Every year, about 30 Americans die because they are struck by lightning. About45,000 die as a result of lack of health insurance. Excluding the 9/11atrocities, which killed 2,997 people and injured more than 6,000, fewer than500 people died in the United States of terrorist attacks between 1970 and 2010.Every single month, millions of Americans lose their jobs. Yet we know moreabout people's perception of the risk of being hit by lightning or beingthe victim of a terrorist attack than their perceptions of the risk of losingtheir health insurance or their job.

In fact, the rich literature on risk perceptions largely ignores risks covered bysocial policy programs, such as old age, ill health, accident, and unemployment.Because much of what governments do is the management of social policy risks,and because risk perceptions motivate political behavior, this is no smallomission. Two questions are particularly relevant: How do citizens form riskperceptions? And are risk perceptions accurate? With respect to social policyrisks, neither of these questions has been explored systematically in theexisting literature. This chapter will address them, in reverse order.

The next section explores whether risk perceptions are accurate. Using theSurvey of Risk Perceptions and Insecurity (SERPI), anoriginal panel survey, I find that subjective risk perceptions are fairlyrealistic, at least in relative terms. I then analyze the correlates of riskperceptions and find that one of the best predictors of the subjectiveprobability of experiencing a bad event at time t is the actualoccurrence of that bad event at future time t + 1. Next,I turn to the question of how risk perceptions are formed. Since people seem tohave a good sense of whether they are exposed to higher or lower risk, I analyzethe characteristics of social policy risks in order to explore how people canform reasonably accurate risk perceptions. I find that social policy risks (inthe United States) are correlated in three ways: over time, across domains, andwithin networks. I conclude by showing that respondents draw on the clusteringof shocks over time and within network as sources of information in order toform their own risk perceptions.