Homework answers / question archive / Activity1 :In Male incarceration, the marriage market, and female outcomes, what is an outcome of incarceration that may have unintended consequences? How does marriage impact other aspects of social and economic interaction? Can you provide other qualitative costs to society based on the topic of this article? For Willingness to pay for crime control, explain willingness to pay

Activity1 :In Male incarceration, the marriage market, and female outcomes, what is an outcome of incarceration that may have unintended consequences? How does marriage impact other aspects of social and economic interaction? Can you provide other qualitative costs to society based on the topic of this article? For Willingness to pay for crime control, explain willingness to pay

Sociology

Share With

Activity1 :In Male incarceration, the marriage market, and female outcomes, what is an outcome of incarceration that may have unintended consequences? How does marriage impact other aspects of social and economic interaction? Can you provide other qualitative costs to society based on the topic of this article?

For Willingness to pay for crime control, explain willingness to pay. Are there limitations to using this method?

Limit your responses to 100-150 words (total).

Activity 2: In The criminal and labor market impacts of incarceration Actions, what is the argument being made? How can the evaluation/focus of this article assist with determining sentence structures? In other words, is there an economic cost of incarceration that is not presently being considered?

The Causes of crime is an old article. Do you agree with the view that the economy is the cause of crime? Why or why not?

Please limit your summaries/comments to 150 words (total)

MALE INCARCERATION, THE MARRIAGE MARKET, AND FEMALE OUTCOMES Author(s): Kerwin Kofi Charles and Ming Ching Luoh Source: The Review of Economics and Statistics, Vol. 92, No. 3 (August 2010), pp. 614-627 Published by: The MIT Press Stable URL: http://www.jstor.org/stable/27867563 Accessed: 08-05-2018 10:53 UTC JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org. Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://about.jstor.org/terms The MIT Press is collaborating with JSTOR to digitize, preserve and extend access to The Review of Economics and Statistics This content downloaded from 155.33.16.124 on Tue, 08 May 2018 10:53:42 UTC All use subject to http://about.jstor.org/terms MALE INCARCERATION, THE MARRIAGE MARKET, AND FEMALE OUTCOMES Kerwin Kofi Charles and Ming Ching Luoh Abstract?This paper studies how rising male incarceration has affected women through its effect on the marriage market. Variation in marriage market shocks arising from incarceration is isolated using two facts: the tendency of people to marry within marriage markets defined by the changes over the past three decades, when the number of Americans behind bars more than tripled (Maguire & Pas tore, 2000). High levels of incarceration lower the number interaction of race, location, and age and the fact that increases in of men freely interacting in society. Rising incarceration incarceration have been very different across these three characteristics. Using a variety of estimation strategies, including difference and fixed rates may thus lower women's marriage probabilities simply effects models and TSLS models in which we use policy parameters to because there are fewer potential husbands to go around. instrument for within-marriage market changes in incarceration, we find evidence that is, on the whole, consistent with the implications of the standard marriage-market model. In particular, higher male imprisonment appears to have lowered the likelihood that women marry, modestly reduced the quality of their spouses when they do marry, and shifted the gains from marriage away from women and toward men. The evidence suggests that women in affected markets have increased their schooling and labor supply in response to these changes. The seminal work of Becker (1973, 1974, 1981) on marriage markets suggests more subtle effects as well. In particular, a reduction in the number of men should shift the gains from marriage away from wives and toward husbands, and from women to men more generally. Women should thus be more likely to marry men whose marital advances they would have previously rejected, when they do marry at I. Introduction has fallen consistently over theentering past half-century and THE fraction of persons traditional marriages today is at historic lows (Stevenson & Wolfers, 2007). Several explanations for these patterns have been proposed in the literature, including changes in reproductive technol ogy such as the introduction of the birth control pill (Goldin & Katz, 2002), improving relative labor market opportuni ties for women (Blau, Kahn, & Waldfogel, 2000), legal changes such as the 1973 Supreme Court ruling in Gomez v. Perez mandating that men be financially responsible for their illegitimate children (409 U.S. 535), and changes in household technology, which, by rendering household chores less onerous, may have reduced the cost of remaining single (Greenwood, Seshadri, & Yorukoglu, 2005). While all of these factors may be important, we study another possibility: the role of rising male incarceration. Whether measured in totals or as a fraction of the popu lation, more Americans are incarcerated than in all but a few other countries in the world.1 The remarkably high levels of incarceration observed today are mainly the results of Received for publication February 23, 2006. Revision accepted for publication September 10, 2008. * Charles: University of Chicago and NBER; Luoh: National Taiwan University. all. These effects tend to lower women's economic well being. Women confronting high male incarceration rates might thus be expected to take actions, such as working more or investing in additional schooling, that augment their earning power and economic independence.2 Have recent increases in incarceration produced these effects? Although rising incarceration levels have coincided with a nationwide reduction in marriage, it is not clear that the two phenomena are causally linked. Higher incarcera tion might partly reflect unobserved changes in male behav ior that could have independently affected marriage out comes. Alternatively, changes in social conventions about marriage might have coincidentally occurred at the same time as rising incarceration levels, rendering any connection between incarceration and marriage outcomes spurious. Without some way of controlling for various confounding factors of this sort, it is difficult to say anything about the possible causal effects of incarceration* Our main innovation for dealing with these difficulties is to exploit the fact that the overwhelming majority of mar riages occur between men and women in distinct "markets," defined by the interaction of race, age, and geographic region. Because the increase in incarceration has varied tremendously over these three categories, rising incarcera tion has lowered the relative presence of men by very This paper has benefited from comments and conversations with Robert different amounts in different marriage markets. We are therefore able to use variation across different markets at a Ronald Ehrenberg, Jeffery Grogger, Jonathan Guryan, Patrick Kline, point in time and within markets over time to identify the Axelrod, Rebecca Blank, John Bound, Charles Brown, John Dinardo, Robert Lalonde, Justin McCrary, Derek Neal, Gary Solon, Matthew Shapiro, and Melvin Stephens Jr. We also thank seminar participants at the University of Michigan, the University of Chicago, Cornell University, McGill University; Syracuse University, the University of Texas at Austin, and the Labor Workshop at IZA. 1 The U.S. Department of Justice (2003) reports that more than 2.1 million Americans were held in jails or prisons in 2003. This reflects a total rate of incarceration of about 715 Americans per 100,000: a rate of more than 1,300 per 100,000 for men, and one of about 113 per 100,000 women. These rates are higher than comparable rates in all OECD countries (see OECD, 2001). effect of interest. Standard panel data models account for the contaminat ing effect of latent factors that are fixed over time but cannot account for changes in unobserved factors within marriage 2 This paper is interested in the effect of male incarceration on women's outcomes. Using the standard marriage market model to understand the impact of higher incarceration on unincarcerated men does not lead to the same sharp empirical predictions, for reasons discussed later in the paper. The Review of Economics and Statistics, August 2010, 92(3): 614-627 ? 2010 The President and Fellows of Harvard College and the Massachusetts Institute of Technology This content downloaded from 155.33.16.124 on Tue, 08 May 2018 10:53:42 UTC All use subject to http://about.jstor.org/terms MALE INCARCERATION AND FEMALE MARRIAGE OUTCOMES 615 markets over time. Moreover, their application can be ex pected to magnify measurement problems associated with the fact that the incarceration rates we observe are noisy indicators of the true relative supply shocks experienced by particular marriage markets.3 Using nationally representa tive data from the National Judicial Reporting Program on source of the disappearance of high-earning men that is the focus of that literature. However, to the degree that men's presence in the market is an important aspect of what makes them "marriageable," the paper's connection to that other literature is obvious. Although it focuses on a distinct question and studies a broader range of outcomes, the paper the actual sentences felons convicted in state courts receive, is also related to work by Gould and Paserman (2003) and Loughran (2002), who both study the effect of male wage we show that the state- and race-specific adjudication of drug cases and their relative prevalence in state caseloads inequality, and Blau et al. (2000), who study the effect of explain much of the within-marriage market changes in men's and women's average labor force outcomes on mar incarceration.4 We perform a series of two-stage least riage rates. The remainder of the paper is organized as follows. squares regressions, in which we use these policy variables as instruments for within-marriage market changes in incar Section II summarizes changes in incarceration over the ceration, finding results that are quite similar to, if a bit past few decades, with particular attention paid to how these patterns have differed for different types of men. We then larger than, the basic panel model estimates. On the whole, our results strongly support the idea that discuss marriage markets and briefly review the predictions rising male incarceration causally affected women's out from the theoretical literature on marriage that guide the comes in precisely the manner predicted by the standard marriage market model. Our confidence in these results is bolstered by the similarity of the results under alternative estimation strategies, including the addition of a variety of empirical work. Section III presents the basic empirical framework and describes the empirical methods we use. Section IV presents the results, extensions, and robustness tests. Section V concludes. controls and the use of an alternative definition of marriage markets, and by the range of outcomes for which we find effects consistent with the causal interpretation. The paper contributes to several different literatures. One II. Basic Facts about Incarceration and Marriage Markets and Predicted Theoretical Effects connection is to the literature studying the relationship A. Imprisonment over the Past Thirty Years between the number of men relative to women in a market and marriage market outcomes.5 We estimate effects similar to these papers, but unlike previous work, we study how marriage markets are affected by a policy, incarceration, that Most of the analysis in the paper is conducted on data from the various census IPUMS from 1970 to 2000. The 1970 and 1980 censuses identify inmates in jails or prisons.6 However, in 1990 and 2000, the census indicates only is of ongoing, controversial interest, and we also study a broader range of outcomes. Other authors have been inter ested in how the characteristics of men in a marriage market whether a respondent is institutionalized. We treat the young men characterized as institutionalized as being incarcerated. Several things justify this decision. One is that given the set affect women's mean marriage rates. One important strand of institutions used by the census to define the institution of this literature originates with Wilson's (1987) conjecture alized population, mental institutions are the only other kind that low marriage rates among black women might be of institutions in which young men could logically be because of a small supply of "marriageable" men?young located.7 Work by Grob (2000) shows that the number of men with high, stable earnings. Wood (1995) finds only weak evidence in support of the notion, but, more important, persons in mental institutions has plummeted in the past few decades, meaning that in later years, "institutionalized" we illustrate that incarceration is concentrated among men effectively means "incarcerated." Reassuringly, the patterns whose labor markets earnings would likely have been low of incarceration from the definitions we use are consistent and unstable. The phenomenon we study thus cannot be the with the aggregate information on incarceration from the Bureau of Justice Statistics (BJS).8 3 This measurement problem arises both because the "markets" we define only imperfectly measure the pools within which men and women date and marry and because it is not possible to say with certainly what 6 Jails in the United States are institutions that generally house individ uals with incarceration terms a year or less. Prisons house persons with spatial market an incarcerated man would have belonged to were he not in jail. longer terms of imprisonment. We will not distinguish between these terms in the paper. 