Outlier detection in contingency tables using decomposable graphical models

For high-dimensional data, it is a tedious task to determine anomalies such as outliers. We present a novel outlier detection method for high-dimensional contingency tables. We use the class of decomposable graphical models to model the relationship among the variables of interest, which can be depicted by an undirected graph called the interaction graph. Given an interaction graph, we derive a closed-form expression of the likelihood ratio test (LRT) statistic and an exact distribution for efficient simulation of the test statistic. An observation is declared an outlier if it deviates significantly from the approximated distribution of the test statistic under the null hypothesis. We demonstrate the use of the LRT outlier detection framework on genetic data modeled by Chow–Liu trees.     

Early teen marriage and future poverty

Both early teen marriage and dropping out of high school have historically been associated with a variety of negative outcomes, including higher poverty rates throughout life. Are these negative outcomes due to preexisting differences, or do they represent the causal effect of marriage and schooling choices? To better understand the true personal and societal consequences, in this article, I use an instrumental variables (IV) approach that takes advantage of variation in state laws regulating the age at which individuals are allowed to marry, drop out of school, and begin work. The baseline IV estimate indicates that a woman who marries young is 31 percentage points more likely to live in poverty when she is older. Similarly, a woman who drops out of school is 11 percentage points more likely to be poor. The results are robust to a variety of alternative specifications and estimation methods, including limited information maximum likelihood (LIML) estimation and a control function approach. While grouped ordinary least squares (OLS) estimates for the early teen marriage variable are also large, OLS estimates based on individual-level data are small, consistent with a large amount of measurement error.Historically, individuals were allowed to enter into a marriage contract at a very young age. In Ancient Rome, the appropriate minimum age was regarded as 14 for males and 12 for females. When Rome became Christianized, these age minimums were adopted into the ecclesiastical law of the Catholic Church. This canon law governed most marriages in Western Europe until the Reformation. When England broke away from the Catholic Church, the Anglican Church carried with it the same minimum age requirements for the prospective bride and groom. The minimum age requirements of 12 and 14 were eventually written into English civil law. By default, these provisions became the minimum marriage ages in colonial America. These common laws inherited from the British remained in force in America unless a specific state law was enacted to replace them (see “Marriage Law,” Encyclopædia Britannica 2005; http://www.britannica.com).While Roman, Catholic, English, and early American law may have allowed marriage at 12 for girls and 14 for boys, many questioned the advisability of such early unions. Researchers and policymakers around the turn of the twentieth century recognized that teens may be especially ill-prepared to assume the familial responsibilities and financial pressures associated with marriage.¹ As a result of the changing economic and social landscape of the United States, in the latter part of the nineteenth century and throughout the twentieth century, individual states began to slowly raise the minimum legal age at which individuals were allowed to marry. In the United States, as in most developed countries, age restrictions have been revised upward so that they are now between 15 and 21 years of age.During this same time period, dramatic changes were also occurring in the educational system of the United States (see Goldin 1998, 1999; Goldin and Katz 1997, 2003; Lleras-Muney 2002). Free public schooling at the elementary level spread across the United States in the middle of the nineteenth century, and free secondary schooling proliferated in the early part of the twentieth century. As secondary schooling became more commonplace, states began to pass compulsory schooling laws. States often also passed child labor laws that stipulated minimum age or schooling requirements before a work permit would be granted. These state-specific compulsory schooling and child labor laws are correlated with the legal restrictions on marriage age, indicating that it might be important to consider the impact of all the laws simultaneously.There are at least two rationales often given for the use of state laws as policy instruments to limit teenagers’ choices. The first argument is that teens do not accurately compare short-run benefits versus long-run costs. If teens are making myopic decisions, restrictive state laws could prevent decisions they will later regret. It is also argued that the adverse effects associated with teenagers’ choices impose external costs on the rest of society. If these effects can be prevented, external costs (such as higher welfare expenditures) would also argue for restrictive state laws. Both teenage marriage and dropping out of high school are closely associated with a variety of negative outcomes, including poverty later in life. To assess the relevance of either argument, however, it is important to know whether the observed effects are causal.Any observed negative effects may be due to preexisting differences rather than a causal relationship between teen marriage (or schooling choices) and adverse adult outcomes. Women who marry as teens or drop out of school may come from more disadvantaged backgrounds or possess other unobserved characteristics that would naturally lead to worse outcomes. For example, teens choosing to marry young might have lower unobserved earnings ability, making it hard to draw conclusions about the causal relationship between teenage marriage and poverty.To identify the effect of a teenager’s marriage and schooling choices on future poverty, I use state-specific marriage, schooling, and child labor laws as instruments. Variation in these laws across states and over time can be used to identify the causal impact that teen marriage and high school completion have on future economic well-being. Although compulsory schooling laws have been used as instruments in a variety of settings, this appears to be the first time marriage laws have been used as instruments. The idea of the marriage law instrument is that states with restrictive marriage laws will prevent some teenagers from marrying who would have married young had they lived in a state with more permissive laws.Using the marriage, schooling, and labor laws affecting teens as instruments for early marriage and high school completion, I find strong negative effects for both variables on future poverty status. The baseline instrumental variables (IV) estimates imply that a woman who marries young is 31 percentage points more likely to live in poverty when she is older. Similarly, a woman who drops out of school is 11 percentage points more likely to be living in a family whose income is below the poverty line. The IV results are robust to a variety of alternative specifications and estimation methods, including limited information maximum likelihood (LIML) estimation and a control function approach. In comparison, the ordinary least squares (OLS) estimates are very sensitive to how the data is aggregated, particularly for the early marriage variable. OLS estimates using grouped data are also large, while OLS estimates using individual-level data indicate a small effect for early teen marriage. Auxiliary data indicate a large amount of measurement error in the early marriage variable, suggesting the presence of attenuation bias in the individual-level OLS estimates.The remainder of the article proceeds as follows. I first briefly review the negative outcomes associated with teenage marriage and dropping out of school and discuss alternative perspectives on why teens might make these decisions. The following section describes the data and presents OLS estimates. The next section discusses the early marriage, compulsory schooling, and child labor laws that will be used as instruments. I then present the instrumental variable estimates and conduct several specification and robustness checks, including a discussion of measurement error issues and a reconciliation with the literature on teenage childbearing.     

