and votes: the effects of election spending by candidates in federal elections, 1997 to 2006 Increased competitiveness in a federal riding associated with increased spending for candidates who are serious contenders.
By ALICE FUNKE and W.T. STANBURY
In this piece, we examine the relationship between candidates' outlays on "election expenses" and the percentage of votes they obtained, and use some statistical techniques to answer some interesting questions: 1. How closely is spending related to outcome? 2. How is spending different in close races? and 3. How much of an increase in a candidate's share of the vote could be expected for an increase in spending? As noted in our previous piece in The Hill Times
(May 4, 2009), candidate (and party) spending on election expenses has been subject to legal limits since 1974.The relationship between campaign spending and votes
To examine the relationship between candidate spending and vote shares, we used a simple correlation. A correlation coefficient is a number that shows whether two things are related, and if so how strongly, but it can't tell you which caused the other. Higher correlations (which vary in absolute terms between zero and one) indicate stronger relationships. Negative numbers mean they are related but move in opposite directions (i.e., more of one means less of another). And a value near zero means there is very little relationship at all.
When we combine all candidates for each election there was a fairly high positive correlation (0.84) between the per cent of the limit spent and the percentage of votes obtained (Table 1
). We must be careful not to jump to conclusions about why, i.e., we can't conclude that higher spending levels led to more votes based merely on the observed positive correlations. It could be instead that the perception of "win-ability" influenced spending levels, or even that strength of support influenced fundraising capacity which, in turn, affected spending. To further analyse the nature of the relationship we used regression analysis, the results of which are reported below.
If we look at the correlations by party, combining all four elections, we find that the correlations were lower except for the NDP (0.84). For the major parties, the lowest correlation coefficient was for the Conservatives (0.44 for two elections) and the Liberals (0.51) and the BQ (0.52). By comparison, the correlation for Reform/Canadian Alliance was 0.75 for two elections (Table 1
When we consider the correlation between election expenses and votes for each party and each election separately, we find it varied greatly. In 1997 for the Liberal candidates it was only 0.23. In 2006 for the Conservatives, it was 0.26. Yet in 1997, the Liberals won a majority government (155 of 301 seats) and in 2006, the Conservatives won a minority government (124 of 308 seats).
At the other end of the spectrum, correlation between spending and votes for the Progressive Conservatives in 2000 was 0.81. Yet they won only 12 seats! This could suggest that had the low level of average
spending (23.7 per cent of the limit) been increased, the PCs might have won more votes and perhaps more seats. Or it could mean that PC candidates planned their campaign spending based on their perceived chances of winning the seat.
In each of the four elections the correlation between spending and votes for NDP candidates was 0.85 to 0.89. Yet the NDP elected only 13 to 29 candidates in each election (average of 21). The problem for the NDP is that, on average, its candidates spend only 20.1 to 28.4 per cent of the limit. Again, more spending might have resulted in more votes, and possibly more seats. But it might equally be true that the NDP was very good at focusing riding-level spending where candidates had a good chance of being elected.
We interpret the results in Table 1
as follows: The higher the correlation (especially where overall spending levels are lower), the more targeted the parties were in their campaign efforts. This is based on our reading the correlations together with the average per cent spent and its standard deviation (not reported here). We found that if there was a low average per cent spent and a high standard deviation, there was a high correlation between per cent spent and percentage of votes obtained. This could be described as the campaign model of a smaller party targeting ridings in the hopes of winning seats in a first-past-the-post electoral system.
In cases where the average per cent spent was higher, with a low standard deviation, we found a low correlation between per cent spent and per cent of vote obtained. That could be described as the campaign model of the winning party: if the money is available, spend as much as possible, in as many ridings as possible, in the hope of seat gains or to make it harder for opponents to compete.
The lower correlations for the Liberals in 1997 and the Conservatives in 2006 suggest that they had such well-funded campaigns they could spend a lot everywhere. The Bloc Québécois elected more candidates in 2004 and 2006, even though they spent a lot less. The details of the so-called "sponsorship scandal" and the subsequent Gomery inquiry certainly gave them a good issue to run on in those two elections. Furthermore, most of the BQ candidates were by that time longstanding incumbents (the benefits of incumbency were discussed by us in The Hill Times
, April 27, 2009), and few of them found themselves in close races.
The Greens are a puzzle. There is only a weak correlation (0.57 to 0.64) between their candidates' election spending and outcomes (defined as percentage of vote) over the period studied, and we did not find any intermediate explanatory variables based on data available to us. Their own members are closely studying the relationship between spending and outcomes, but perhaps the answer is that they do not yet have sufficient history and data to target properly, have not maximized the effectiveness of their spending, or else cannot agree on a "rising tide" versus a "beachhead" strategy in the first-past-the-post context.
