2. Economic Voting Hypothesis
The economic voting hypothesis is based on the assumption that voters
hold their government responsible for the current state of the economy.
Consequently, they support the governing party if they are satisfied with
the economy, but back opposition parties if economic conditions are
deteriorating. Voters' support is usually measured with two sources
of information: electoral outcomes and opinion polls. The economic voting
hypothesis has been investigated with these two sources, each leading to
the development of a specific model: the “vote function” for
electoral outcomes and the “popularity function” for opinion
polls. From a theoretical standpoint, there are no major differences
between the two functions: both take into account the same economic
determinants and make similar predictions. However, the effect of economic
conditions is not necessarily of the same magnitude for both functions.
Economic conditions are believed to exert lesser influence on electoral
outcomes because politicians can bring non-economic issues to the
attention of voters during electoral campaigns (Nannestad and Paldam, 1994).
Several macro-economic indicators have been used to empirically test
the economic voting hypothesis. Over the years, two variables have emerged
as major economic determinants of electoral and popularity outcomes: the
rate of unemployment and inflation. The impact of economic growth,
measured, for instance, by gross domestic product, has also been
investigated, but findings vary significantly from one study to the
next—an indication that, at best, its influence might hold for a
specific country or a given period of time. Other macro-economic
indicators have shown even greater instability, so generalizations about
their real impact are difficult to establish (Lewis-Beck and Paldam, 2000; Lewis-Beck and Stegmaier, 2000; Nannestad and Paldam, 1994; Paldam,
1997). More recently, fiscal policy outcomes were added to the vote
and popularity functions. The empirical evidence seems to suggest that
taxes and unbalanced budgets influence voters' support toward
incumbents (Kone and Winters, 1993; Lowry et al., 1998; Niemi et al.,
1995).
The economic voting hypothesis has also been studied in Canada using
the vote and popularity functions. Overall, findings seem to indicate that
economic conditions influence voters' support toward federal
political parties. However, the real pattern of economic voting is
unclear, as inconsistencies are often found from one study to the next.
For instance, the effect of the rate of unemployment on voters'
support is confirmed by some (Cutler, 2002;
Happy, 1992; Nadeau and
Blais, 1993; Pétry and Harmatz,
1995), but refuted by others (Carmichael,
1990; Happy, 1986, 1989; Johnston, 1999; Monroe and Erickson, 1986). Findings on the effect of
inflation are equally mixed. Several explanations have been presented in
an attempt to reconcile these contradictory results. For instance, it has
been demonstrated that substituting different indicators for the same
macro-economic variable can lead to different conclusions (Happy, 1986, 1989, 1992; Nadeau and Blais,
1993). The impact of economic determinants was also found to
fluctuate over time and between some provinces (Carmichael, 1990; Godbout and
Bélanger, 2002; Guérin and Nadeau,
1998; Johnston, 1999). At the provincial
level, in Quebec, a number of studies have investigated the relationship
between the economy and political support. Once again, the empirical
evidence seems to support the economic voting hypothesis, though some
discrepancies can be found (Albert, 1980; Crête and Simard, 1984; Guérin and Nadeau, 1995).
Although, at first sight, the explanation presented by the economic
voting hypothesis seems straightforward, it has, nonetheless, provoked
several questions about the behaviour of voters over the years. For
example, the assumption that voters are retrospective, that is, that they
use the current or past state of the national economy to evaluate
government performance, has been questioned. If the economic theory of
rational expectation is correct, voters should be expected to behave in a
prospective manner, that is, only taking into account future expected
outcomes. The use of aggregate economic measures has also been called into
question: Are voters egocentric individuals, more concerned about their
personal financial well being, or sociotropic, responding to the state of
the national economy (Kinder and Kiewiet, 1979)?
Overall, empirical studies reported evidence of both retrospective and
prospective behaviour, but found that voters are more sociotropic than
egocentric (Fiorina, 1997; Lewis-Beck, 1988; Lewis-Beck and
Stegmaier, 2000). Canada seems to be no exception (Alvarez et al., 2000; Archer and
Johnson, 1988; Clarke and Kornberg, 1992;
Godbout and Bélanger, 2002; Nadeau et al., 2000).