4 This is consistent with the commonly held notion that the "war on drugs" accounts for much of the run-up in incarceration. We examine but 7 In 1990 and 2000, institutionalized persons are in jails and prisons, ultimately do not use variation associated with announced sentencing mental institutions, institutions for the elderly handicapped and poor, guidelines within states, as the data from actual sentences received and nursing and convalescent homes, homes for neglected, or dependent children; other institutions for children; deaf/blind schools; schools for time served within states do not line up at all well with these announced guidelines. "feeble-minded," sanataria, poorhouses and almshouses; poor farm houses 5 Examples include Guttentag and Secord (1983), Cox (1940), Freiden (1974), Chiappori, Fortin, and Lacroix (2001), South and Trent (1988), South and Lloyd (1992), Brien (1997), Fosset and Kiecolt (1993), and Angrist (2002). and workhouses; homes for unmarried mothers, widows, and single women; and detention homes. 8 We do not use these data in our analysis because it is not possible with these data to do the disaggregations so central to our analysis. This content downloaded from 155.33.16.124 on Tue, 08 May 2018 10:53:42 UTC All use subject to http://about.jstor.org/terms 616 THE REVIEW OF ECONOMICS AND STATISTICS Figure 1.?Fraction of Men Aged 20-35 of Different Races and Born in Different States Who Are Incarcerated in Census Year: Mean, 10th, and 90th Percentile across All States of Birth Whites 0.2 0.15 0.1 0.05 ' "6 ^ * 0 1970 1990 1980 2000 Year -Mean - - O - - 10th -90th Hispanics 0.2 0.15 0.1 0.05 - -A - 1970 1980 2000 1990 Year -Mean 10th 90th Blacks 0.2 - 0.15 0.1 . - -o 0.05 1970 1990 1980 2000 Year -Mean The paper studies men aged 20 to 35. In the analysis to follow, we split these men into younger (20 to 27) and older (28 to 35) groups, but in this initial section, we present average incarceration figures for the all men ages 20 to 35. Census respondents report not only their age but also their race and state of birth. We focus on three race categories: whites, blacks, and nonblack, nonwhite Hispanics. For each state of birth, we compute the proportion of men of a given 10th - -90th The middle line in each graph shows the mean, across all states of birth, of the incarceration rates for men on the type indicated in the graph heading. The top line in each figure shows the 90th percentile (the fifth highest state) of the distribution of incarceration rates across states. The bottom line shows the 10th percentile (the fifth lowest state) of the same distribution. We exclude from these graphs race and state cells for which the number of observations in the race who are incarcerated in a given census year.9 Figure 1 graphically summarizes these numbers.10 census was too small to compute mean estimates. So, for 9 Later we discuss the choice of state of birth rather than state of residence. be incarcerated. The correlation between the male and female incarcera tion rates across all markets is 0.36. In all years, incarceration rates for men vastly exceed those for women; in 2000, for example, the male rate of 3.58% is more than ten times larger than the female rate of 0.35%. 10 The numbers on which the graph is based are presented in table A.l in the appendix. Throughout the paper, the incarceration rate is the rate among men in the marriage market, including any married men who might This content downloaded from 155.33.16.124 on Tue, 08 May 2018 10:53:42 UTC All use subject to http://about.jstor.org/terms MALE INCARCERATION AND FEMALE MARRIAGE OUTCOMES 617 the sum of the output of an unmarried man and unmarried example, the graphs do not include values for young His panics born in North Dakota, or young blacks born in New woman.14 Equilibrium in the marriage market satisfies two Hampshire, as these cells are vanishingly small. In the conditions. First, the marital sorting of men and women in analysis that follows, we weight all regressions by the cell equilibrium is such that total societal output is maximized. size for which the incarceration rate is measured. Although Second, the law of one price holds in equilibrium?meaning the basic patterns documented above accurately reflect that persons of the same type or quality consume the same changes in incarceration over time, there is likely to be error output in equilibrium.15 If male and female qualities are in the measured rate for any particular type of men (and complements in the production of household output, as is especially racial minorities from some states).11 usually assumed, these conditions imply positive assortative These figures, all drawn with the same scale, reveal mating in equilibrium.16 several interesting facts. First, for all races, most of the Becker's original model summarizes the predicted theo increase in incarceration between 1970 and 2000 occurred retical effect of a change in the relative number of men after 1980, and especially after 1990. The trends in the using the example of a hypothetical large-scale immigra mean, and for the 90th and 10th percentiles, indicate that tion. Angrist (2002) focuses on the same predictions in his this post-1980 increase occurred across the country rather study of actual nineteenth-century immigration to the than in a select set of states, although cross-state differences United States. In the context of incarceration, the marriage for men of a given race tended to widen over time. For market model yields several predictions. Since higher in example, the difference in incarceration rates among blacks carceration lowers the number of men relative to women, its in the high- versus low-incarceration states was 6 percent first and most obvious marriage-related effect should be to age points in 1990 and grew to 10 percentage points in lower the incidence of marriage among women overall. Second, because the men withdrawn from the marriage 2000. Finally, and most dramatical, the graphs reveal how markedly the incarceration experience of young men has differed across races, both over time and in any particular year. Most notably, the mean incarceration rate of young black men across the various states reached a staggering 18% in 2000, while the comparable rate for Hispanics was market would have tended disproportionately to marry lower quality women, higher incarceration should raise the likeli hood of nonmarriage more for them than it does for their higher-quality counterparts. Third, the male scarcity arising from incarceration should raise the odds that high-quality women "marry down" when they do marry, or lower the numbers for whites.12 Finally, although we do not show it in probability odds that lower-quality women "marry-up," this figure, we find that the distribution of education is very conditional on them marrying at all. different from that observed among men who are not in Increased male scarcity should have other implications prison, suggesting that incarceration draws disproportion for women. Women who do not marry when men are scarce 10%. These numbers completely dwarf the comparable ately from the lower portion of the schooling distribution for all races.13 B. Theoretical Overview: Incarceration and Standard Marriage Market Model What are the predicted effects of the incarceration "shocks" for women? Becker's (1973) seminal and widely known model of the marriage market suggests some possi ble answers. In Becker's model, there are gains from mar riage, meaning that married households produce more than 11 We compute but do not present incarceration rates by census divi sion?collections of roughly five or six states apiece. Reassuringly, the basic patterns of incarceration, by race, are quite consistent with the state numbers. 12 These numbers reflect the raw means of the incarceration rate of particular kinds of men across the fifty states and thus do not reflect the fraction of those men in the national population who are incarcerated. This latter is the weighted mean of the state incarceration rates and shows the same basic trends over time, although the levels are lower. For example, around 12.5% of young black men in the country were incarcerated in 2000. 13 For example, among black incarcerated men, 48% have less than a high school education, and only 15.5% have a year or more of college education. The comparable numbers for unincarcerated black men over the same time period are 19.8% and 37%, respectively. Similar differences exist for white and Hispanic men. are hurt because they do not receive the production benefits associated with marriage. Women who do marry when men are scarce are hurt as well; the law of one price implies that their gains within marriages should have fallen.17 By low ering female economic well-being, greater male scarcity should thus raise women's incentives to increase their eco nomic independence by means such as investing in addi tional schooling or working for pay. As far as we are aware, these secondary predictions of the marriage market model have not previously been tested. 14 Also see Becker (1974, 1981) for discussions and extensions of this model. 15 Strictly speaking, this implication of the standard model requires that there be no rents. Since the various gains from marriage likely produce rents, it might thus be more accurate to say that each spouse does in a marriage at least as well as he or she could fare outside it. 16 Equilibrium would exhibit negative assortative mating in the unlikely event that male and female quality were substitutes rather than comple ments in household production. Various authors have shown that in the United States, there is positive assortative mating along the dimension of schooling?the index of quality we study in this paper. 