** **

Results

Table 1 shows the results for each of four models. Each model considers the

effects of a different indicator of decentralization. As expected, in all the models, the

coefficient for previous incumbent vote percentage is both large and highly significant.

Knowing the strength of electoral support from the previous election explains a large

percentage of the current incumbent support.

(Table

1

about

here)

The variables of greatest theoretical interest are the direct effects of unemployment

and the horizontal and vertical clarity interaction terms with unemployment.

6

In model 1,

the direct effect of unemployment is statistically significant and the coefficient is well

below zero (-0.32). By contrast, the coefficient for unemployment in the least horizontally

clear cases is positive and significant (p<.1). This result indicates that rates of

unemployment have smaller effects on incumbent vote share under conditions of low

horizontal clarity and confirms prior findings.

Similar to the coefficient for the horizontal clarity-unemployment interaction term,

it is expected that the federal interaction term should be positive. While not reaching

statistical significance, the coefficient (0.11) is in the expected direction and of moderate

magnitude. Results using the regional election interaction term (in Model 2) are very

similar to those of the first model in terms of coefficient direction and magnitude with the

exception that the horizontal clarity interaction term loses significance.

Models 3 and 4 perform less well. The effect of unemployment in conditions of

low horizontal clarity continues to be in the expected direction (positive). In contrast to

Models 1 and 2, the effect of unemployment under different criteria of vertical clarity in

models 3 and 4 is actually strengthened.

While the results from the regression models convey information about the role of

clarity in weakening the effects of unemployment, a more complete understanding of the

effects of unemployment on incumbent vote share under different conditions of clarity can

be ascertained by developing linear combinations of the unemployment coefficients from

each model.

7

Following the argument, it is expected that unemployment will have the

greatest effect (both in terms of coefficient magnitude and statistical significance) in the

most horizontally and vertically clear cases. By extension, the effects of unemployment

should be weakest in the least horizontally and vertically clear cases. Finally, relative to

these extremes, the effects of unemployment are expected to be moderate in mixed

conditions of clarity.

(Table

2

about

here)

Table 2 shows the results of linear combinations developed as functions of the

models estimated in Table 1. As seen from the linear combinations of federal institutions

in Model 1, negative effects of unemployment on incumbent vote share are greatest in the

5

The models as specified only allow for the additive effect of clarity. In theory, clarity may operate

multiplicatively as well. However, when modelled multiplicatively, the models break down-mainly due to

collinearity problems. Additionally, tests using individual-level data showed no multiplicative effect of

horizontal and vertical clarity (Anderson, 2004).

6

All analyses were conducted using Stata. To develop linear combinations of coefficients, the ‘lincom’

command was used. This post-estimation command produces coefficients, standard errors and t-statistics.