In addition, some have suggested that economic voting is also shaped
by how voters perceive the incumbent's responsibility over the
economy. Powell and Whitten (1993) proposed the
“clarity of responsibility” hypothesis, which states that not
all governing parties are held equally accountable for economic
performance. For instance, coalition or minority governments can be
thought of as having less control over the economy, since they must work
with other political parties if they wish to remain in power. Concerns
were also raised about the magnitude of voters' reaction to economic
conditions. For example, Bloom and Price (1975)
proposed the “asymmetry” hypothesis, which alleges that voters
punish governments more severely for deteriorating conditions than reward
them for improving ones. Stevenson (2002)
extended this assumption by suggesting that larger economic variations
have more impact than smaller ones. Government ideology may also make a
difference (Lowry et al., 1998; Powell and Whitten, 1993). It is commonly believed
that left-wing parties pay more attention to the rate of unemployment and
redistribution of income, while right-wing parties are more concerned
about inflation and fiscal restraint.
So far only economic factors have been presented as potential
determinants in the vote and popularity functions. However, it is widely
acknowledged that political variables should also be taken into account,
in conjunction with economic indicators (Nannestad and
Paldam, 1994). Until now, one of the most consistent empirical
findings concerns the “cost of ruling” explanation (Nannestad and Paldam, 2002). Support for the incumbent
seems to decay over time, usually after a honeymoon period (Powell and Whitten, 1993). In Canada, Pétry and
Harmatz (1995) found that the federal government
loses, on average, one half of a point in popularity lead per quarter
during its term in office, while Nadeau (1990)
estimated that it gains at least six percentage points in the two quarters
following a general election. The presence of another level of government
might also have an impact. Several studies on American states have
reported that the president's ideology influenced gubernatorial or
legislative electoral outcomes. Voters seem to attribute more
responsibilities to the president over regional economic performances than
to state legislators (Niemi et al., 1995;
Peltzman, 1987, 1992).
Furthermore, the commitment of voters to a national party seems to
generate a partisan predisposition at the state level (Chubb, 1988; Leyden and Borrelli,
1995; Lowry et al., 1998).
Overall, there is no lack of testable propositions that can be derived
from the economic voting hypothesis and applied to the Canadian provincial
case. We will take on the task of assessing to what extent they can
explain the behaviour of voters on the provincial scene. Because economic
voting has been found to have a greater influence on the popularity
function than on the vote function, our analysis will focus on the popular
support of the incumbent government as expressed by opinion polls.
3. Popularity Function
The general form of the popularity function may be written as:
where POPLEAD is the dependent variable, Σ
βj POLj and Σ
βl ECOl are two
distinct sets of explanatory variables, the α and β values are the
coefficients to be estimated and υ is the residuals term. The data
used to estimate the popularity function in this study combined
cross-section and time-series annual observations (i = 1,
2,…n units, or provinces, and t = 1,
2,…T periods). The pooled design is used to increase the
number of observations, which in turn enhances the quality of statistical
estimations. However, the use of a pooled structure can generate residuals
that are not independently and identically distributed. Preliminary
statistical tests have revealed the presence of heteroscedasticity, serial
and contemporaneous correlation of residuals.1
Lagrange and Breusch–Pagan–Lagrange multiplier
tests were used to detect heteroscedasticity, an F-test was used
for fixed effects, and a pooled Durbin–Watson d-test was
used for autocorrelation (Sayrs, 1989; Greene, 2000).
Consequently, the estimation
procedure is ordinary least square (OLS) with unit fixed effects and
correction for first-order autocorrelation of residuals (AR1).
2 One method that can be used to solve the problem
of residual autocorrelation is to include a lagged dependent variable on
the right side of the equation. However, this procedure is problematic in
the present case because incumbents may not have been in power during the
previous year.
In addition, standard errors were computed with
the panel-corrected standard errors (PCSE) method proposed by Beck and
Katz (
1995).
3.1. Dependent Variable
POPLEAD is the popularity of the incumbent provincial government. It
measures voting intentions of decided voters as a percentage, as reported
by polling firms. Respondents were asked the following question: “If
a provincial election was held today, which political party would you vote
for?” Since the number of major political parties varies from one
province to the next, absolute percentages of voting intentions do not
seem to provide the most accurate measure: a party obtaining the support
of 45 per cent of the electorate does not enjoy the same level of
popularity in Ontario (where the Conservative party, the Liberal party and
the New Democratic party (NDP) each usually receive more than 20 per cent
of the votes in general elections), in Quebec (which traditionally has two
major players: the Liberals and the Parti Québécois), or in
Alberta (where the political scene has been somewhat dominated by the
Conservative party). Therefore, we measured popularity of the incumbent as
its popularity lead over the main opposition party (the
second-highest-ranking party in voting intentions). The appendix provides
further information on the conceptualization of this variable (as well as
all other explanatory variables used in this study). Nonetheless, some
additional comments are in order concerning the sources used to measure
voting intentions on the provincial political landscape.