17 This is represented in the theory as women having to make greater transfers to husbands. This notion of transfers likely subsumes both very serious concerns, like the distribution of work effort in the family or sexual fidelity, and less serious matters, like which spouse washes dishes or whose entertainment choices will be honored. This content downloaded from 155.33.16.124 on Tue, 08 May 2018 10:53:42 UTC All use subject to http://about.jstor.org/terms 618 THE REVIEW OF ECONOMICS AND STATISTICS III. Empirical Framework A. Basic Setup We first estimate the fixed effects model, equation (1), including marriage-market fixed effects, and a full set of time and marriage-market fixed effects. The estimates of ?2 The analysis seeks to estimate the causal effect of the incarceration rate of men in a marriage market on marital and other outcomes for the women in those marriage mar kets, holding all other factors constant. To a first approxi mation, marriages in the United States occur within specific race, regional, and age cells.18 In what follows, we focus on markets consisting of men and women of a specific race, born in a given state, and who are aged either 18 to 25 for women and 20 to 27 for men or 26 to 33 for women and 28 to 35 for men. Although we have millions of individual level observations from the IPUMS, all of the variation in the analysis occurs at the race, state, and age level across census years. To make our estimation approach transparent, we therefore collapse all of the data down to the marriage market (race, state, age) and time level, yielding a total of from these models, as well as those simple difference models we also run, account for unobserved features of a marriage market that are fixed or change relatively slowly over time. However, these estimates do not account for latent determinants of females outcomes that change over time within marriage markets. More important, panel data methods tend to exacerbate any attenuation bias caused by the fact that we do not observe a marriage market's actual incarceration rate 7* s? only an error-ridden version Jm,t.20 To attempt to deal with these problems?endogeneity bias that survives the application of standard panel data methods and attenuation bias likely magnified by the appli cation of those methods?we also present two-stage least squares (TSLS) estimates. An appropriate instrument is something that is strongly correlated with the change in a around 1,200 observations?300 marriage markets across marriage market's incarceration rate and conditional on four census years. We assume that the structural relationship between out come Yma for women in marriage market m in time and the male incarceration /* t for men in that market is given by Ym,t = ?o + ?iXWt, + ?*/t, + Tm + em,,. (1) In equation (1), XmJ is a set of observed characteristics about the marriage market in census year t. The vector Tm reflects fixed features of the marriage market that affect marriage market outcomes, and the error term EmJ repre sents all unobserved determinants of outcomes Y. We are interested in estimating the parameter ?2. B. Instrumental Variables for Dealing with Endogeneity and Measurement Error Concerns Two problems frustrate the straightforward estimation of the parameter ?2. The first is the endogeneity bias caused by possible correlation between the incarceration rate in a marriage market and some unobserved feature of the mar ket, perhaps related to general social breakdown. For exam ple, high levels of criminality among men of a certain type would lead to higher incarceration rates for those men but would also increase women's reluctance to marry them whether they were in jail or not. The other problem is that we do not observe a marriage market's actual incarceration rate f, only an error-ridden version Jm t} Estimates of ?2 will thus, in general, also suffer from attenuation bias. observables, independent of latent behavioral factors oper ating differentially over time across various marriage mar kets. In our view, these conditions are likely satisfied by the various policy changes associated with the criminalization and greater punishment of drug offenses: the war on drugs. The late 1980s saw the launching of the "Just Say No" campaign and the creation of the Office of the Drug Czar. Around the same time, a series of laws was passed at the federal and state levels designed to decrease the variance and increase the severity of sentencing for drug offenders and other felons. The fact that most of the increases in incarceration occurred after 1980, and especially after 1990, strongly suggests that these policy changes had a significant effect on the number of people behind bars.21 To formally assess the importance of policy changes related to the War on Drugs, we focus directly on the actual composition and adjudication of cases in state criminal caseloads because it is the actual sentences that guidelines and other measures were designed to affect. We use data from the National Judicial Reporting Program Series (NJPRS), a data series put together by the Bureau of Justice Statistics division of the U.S. Justice Department, with the first wave of data collected in the late 1980s. Every two years, the series tabulates data about persons tried in state courts?the overwhelming majority of all criminal cases. which men and women date and marry, and also because of issues such as how the census defines and measures institutionalized populations. 20 See Freeman (1984) for discussion of panel methods and measure ment error. 21 At the federal level, the Sentencing Reform Act and the Violent Crime Hispanics are less dramatic, but the basic pattern is evident for this group Control and Law Enforcement Act were passed by Congress in 1984 and 1994, respectively. Shortly after, various state laws were enacted with features like mandatory minima, limitations on parole, and so-called three-strikes rules. There is some debate about whether the mere passage of laws affected incarceration; state-specific implementation likely mat 19 As noted earlier, this mismeasurement arises because the marriage markets we define only imperfectly characterize the dimensions along tered importantly. For example, work by Johnson (2005) and others shows that prosecutors have considerable discretion with respect to the precise charge a defendant faces and over the plea-bargaining process. 18 Using data from several census years we confirm this familiar pattern. For example, we find that more than 90% of black (white) marriages are to other blacks (whites) of a closely related age. The numbers for as well. This content downloaded from 155.33.16.124 on Tue, 08 May 2018 10:53:42 UTC All use subject to http://about.jstor.org/terms MALE INCARCERATION AND FEMALE MARRIAGE OUTCOMES 619 The data are collected from prosecutors' offices and state court records from 100 of the largest counties in the country. the instrument. We cannot be sure that rising criminalization of drugs within marriage markets is not correlated at all with and the ultimate dispensation of their cases. We focus on behavior within the states. However, two considerations The survey collects information race, age, and sex of changes in behavior. Indeed, states that raised the penalties defendants; the crimes with which they were charged; for drug use were likely responding to some changes in data from 1990, since data on drug possession charged suggest that this correlation is not a large concern in the were not collected separately in the first two waves of the context of our TSLS estimates. First, recall that our first study. Fortunately, as was shown earlier, the period from stage results control directly for changes in personal and 1990 to 2000 represents nearly 60% of the increase in property crime?factors likely correlated with rising drug incarceration since 1970 for all of the marriage markets use.23 Second, self-reported drug use shows little evidence we study. As we show in table A.l in the appendix, we find that across all counties and states, the fraction of all state of higher drug use among the cohorts of men facing higher incarceration risk for drugs. Perhaps the best available age- and race-specific data prosecutions that were for drug charges rose dramatically on offending are those for self-reported drug use in the over the 1990s, while that for every other type of serious Monitoring the Future study (MTF).24 Since 1975, the charge either remained constant or fell substantially. At the MTF has inquired about drug use among a nationally beginning of the decade, about 21% of men facing trial in representative sample of high school juniors and seniors. state court were being tried on some type of drug charge. By Of course, there is not a perfect relationship between high 2000, the number had risen to 38%?23% of all cases being school drug use and criminal behavior. Nonetheless, the for drug trafficking charges and 15% for drug possession. notion that adolescent drug use is a pathway into more The data indicate that conviction probabilities for all of fenses remained relatively constant over the decade. The facts together meant that by 2000, almost 38% of men in prison had been convicted of some drug offense?nearly a serious crime is widely established in the social science male incarceration rate to the average of the drug caseload measure between 1990 and 2000. Importantly, in addition to other observables about marriage markets, the regressions control for changes in the levels of personal and property crime in the state. This ensures that what we estimate is falling. The same pattern is evident for all groups of men, and across the country.26 This evidence strongly supports the notion that the increases in incarceration arising from the criminalization of drugs in the War on Drugs princi pally reflect the effect of criminalization policy rather literature.25 If changes in incarceration patterns derived principally from changes in behavior, we would expect some sort of positive association between the incidence doubling in that fraction over the decade. The share of of imprisonment for particular types of men and their self-reported drug use. In fact, as graphically summarized prisoners convicted of every other kind of offense fell in figure A.l in the appendix, we find that incarceration dramatically over the same period. rates for the different cohorts of men of a particular To gauge how the increased criminalization of drugs region and race type have been rising at the same time associated with the war on drugs affected male incarceration within marriage markets, we relate the change in a market's that high school drug use for those men has been flat or purely the effect of state and race differences in how drug than behavioral changes. offenses are treated within a particular state rather than IV. anything having to do with criminal activity (behavior) in the state. As shown in table A.l the evidence that drug Results A. Female Marital Outcomes caseloads positively affected the 1990-2000 change in male incarceration is strong. The effects are large and strongly Table 1 presents various estimates of the effect of male statistically significant, and the associated partial F-statis incarceration on female marital outcomes. The results in the tics pass recommended "weak instrument" thresholds (see first row are for the basic fixed-effects model given by Stock, Wright, & Yogo, 2002).22 equation (1). This specification includes a vector of fixed For drug caseload measures over the 1990s to be valid instruments for marriage market changes in incarceration 23 Indeed, why drug use changes across different cohorts remains an over the 1990s, it must be the case that they are correlated open question. See Charles and Stephens (2006). with incarceration only through the policy channel of the 24 In the 1984 wave and in a more limited follow-up in the 1988 wave, the NLSY-79 also asked questions about current and historical drug use. criminalization of drugs. Correlation between the caseload The MTF drug use data are widely considered the most reliable informa policy measures and increases in, for example, drug use tion on drug use and cover several cohorts of young persons. But in the changes within marriage markets threatens the validity of NLSY data also, we find that levels of incarceration for drug offenses are only weakly correlated with reported drug use. 25 See, for example, Markowitz (2000), Parker and Auerhahan (1998), 22 A variety of alternative specifications, using other measures of the adjudication of drug cases over the ten-year period, yield very similar results. and Baumer et al. (1998). 26 Note that we cannot obtain reliable drug use information for Hispanics from the MTF. This content downloaded from 155.33.16.124 on Tue, 08 May 2018 10:53:42 UTC All use subject to http://about.jstor.org/terms THE REVIEW OF ECONOMICS AND STATISTICS 620 Table 1. -Fixed Effects, First Difference and TSLS Estimates of Effect on Male Incarceration on Women's Marital Outcomes (Robust Standard Errors in Parentheses) Wife's Education > Ever Married? All Women with Women with Women No College Any College Wife's Education < Husband's Education? Husband's Education? All Wives All Wives Mean of dependent variable 0.58 -0.084 -0.03 0.18 0.19 Fixed effects model: 1970, 1980, 1990, 2000 Controls: Marriage market fixed effects, Time, Time -0.0111 -0.0131 -0.0081 0.00219 -0.00934 X Race, Time X Age, Time X State [0.0016]** [0.0020]** [0.0012]** [0.00070]** [0 Observations 1,195 1,191 1,187 1,177 1,177 R2 0.98 0.97 0.99 0.88 0.83 Mean of dependent variable -0.059 -0.084 -0.03 0.012 -0.045 Difference model: 1990-2000 Controls: Maximum state welfare payment, total -0.0038 -0.0039 -0.0037 0.0006 0.0003 property and personal crime, median male wage [0.0005]** [0.0008]** [0.0004]** [0.0006] [0.0005] Observations R2 TSLS 0.79 306 0.71 model of 303 306 0.68 300 0.3 1990-2000 300 0.9 difference Controls: Same as in difference model instrument: -0.0122 ? ? 