17
personal conflicts among core values should lead an individual to more complex forms of reasoning (e.g.,
Tetlock 1986); in turn, this should inhibit the person’s ability to express his/her policy stands in a single
response to an issue question on a public opinion survey. Similarly, Alvarez and Brehm (e.g., 2002),
argue that ambivalent feelings about values have an adverse effect on the degree to which people express
crystallized, consistent feelings about issues, because a number of conflicting considerations come more
readily to mind when people formulate their own responses to public policy controversies. In either case,
value ambivalence would not necessarily determine an individual’s stand on a policy issue. However, it
would have an adverse effect on the strength of the relationship between his/her value choices (or lack
thereof) and subsequent issue attitudes.
The preceding hypotheses can be tested by performing a heteroskedastic regression analysis of
the impact of value preferences and ambivalence on an important political issue: Government spending.
The latter is an important political issue. Disagreements over the government’s ability and/or willingness
to fund programs cut to the heart of the basic distinction between liberal and conservative ideologies and
modern partisan alignments (e.g., Jacoby 2000). Government spending also seems to lie at the core of
predominant trends in macro-level public opinion— the characteristic that has come to be known as the
“public mood” (Jacoby 1994; Stimson 1999). Thus, the political prominence and substantive importance
of government spending make it important to understand public opinion on this issue.
Heteroskedastic regression uses a maximum likelihood approach to estimate simultaneously two
functions (King 1989): First, the expected value of the dependent variable is estimated to be a linear
function of the independent variables— just as in a traditional, OLS, regression. Here, the independent
variables will be composed of relative value preferences along with other, more-or-less standard
determinants of issue attitudes. Second, the disturbance variance is expressed as a function of another set
of explanatory variables. This second set of variables will contain a measure of value ambivalence and
several indicators of political sophistication. The former should make political attitudes more difficult to
predict and, thereby, increase the disturbance variance in the regression equation. The latter should
Convention