At first glance, Fair’s data seems toprove that voters are capable of evaluatingeconomic circumstances at the aggregatelevel. He admits that voters do not know the”exact numbers,” but he believes they cansense the direction of the economy throughtheir daily interactions. Fair also suggeststhat voters are strongly influenced by themedia, and the “good news quarters”variable supposes that they can rememberthe number of quarters in which the mediahas praised the economy’s growth.
9 He restshis theory on the statistical accuracy of hismodel, which has come within one or twopercentage points of the incumbent’s voteshare in most elections, and he insists thatthese results are due more to marginal voters’ retrospective macroeconomicevaluations than to campaigns effects.11Fair’s model also makes dangerousassumptions. One keystone of his argumentis that marginal voters rely on retrospectivejudgments to guide their vote choice, butrecent data suggests that this behavior is notas prevalent as he claims. During the 1992campaign, Marion Just and her colleaguesextensively interviewed voters of allbackgrounds, and found that theirretrospective considerations of candidatesdeclined with each interview. Respondentswere uncomfortable with the “up-or-downvote on the incumbent” that Fair’s modelassumes, and they looked to the campaignsfor information about challengers. Manyrespondents also indicated that candidates’,even incumbents’, past actions were notsufficient to predict future actions, and soprospective considerations were also used.
17This data seems to have been reaffirmed in2000, when Al Gore should have held asolid edge in retrospective considerationsdue to the economic prosperity of theClinton era. Gore’s campaign, however, cutalmost all ties with the President because ofhis moral failings, and voters did notassociate Gore with the past success ofClinton’s administration.18 Instead, votersseemed to rely heavily on the current eventsof the campaign, as seen through mediacoverage and television advertising.19While Fair shows impressivestatistical correlations, and makes clear thateconomic conditions play some role inelectoral outcomes, his model has technicalflaws.
For one thing, he includes no marginof error. Jay Greene of the AmericanProspect estimates Fair’s error to be around8% for the 1992 election, which, if true,means that a victory by either candidatecould have satisfied his model.12 Also, thefact that his data is historically accurate doesnot prove its predictive capabilities. Greenpoints out that constructing the model afterthe fact allows Fair to fit the variables topast conditions, and until the model is testedextensively in the future, it cannot beconsidered a reliable forecast.13 In fact, themodel has failed to accurately choose thewinner of the two most recent elections, andthis trend may be likely to continue becauseof two glaring omissions. First, Fair’smodel only predicts the two-party voteshare, despite the fact that two of the pastthree elections have been heavily influencedby third party candidates.
Not surprisingly,his model was the least accurate in 1992,when a third party candidate got asignificant portion of the vote share.14 Also,his model discounts the role of the ElectoralCollege, a point which he openly admits anddismisses as immaterial.15 Rebecca Morton,however, has shown that the ElectoralCollege encourages targeted campaigning,which affects voter preferences in key statesand can give candidates an electoral victorywithout winning Fair’s prized two-partyvote.16