0.0019 -0.0014 1990-2000 drug caseload [0.0023]** [0.0013] [0.0011] Note. Robust standard errors are in parentheses. Regression are performed at the level of the marriage market: individual-level census data are aggregated to race, st corresponds to equation (1) in text. Difference models to 2000-1990 differenced version of equation (1). For any particular outcome, mean is taken over all women on in there are no women of particular type in market, the cell is dropped. See text for additional details. theset median wage of all men in the market, with th effects for each marriage market and a full of interac of This men in the market set to 0.28 tions between time, age, state, and race. model is We have estimated versions of the difference mo estimated over the four census years. (For graphical illus the supplemental 1980-2000 period, and a version in which we us tration of marriage market strategy, see the intercensal year changes in the data (1970-198 material available online at http://www.mitpressjournals.org/ 1990, and 1990-2000), and find results that in eve arebottom very similar Two sets of estimates are presented in the panel to those we show for model restricte change only. We present the results of the table. The results in the second row1990-2000 are for a simple particular model because our final result difference model over the 1990-2000 period, the difference decade in theWe third row of the tables, are TSLS estima with the largest increase in male incarceration. relate instrument for the 1990-2000 change in male incar the 1990-2000 change in the relevant outcome variable using the marriage market's drug caseloads betwe for women in a marriage market to the change in the market's male incarceration rate over the This anddecade. 2000. As noted above, this information on cri doi/suppl/10.1162/rest_a_00022.) cases is available regression includes controls for changes in the welfareafter 1990. In all regressions, the ceration rate variable is measured in percentage po payments, total property crime, and total personal crime across the first row of table 1 reveals t in the state.27 In this and later regressions, Reading we also control point for the labor market conditions men in the estimates marriagefrom a standard fixed-effects spec that market face. One obvious problem here is imply that we domale not incarceration affects women's mar comes precisely observe what the wages (or employment) would have as the marriage market model wo The strongly statistically significant esti been for men who are incarcerated. To gest. get around this ?0.011 the first column implies that a 1 percenta problem, we follow a strategy used by Neal andin Johnson in male incarceration in a given marriage mar (1996). Specifically, we assume that in aincrease given market, associated a reduction in the probability of a incarcerated men are likely drawn from the left tail with of the ever having married by 1.1 percentage points. The offer-wage distribution. Under this assumption, the me deviation of male dian offered wage of men in the market, a measure of incarceration rates across all ma 2.8 percentage points. The point estimate in the first men's labor market opportunities in the market, is simply of -0.011 therefore implies that a 1 standard d ineach a marriage market's male incarceration 27 We get property and personal crime numbers for increase each state in of the four census years from the Bureau of Justice Statistics (http://www. associated with a reduction in the likelihood of a w ojp.usdoj.gov/bjs). The specific welfare measure we use is the amount of welfare payments a woman with two children, who did not work in the If x% of men in a market are incarcerated, our measure of t market, would receive in welfare transfers. These 28 figures have been offered male wage in the market is simply the (50 - x)th percent calculated for all the years from 1970 to the late 1990s by Rebecca Blank observed wage distribution. and Robert Schoeni. We thank them for providing us with these data. This content downloaded from 155.33.16.124 on Tue, 08 May 2018 10:53:42 UTC All use subject to http://about.jstor.org/terms MALE INCARCERATION AND FEMALE MARRIAGE OUTCOMES 621 that market ever having married of about 3.1 percentage points. This represents slightly more than a 5% reduction in the likelihood of marriage, relative to the mean of 58%. We showed earlier that because incarceration tends to be twelve years of schooling, who marry dropouts, and of women with two or fewer years of college, who marry me with exactly a high school diploma. Interestingly, we find no changes in the propensity to marry down among women concentrated among men in the lower part of the education who are college graduates. This result is reassuring, as this distribution, ever-married rates should be most affected for is the group that theory suggests should be least affected by lower-skilled women than for their higher-skilled counter changes in male incarceration. All of the results in the top row are consistent with the parts when incarceration is high.29 The two entries in the second and third columns of the first row bear out this predictions of the marriage-market model. The second row in the table presents results for 1990-2000 difference mod prediction. The second and third estimates in the first panel, of ?0.013 for the fraction of women with only high school els. Recall that unlike the baseline specification in the first row, these regressions control for changes in observable training who have ever been married is larger than (although only weakly statistically different from) the estimated asso features of a marriage market such as state welfare pay ciation of ?0.0081 for the fraction of women with any ments, property and personal crime, and the median mal college training who have ever been married. There is some wage. The most noteworthy point about the differenc evidence that higher incarceration is associated with larger results is that they are of much smaller magnitude than the reductions in ever-married probabilities for lower-skilled baseline results in the top row. Most of the results remain women. statistically significant and of the expected sign, but the sorting results are now statistically indistinguishable from 0. The final two entries in the first panel investigate the association across markets between male incarceration andthese difference estimates so small? One possi Why are the sorting patterns of women and men who dobility marry. is In the role of the various control variables. We exam these regressions, people are sorted into three education ine the results with the variables added one by one and fin that ithigh is the addition of controls for the median male wages attainment categories: less than high school graduate, school graduate, and a person with any college training. that most We appreciably lowers the point estimates. This sug estimate two sets of models to assess sorting. The first, gests that some of the marriage-related outcomes for women shown in the fourth column, measures how incarceration derive not from the fact that men are incarcerated, but rather rates affect women's propensity to "marry down." from That the is,fact that men in markets with rising incarceration would not make good spouses because of their poor labor this regression asks what fraction of wives have spouses market prospects. Another, and we believe more important, with schooling less than theirs. The model in the fifth for the smaller estimates in the second row is what column measures wives' propensity to "marry up"reason and asks, wea earlier discussed about the measurement error associate What fraction of wives in a marriage market have husband with linking whose schooling is greater than theirs? If the insights from people to marriage markets. row of the table presents the TSLS estimates the standard marriage market model are correct, the The shiftthird in discussed bargaining power from women to men associated with a above. In these regressions, the change in a reduction in the number of men in the market should have marriage market's incarceration rate is instrumented for using the market's drug caseload. In principle, these results a nonnegative effect on the incidence of marrying down. accountthe for any endogeneity bias that survives the applica Furthermore, it should either lower or leave unchanged of the panel data methods in the first two rows of th likelihood of marrying down among women who tion do marry. table, well as the effect of any attenuation bias associate The strongly statistically significant results in the lastastwo mismeasurement of the marriage markets and the columns are consistent with these predictions. with Thesethe asso incidence ciations imply that a 1 standard deviation higher level ofof male incarceration within them. incarceration is associated with a 4% increase in the inci all of the TSLS estimates of the effects of Overall, incarceration on women's marital outcomes are of the pre dence in marrying down and a 5% decrease in the incidence dicted sign. They are generally larger than the difference of women marrying up, relative to the means of these outcomes. results in the second row and closer in magnitude to the baseline fixed-effects estimates in the top row. The sorting We do not present these numbers in the table but estimate results in this specification are now of the expected sign but the propensity to marry up or marry down among women not statistically significant. This pattern of results sug with different levels of schooling. We find that are marrying gests that the conclusions one would draw about the effect down is especially pronounced among women with exactly of incarceration on women's marital outcomes based on th 29 Indeed, positive assortative mating implies that we should results expect of the simple associational relationships are probably same effect even if only high-skilled men were removed from the market upward biased, most likely because of the influence o by incarceration, since such men as remained who wished to be married unmeasured factors specific to would remain in the marriage market should always wish to marry the highest-quality woman they can get. appears to have been relatively This content downloaded from 155.33.16.124 on Tue, 08 May 2018 10:53:42 UTC All use subject to http://about.jstor.org/terms markets. However, this bia small. Panel data estimates, 622 THE REVIEW OF ECONOMICS AND STATISTICS Table 2.?Estimated Effect of Male Incarceration Rate on Women Ever Married, and Implied Fraction of 1980-2000 Change in Marriage Explained by Incarceration Fixed Effects Difference Model TSLS Specification 1980 and 2000 in Share of: Fraction of Fraction of Fraction of Percentage Point Change Between - - - Change Change Change Men Incarcerated Women Married Point Estimate Explained Point Estimate Explained Point Estimate Explained All races 2.057 -12.95 -0.0111 18% -0.0038 6% -0.0122 19% Whites 0.742 -11.31 -0.0694 46% -0.0325 21% -0.1108 73% Blacks 7.362 -17.22 -0.0064 27% -0.0031 13% -0.0043 18% Hispanics 3.637 -16.49 -0.0124 27% -0.0042 9% -0.0045 10% Note. Estimates are race-specific results from the alternative. All point estimates in the table are statistically significant at the 5% level. Implied effects equal: 100 X rate/change in marriage rate). See text for additional details. from either fixed-effects or difference models, In all account three models, for somewhere between 10% and much of this bias, but the estimates theythe yield changes seem to inbe marriage for blacks and Hispan significantly attenuated because of measurement counted for error byin male incarceration. Strangely, tw classifying men to marriage markets. three models yield much larger effects for whites. Pr In table 2, we ask how much of the changes why the in female estimates for whites?and especially that ever-married rates between 1980 and 2000TSLS can beestimates?are explained so large is not at all cle by estimates from the various models. Because the effect possibility is that white marriage markets with risi might differ sharply by race, we estimate