Only since the mid-1990s have polling firms shown a genuine interest
for the estimation of voting intentions on the provincial scene on a
regular basis (for instance, Léger Marketing in Quebec, Compas in
Ontario and, more recently, Corporate Research Associates in the Atlantic
provinces).3
Surprisingly, Gallup has been
measuring federal voting intentions since the 1940s.
Consequently, the data drawn from provincial surveys covers a very narrow
period of time—too short to be used in this study. On the other
hand, several polls conducted on a national scale have asked specific
questions about provincial voting intentions. Those national polls used a
sample size large enough to make reliable statistical inferences about
voting intentions at the national level, but their accuracy declines
sharply when answers are disaggregated by province. For example, a
national survey conducted among a sample of 1500 respondents is estimated
to be accurate within 2.5 percentage points (19 times out of 20), but the
margin of error increases to 4.6 points for a subsample of 450 (which is
representative of the number of Quebec and Ontario residents polled in
national surveys), 6.9 points for a subsample of 200 (typically used for
Alberta and British Columbia) and 9.8 for a subsample of 100 (Manitoba and
Saskatchewan). The problem of a small sample size can be overcome,
however, because more than one polling firm conducted national surveys on
a regular basis in Canada. Therefore, by aggregating answers provided to
the same question during the same period of time by different sources, we
can effect a meaningful increase in the overall sample size. We were able
to collect data from four independent polling organizations to obtain good
quality estimations for several provinces over the last two decades. The
data were first computed on a quarterly basis, to ensure that enough data
were available year-round, and then averaged on an annual basis.
4 Only 15 of the 432 quarters covered in this
study have missing observations.
Regrettably, sample sizes
remained too small for the Maritime provinces (frequently less than 200
per year), so they had to be excluded from our analysis. The observations
used in this study cover the following period: 1982–2001 for Quebec
and Ontario, 1986–2001 for Manitoba and Saskatchewan, and
1984–2001 for Alberta and British Colombia.
5 The annual measure of the popularity lead is accurate within
2.2 percentage points or less for Quebec and Ontario; 2.9 percentage
points for Alberta and British Colombia; and 4.7 percentage points for
Manitoba and Saskatchewan.
In total, 108 annual observations are
used.
3.2. Independent Variables
Following previous research, our provincial popularity function
includes political and economic variables. The political variables pertain
to the depreciation and honeymoon effects, and the presence of the federal
government. The depreciation effect, which postulates that government
popularity decays constantly over time, is represented by YEARS—a
variable that measures the incumbent's number of years in office
since the last general election. The honeymoon effect, which assumes that
newly elected parties enjoy a burst of popularity at the beginning of
their legislative mandate, is capture by CHANGE—a dummy variable
that takes the value 1 if a new party has come into power.6
The arrival of a new premier is also included in
the variable CHANGE to represent a honeymoon effect in provinces where a
political party has been in power for a long period of time. This is
notably the case for Alberta, where Conservatives have ruled without
interruption since 1971, but under the leadership of different premiers
(P. Lougheed, D. Getty and R. Klein). Only premiers who have been
subsequently elected are considered in our measure.
Several studies have revealed that electoral outcomes in American
states are affected by partisan attitudes at the federal level. However,
it is not certain that the presence of the federal government can have an
effect on the provincial political scene. The Canadian political party
system is highly decentralized, and provincial and federal parties,
identified by the same label, generally work independently from one
another (Cross, 2004). In contrast, however,
some studies have shown that provincial political features influence
voters' support toward the federal government. For instance, the
province of origin of party leaders was found to have a significant effect
on the popularity of federal parties (Nadeau and Blais,
1993; Pétry and Harmatz, 1995). In
addition, the empirical evidence has indicated that provincial
Conservative party supporters were more likely to vote for the federal
Progressive Conservative party than others (Clarke and
Kornberg, 1992). A similar pattern was uncovered for Liberal and
Conservative partisans in Quebec (Nevitte,
1984). Consequently, voters' attitudes toward federal and
provincial parties might be linked to some degree. These considerations
led us to incorporate an additional political variable in our
model—the popularity lead of the federal government in each province
(FED_ POPLEAD)—as measured by opinion polls. It is therefore
expected that voters' support toward the federal government is
correlated to provincial support. However, the influence of the federal
popularity lead might not point in the same direction for all provincial
incumbents. For instance, it could be anticipated that an unpopular
Conservative federal government might be beneficial to non-Conservative
provincial governments, but harmful to Conservatives provincial
incumbents. For this reason, an interaction effect is added to the model,
based on IDEOL, a dummy variable that takes the value 1 when both
provincial and federal parties share a similar ideological label (if both
are Liberal or Conservatives) and 0 otherwise.
