Learning discrete decomposable graphical models via constraint optimization

Statistical model learning problems are traditionally solved using either heuristic greedy optimization or stochastic simulation, such as Markov chain Monte Carlo or simulated annealing. Recently, there has been an increasing interest in the use of combinatorial search methods, including those based on computational logic. Some of these methods are particularly attractive since they can also be successful in proving the global optimality of solutions, in contrast to stochastic algorithms that only guarantee optimality at the limit. Here we improve and generalize a recently introduced constraint-based method for learning undirected graphical models. The new method combines perfect elimination orderings with various strategies for solution pruning and offers a dramatic improvement both in terms of time and memory complexity. We also show that the method is capable of efficiently handling a more general class of models, called stratified/labeled graphical models, which have an astronomically larger model space.     

Gender equality in ‘the most equal country in the world'? Money and marriage in Sweden1

This article analyses how Swedish couples perceive the sharing of money and consumption between themselves and their partner. Interviews were conducted with ten Swedish married couples. Each spouse was interviewed separately about their incomes, financial organization, patterns of consumption, views about money and decision-making. Regardless of whether they pooled their incomes or kept money separately, all were in agreement about the importance of equal sharing and access to money and consumption. Despite stated goals of gender equality, however, consumption was not perceived as being shared equally. Two factors central to understanding this were the ways that daily finances were managed and the fact that women had responsibility for the daily finances of the family. Another important aspect was the ways that items of consumption were defined. Food and children's clothes were areas that were in a ‘grey zone’ regarding which money was used to pay for them, and they often fell to the woman. This practical responsibility and associated awareness of the family economy serve as obstacles to women's sense of entitlement and access to money for personal discretionary spending, a problem not experienced by men.     

Assisted living: will it reduce long-term care costs?

In their quest to reduce nursing home care expenditures, the various states in this country have looked to assisted living as a potentially preferred and lower-cost housing alternative for their Medicaid patients. For an assisted-living program to save costs, states must recognize that some assisted-living residents will not come from nursing homes, but rather from private residences, resulting in cost increases. This article argues that this "woodwork effect"--new clients appearing--is likely to be smaller than the level reported in the home and community care demonstrations, but that the numbers are difficult to predict with the possibility of divestiture. It also argues that the true savings from substitution, or of one form of care for another, depend on the nursing home reimbursement system in effect at the time.     

Work and Non-Pathological Gambling

Most economists believe that people would value an additional $1,000 in income more if they were poor than if rich, but if so, people should not gamble according to standard expected utility theory. Thus, economists have been challenged to explain the pervasiveness of gambling in human behavior. A recently proposed solution to this theoretical challenge (Nyman 2004; Nyman et al. in Journal of Socio-Economics 37:2492–2504, 2008) suggests that, because having to work for one’s income is a fact of life in market economies, many individuals view the winnings from gambling not only as additional income, but as additional income for which one does not need to work. As a result, individuals, and especially those who are disadvantaged in the labor market, attach a utility premium to gambling winnings and gamble because of that. This utility premium would explain the pervasiveness of gambling in society, especially among the economically disadvantaged. This paper reviews the economic approaches to explaining non-pathological gambling, presents an overview of the new theory, and uses data from the National Epidemiological Survey of Alcohol and Related Conditions from 2001 to test it. The results indicate that the respondent’s work characteristics explain the decision to gamble in a way that is consistent with theory.     

Structure Learning of Contextual Markov Networks using Marginal Pseudo‐likelihood

Markov networks are popular models for discrete multivariate systems where the dependence structure of the variables is specified by an undirected graph. To allow for more expressive dependence structures, several generalizations of Markov networks have been proposed. Here, we consider the class of contextual Markov networks which takes into account possible context‐specific independences among pairs of variables. Structure learning of contextual Markov networks is very challenging due to the extremely large number of possible structures. One of the main challenges has been to design a score, by which a structure can be assessed in terms of model fit related to complexity, without assuming chordality. Here, we introduce the marginal pseudo‐likelihood as an analytically tractable criterion for general contextual Markov networks. Our criterion is shown to yield a consistent structure estimator. Experiments demonstrate the favourable properties of our method in terms of predictive accuracy of the inferred models.     