the various or model ceration, markets in which the war on drugs hit es separately by race. The first two columns ofhard, the table differ depict from other marriage markets in some fu the change in male incarceration and female ever-married tal way that we do not capture. The apparent impla rates between 1980 and 2000, overall and for each race for whites notwithstanding, we of some of results rest and of the results (for groups in which incarcerati separately. In all cases, incarceration has risen, marriage for women fallen sharply. The next three sets of columns inmost sharply) credible across specif have risen the the table show that for all models and for each race, thethe mean across races and across s Overall, taking tions, it but appears point estimate for male incarceration is negative verythat somewhere around 13% of th different in magnitude across races. More in striking are across the marriage the population appears to the r male incarceration. differences across races and across specifications in the implied fraction of the change in marriage accounted for by the change in male incarceration. TheseB. implied Other effects Female Outcomes equal the relevant point estimate multiplied by the ratio of Table which has the same layout as table 1, pres the change in incarceration to the change in the 3, marriage rate, all times 100. results for the other female outcomes. We earlier argu Table 3.?Fixed Effects, First Difference, and TSLS Estimates of Effect on Male Incarceration on Women Work for Any College? Years Schooling All Women All Women Pay? Fixed-effects Model: 1970-2000 data 0.0062 Controls: Marriage market fixed effects, time, time X 0.0524 0.0122 race, time X age, time X state [0.0069]** [0.0017]** Observations 1,195 1,195 R2 0.9 0.91 Participant? All Women All Women 0.64 Mean of dependent variable 12.92 0.48 Labor Force [0.0006]** 1,195 0.83 0.017 0.69 0.0059 [0.0007]** 1,191 0.75 Child Out of Wedlock? Divorce? All Women All Women 0.18 0.0752 [0.0074]** 1,191 0.58 0.415 0.016 Mean of dependent variable 0.291 0.073 Difference model: 1990-2000 0.0035 0.0396 0.0043 Controls: Maximum state welfare payment, total 0.0119 0.0023 [0.0005]** [0.0039]** [0.0007]** property and personal crime, median male wage [0.0021]** [0.0006]** 306 306 306 Observations 306 306 R2 0.9 0.9 TSLS model of 1990-2000 difference Controls: Same as in difference model instrument: 0.0182 0.0067 1990-2000 drug caseload [0.0044]** [0.0016]** 0.47 0.0031 [0.0011]** 0.42 0.0057 [0.0008]** 0.16 -0.0042 [0.0007]** 1,187 0.89 -0.01 -0.0064 [0.0007]** 302 0.43 0.58 0.1555 [0.0251]** -0.0102 Note. Robust standard errors are in parentheses. Regressions are performed at the level of the marriage market: individual-level census data are aggregated to race/state/age cells. Fixed-effects specification corresponds to equation 1 in text. Difference models to 2000-1990 differenced version of equation (1). For any particular outcome, mean is taken over all women on indicated sample in race/state/age cell. If there are no women of a particular type in market, the cell is dropped. See text for additional details. This content downloaded from 155.33.16.124 on Tue, 08 May 2018 10:53:42 UTC All use subject to http://about.jstor.org/terms [0.0013]** MALE INCARCERATION AND FEMALE MARRIAGE OUTCOMES 623 Table 4.?Two Stage Least-Squares Estimates of Effect of 1990-2000 Change in Male Incarceration Rate, for Specific Alternative Specifications Indicated in Table Dependent Variable: 1990 to 2000 Mean Difference among Women in Marriage Market: Wife's Wife's Education Education Labor Ever > Husband's < Husband's Years Any Work for Force Child Out Married Education? Education? School College? Pay? Participant of Wedlock D Extension/Alternative Specification Marriage market measured -0.0114 0.0019 -0.0017 0.0199 0.007 0.0029 0.0042 0.1469 -0.0102 by state of residence [0.0023]** [0.0014] [0.0010] [0.0052]** [0.0018]** [0.0013]* [0.0011]** [0.0249]** [0.0016]** Controls for total number of -0.0135 0.0012 -0.0018 0.0215 0.0081 0.0036 0.0064 0.1695 -0.0103 men and women in [0.0026]** [0.0018] [0.0012] [0.0045]** [0.0018]** [0.0012]** [0.0008]** [0.0291]** [0.0015] market Note. The instrument is the same as in previous tables in the relevant market. Robust standard errors are in brackets. Each point estimate represents results from a separate TSLS specification. Each regression contains controls for the change in maximum welfare payments, the level of property crime, the level of personal crime in the state, and the median offered male wage between 1990 and 2000. See text for additional details. because higher male incarceration produced negative mar riage outcomes for women, there should be increased in centive among women to undertake actions that raise their levels of economic independence. We find that increases over time within a marriage market are associated with increases in female schooling and employment.30 For example, the strongly significant point estimate of 0.012 from the fixed-effects model implies that a standard deviation increase in the incar ceration rate of 2.8 percentage points raises the fraction of women in the market with any college training by about 7%. This effect, like that for the fraction of women who work for pay and for the share of women who are in the labor force, is relatively modest. But it is still very consistent with the notion that higher male incarceration, by virtue of its negative effect on marriage outcomes, induced women to engage in behaviors that raised their levels of economic independence. The first row also presents fixed-effects estimates for the share of women in a marriage market who have had an out-of-wedlock birth. A decline in the number of available men should raise the As in table 1, the TSLS estimates are of the sign predicted by theory for all of the outcomes, but they are much larger than those from the simple panel data methods. We argue throughout that this pattern is precisely what one would expect if the instrumental variables methods corrected for attenuation bias in the standard estimates. An alternative explanation is, of course, that the instrumental variables method induces some positive bias into the estimates, prob ably because drug caseloads are correlated with some un observed behavioral change in the relevant markets. We are dubious that this explains our results, partly because all of the estimated effects are of the sign predicted by theory and also because one would expect that the sorts of biases that come readily to mind should, for some of these outcomes, have produced biases going in the opposite direction. For example, areas with rising drug caseloads may be places where there is a general increase in social dysfunction. But such dysfunction should cause women in those markets to work less or be less likely to obtain schooling. That we find results that override this bias suggest that the relationships we have documented are real. bargaining power of men who are not incarcerated. One consequence of this change, we reason, might be an increase in less committed relationships. The results C. Some Extensions and Robustness Tests about out-of-wedlock birth bear out this prediction. The last entry in the table shows the results for divorce?the share of women who were married at some point but no longer are. We find that higher incarceration is system main results above.31 The first extension concerns the spatial atically associated with a lower incidence of divorce. This result too is consistent with the marriage-market model: the presence of fewer men means that fewer women marry at all, but those who do are more selective and their marriages are more stable as a result. 30 We study female schooling and labor force participation as outcomes. We also investigate how the estimated effects for the share of women ever marrying are affected by adding these measures as controls in the ever married regressions. We find that the point estimates for male incarcera tion in these models are either unchanged or slightly larger. Before concluding, we present two extensions to our definition of marriage markets. In all of the analysis con ducted thus far, we characterize markets by respondents' states of birth rather than their state of residence at the time 31 Although our focus in this paper is on how women are affected by rising male incarceration, for completeness, table A.3 in the appendix shows some results for unincarcerated men. These estimates cannot be taken as a test of the marriage model since observed outcomes for unincarcerated men are mechanically affected by the share of men incar cerated. For example, rising incarceration leads necessarily to an observed higher incidence of schooling among men who are not in prison since incarceration draws more less educated men out of the market. Moreover, all experience suggests that men who are especially scarce in a marriage market may be loath to marry even though they could. This content downloaded from 155.33.16.124 on Tue, 08 May 2018 10:53:42 UTC All use subject to http://about.jstor.org/terms 624 THE REVIEW OF ECONOMICS AND STATISTICS of the survey. We earlier noted that our main reason for making this choice was that inmates might be systematically misclassified with respect to the marriage market to which they would belong were they free. How sensitive are our results to using state of birth rather than state of residence? The first row of table 4 presents the results from the TSLS regression of the effect of the change in incarceration on various outcomes for women, but with the marriage market measured by state of residence rather than state of birth as in the earlier tables. To conserve space, we report only the estimated coefficient and (robust) standard error on the incarceration rate term from each of the twenty regressions. The results indicate that for all outcomes, the estimated effects when measuring marriage markets by state of resi dence are very close to those from the preferred state of birth estimates. For example, for the fraction of women ever-married outcome, the preferred estimate from this instrument set for state-of-birth marriage markets is ?0.018, while that from using state-of-birth marriage mar kets is ?0.011. Our results appear quite robust to the next most obvious alternative spatial definition of marriage mar kets. In the second row of table 4, we control directly for the number of men and women in a market. All of the main Our various empirical strategies for breaking the possible endogenous relationship between marriage market out comes and incarceration make use of two facts. First, we show that the increase in incarceration has not been uniform across all types of men. Instead, there has been tremendous variation in rates of incarceration across men of different races, locations, and ages, and also great variation within each of these categories. Second, we show that most mar riages occur within relatively narrow marriage markets, defined by the interaction of race, age, and location. Taken together, these two facts imply that different types of women have been exposed to dramatically different shocks relating to the relative presence of men in their respective marriage markets. Using data on marriage markets over time, we estimate a series of panel data models that exploit variation in marriage markets over time. These models have two possible shortcomings: changes in incarceration might be associated with changes in unobserved factors within markets, and panel methods likely exacerbate any measure ment error problems that confound our estimate. With self-reported data on drug offending, we find sug gestive evidence that changes in incarceration over time appear to be due not to changes in male behavior but rather results are weighted by the size of the marriage market, but to changes in policy. We then isolate variation in the change the previous regressions did not control directly for these two measures of marriage market size. A concern might be to a specific policy: the handling of drug cases following the that the estimated effect of incarceration rate could in fact be due to changes in cohort size. Men tend to marry younger women, so population growth or decline can have indepen dent effects on women's marriage outcomes. For example, if there is population growth, women confront a situation where men of the sort they are likely to marry (older men) are scarce. If this growth coincides with rising incarceration and we did not control for