Economic variables relate to provincial economic performances. Two
groups of economic indicators are used. First, three
macro-economic indicators that have proved to be significant in
previous studies are incorporated in the model: the rate of unemployment
(UR), inflation (INFLATION) and personal income (INCOME), as a measure of
economic growth. Although personal income does not appear to have a
meaningful effect on economic voting in general, it was shown to be
significant in Canada (Crête and Simard,
1984; Johnston, 1999). Following standard
procedures, inflation and economic growth are measured as annual
percentage variation and unemployment as absolute level of percentage rate
(Powell and Whitten, 1993).7
We also tested if other measures of inflation and personal
income would provide better estimates (such as the level of CPI, the level
and annual variation of the GDP implicit price index, total and per capita
personal income, in nominal and in real terms), but none turned out to be
statistically significant.
Secondly, two variables linked to
fiscal policy outcomes are added to the provincial popularity
function: the annual percentage variation of provincial income tax (TAX),
which was reported to have an effect in Canada (
Albert,
1980;
Happy, 1992), and provincial
surplus or deficit in percentage of total provincial public expenditures
(budgetary balance or BB). Since our measures of economic indicators are
based on current provincial events, it is implicitly assumed that voters
are both retrospective and sociotropic.
8 Studies have generally shown that retrospective voters are
highly myopic, using economic events that have occurred only in close
proximity to the current situation, generally within a quarter, to
evaluate the incumbent's economic performance (Nannestad and Paldam, 1994; Lewis-Beck and Stegmaier, 2000). Since our analysis
relies on annual data, it does not seem relevant to use lagged explanatory
economic variables to represent voters' myopia. Nonetheless, we have
tested the popularity function with lagged economic variables (t
− 1) and found that none turned out to be statistically
significant.
We estimated the provincial popularity function with the two following
equations:
Equation (2) is the popularity function when the set of economic
variables includes macro-economic indicators only. This equation will
serve as our baseline model, so that comparisons can be established with
previous studies and among competing explanations. Equation (3) adds
fiscal policy outcomes. In both equations, the impact of the federal
government popularity lead is introduced with an interaction effect that
assumes a different intercept (IDEOL) and a different slope (IDEOL ×
FED_POPLEAD) between provincial incumbents sharing a similar ideological
orientation with the federal government and the others. It is unclear why
a provincial ruling party that shares a similar ideological label with the
federal government should be more or less popular than other provincial
governments per se. However, the potential effect of IDEOL alone on
provincial government popularity deserves to be investigated more closely,
so we will test if it has an influence on provincial government popularity
lead.9
We are grateful to an anonymous
reviewer for pointing out this possibility.
The theoretical expected values and/or signs are: for
depreciation, b1 < 0; for the honeymoon effect,
b2 > 0; for the presence of the federal government,
b3 is undetermined, b4 < 0, and b4 +
b5 > 0; for the macro-economic indicators, b6 and
b7 < 0 and b8 > 0; and for fiscal policy
outcomes, b9 < 0 and b10 > 0, since a positive
sign for BB represents a surplus and a negative sign a deficit (a surplus
increase (decrease) or a deficit decrease (increase)
will lead to a popularity lead increase (decrease)).
4. Findings
The estimates for equations (2) and (3) are reported in Table 1. Starting with our baseline model (eq. (2), in
Table 1, column 1), we see that the rate of
unemployment is the only macro-economic indicator that exhibits an
estimated coefficient with the correct sign and reaches statistical
significance (p < 0.10). An increase of 1 per cent in the rate
of unemployment reduces the popularity lead of the provincial incumbent by
1.80 percentage-points. The estimate for inflation has the wrong sign and
is statistically insignificant, as its large standard error and p
value indicate, thus suggesting that its effect is equal to zero. The
division of powers set by the Canadian Constitution might explain this
outcome. The federal government has exclusive power to legislate over
monetary issues. Consequently, Canadian voters should not expect
provincial governments to implement price control policies. Our findings
support this view. The estimate for personal income displays the expected
sign, but its value and standard error show that it has virtually no
impact on provincial government popularity. This finding is perhaps an
indication that Canadian voters attribute equal responsibility to federal
and provincial governments for personal income fluctuations. Therefore
they do not believe that provincial governments should be rewarded or
blamed for changing conditions.10
The
question of multicolinearity among the macro-economic indicators must also
be addressed since high levels of correlation can generate biased
estimates. Additional tests were performed, where each economic variable
was regressed on all other explanatory variables. The resulting
R2 values never exceeded 0.45 (and were usually below
0.20), so multicolinearity problems can be ruled out.
