we label the Governing Effect variable, represents the In Government variable from
Equation 1, and equals 1 if the party was in government at the time of the current
election, and zero otherwise. We include this term in our specification so that we can
estimate the degree to which the valence scores of parties differ depending on whether
they are in government or not. The parameter estimate b
2
for the Governing Effect
variable is essential for our evaluation of the Governing Effect Hypothesis (H2). If the
parameter estimate b
2
is negative and statistically significant, this will provide evidence
that governing parties systematically receive more negative valence scores than parties
outside of government over the course of an inter-electoral period, thus supporting H2. If
the estimate for b
2
is not statistically significant (or if the estimate is positive and
statistically significant), we will not reject the null hypothesis, that governing parties do
not receive more negative valence scores than non-governing parties.
We also include the party’s vote share in the previous election (the term [VSj(t-
1)] in equation 2), as a control variable since it is plausible that larger parties receive
lower negative scores than smaller parties, since they have more members and therefore
more “opportunities” for scandalous behaviour, and/or acts of perceived incompetence.
In addition, given that party heterogeneity is likely to increase with size, intra-party
disagreements for example, would seem especially likely to occur with larger parties. If
this is the case, we would expect the parameter estimate b
3
to be negative and statistically
significant. Finally, we also include a control variable, Time Effect (the term T(t) in
Equation 2), as we also expect that the longer the time period between elections, the more
negative parties’ valence scores will be since the additional time will translate into
additional “opportunities” for negative valence events such as scandals, disagreements,
and acts of incompetence to take place. Assuming this is the case, the parameter estimate
b
4
will be negative and statistically significant.
Evaluating the Governing Effect Hypothesis
As with the model specified to test H1, our data is treated as Time-Series Cross-
Sectional data. Our analysis covered 217 valence scores for 44 parties from the nine
countries included in our study. Given our earlier concerns regarding performing
regression on pooled data, we again test for party-specific heterogeneity, which could
bias our estimates. Initially, these tests suggest that party-specific effects are present in
19
Convention