population, our analysis would incorrectly attribute all of the reduction in marriage to the increase in imprisonment. Controlling for marriage-market size also accounts for any differential trends in large and small markets. Finally, controlling separately for the total number of men and number of women represents a more flexible formulation of a single control for the sex ratio, which some of the previous literature on marriage empha sizes. Our results suggest that controlling for the (the change in) marriage-market size, the results in the second row all remain of the predicted signs and are roughly of the same size and significance level as the results in table 1. On the whole, these results are qualitatively quite similar to those presented earlier in the paper and in no way change our main conclusions. V. Conclusion In this paper we study how women have been affected by rising male incarceration levels over the past thirty years. in incarceration rates within marriage markets attributable initiation of the war on drugs. In a series of TSLS regres sions, we then use these policy measures to account for any endogeneity associated with the change in incarceration, as well as for the effect of measurement error. Our results show strongly that higher levels of male incarceration lower female marriage. Our findings regarding martial sorting are less definitive; whereas many of the point estimates indicate a tendency for women to marry men of inferior quality when they do marry, as implied by the standard marriage-market model, the effects are either not statistically significant or only weakly so. Evidence for divorce does show that marriages that do form in the face of rising male scarcity from incarceration are more durable, presumably of greater initial selectivity. We find that women increase both their schooling and labor supply in the face of higher male incarceration, actions that we argue represent rational reactions to negative marriage-market shocks. Fi nally, whereas observable controls for market characteristics such as state welfare payments have little effect on our estimates, the evidence strongly suggests that declining male labor market opportunities in markets with rising male incarceration play an important role in explaining recent female marriage market outcomes. The results are stable across a variety of specifications and robust to alternative spatial definitions of marriage markets, but we nonetheless note that some sort of generalized social dysfunction in marriage markets may partially explain our results. This content downloaded from 155.33.16.124 on Tue, 08 May 2018 10:53:42 UTC All use subject to http://about.jstor.org/terms MALE INCARCERATION AND FEMALE MARRIAGE OUTCOMES 625 Our results identify an important unintended effect of Gould, Eric, and E. Daniele Paserman, "Waiting for Mister Right: Rising Inequality and Declining Marriage Rates," Journal of Urban Eco increased incarceration. Whether this policy is socially ben nomics 53 (2003), 257-281. eficial also depends, of course, on the degree to which Greenwood, Jeremy, Ananth Seshadri, and Mehmet Yorukogu, "Engines these, but we argue that the results presented here should be Health and Human Services, Substance Abuse and Mental Health Services Administration, 2000). Guttentag, Marcia, and Paul F. Secord, Too Many Women? The Sex Ratio Question (Thousand Oaks, CA: Sage, 1983). Johnson, Brian D., "Contextual Disparities in Guidelines Departures: Courtroom Social Contexts, Guidelines Compliance, and Extra legal Disparities in Criminal Sentencing," Criminology 43 (2005), of Liberation," Review of Economic Studies 72 (2005), 109-133. imprisonment achieves its direct aim of lowering crime and Grob, Gerald N., "Mental Health Policy in 20th Century America," in punishing criminals, and the relative societal valuations of Ronald M. Manderscheid and Marilyn J. Henderson (Eds.), Mental Health, United States, 2000 (Rockville, MD: U.S. Department of these ends. Our work has nothing to say about either of part of the debate about the societal wisdom of increased incarceration. REFERENCES 101-136. Angrist, Joshua, "How Do Sex Ratios Affect Marriage and Labor Mar Loughran, David S., "The Effect of Male Wage Inequality on Female Age at First Marriage," this review 84 (2002), 237-251. kets? Evidence from America's Second Generation," Quarterly Journal of Economics 117 (2002), 997-1038. Maguire, Kathleen, and Ann L. Pastore, Sourcebook of Criminal Justice Statistics (Washington, DC: U.S. GPO, 2000). Baumer, E., J. L. Lauritsen, R. Rosenfeld, and R. Wright, "The Influence of Crack Cocaine on Robbery, Burglary, and Homicide Rates: A Markowitz, Sara, "The Role of Alcohol and Drug Consumption in Deter Cross-City Longitudinal Analysis," Journal of Research on Crime mining Physical Fights and Weapon Carrying by Teenagers," and Delinquency 35 (1998), 316-340. National Bureau of Economic Research working paper no. 7500 (2000). Becker, Gary, "A Theory of Marriage: Part I," Journal of Political Neal, Derek A., and William R. Johnson, "The Role of Premarket Factors Economy 81 (1973), 813-846. - "A Theory of Marriage: Part II," Journal of Political Economy 82 in Black-White Wage Differences," Journal of Political Economy (1974), S11-S26. 104 (1994), 869-895. -A Treatise on the Family (Cambridge, MA: Harvard University OECD, Society at a Glance: OECD Indicators (OECD, 2001). Press, 1981). Parker, R., and K. Auerhahan, "Alcohol, Drugs and Violence," Annual Blau, Francine, Lawrence Kahn, and Jane Waldfogel, "Understanding Review of Sociology 24 (1998), 291-311. Young Women's Marriage Decisions: The Role of Labor and South, Scott J., and Kim M. Lloyd, "Marriage Opportunities and Family Marriage Market Conditions," Industrial and Labor Relations Formation: Further Implications of Imbalanced Sex Ratios," Jour Review 53 (2000), 624-647. nal of Marriage and the Family 54 (1992), 440-451. South, Scott, and Katherine Trent, "Sex Ratios and Women's Roles: A Markets," Journal of Human Resources 32 (1997), 741-778. Cross-Sectional Analysis," American Journal of Sociology 93 (1988), 1096-1115. Charles, Kerwin Kofi, and Melvin Stephens Jr., "Abortion Legalization and Adolescent Substance Use," Journal of Law and Economics 49 Stevenson, Betsy, and Justin Wolfers, "Marriage and Divorce: Changes Brien, Michael, "Racial Differences in Marriage and the Role of Marriage (2006), 481-505. and Their Driving Forces," NBER working paper no. 12944 Sociological Review 6 (1940), 937-947. (2002), 518-529. (2007). Chiappori, Pierre-Andr?, Bernard Fortin, and Guy Lacroix, "Marriage Market, Divorce Legislation, and Household Labor Supply," Jour Stock, James H., Jonathan H. Wright, and Motohiro Yogo, "A Survey of Weak Instruments and Weak Identification in Generalized Method nal of Political Economy 110 (2001), 37-72. Cox, Oliver C, "Sex Ratio and Marital Status among Negroes," American of Moments," Journal of Business and Economic Statistics 20 Fosset, M. A., and K. J. Kiecolt, "Mate Availability and Family Structure U.S. Department of Justice Bureau of Justice Statistics, Key Facts at a among African Americans in U.S. Metropolitan Areas," Journal of Glance: Correctional Populations (Washington, DC: U.S. Govern Marriage and Family 55 (1993), 288-302. ment Printing Office, 2003). http://www.ojp.usdoj.gov/bjs/glance/ tables/corr2tab.htm. Freeman, Richard, "Longitudinal Analyses of the Effects of Trade Unions," Journal of Labor Economics 2 (1984), 1-26. Wilson, William J. The Truly Disadvantaged: The Inner City, the Under Freiden, Alan, "The United States Marriage Market," Journal of Political class, and Public Policy (Chicago: University of Chicago Press, 1987). Economy 82 (1974), S34-S53. Goldin, Claudia, and Lawrence Katz, "The Power of the Pill: Oral Wood, Robert, "Marriage Rates and Marriageable Men: A Test of the Contraceptives and Women's Career and Marriage Decisions," Wilson Hypothesis." Journal of Human Resources 30 (1995), 163-193. Journal of Political Economy 110 (2002), 730-770. This content downloaded from 155.33.16.124 on Tue, 08 May 2018 10:53:42 UTC All use subject to http://about.jstor.org/terms 626 THE REVIEW OF ECONOMICS AND STATISTICS APPENDIX Table A. 1.?Prosecutions and Ultimate Dispensation of All Criminal Cases Brought in State Court Homicide Rape Robbery Aggravated Assault Burglary Larceny Drug Trafficking Drug Possession All Other Felonies A. Distribution of All Felony Cases Brought in State Court, by Main Charge Faced by Defendant 1990 4.4% 6.5% 9.5% 10.4% 12.5% 10.7% 13.0% 8.4% 24.5% 1992 5.1% 7.0% 9.6% 11.0% 11.6% 9.6% 11.4% 8.2% 26.4% 1994 4.9% 6.7% 8.0% 11.8% 9.4% 8.7% 11.9% 7.7% 30.8% 1996 1.5% 3.3% 6.3% 7.7% 10.4% 10.5% 23.6% 13.6% 23.2% 1998 1.3% 3.1% 5.5% 7.7% 9.4% 10.3% 23.6% 15.2% 24.0% 2000 1.1% 3.4% 5.6% 8.6% 9.0% 10.0% 22.8% 15.0% 24.5% B. Fraction of Cases in Which Defendant Convicted, by Main Charge Faced at Trial and Year 1990 1992 1994 1996 1998 2000 92.1% 92.8% 95.8% 93.6% 93.3% 92.0% 66.0% 67.3% 73.2% 63.5% 67.4% 65.4% 73.4% 74.2% 77.7% 73.8% 76.3% 74.6% 46.8% 45.8% 54.0% 43.5% 48.3% 45.9% 55.7% 54.3% 54.7% 49.0% 56.1% 55.0% 43.0% 41.0% 42.3% 35.2% 43.3% 38.9% 50.4% 50.5% 49.6% 42.6% 46.2% 45.6% 39.1% 36.7% 37.1% 32.2% 38.5% 39.8% 39.6% 37.9% 42.6% 34.5% 39.1% 37.8% C. Fraction of Convicted Defendants Sentenced to Prison/Jail Term, by Charge at Trial and Year 1990 1992 1994 1996 1998 2000 96.6% 97.1% 99.2% 96.4% 97.9% 97.6% 88.3% 89.1% 90.7% 85.3% 83.9% 85.4% 93.2% 91.7% 93.2% 92.5% 91.2% 92.8% 79.5% 78.6% 84.6% 81.1% 78.2% 78.2% 81.6% 80.2% 82.8% 78.7% 80.6% 81.2% 75.7% 75.1% 78.3% 73.9% 72.2% 70.8% 79.8% 81.5% 80.9% 81.2% 75.4% 74.2% 71.4% 69.2% 74.7% 80.1% 74.7% 73.2% 73.3% 71.4% 77.8% 72.3% 69.9% 71.9% D. Distribution of Main Charge at Trial, of All Persons Convicted and Sentenced to Prison/Jail 1990 5.2% 6.4% 11.4% 10.4% 14.1% 10.4% 13.8% 8.0% 20.3% 1992 6.2% 7.3% 11.6% 10.9% 12.5% 9.1% 12.4% 7.2% 22.8% 1994 5.7% 6.7% 9.1% 11.8% 9.7% 8.2% 12.1% 7.9% 28.8% 1996 2.0% 3.2% 6.9% 8.0% 10.8% 10.0% 25.3% 13.8% 20.1% 1998 1.7% 3.0% 6.5% 8.2% 10.1% 10.0% 25.0% 14.3% 21.2% 2000 1.5% 3.4% 6.9% 9.0% 9.9% 9.1% 24.2% 13.4% 22.7% E. Mean Maximum Sentence (Months), of All Persons Convicted and Sentenced to Prison/Jail 1990 244.8 160.6 117.3 73.5 78.6 51.2 72.2 1992 247.9 141.3 116.7 81.2 75.7 53.7 69.3 1994 277.2 141.7 112.7 84.1 67.1 43.3 62.1 1996 269.6 104.7 96.3 67.8 55.0 37.7 50.7 1998 281.8 108.9 106.4 66.2 50.5 34.5 49.2 2000 255.3 96.1 97.6 60.4 50.3 31.9 48.8 48.8 47.3 40.6 34.9 31.7 32.5 52.2 50.3 45.3 40.7 36.9 35.2 Note. Data are from multiple years of the National Judicial Reporting Program Series (NCJRPS). See text for details. Table A.2.?Effect of Incidence Drug Crimes in Overall State Caseloads between 1990 and Rates for Race/State/Age Cells (Robust Standard Errors in Parenthe 0.2339 Constant [0.2776] 0.0915 Drug caseload [0.0206]** Violent crime -0.0022 -4.0293 Welfare payment -0.8129 [2.0044]* [0.6261] 222 Observations R2 F-statistics: drug caseload = 0 0.45 19.80 0.0674 0.1032 [0.0223]** 25.4442 [10.6321]* Property crime Median wage 2.6014 [0.4575]** [0.2818] 222 0.47 21.33 [0.0191]** 23.8031 [11.1887]* -2.8992 [2.0882] -0.6311 [0.5884] -0.3586 [0.0600]** 222 0.57 12.50 Note. Regression are performed at the level of the marriage market: individual-level census data are aggregated to race, state, age cells. For any particular outcome, mean is taken over unincarcerated men in race/state/age cell. Robust standard errors are in brackets. See text for discussion. * Significance at the 5% level. ** Significance at the 1% level. Table A.3.?TSLS Estimates of Relationship between Male Marriage Rate and Outcomes for Unincarcerated Men in Marriage Market Ever Years Any Work for Labor Force Child Out of Married? Schooling College? Pay? Participant? Wedlock? Divorce? Estimated effect of male -0.0099 0.0383 0.0111 -0.0111 -0.0184 0.0034 -0.008 incarceration rate [0.0021]** [0.0068]** [0.0023]** [0.0016]** [0.0012]** [0.0020] [0.0011] Note. Regressions are performed at the level of the marriage market: individual-level census data are aggregated to race, state, age cells. For any particular outcome, mean is taken over unincarcerated race, state, age cell. See text for discussion. This content downloaded from 155.33.16.124 on Tue, 08 May 2018 10:53:42 UTC All use subject to http://about.jstor.org/terms MALE INCARCERATION AND FEMALE MARRIAGE OUTCOMES Figure A.l.?Relationship between Adolescent Drug Use and Rate of Adult Incarceration, by Race and Census Region, for Different Generations of American Men