The Organization of Wariness of Heights in Experienced Crawlers

Most infants with more than 6 weeks of crawling experience completely avoid the deep side of a visual cliff ( Campos, Bertenthal, & Kermoian, 1992 ; Gibson & Walk, 1960 ). However, some experienced crawlers do move onto the transparent surface suspended several feet above the ground. An important question is whether these nonavoiders lack wariness of heights or whether they have a qualitatively different way of showing their wariness than do avoiders of the deep side. The current study addressed this question by measuring heart rate (HR) acceleration upon being lowered on the deep and shallow sides of the visual cliff, latency to crawl toward the mother, and tactile exploration of the cliff surface. Nonavoiders and avoiders had indistinguishable patterns of HR acceleration, showing greater HR acceleration when lowered onto the deep than when lowered onto the shallow side of the cliff. Nonavoiders also showed more tactile exploration and longer latencies than did a comparable group of infants tested on the shallow side. This study illustrates how the same emotion, wariness of heights, can be shown by qualitatively different behaviors, all serving the same function of protecting the individual from falling over a drop‐off.     

Inner‐City Children's Exposure to Community Violence: How Much Do Parents Know?

This study examines the psychological impact of children's exposure to violence and the influence of mothers' knowledge about their children's encounters with violence. Our sample consists of a poor, multiethnic sample of 104 fourth‐ or fifth‐grade children and their mothers. Children in this sample were exposed to rather high levels of community violence, and on the whole, mothers greatly underestimated their children's exposure to violence and feelings of psychological distress. Hierarchical regression analyses indicated that children's exposure to violence was associated with greater psychological distress. Our findings suggest that the detrimental effects of community violence are present for all children, irrespective of their racial background. Further, greater mother‐child agreement about children's exposure to violence was related to better psychological functioning. The implications of these results for effective parenting strategies and community‐based interventions are discussed.     

Gambling and the Reasoned Action Model: Predicting Past Behavior,Intentions, and Future Behavior

Gambling is a serious concern for society because it is highly addictive and is associated with a myriad of negative outcomes. The current study applied the Reasoned Action Model (RAM) to understand and predict gambling intentions and behavior. Although prior studies have taken a reasoned action approach to understand gambling, no prior study has fully applied the RAM or used the RAM to predict future gambling. Across two studies the RAM was used to predict intentions to gamble, past gambling behavior, and future gambling behavior. In study 1 the model significantly predicted intentions and past behavior in both a college student and Amazon Mechanical Turk sample. In study 2 the model predicted future gambling behavior, measured 2 weeks after initial measurement of the RAM constructs. This study stands as the first to show the utility of the RAM in predicting future gambling behavior. Across both studies, attitudes and perceived normative pressure were the strongest predictors of intentions to gamble. These findings provide increased understanding of gambling and inform the development of gambling interventions based on the RAM.     

Dynamics of social assistance: the Norwegian experience in comparative perspective

In the article we ask how long Norwegian recipients of social assistance stay on aid over an eight‐year period? We focus on the populations selected, the design of the study and how spells are recorded and measured. The data set includes nearly the entire 1995 cohort of social assistance claimants (n = 155,000), and contains individual information over eight years (1992–1999), which is the maximum length of a single continuous spell. The study combines a retrospective and a prospective multiple cohort design. The data analysis shows that there is no simple answer to our question; the answer depends on the sample selected, the design, number of spells counted and the chosen measure of central tendency. We find that median duration times vary from two months to 40 months. The conclusion is that a point‐in‐time sample has clearly the longest duration. It makes a big difference whether one measures one or more spells. Because cycling is a common phenomenon, the sum of all spell durations is much higher than that of the first continuous spell. A retrospective design severely underestimates spell durations since many spells are right‐hand censored. As there is no single, adequate answer to our initial question, it is recommended that analysts of social assistance dynamics carefully specify their sample, design and spells counted.     

Partnerships with children in child welfare: The importance of trust and pedagogical support

Within the context of Norwegian Child Welfare Services, children's best interests are often promoted through inter‐professional collaboration. Although children have the right and desire to participate, research reveals that professionals do not listen to them. On the basis of qualitative interviews with 10 children about their experiences collaborating with professionals, we have identified ways in which professionals can facilitate children's participation. The findings show that trusting relationships, emotional support, and pedagogical approaches increase children's participation in their interactions with professionals. The results show the importance of including a relational understanding of participation as a theoretical concept in child welfare and an awareness that power and dominance are in play.