Economic Conditions and Government Popularity
Turning to the influence of political factors, we found evidence of
both a depreciation and a honeymoon effect. The estimates for YEARS and
CHANGE exhibit the expected sign and are statistically significant. Newly
elected parties see their popularity rising at the beginning of their term
(leading the main opposition party by 9.6 percentage points during their
first year in power), but lose support in subsequent years (their
popularity lead declines by 1.9 percentage points per year). Our findings
also reveal a relationship between federal and provincial
governments' popularity lead. As predicted, the sign of the estimated
coefficient is negative when both share a similar ideological orientation
(−0.18) and positive otherwise (−0.18 + 0.62 = 0.44 with a
standard error of 0.14 and p < 0.01). What is somewhat
unexpected is the difference of magnitude between the two estimates:
provincial governments identified by the same label as the federal
government gain or lose much more (2.5 times more) than other provincial
governing parties. This implies, for instance, that an unpopular
(popular) federal Conservative government harms (helps)
provincial Conservative governments much more than it helps
(harms) non-Conservative provincial governments. Our results also
show that the estimated coefficient for the dummy intercept IDEOL alone is
negative but not statistically significant. Therefore we cannot conclude
that sharing a similar ideology with the federal government influences
provincial government popularity by itself. Because no theoretical
argument seems to support the inclusion of the dummy intercept IDEOL in
our model, and since the addition of an irrelevant independent variable
might produce inefficient OLS estimators (although the estimators are
unbiased), equation (2) was re-estimated with the assumption of a common
intercept. Results are presented in Table 1,
column 2. We can see that all new estimated coefficients and standard
errors are very similar to those reported in Table
1, column 1, thus providing support for our previous
conclusions.
Column 3 in Table 1 shows the estimated
provincial popularity function when fiscal policy outcomes are added to
the model. As far as political variables are concerned, the new estimates
do not point to different conclusions. The estimated coefficients for the
depreciation and the honeymoon effects, and for the popularity lead of the
federal government, have values very similar to the ones previously
reported, while the dummy intercept IDEOL remains statistically
insignificant. Once again, removing the dummy intercept from the
popularity function does not alter our findings (estimates shown in Table 1, column 4 are nearly identical to those
reported in column 3).
Looking at macro-economic indicator estimates, a different picture
emerges to some extent. Although the estimated coefficient for the rate of
unemployment continues to exhibit a negative sign (−0.65 in column 3
and −0.68 in column 4), it is no longer statistically significant by
conventional standards (standard errors are now higher than estimates).
Inflation and personal income also do not reach statistical significance.
Concerning fiscal policy outcomes, we found the estimate for income tax to
have the proper sign but it is not significant, thus showing a lack of
influence on provincial government popularity lead. This situation might
be linked to the information available to voters: with the exception of
one province (Quebec), provincial income taxes are collected by the
federal government, thus possibly making it difficult for voters to
distinguish exactly the responsibility of each level of government on this
issue. Finally, budgetary balance is found to play an important role in
the popularity function: its estimate (0.47) has the correct sign and is
significant at the 0.01 level. Also, when budgetary balance is accounted
for, the overall explanatory power of the model rises from about 0.50 to
0.64,11
The Buse R2 is a
goodness-of-fit measure similar to the conventional
R2, but it cannot be guaranteed to be a non-decreasing
function of the number of explanatory variables (Whistler et al., 2004). The conventional
R2 cannot be computed when pooled observations are
corrected for AR(1) errors.
an increase that seems difficult to
attribute solely to the presence of two additional variables.
Consequently, voters seem to be concerned about unbalanced budgets and
hold provincial governments accountable for changing conditions. However,
the impact of the rate of unemployment has declined sharply, up to the
point that it now seems irrelevant in explaining provincial government
popularity.