This content downloaded from 155.33.16.124 on Tue, 08 May 2018 10:53:42 UTC All use subject to http://about.jstor.org/terms 627 Willingness-to-Pay for Crime Control Programs* November 2001 Mark A. Cohen** Vanderbilt University and University of York mark.cohen@owen.vanderbilt.edu Roland T. Rust University of Maryland rrust@rhsmith.umd.edu Sara Steen University of Colorado—Boulder steen@colorado.edu Simon T. Tidd Vanderbilt University simon.tidd@owen.vanderbilt.edu * This project was supported by Grant No. 1999-CE-VX-0001, awarded by the National Institute of Justice, Office of Justice Programs, U.S. Department of Justice. Points of view are those of the authors and do not necessarily represent the official position or policies of the U.S. Department of Justice. We are grateful for comments and suggestions received from Jens Ludwig, Phil Cook, and participants in a seminar at the University of York (UK) and workshops at the annual meetings of the American Society of Criminology and the Association of Public Policy Analysis and Management. Special thanks to the survey team at Roper Starch: Kevin Bray, Project Director; Kathleen Barringer, Project Manager; Robert Benford; and Nicolas A. Holt, PhD. Additional research assistance was provided by Gabrielle Chapman, Achintya Ray and Mihir Shah. ** contact author for permission to cite and for most recent draft. Willingness-to-Pay for Crime Control Programs Abstract This paper reports on a new methodology to estimate the “cost of crime.” We adapt the contingent valuation method used in the environmental economics literature to estimate the public’s willingness-to-pay for reductions in crime. In a nationally representative sample of 1300 U.S. residents, we found that the typical household would be willing to pay between $100 and $150 per year for crime control programs that reduced specific crimes by 10% in their communities. In the aggregate, these amounts imply a marginal willingness-to-pay to reduce crime of about $31,000 per burglary, $75,000 per serious assault, $253,000 per armed robbery, $275,000 per rape and sexual assault, and $9.9 million per murder. Consistent with economic theory and rational behavior, willingnessto-pay generally increases with both income and the risk of victimization. The new estimates are between two and ten times higher than prior estimates of the cost of crime to victims and are thought to more fully represent the true cost of crime to society. By focusing exclusively on costs to victims of crime and the criminal justice system, previous studies have ignored many other social costs of crime. I. Introduction In his seminal paper on the economics of crime, Gary Becker (1968) noted that zero crime is not optimal. Instead, he described the conditions for optimality in the supply and demand for crime by noting that the ultimate goal is to equate the marginal costs of crime control activities to the marginal benefits of crime reductions. Despite the many theoretical insights he brought to the table, Becker concluded, “although many more studies of actual policies are needed, they are seriously hampered on the empirical side by grave limitations in the quantity and quality of data on offenses, convictions, costs, etc.…(p. 209).” Since that time, both the quality and quantity of government data on offenses and the cost of crime control have grown. 1 Perhaps the largest gap in our knowledge, however, is data on the marginal benefits of crime reduction. Benefit-cost analysis is a well-developed methodology that has become an important component of regulatory and policy development for many government agencies. Since the early 1980’s, Federal government regulatory agencies have been required to conduct benefit-cost analyses on major regulatory initiatives. These requirements have been adopted through Executive Order and implemented by the Office of Management and Budget. 2 Recent proposals in Congress would legislatively mandate 1 For example, while the FBI has reported crimes known to police for many years, the first nationally representative survey of victims (including those not reported to police) did not occur until 1973 with the National Crime Survey (since renamed the National Criminal Victimization Survey). Similarly, many of the data available on state and local expenditures date from the early 1980s. 2 President Reagan promulgated the first such requirement in 1981, Executive Order 12291 (46 Federal Register 13193). In 1993, President Clinton issued Executive Order 12866 (58 Federal Register 51735). Although these Executive Orders cannot supercede 1 similar requirements. 3 Thus, benefit-cost analyses have become a routine tool in the development of environmental, health and safety regulations. Unlike regulatory programs, criminal justice programs have rarely been examined using a benefit-cost framework. This is largely due to the paucity of data on the costs of crime (or benefits of crime reduction). Early approaches to estimating the costs of crime focused on differences in property values in low crime versus high crime areas using hedonic pricing methods (Thaler, 1978; Hellman and Naroff, 1979; Rizzo, 1979). These prior studies - which are location specific and not national in scope - suffer from identification problems, as it is empirically difficult to distinguish the independent effects of crime, run-down housing, pollution, and other negative factors that are likely to depress housing prices. Even if we ignore this problem, the hedonic pricing method is likely to underestimate the social cost of crime since it only includes the capitalized effect of lower property values by private dwellings near a high crime area. At the margin, individuals trade-off their personal valuation of the risk of living in a high crime area against other goods and services they demand. Since individuals have different risk preferences and wealth constraints, houses in high crime areas will tend to be owned by those who are less wealthy and more risk tolerant - reducing the risk premium that otherwise would reduce the value of houses due to crime. Thus, relying on housing price differentials to infer an average cost of crime to all members of a community inherently results in estimates that are biased downward. statutory provisions, they have had a dramatic effect on the manner in which regulatory agencies draft and analyze proposed rules. 3 For example, see Senate Bill S. 981, 105th Congress (1997), which would require all major rules to be accompanied by a benefit-cost analysis. 2 In addition, crime imposes other costs on society that are unlikely to be fully capitalized in home prices - such as the increased cost of security to businesses, reduced property values to commercial enterprises, higher medical costs paid by third-parties such as government and insurance agencies, unemployment insurance costs, and lost productivity. In addition to underestimating the cost of crime as a result of data limitations, researchers who have employed this technique have been unable to disentangle the costs of individual types of crime and instead estimate the cost of an “index crime.” Since some policies are directed towards specific types of crime, crimespecific cost estimates are needed. Cohen (1988) developed a methodology for estimating the cost of individual crimes based on jury awards and economic studies of the value of a statistical life. That approach was also used in a study commissioned by the National Academy of Sciences (Cohen, Miller and Rossman, 1994), and in subsequent NIJ- funded research (Miller, Cohen and Wiersema, 1996; Cohen, 1998). We refer to these studies as “victim cost” studies since they focus on the cost of crime to victims. The victim cost studies have been widely cited in the press4 and in recent academic studies that use the cost estimates in benefit-cost analyses of crime control programs. Examples include increased prison population (DiIulio and Piehl, 1991; Levitt, 1996); prevention versus prison (Donahue and Siegelman, 1998); police hiring (Levitt, 1997); juvenile justice policy (Levitt, 1998); rehabilitation (Gray, 1994); priva te security expenditures (Ayers and Levitt, 1998); and legalized abortion (Donahue and Levitt, 2001). 4 See for example, “Study Reveals High Cost of Crime in the U.S.” New York Times, April 22, 1996; and “Prison: Where the Money Is,” New York Times, June 2, 1996. 3 Despite its growing acceptance and use by other researchers, the “victim cost” methodology is not without controversy - both on theoretical and empirical grounds. The main theoretical criticism has been that the victim cost approach is based on an ex post compensation criterion, whereas benefit-cost analysis is generally conducted on an ex ante willingness-to-pay (WTP) approach (see Cook and Graham, 1977). Conceptually, WTP is appropriate for benefit-cost analysis since decisions about whether or not to fund a project are ex ante decisions that require actual expenditures. Since the amount people are willing to pay to avoid a social ill is generally less tha n the amount of money they would require to voluntarily accept it, there is concern that the “victim cost” method overestimates the cost of crime. On the other hand, since the “victim cost” methodology is largely based on the costs to individual victims, it is likely to understate costs by excluding the cost of crime to potential victims and to society at large (see Nagin, 2001a and 2001b). This paper reports on a new approach to valuing crime based on the “contingent valuation” (CV) methodology developed in the environmental economics literature to estimate WTP. The CV methodology has been used extensively to place dollar values on nonmarket goods such as improvements in air quality, saving endangered species, and reducing the risk of early death - social benefits that do not have direct market analogs. There have been literally hundreds of CV studies, meta-analyses and textbooks written on the subject. 5 Although being used in many different policy contexts, contingent valuation has not generally been employed in criminal justice research. One exception is Ludwig 5 For an overview of the contingent valuation method, see Mitchell and Carson (1989). Also see Portney (1994). 4 and Cook (2001), who use this method to estimate that the average household would be willing to pay about $200 per year to reduce gun violence caused by criminals and juvenile delinquents by 30%, which translates into about $1 million per injury. 6 We employ a similar methodology to study the public’s willingness-to-pay to reduce the crimes of burglary, armed robbery, assault, rape or sexual assault, and murder. The next section of the paper describes our survey design process and interview methodology. Section III presents our main survey results on the public’s willingness-topay for reduced crime. Section IV compares these estimates to previous estimates of the cost of crime and provides further insights into the nature of our results. Concluding remarks are reserved for Section V. II. Survey Design and Methodology While conceptually, WTP is the correct approach to valuing crime, the CV methodology is necessarily survey-based due to a lack of direct markets to observe. The CV methodology has been extensively critiqued in the literature and some disagreement exists about its usefulness. The main criticism is that for various reasons, survey respondents might not reveal their true preferences in a survey design. There are several reasons for this concern, including lack of familiarity with the public good in question and lack of an incentive to tell the truth. In the case of crime reductions - unlike valuing a spotted owl or improved visibility of a national park - people commonly deal with the good in question and trade-off dollars or direct actions in response to crime rates. For example, evidence suggests that people respond to crime by purchasing security equipment or moving to lower crime areas (see e.g., Cullen and Levitt, 1999). In response to many of the other concerns about truthful reporting, we have attempted to 6 See also Cook and Ludwig (2000). In addition, Zarkin, Cates and Bala (2000) report on a pilot study in which they use the CV method to value drug treatment. They estimate that the typical household would be willing to pay between $15 and $37 per year for a program that would successfully treat a significant number of drug abusers in their community. This translates into between $28,000 - $69,500 per drug abusers, which compares favorably to the cost of treatment, approximately $12,500. 