position of Clinton and Dole on government services (69.5%), job assurance (64.5%), health

insurance (65.5%), environmental regulation (51.9%), affirmative action (60.8%), and abortion

(58.4%). Respondents in the lowest knowledge quartile averaged 4.7 correct answers out of 16

possible, while respondents in the highest knowledge quartile averaged 14.2 correct answers.

*Simulation Method *

The method I use for simulating “fully informed” preferences extends the simulation

approach developed by Bartels (1996) and by Delli Carpini and Keeter (1996). I use logistic

regression to avoid the restrictive assumption that political information must have a linear

relationship with preferences. Delli Carpini and Keeter use ordinary least squares to estimate

parameters in their simulation; Bartels uses a probit model to estimate parameters, but later

transforms those parameters in a way that assumes linearity of information effects. I also estimate

“fully informed” preferences for people who give “don’t know” and “no opinion” responses in

the survey data, following the assumption that as information levels rise, the proportion of people

who give opinions or turn out to vote should also rise. The logit model I use for simulating “fully

informed” opinions in dichotomous questions is structured as follows:

prob *Y*

*i*

=

1

(

)

=

α + β

1

*I*

*i*

+ ∑

β

*k*

*D*

*ik*

+ ∑

δ

*k*

*I*

*i*

∗

*D*

*ik*

(

)

+

*e*

*i*

,

where *Yi *is respondent *i’*s policy preference (e.g., 1 for “favor”, 0 for “oppose”), *Ii *is respondent

*i’*s score on a scale of political information, *Dik* is respondent *i’*s score on the *k*th demographic

characteristic, *Ii ***Dik* is the product of respondent *i’*s information score multiplied by respondent

*i’*s score on the *k*th demographic characteristic, and *ei *is the error term for the *i*th observation. In

this equation,

β

1 is the coefficient for the information variable,

β

*k *is the coefficient for the *k*th

demographic characteristic, and

δ

*k *is the coefficient for the *k*th interaction term. An ordered logit

model is used when estimating “fully informed” opinions in trichotomous questions.

After completing the four-step simulation procedure described in Althaus 1998 (and

elaborated in Althaus 2003), the mean of the *Yi *probabilities from hypothetically “fully