Note that since the public subsidies for parties came into effect on Jan. 1, 2004, there is also an incentive for parties to try to increase their total popular vote, even if they win no more seats. Thus, there is a reward for increasing your raw vote even in non-winnable ridings. The strategic question for parties is whether the cost of higher spending to get more votes exceeds the increased subsidy over the period to the next election. Moreover, it may pay safe incumbents to spend less, in order to maximize the return on investment for their party in their ridings.Spending and outcomes for incumbents and challengers
Incumbent MPs who chose to run again had a high rate of success (average of 86.1 per cent) in the four federal elections between 1997 and 2006. (For more detail on the re-election rates for incumbent MPs, see our piece in The Hill Times
, April 27, 2009.) In each of the four elections between 1997 and 2006, incumbents who were re-elected spent on average slightly less
(from 2.1 to 4.2 percentage points) than did challengers (i.e., non-incumbents) who were elected. (Note that the data in Table 2
combines the two categories of successful challengers: those who defeat an incumbent seeking re-election; and those who win an open seat for their party.)
Challengers who were unsuccessful
spent an average of only 21.1 per cent to 28.7 per cent of their limit on election expenses. This was typically less than one-third what winners spent whether Incumbents or Challengers (Table 2
).Spending in close races
It is reasonable to expect that in close races the leading candidates will spend a higher percentage of the limit on election expenses, and indeed this is what we found. We defined three types of mutually exclusive close races: two-way, three-way, and four-way. See the definitions below Table 3
Over the four federal elections between 1997 and 2006, there were 121 (9.9 per cent) close two-way races, 80 (6.6 per cent) close three-way races, and 24 (two per cent) close four-way races. Thus "not close races" comprised 81.5 per cent of all races (Table 3
). Note that for all four elections combined, each race had an average of 5.6 candidates, but there were some races with as few as three candidates.
Looking at all 1,218 races without regard to how close they were, the winner spent 77.7 per cent of the limit compared to 66.4 per cent for the second place candidate and 31.8 per cent for the third place candidate (Table 3
In all four types of races, the winner still on average outspent all rivals, although the difference between the winner and the second place candidate was minute (81.5 per cent vs. 81.1 per cent of the limit) in close two-way races. In the 121 close two-way races, the tiny difference in average per cent spent means that in many cases (65 of 121), the winners were not
the highest spenders.
In the close three-way races, the third place candidate spent, on average, 66.7 per cent of the limit compared to 73.2 per cent from the second place candidate and 78.2 per cent by the winner (Table 3
). These differences were small enough that again slightly more lower-spending candidates were able to win in the close three-way races (45 of 80). Interestingly, when we broke these numbers down further (not reported here), we noticed that an incumbent was more likely to be reelected in a close three-way race with two high-spending opponents, but was slightly less likely to be reelected with just one high-spending opponent.
Close four-way races (only two per cent over the four general elections studied) did not result in those winners spending more on average than in close two-way races, but they did spend just slightly more than the winner of close three-way races (78.4 per cent vs. 78.2 per cent) (Table 3
). Close four-way races did not result in a high per cent spent by all of the top contenders. The winner and #2 were close (78.4 per cent vs. 72.7 per cent), while #3 (56.5 per cent) and #4 (43.3 per cent) were far behind, although their spending was significantly higher than the average unsuccessful challenger. And in 13 of 24 of the close four-way races, the winner was not
the candidate who spent the most.
As expected, the winners who spent the least to win
were found in "not close races": 77.2 per cent of the limit on average. And in such races, the gap between the winner and the second place candidate was greatest (13.3 percentage points), with the highest spender winning the race much more often (614 of 993). Included here were Conservative incumbents in Alberta, Bloc incumbents in Quebec, and longstanding incumbents of several parties in the Atlantic provinces. By comparison, in close three-way races the spending gap between the winner and the third place
candidate was 11.5 percentage points (Table 3
).What factors determine vote share, and by how much
To separate out the effects of several variables, we used a linear regression equation for all candidates in the four general elections combined (N=6799, R squared=.832, Standard Error= .076) with "Per Cent of the Vote" (the percentage of the valid ballots obtained, a.k.a. "vote share") as the dependent variable (Table 4
). Details of the regression analysis are explained below the table, but here are the major findings.