5 follow guidelines established by a distinguished panel of social scientists (Arrow et al., 1993) commissioned by the National Oceanic and Atmospheric Administration (NOAA) to assess the contingent valuation methodology. This panel was brought together because NOAA had drafted regulations calling for the use of this methodology when estimating natural resource damages in legal proceedings involving compensation for damaged public property. The panel concluded that CV is a valid approach and provided a set of guidelines for conducting a reliable survey. The survey instrument was drafted using an extensive design process including consultation with a panel of experts, three focus groups, 12 hour-long cognitive interviews, and professional interviewer training and monitoring. The survey design process is described in detail in Roper Starch (2000). Table 1 identifies the key recommendations of the NOAA panel that apply to our study and how we have attempted to implement each one. While most of the recommendations were followed, one exception is noteworthy. Arrow et al. (1993) caution CV researchers to thoroughly describe the program under consideration. However, the Arrow et al. (1993) report is primarily concerned with environmental amenities and programs designed to mitigate environmental harm. Thus, for example, researchers might be interested not only in the value of saving a particular endangered species, but also in the value of a “no mining” or “no logging” option that provides more habitat space. We are not interested in one particular crime control policy. Instead, we are interested in valuing crimes themselves. In the focus groups we used to pretest questions, one of the key concerns expressed by participants was that survey respondents would not be able to separate out their desire for reduced crime from the mechanism by which crime reductions take place. For example, although everyone might agree that fewer assaults would be a good thing, there would be significant disagreement over whether a policy mandating life in prison for third time offenders should be implemented if it is shown to deter assaults. In evaluating preliminary survey questions, some participants noted that they had trouble separating their cynicism for the ability of the government to effectively reduce crime from their willingness to pay. Thus, the final sur vey was worded carefully to ensure that a crime control policy was not specified. Instead, respondents were told that a crime prevention strategy had worked last year and that the program had community support. Respondents were asked if they would be willing to vote for a proposal that would require each household in their community to pay a certain amount that would be used to 6 prevent one in ten crimes in their community. 7 They were randomly given three out of five crimes, and the order of each question was randomized. The crimes were: (1) burglary, (2) serious assault, (3) armed robbery, (4) rape or sexual assault, and (5) murder. Given the time limitations of our survey, we identified five of the most commonly understood and important crimes. However, these crimes were not defined for the respondents, and no information was provided on the prevalence, risk of victimization, average tangible losses or severity of injuries normally associated with the violent offenses. Instead, respondents were asked simply to respond based on their understanding of these crimes. The actual text of the survey follows: Now I want to ask you how much of your own money you would be willing to pay to reduce certain crimes. In each case, I am going to ask you to vote “yes” or “no” to a proposal that would require your household and each household in your community to pay money to prevent crime in your community. Remember that any money you agree to spend on crime prevention is your money that could otherwise be used for your own food, clothing, or whatever you need. Unlike the previous question, where the government was planning to give you money back, now I want you to think about actually taking more money out of your pocket. Last year, a new crime prevention program supported by your community successfully prevented one in every ten [INSERT CRIME] from occurring in your community. Would you be willing to pay [INSERT AMOUNT] per year to continue this program? The amounts inserted into the text were randomized between $25 and $225 (in $25 intervals). The maximum annual cost of $225 was selected based on focus 7 The term “community” was purposefully left ambiguous since we wanted respondents to value crime reduction that affects them in some manner - whether through their own household, their families, friends or coworkers. Further, crime reductions might spur economic development in their communities. While use of a term like “neighborhood” would be too limiting, reducing crime in a large “city” might be too broad and would dilute the amount some people are willing to pay. 7 group discussions that indicated $200 would be the maximum amount they would consider paying for such programs. Once an amount was chosen for a particular respondent, that same amount was used for all three crime types for that respondent. If the respondent answered “yes” to the amount, it was increased by $25 and the respondent was asked, “Would you be willing to pay…?” If the initial answer was “no,” the amount was reduced by $25 and the question was asked again. (In the case that the initial bid level was $25 and the initial answer was “no,” the respondent was asked on follow- up if she would be willing to pay $10.) Following the second bid level, the respondent was asked, “And can you please explain why you [would be willing/would not be willing/don’t know if you’d be willing] to pay $[insert amount]?” The verbatim response was recorded. After the first crime question was finished, the following was read: Now please disregard the crime prevention strategy that we just discussed and think of this. Last year, a new crime prevention program supported by your community successfully prevented one in every ten [INSERT CRIME] from occurring in your community. Would you be willing to pay [INSERT AMOUNT] per year to continue this program? The process described above was then repeated for the second and third crimes. The respondent was specifically asked to disregard the earlier question in order to eliminate any ‘income effects’ associated with their earlier response. That is, a respondent might be willing to pay $200 per year to prevent murders and assaults individually, for example, but might not be willing to pay $400 combined to prevent both. To determine whether or not we could add the three bid levels together, or if there were any income effects associated with adding their responses, we asked a final follow-up question at the end of the third crime type: 8 I realize that I asked you to evaluate each crime prevention strategy individually. However, now I’d like you to think of adding all of the money you have spent on each strategy together. You said that you'd pay up to [INSERT AMOUNT] to prevent one in ten armed robberies, up to [INSERT AMOUNT] to prevent one in ten serious assaults, up to [INSERT AMOUNT] to prevent one in ten burglaries, up to [INSERT AMOUNT] to prevent one in ten rapes or sexual assaults, and up to [INSERT AMOUNT] to prevent one in ten murders in your community. Now, if I were to add all that up it comes to [INSERT AMOUNT]. Would you be willing to pay this amount out of your own pocket to prevent all of the crimes we have just talked about? Telephone interviews were conducted with a sample that is representative of the entire United States population of adults age 18 or over. A random digit dial sample of 4,966 phone numbers yielded a total of 1,300 completed interviews. Depending upon the definition used, the “response rate” ranged from 43% to 58%. 8 The data are weighted to adjust for probabilities of selection and to adjust for nonresponse on age, sex, education, and race. Results of this study can be projected to the English-speaking population of people who are 18 years of age or older living in households in the fifty United States, including the District of Columbia. 9 8 The 43% rate is based on a very conservative methodology accepted by CASRO (Council of American Survey Research Organizations). It includes estimates of the percentage of the sample with unknown usability that would become usable and the percentage of sample with unknown eligibility that would become eligible if time were unlimited and the study continued indefinitely. The CASRO method does not estimate the percentage of eligible sample with unknown cooperation that would cooperate if time were unlimited and the study continued indefinitely. Thus, the denominator in the calculation of response rate is increased by these estimates, but the numerator is fixed. Another common method is to compare actual completions to refusals. In that case, our 1,300 respondents are compared to 928 who we were successful in contacting but refused to participate, resulting in a response rate of 58%. 9 A comparison of our sample to the U.S. Census, however, indicates that we might have underrepresented Hispanics and the very poor. Whereas 10.0% of the 18 and older U.S. population is Hispanic, only 6.4% of our sample (and 4.8% of our weighted sample) is Hispanic. One part of that difference is apparently due to language barriers. Sixty- four individuals who were originally contacted were deemed ineligible due to language barriers. If all of these individuals were Hispanic, for example, that would represent 4.9% of our sample, which would bring our sample up to the estimated population ratio. There is also some noticeable difference in reported household income. The main difference appears to be in the percentage of our sample that report household income below $15,000. While 16.5% of the U.S. household population reportedly has an income under 9 The survey was designed with numerous checks to ensure that respondents understood the questions, could respond with some rationality and consistency, and were not biased by the wording of previous questions. We also tested for (and rejected) any potential interviewer bias, 10 temporal changes in responses, and potential bias due to external media attention on crime issues that might have occurred around the time of the survey. 11 III. Willingness-to-Pay Estimates $15,000, only 9.0% of our sample report that level of income (9.6% of the weighted sample). One potential reason for this difference is that 13.8% of our sample refused to provide detailed household income information (as compared to the typical refusal rate for most of the questions in our survey that was well below 5%). If these refusals are clustered at the low end of the income distribution, our sample might look much more like the U.S. population. Finally, we note that since the lowest income families will be those without telephones, and 5.9% of U.S. households do no t have telephones, this could account for the bulk of the difference in our sample. 10 To assess whether any particular interviewer had systematically different responses, we included a dummy variable for each interviewer in regression equations explaining responses. None of the interviewer dummy variables had any significant explanatory power in these regressions. 11 To assess news media effect, we searched and coded all crime-related stories that appeared in national or major regional newspaper headlines by date. Including these dates as dummy variables in regression analyses explaining responses led to no significant explanatory effects. Similarly, we included a time trend in regression equations and found no change in responses over time. Interestingly, we found that there was some slight increase in the number of respondents in the highest income categories and in urban areas over time. We attribute this to our interviewing technique whereby we were persistent in calling back people until we found someone home. We view this as a positive feature of our survey design as it results in a more representative respondent population. 10 Table 2 (and Figure 1) reports on the percentage of respondents who indicated in response to the initial “bid” level that they were willing to pay the specified amount to reduce each particular type of crime. 12 As shown in Table 2, the majority of respondents were willing to pay up to $100 per year for these crime reduction programs. At the lowest...