the lack of correlation many scholars find between electoral success and negative
advertising may be caused by an underlying problem with the campaign or candidate
rather than anything inherent in negative advertisements themselves. Simply controlling
for early polls will not solve the problem because campaigns may be acting upon private
information about the state of the campaign. These types of unobserved and unmodeled
processes plague observational research.
Laboratory experiments avoid this unobserved heterogeneity by randomly
assigning subjects to view different types of carefully selected advertisements. The
random assignment assures that, on average, the people exposed to a negative political
advertisement are identical to the people viewing positive political advertisements (e.g.,
they reside in similar political contexts, share similar tastes with regards to politics, and
possess similar behavioral patterns). Consequently, selection bias is not a problem in
properly implemented experimental research designs because exposure to a negative
campaign message is randomly determined rather than strategic.
Laboratory experiments also provide researchers a great deal more control over
the measurement of the independent and dependent variables. The researcher does not
need to infer what type of ads the subject viewed, because she showed the ads to the
subject. The researcher also has control over the content of the advertisement. Rather
than coding hundreds of messages and pigeonholing them into a particular category, the
researcher can carefully select political advertisements that are archetypes of the desired
concepts. In short, randomization and control over the experimental inputs affords the
3
Measurement error in the independent variables of interest is also a problem for observational studies.
Not all political messages are created equally, yet content analyses of campaign messages implicitly
assume that citizens weight all campaign messages equally by placing a diverse set of messages in discrete
categories, thereby creating an illusion of uniformity.
8
Convention