other variables are set equal to their median values, produce a point estimate of the impact of
changing a given independent variable on the dependent variable. On the other hand, our coefficients
are estimated with uncertainty. We therefore estimated our impact effects with 95% confidence
intervals, to reflect the fact that different samples would have produced different coefficient
estimates.29 The results are given in Table 10.
The first panel simply presents the simulated probabilities that an observation will take on
any of the five values of the dependent variable, protect. As can be seen, there is nearly a 20%
probability that protect takes on the value 3, a 38% probability that it takes on the value 4, and a31%
probability that it takes on the value 5. The panels immediately below indicate the impact of changing
our nationalism variables on all five probabilities; in the following discussion we focus on the
probability that protect takes on the value five. Increasing patriotism from its 40th to its 60th percentile
increases the probability that protect take on the value 5by 2 percentage points, with a95%
confidence interval of 1.79 to 2.28 percentage points. Changing chauvinism from its 40th to 60th
percentile increases the probability that protect takes on its highest value by 5.8 percentage points. If
these two variables are changed from their 20th to their 80th percentiles, the impact on protect is
enormous: the probability of the most protectionist response occurring increases by 10.1 percentage
points in the case of patriotism, an increase of 32% over the benchmark simulated probability. In the
case of chauvinism, the impact is to increase the probability that protect equals five by 19.8
percentage points, an increase of 64% (with a 95% confidence interval of 18.4 to 21.2 percentage
points). These are huge effects by any standards.
No other variable has an impact as big as chauvinism, although some come close. The
interaction term between high-skill and GDP per capita in the equation indicates that the effect of
being high-skilled on protectionist attitudes depends on how rich the country is. The next three panels
therefore indicate the impact ofbeing high-skilled in economies with GDPs per capita of $5000,
29 All results were produced using Clarify, as described in Tomz, Wittenberg and King (1999) and
King, Tomz and Wittenberg (2000).
22