The Candidate Evaluation Experiment: Testing Proximity and Directional Predictions on Policy Issues and Ideology

Download data and study materials from OSF

Principal investigator:

Ryan Claassen

Kent State University



Sample size: 503

Field period: 9/29/2006-10/5/2006


Endogeneity problems and observational equivalence in observational studies render many tests of the proximity and directional models inconclusive, but a small and growing body of experimental evidence is providing new leverage. This experiment reexamines the role of ideology structuring candidate evaluations in the general population and tests the models in two new policy areas: opinion about military spending and opinion about abortion. Following self-placement on the attitude scales, an interactive computer survey solicits subjects’ evaluations of hypothetical candidates described by positions on each scale. To ensure the models make different predictions for each individual (a problem in observational studies), subjects are randomly assigned to evaluate candidates on the same side of the policy space, but 1-5 units more distant. For each issue, the experiment tests whether subjects use directional or proximity logic comparing their positions to those of hypothetical candidates.


1) Because directional evaluation is represented thusly,
Directional Evaluationij = (Pi * Cj)
where Pi represents citizen i’s position and Cj represents candidate j’s position, the directional model predicts citizens will evaluate candidates on the same side of the policy space positively (e.g. 2*2=4 and -2*-2=4) and, hence, the directional model predicts warmer evaluations as the distance between subjects and hypothetical candidates (on the same side of the center) increases.
2) Because proximity evaluation is represented thusly,
-Proximity Evaluationij = |Pi – Cj|
the proximity model predicts cooler evaluations as the distance between subjects and hypothetical candidates increases.

Experimental Manipulations

To create the desired evaluation situations, the hypothetical candidate evaluation prompts were designed to generate a candidate who was on the same side of each scale neutral point (e.g. 0), but more extreme, with how much more extreme determined at random. If the subject’s location was a positive integer, the candidate’s location was determined by drawing an integer from a uniform distribution of integers from 1 to 5 and adding the integer to the subject’s location. If the subject’s location was a negative integer, the candidate’s location was determined by drawing an integer from a uniform distribution of integers from -1 to -5 and adding that integer to the subject’s location. To make the process a bit more concrete, a subject with an abortion position of -1, for example, was asked to evaluate a candidate with an abortion opinion ranging from -2 to -6.
In addition, subjects were asked to evaluate a candidate on the opposite side (of themselves, relative to the first candidate), but one unit closer. And, for the ideology scale, subjects were asked to evaluate a third candidate who was described as being even more extreme than the first candidate by 1-5 units, with “how much” more extreme again determined at random.


Candidate evaluations on each of the scales. Abortion: q2 and q3. Military spending: q5 and q6. Ideology: q8 – q10.

Summary of Results

The experiment was designed to distinguish directional from proximity candidate evaluation behavior. My results indicate that ideology-based evaluations turn on proximity motivations, as do opinions about the proper level of military spending, though the latter association is quite weak. However, abortion-based evaluations appear to turn on directional motivations. One interesting explanation for directional behavior is that of policy balancing. In the case of abortion, this suggested to me that I would observe more directional behavior from those on the “oppose abortion” side of the scale. To test this possibility, I stratified the sample into those who were on the “oppose” side and those who were on the “support” side. Consistent with balancing theory, the results for the “oppose” side were even more robustly supportive of directional logic. But among those who support abortion I do not find significant evidence of directional behavior.


Ideological considerations and military spending (perhaps budgetary policy, more generally) appear to follow proximity logic, and so elections that turn on ideology and other proximity oriented themes should adhere to the basic moderation-rewarding precepts of the spatial theory of elections. However, if observations of recent political polarization require an issue that both inspires directional behavior while being highly salient; then the abortion results reported here may well supply an explanation for recent evidence of centrifugal tendencies in American politics. For those concerned about political polarization; however, balancing offers a silver lining. Even a directional issue, such as abortion, is only directional for those attempting to balance policies they find disagreeable.

Additional Information

Use subjects’ self placements and the locations of hypothetical candidates to operationalize directional and proximity comparisons.


Claassen, R. L. 2009. "Direction Versus Proximity." American Politics Research 37:227.