Overall, our empirical analysis indicates that several economic
factors do not have a significant impact on provincial government
popularity. Although some theoretical arguments can be provided to justify
this absence of a relationship for inflation, personal income and income
tax, the empirical evidence is more puzzling for unemployment since, in
other studies, it has generally been found to have an effect. Furthermore,
it does not seem unreasonable to think that provincial governments have
the means to intervene as far as job-related matters are concerned and
that voters are aware of this situation. Therefore the issues deserve to
be looked at more closely.
As stated above, several supplementary theses have been formulated
over the years to improve the explanatory power of the popularity
function. The clarity of responsibility hypothesis is one of them. As was
pointed out: “minority ruling parties can always claim that their
best efforts were blocked by other parties and that responsibility of
policy failures must be shared by them” (Powell
and Whitten, 1993). Consequently, voters might hold minority
governments only marginally responsible for the state of the economy if
they believe that incumbents are under pressure to make compromises with
opposition parties. If voters act in accordance with this assumption,
economic performance should therefore have a smaller effect on the
popularity of minority governments when compared to majority governments.
To test this hypothesis, an interaction effect is added to the popularity
function based on MINORITY, a dummy variable that takes the value 1 if the
provincial government does not hold a majority of seats in the provincial
legislature. This distinction between minority and majority governments is
expected to yield different intercepts and slopes for the two groups. The
following equations were used in the estimations:
For macro-economic indicators only:
For macro-economic indicators and fiscal policy outcomes:
The theoretical expected values and/or signs remain the same for
depreciation, the honeymoon effect, the presence of the federal
government, macro-economic indicators and fiscal policy outcomes. Minority
governments are expected to have a smaller popularity lead, since they are
usually elected with a small share of the vote. Therefore, the prediction
is b5 < 0. As for the variables linked to the clarity of
responsibility hypothesis, the expected signs and values are:
b7 > 0 and b6 + b7 ≤ 0 for
unemployment, b9 > 0 and b8 + b9 ≤
0 for inflation, b12 > 0 and b11 + b12
≤ 0 for income taxes, and b14 < 0 and b13 +
b14 ≥ 0 for budgetary balance. Personal income is expected
to be equally beneficial to all governments (Powell and
Whitten, 1993), so no interactive effects were added for this
variable.12
However, we also tested the
model with the inclusion of an interactive MINORITY variable for personal
income but found no significant relationship.
Results for equations (4) and (5) are reported in Table 2. Columns 1 and 3 show estimates when a
different intercept is used for minority and majority governments.
However, our findings do not present a clear indication that minority
governments are less popular than other ruling parties during their entire
term in office (everything else being equal). On the one hand, although
statistically significant, the estimate presented in column 1 is difficult
to justify. Its value suggests that the popularity lead of minority
governments is inferior by about 100 percentage points (−101.35) to
the popularity lead of other incumbents—an unrealistic situation by
all means. On the other hand, when fiscal policy outcomes are added to the
popularity function, the estimate for the dummy MINORITY does not reach
statistical significance. Removing the dummy variable MINORITY from the
model leads to estimates displayed in columns 2 and 4 of Table 2. Overall, the empirical evidence does not
support the “clarity of responsibility” hypothesis for both
equations: estimated coefficients for the rate of unemployment, inflation,
income tax and budgetary balance all display signs or values contrary to
expectation, while personal income remains statistically insignificant. In
contrast, all estimates related to political variables show values similar
to those reported in Table 1, thus suggesting
that our previous findings are robust (a result that will prove to be
consistent in all subsequent regression analyses presented below).
The “Clarity of Responsibility” Hypothesis
The asymmetry hypothesis is another explanation that might offer
greater insight into the problem. It postulates that voters' support
for the incumbent is influenced by the political issues of the day or by
party affiliation in times of prosperity, but by economic performance
during economic recession (Bloom and Price,
1975). If this hypothesis holds, voters should punish governments
for deteriorating conditions, but should not reward them for improving
conditions, or at least not as much as they punish them. To test this
explanation, the following dummy variables, indicating the presence of
improving economic conditions, are added to the popularity function:
DECREASE, coded 1 if the rate of unemployment has decreased since the
previous year; LOW, coded 1 if the rate of inflation is under 4 per cent
(the average rate of inflation for the entire period under investigation);
POSITIVE, coded 1 if the annual variation of personal income is positive;
NEGATIVE, coded 1 if the annual variation of income tax is negative; and
SURPLUS if the budgetary balance is a surplus. The popularity functions to
be estimated are now as follows:
For macro-economic indicators only:
For macro-economic indicators and fiscal policy outcomes:
It is expected that improving conditions will have little or no effect
on government popularity, contrary to deteriorating circumstances.