High incidence of disability pension with a psychiatric diagnosis in western Sweden. A population-based study from 1980 to 1998

Regional differences in Sweden in the prevalence of disability pension with a psychiatric diagnosis are unexplained, in spite of the significant impact on the population's health, rehabilitation systems, and the health care system. The purpose of this study was to describe the pattern of disability pensions with a psychiatric diagnosis and to analyze the impact of age and gender. We examined the incidence rates in one urban and one semi-rural region and compared these to national rates. The study sample was drawn from employed persons between 16-64 years of age who, because of their sickness insurance coverage, would be eligible to access disability pensions should it be necessary. Analysis of annual incidences and standardized morbidity ratios were made for 1980, 1985, 1990, 1995, and 1998. Data on disability pension cases were collected from the National Social Insurance registers. In the urban region we found that the proportion of men and women clearly outnumbered the national average: approximately twice the number of persons between 16-64 years of age with a psychiatric diagnosis were receiving a disability pension. In the semi-rural region there were fewer men overall on disability pensions with psychiatric disorders, but in 1980, 1985, and 1995 women clearly outnumbered men. Access to psychiatric care, unemployment, alcohol dependence, and previous sickness absence are suggested as possible factors that might affect the rates of disability pension in different geographical settings.     

Does Telephone Audio Computer-Assisted Self-Interviewing Improve the Accuracy of Prevalence Estimates of Youth Smoking? Evidence from the UMass Tobacco Study

Despite their advantage for obtaining representative samplesof adolescents, telephone surveys have been regarded as an inferiormethod for collecting data on youth tobacco use because theyyield lower estimates than school-based, self-administered surveys.Although no gold standard for smoking estimates exists, thelower estimates in telephone surveys have been attributed tounderreporting due to youths’ concern that parents orothers may overhear their responses. Telephone audio computer-assistedself-interviewing (T-ACASI) is a cost-effective method for obtaininga representative sample of youths and provides increased privacyfor the respondent. We hypothesize that using T-ACASI wouldencourage youths to report more fully smoking behavior comparedto traditional interviewer-administered telephone methods. Ouranalysis further assesses whether respondent age, gender, race/ethnicity,and parental attitude toward smoking moderated the relationshipbetween survey mode and smoking reports. Using data from a statewidetobacco use survey that randomly assigned youth respondentsto either T-ACASI or interviewer-administered modes, we findthat youths were more likely to report smoking behaviors inT-ACASI mode and that this was especially true for girls, particularlythose who believed their parents would disapprove strongly oftheir smoking. The findings suggest that traditional telephonesurveys may underestimate smoking prevalence in most girls bya factor of two, and that a technique for ensuring privacy forthese respondents is an important component of effective telephonesurvey methodology.     

Maternal neural reactivity during pregnancy predicts infant temperament

Maternal biological systems impact infant temperament as early as the prenatal period, though the mechanisms of this association are unknown. Using a prospective, longitudinal design, we found that maternal (N = 89) amplitudes of the late positive potential (LPP) in response to negative stimuli during the second, but not the third, trimester of pregnancy predicted observed and physiological indices of temperamental reactivity in infants at age 4 months. Maternal LPP was positively associated with observed infant fear and negatively associated with frontal EEG asymmetry and cortisol reactivity in infants at age 4 months. Results identify a putative mechanism, early in pregnancy, for the intergenerational transmission of emotional reactivity from mother to infant.     

Stratified Gaussian graphical models

Gaussian graphical models represent the backbone of the statistical toolbox for analyzing continuous multivariate systems. However, due to the intrinsic properties of the multivariate normal distribution, use of this model family may hide certain forms of context-specific independence that are natural to consider from an applied perspective. Such independencies have been earlier introduced to generalize discrete graphical models and Bayesian networks into more flexible model families. Here, we adapt the idea of context-specific independence to Gaussian graphical models by introducing a stratification of the Euclidean space such that a conditional independence may hold in certain segments but be absent elsewhere. It is shown that the stratified models define a curved exponential family, which retains considerable tractability for parameter estimation and model selection.     

Something for nothing: A model of gambling behavior

Gambling is an ancient economic activity, but despite its universality and importance, no single explanation for the demand for gambles has gained ascendance among economists. This paper suggests that the demand for gambles is based on the ability to obtain “something for nothing.” That is, the gain from gambling is not merely additional income, but additional income for which the gambler does not need to work. Thus, to fully understand gambling behavior, it must be placed in a labor supply context. The theory is tested empirically using the Survey of Gambling in the U.S. Support for the theory is found.