17
the individual-level analysis demonstrated, these are the respondents who combine educational
attainment, which should make them more likely to be consumers of the news, with political
independence, which should make them more susceptible to the influence of issue ownership.
Other independent variables used to predict presidential approval include high-profile
rally events, administration dummy variables, and objective measures of economic performance.
Consistent with a long line of scholarship in this area, we expect rally events to capture short-
term positive or negative shocks to the series. Moreover, we expect the economic measures,
unemployment and inflation, to be negatively related to approval. In other words, as
unemployment climbs, approval should decline, all things equal.
Following Beck (1992) and Clarke and Stewart (1994), the subsequent equations are
implemented as Error Correction (EC) models. Presidential approval has routinely been
analyzed in its level form with differenced continuous variables to ensure stationarity (Box and
Jenkins, 1976). However, differencing necessarily eliminates any long-run information provided
by levels data. The EC approach adds back such long-run information, thus implementing
statistically the intuitive notion that a potential equilibrium exists between presidential approval
and certain aspects of the economy. In other words, if approval and the economy are in
equilibrium, approval will not remain at high levels while the economy performs consistently
poorly, and vice versa. Thus, according to Beck, implementing the error correction approach
“measures the speed at which approval returns to its equilibrium value (in terms of the
economy)” (69).
Our approach to specifying an EC model of approval is, first, to test for unit roots to
ascertain the order of integration of the continuous variables. Second, we check for cointegration
Convention