Therefore, the theoretical predicted signs and values are: b7
> 0 and b5 + b7 ≤ 0 for unemployment;
b10 > 0, and b8 + b10 ≤ 0 for
inflation; b13 < 0, and b11 + b13 ≥
0 for personal income; b16 > 0 and b14 +
b16 ≤ 0 for income taxes; and b19 < 0 and
b17 + b19 ≥ 0 for budgetary balance. On the other
hand, it is unclear why and how the interaction effect should lead to
different intercepts when economic conditions are improving or
deteriorating, so the predicted signs for b6, b9,
b12, b15 and b18 are undetermined. The
theoretical expected signs and values for variables introduced previously
remain the same.
Estimates for equations (6) and (7) are reported in Table 3, columns 1 and 3 respectively. Although one
estimated dummy intercept variable reached statistical significance (LOW),
we found no evidence that, jointly, all intercept dummies are
significantly distinct.13
The
F-test for H0: b6 = b9 =
b12 vs. H1: b6 ≠ b9 ≠
b12 is F(3,76) = 0.97 (Eq. (6)); and for
H0: b6 = b9 = b12 =
b15 = b18 vs. H1: b6 ≠
b9 ≠ b12 ≠ b15 ≠ b18
is F(5,68) = 0.72 (Eq. (7)).
Therefore both equations
were re-estimated using a common intercept. Results are shown in
Table 3, columns 2 and 4. With the exception of one
variable, the asymmetry hypothesis is not supported by our dataset.
14 We also tested if larger variations have more
impact than smaller ones by using squared economic indicators (as
suggested by Stevenson, 2000), but found no
empirical evidence that they provide a better measure to test the clarity
of responsibility hypothesis.
Estimates show no significant
differences between improving and deteriorating economic conditions for
the rate of unemployment, personal income and income tax (b
7,
b
13, b
16 are not statistically significant), while
inflation is positively correlated with popularity lead, which is contrary
to expectations. On the other hand, the asymmetry hypothesis seems to be
validated for budgetary balance outcomes (eq. (7)). Deficits are found to
be damaging for incumbents (the estimate is significantly positive in both
columns 3 and 4), but surpluses have no meaningful impact (0.48 −
0.45 = 0.03 with a standard error of 0.33 in column 3, and 0.61 −
0.59 = 0.03 with a standard error of 0.30 in column 4).
15 The estimate for SURPLUS × BB is not statistically
significant (p = 0.155, which is above the 0.10 level in column 4 of Table 3), but its standard error remains inferior to
its estimated coefficient and its large value might be explained by a
small sample size.
Consequently, voters seem to react differently
in the presence of provincial budget deficits and surpluses.
Lastly, the partisanship hypothesis suggests that economic voting
shows up differently for governing parties of different ideological
orientation. For instance, voters may expect left-wing governments to deal
with the unemployment situation and right-wing governments with inflation.
The responsibility of centrist governing parties is less clear: whether
voters hold them more responsible than other parties remains an open
question. As for fiscal policy outcomes, taxes are expected to increase if
a left-wing party is in power and decrease if it is a right-wing party
(once again the assumption about centrist governments is unclear). The
impact of public deficit is more contentious. It has long been claimed
that left-wing parties are less fiscally responsible than right-wing
parties because of their propensity to increase spending. However, most of
the latest empirical findings do not support this view: left-wing and
right-wing governments are equally fiscally disciplined (Franzese, 2002). In fact, some findings indicated that
deficits occur mostly when centrist parties are in power (Alesina and Perotti, 1995; Tellier,
2004).
To test if partisanship matters, provincial governing parties were
grouped into three distinct categories: the Right, consisting of
provincial Conservatives and Social Creditists, the Centre, for provincial
Liberals, and the Left, for provincial NDP and the Parti
Québécois. Incorporating two new dummy variables, RIGHT,
coded 1 if the incumbent belongs to the Right, and CENTRE, taking the
value 1 if Liberals are in power (the Left is the reference group), we now
can estimate the following popularity functions:
For macro-economic indicators only:
For macro-economic indicators and fiscal policy outcomes:
If a partisan effect exists, we should see distinct partial slopes for
macro-economic indicators and fiscal policy outcomes for the three groups.