All other things being equal, for every percentage point increase in election expenses, candidates can expect a 0.28 percentage point increase in their share of votes obtained (significant at .001 level) (Table 4
). This is higher than a previous estimate of 0.21 in the 1980s (D. Keith Heintzman, "Electoral Competition, Campaign Expenditure and Incumbency Advantage," cited in Laschinger & Stevens, Leaders and Lesser Mortals
Being an incumbent is worth an additional 0.19 percentage points in the percentage of votes obtained (significant at .001 level) (Table 4
). This seems small when we consider that the simple re-election rates for incumbents averaged 86 per cent for the four elections. However, the regression result is taking into account five other variables simultaneously, and note that this includes both first-time and long-time incumbents, as well as incumbents in close fights versus those who were easily reelected with large pluralities.
Running for a party which had MPs in the House at dissolution, all other things being equal, adds 0.071 percentage points in the percentage of votes obtained (significant at .001 level). Again, this seems a small advantage compared to the minute election rate for candidates who were not a member of a party with Commons representation.
Where a race was close (the winner getting less than five per cent more votes than the second place candidate), the effect is to add only 0.079 percentage points to the votes obtained by candidates (significant at .001 level). But remember that winners can win close races with smaller vote shares than many challengers obtain in non-close races. In fact, when we computed correlations for different kinds of candidates (not reported here), we obtained insignificant but negative correlations between per cent spent and per cent of the vote received by winners (whether incumbent or newly-elected), indicating that candidates running to win in all but the safest seats typically spend at least 75 per cent of the limit, regardless of the vote share it takes to win in their riding (see Table 2
The year of the election and hence the number of major parties contesting it (we set 2004 and 2006, the post-unite-the-right elections, as zero and 1997 and 2000 as 1), had a negative effect on vote share (significant at .001 level). This makes sense when you consider that fewer parties dividing the electorate translates into higher vote shares for candidates.
We assume that the rest of the variability in vote share is explained by the usual demographic variables, party loyalty, candidate attractiveness, campaign effects and regional voting patterns.Summary and conclusions
From these results we conclude that increased competitiveness in a riding is associated with increased spending for candidates who are serious contenders. Becoming a contender costs money, but once dominant (i.e., an incumbent) in a riding that candidate can scale back on "election expenses" somewhat.
The data do not
support a simplistic prescription that says "spend more to improve your vote share," since if everyone followed that advice everyone's vote shares would decline, and in our first-past-the-post system only one candidate gets to win in each riding. Larger parties who are dominant in an election will spend more everywhere (even in ridings they don't win), partly because they can, and partly to make it harder for their opponents to challenge them financially. Smaller parties win seats by targeting their spending to ridings with good growth prospects.
Candidates who are not trying to win, but are trying to raise their party's vote total, can improve their vote share somewhat through increased spending. However the resulting benefits of an increase in the public subsidy goes to the party rather than the candidate. Further, this will work only where some candidates in a riding raise their spending relative to others. The return on investment for that strategy is a 0.28 per cent of the vote increase for every additional per cent of the expense limit spent.
Using 2006 as the example, spending an additional one per cent of the limit in each of the 308 ridings would have cost a party approximately $250,000, resulting in additional roughly 41,500 votes, worth approximately $315,300 in additional public subsidies over four years (assuming an average of $1.90 over the period following 2006), for a 26.1 per cent return on investment. However, if the next election were held after just two years rather than four (as in fact it was), the same additional 41,500 votes would have been worth only $157,600 in added subsidies, for a negative 36.9 per cent return.
Simple formulae about spending and results will never replace the acumen and experience of political parties and their strategists. However, we hope we have contributed some interesting findings for them to consider in their efforts.Alice Funke is the publisher of the Pundits’ Guide to Canadian Federal Elections (punditsguide.ca). W.T. Stanbury is professor Emeritus, UBC. He conducted research on party and election finance for the Royal Commission on Electoral Reform & Party Finance (Lortie Commission).
: Average Percent Spent, and Correlations Between Percent of Limit Spent and Percent of Total Vote Obtained for Candidates by Party for Four Federal General Elections, 1997 to 2006Table 2
: Average Percent of Election Expenses Limit Spent by Candidates in Relation to Outcome for Four Federal General Elections, 1997 to 2006Table 3
: Analysis of Percent of Elections Expenses Limit Spent by Candidates by Type of Race in Four Federal General Elections Combined, 1997 to 2006Table 4
: Regression Results, All Candidates in Four Federal Elections, 1997 to 2006
Labels: Election Expenses