Amplificació del mitogenoma sencer (Long-PCR)

In sum, the normative implications of the extant evidence on political

information processing are more ambiguous than is often assumed. This of course is not to say that we have reason to believe that citizens meet any normative standard, but rather that it is surprisingly difficult to draw any normative conclusion, the standard itself is a moving target. In following section, I specify a restrictive Bayesian model that will be used as a benchmark against which to judge the results of the experiments presented below.

Suppose Gina, our Bayesian voter, receives an argument with some evidence that the ACA will slow down cost growth. The model states:

𝑃(𝐻|𝐸) = 𝑃(𝐸|𝐻) ∙ 𝑃(𝐻)

𝑃(𝐸) (3.1)

where H is a hypothesis about the Affordable Care Act (e.g., that the net benefit of the law is greater than zero); E is evidence relevant to the hypothesis; 𝑃(𝐻) is Gina’s prior probability that H is true, before observing E; 𝑃(𝐻|𝐸) is her posterior probability, revised in light of E; and 𝑃(𝐸|𝐻)

𝑃(𝐸) is the likelihood function, representing her interpretation

of how the evidence bears on the hypothesis at hand. This term can be rewritten as:

𝑃(𝐸|𝐻)

where 𝑃(𝐸|𝐻) is her belief about the likelihood of observing E when H was true, and

𝑃(𝐸|~𝐻) is the likelihood of observing E when the contrary was true.38 The ratio between 𝑃(𝐸|𝐻) and 𝑃(𝐸|~𝐻)—the likelihood ratio—dictates in what direction, and how much, Gina should update her belief about H after observing E. The likelihood ratio of completely uninformative evidence will be 1, in which case, the posterior will be the same as the prior. And the larger the difference the between the likelihoods, the more will Gina’s posterior tend toward the evidence.

It is worth emphasizing that Bayes’ Theorem does not tell Gina how to construct her likelihood function; it is up to her own judgments, which would reflect not only some

objective properties of E, but also her subjective evaluations of E. To be sure, if the treatment of new evidence is left fully subjective (i.e., there is no constraints on the acceptable treatment of new evidence) a Bayesian model becomes tautological because it can account for any belief change (Taber et al. 2009). But then again, one should not conflate subjectivity per se with bias. Although “[h]ow we determine the boundary line between rational skepticism and irrational bias is a critical normative question, but one that empirical research may not be able to address,” the fact remains that one somehow should draw the line “to resolve the controversy over the rationality of motivated

reasoning” (Taber and Lodge 2006, 768). In order to make a Bayesian model falsifiable, it is necessary to put constraints on likelihood functions based on the properties of E.

38 _{My discussion here focuses on a dichotomous parameter—the net effect of the ACA is} either positive (H) or negative (~H)—, this Bayesian framework can be extended to a continuous parameter—the value of the net effect—using probability density functions.

In a recent study, Guess and Coppock (2015) address this dilemma by imposing the restriction that the difference between the likelihoods, 𝑃(𝐸|𝐻) − 𝑃(𝐸|~𝐻), is “correctly” signed, which does not allow people to have an extreme likelihood function that would lead one to believe, for instance, that evidence of strong deterrent effects is good news for the anti-death penalty position. I build on their research by introducing an additional restriction that individuals’ likelihood functions are constrained not just by the direction but also the strength of evidence. That is, Gina grasps the differences in the diagnostic values of an event that is far more likely to occur when H is true than is not (strong evidence), and an event that is slightly more likely to occur when H is true than is not (weak evidence). More formally, it means Gina’s subjective likelihood ratios are lined up such that 𝑃(𝐸𝑆𝑃|𝐻)

𝑃(𝐸𝑆𝑃|~𝐻) >

𝑃(𝐸𝑊𝑃|𝐻)

𝑃(𝐸𝑊𝑃|~𝐻), where 𝐸𝑆𝑃 and 𝐸𝑊𝑃 are strong and weak pro

evidence for H. Likewise, 𝑃(𝐸𝑆𝑐|~𝐻)

𝑃(𝐸𝑆𝑐|𝐻) >

𝑃(𝐸𝑊𝑐|~𝐻)

𝑃(𝐸𝑊𝑐|𝐻) where 𝐸𝑆𝐶 and 𝐸𝑊𝐶 are strong and weak

con evidence against H.

Imposing this restriction requires defining the objective elements of evidence that would compel Gina toward the position it supports—i.e., what makes a piece of strong evidence strong. I define these elements as the commonly held standards of statistical, external, and construct validity, upon which social scientists readily draw to evaluate the quality of research findings (e.g., Shadish et al. 2002). Accordingly, in the experiments presented below, “strong” evidence describes, for instance, more sizeable effects of the ACA on health care costs, based on more externally valid, and more relevant, data.

No one thinks that the average citizen is as skilled in making such assessments of evidence as trained researchers (Niesbett and Ross 1980). But I would argue that people

routinely calculate the probative values of various kinds of information they encounter, guided by common sense and logic, which allow them to notice the difference between strong versus weak evidence. The theoretical expectation of the (restrictive) Bayesian model specified here is that the following equation holds, regardless of prior attitudes:

𝑃(𝐻|𝐸𝑆𝐶) < 𝑃(𝐻|𝐸𝑊𝐶) < 𝑃(𝐻) < 𝑃(𝐻|𝐸𝑊𝑃) < 𝑃(𝐻|𝐸𝑆𝑃).

If, alternatively, people accept whatever arguments they encounter, without accounting for the uncertainty of the evidence, one will find 𝑃(𝐻|𝐸𝑆𝐶) = 𝑃(𝐻|𝐸𝑊𝐶) < 𝑃(𝐻) < 𝑃(𝐻|𝐸_𝑊𝑃) = 𝑃(𝐻|𝐸_𝑆𝑃). If this holds true, political persuasion is possible, but even to the point where politicians can get away with claiming whatever they want citizens to believe. If people categorically reject counterevidence to defend their priors, one will find P(𝐻|𝐸_𝑆𝐶) = 𝑃(𝐻|𝐸_𝑊𝐶) = 𝑃(𝐻) for proponents (or Democrats), and

P(𝐻) = 𝑃(𝐻|𝐸_𝑊𝑃) = 𝑃(𝐻|𝐸𝑆𝑃) for opponents (or Republicans). If people backlash

against counterevidence, which is the most extreme form of motivated reasoning, one will find P(𝐻|𝐸𝑆𝐶) > 𝑃(𝐻) and P(𝐻|𝐸𝑤𝐶) > 𝑃(𝐻) for supporters (or Democrats), and P(𝐻|𝐸_𝑆𝑃) < 𝑃(𝐻) and P(𝐻|𝐸_𝑊𝑃) < 𝑃(𝐻) for opponents (or Republicans). If this holds true, political persuasion would not be generally possible.

In what follows, I present three experiments that test these competing hypotheses. To the extent that motivated reasoning overwhelms evidence-based opinion revision, “hot button” issues should provide the hardest context under which to detect the effect of evidence (Taber and Lodge 2006, 757). With that in mind, the experiments draw on two

of arguably the most contentious issues in American politics today—the Affordable Care Act (Experiments 1 and 2), and the economic performance of the two major parties (Experiment 3). The experiments examine how participants’ posterior beliefs and attitudes vary in response to arguments coupled with varying degrees of evidence certainty.