Furthermore, it is predicted that the rate of unemployment is determinant
for left-wing incumbents but not for right-wing governing parties
(b7 < 0, b7 + b9 = 0, and
b8 undetermined), while inflation and income tax have the
opposite effect (b10 and b14 = 0, b10 +
b12, and b14 + b16 < 0, and
b11 and b15 are undetermined). On the other hand,
predictions about public deficits for each group are more difficult to
establish (but b17, b18 and b19 ≠ 0 if
a partisan effect exists). Once again, personal income is expected to be
equally beneficial to all governments, so no interaction effect is added
for this variable.16
However, we also tested
the model with the inclusion of an interactive effect between personal
income and party ideology but found no significant relationship.
Equations (8) and (9) also assume that ideology leads to different
intercepts, although the theoretical expected signs are unclear (thus, the
predicted signs for b
5 and b
6 are undetermined).
Once again, the theoretical predicted signs and values for variables
introduced previously remain the same.
Estimates for equations (8) and (9) are reported in Table 4, columns 1 and 3, respectively. For both
equations, none of the dummy intercept variables reached statistical
significance (CENTRE and RIGHT), indicating that ideology alone does not
influence provincial government popularity lead.17
The F-test for H0: b5 =
b6 vs. H1: b5 ≠ b6 is
F(2,70) = −0.13 (Eq. (8)) and F(2,56) = 0.01 (Eq.
(9)).
Therefore, equations (8) and (9) were re-estimated using a
common intercept. Estimates are displayed in
Table
4, columns 2 and 4. Once again, we found no evidence that
inflation, personal income and income tax have a significant effect on the
dependent variable. None of the estimates associated with each of these
three variables is close to reaching statistical significance by
conventional standards. The rate of unemployment and public deficits, on
the other hand, exhibit strong partisan effects. Estimates reported in
columns 3 and 4 suggest that the rate of unemployment influences the
popularity of left-wing governments, while public deficits have an effect
on centrist and right-wing incumbents' popularity. Because some
differences are visible between estimates displayed in the two columns,
the popularity function was re-estimated using only unemployment and
budgetary balance as explanatory economic variables.
18 We also tested for the presence of a distinct ideology
intercept in this last regression by adding the dummy variables CENTRE and
RIGHT, but the estimates for both variables remained statistically
insignificant while overall results were not affected.
The
results are presented in the last column of
Table
4. These results confirm previous findings: unemployment matters,
but only for left-wing governments. An increase in the unemployment rate
by 1 per cent decreases the popularity lead of left-wing governments by
2.59 percentage points. The effect of unemployment is of less magnitude
and not statistically significant for centrist (−2.59 + 3.59 = 1.00,
with a standard error of 2.17) and right-wing parties (−2.59 + 1.47
= −1.12, with a standard error of 0.92). As for the budgetary
balance, public deficits influence the popularity of centrist (0.15 + 2.35
= 2.50, with a standard error of 1.41) and right-wing governments (0.15 +
0.60 = 0.75, with a standard error of 0.25), but not that of left-wing
parties. In addition, surpluses are found to have no significant effect on
popularity, confirming previous findings (0.15 − 0.59 = 0.44, with a
standard error of 0.41).
The Partisanship Hypothesis
Our results indicate that public deficits are more harmful for
Liberals than for Conservatives and Social Creditists: an increase in the
ratio deficit/total expenditure by one percentage point decreases the
popularity lead of Liberals by 2.50 percentage points, compared to 0.75
for right-wing governments. Therefore, as far as public deficits are
concerned, voters punish Liberals more severely than right-wing parties.
How can this outcome be accounted for? The fact that public deficits were
higher under a Liberal incumbency might provide an explanation. As Table 5 indicates, public deficits amounted to 7.6 per
cent of total expenditures, on average, when Liberals were in power,
compared to 5.8 per cent for left- and right-wing parties. However, data
in Table 5 also suggest that the presence of a
high percentage might not be sufficient to provide a satisfactory
explanation. As stated previously, we found that income tax variations do
not influence the popularity of any ruling parties. Still, income taxes
increased much more when right-wing parties were in power, compared to
other parties (on average, 3.9 per cent per year compared to 1.1 and 1.6
per cent for the left-wing and centrist parties, respectively). Why then
do Conservative and Social Credit supporters, who are believed to be
strongly opposed to tax increases, sanction such an outcome? It might be
that dissatisfied right-wing voters are less likely to support other
political parties. Being located on the right of the political spectrum,
it might be more difficult for them to find another political party that
meets their expectations. Being in the centre, Liberal supporters have
more choice. Consequently, the positioning of the Liberal parties in the
middle of the scale might well be the reason why public deficits have such
an impact on their popularity.
Average Annual Economic Performances by Governing
Parties
