Variable Selection for Qualitative Interactions

In this article we discuss variable selection for decision making with focus on decisions regarding when to provide treatment and which treatment to provide. Current variable selection techniques were developed for use in a supervised learning setting where the goal is prediction of the response. These techniques often downplay the importance of interaction variables that have small predictive ability but that are critical when the ultimate goal is decision making rather than prediction. We propose two new techniques designed specifically to find variables that aid in decision making. Simulation results are given along with an application of the methods on data from a randomized controlled trial for the treatment of depression.     

Theoretical stances and environmental debates: Reconciling the physical and the symbolic

Several sociologists arc currently debating the relationship of sociology to the physical environment. Their debates beg a question of more general importance to sociology: How do we organize our thinking about phenomena that are at once physical or material and symbolic or ideal? Our intention is not to add another voice in favor of or opposed to theorizing material, physical, or organic characteristics, but to examine the process of thinking about environments and more generally the realist-idealist divide. Environment (like the body) is unlike typical social science concepts in so far as it is both physical and social. If, for example, status and role are purely social concepts, environment is always more than social. I low do sociologists approach what is always more than social in the study of physical environments? Theorizing environments, we propose, is fashioned by the analytic stance of the investigator as legislative, interpretive, or symbolic realist. The strengths and weaknesses of these stances are discussed, and throughout our discussion empirical work representing each ol them is introduced. A final inquiry examines how sociologists can approach these three stances. Two strategies are identified: to assume each stance mirrors the environment as it actually exists or assume the stances are terminologies for exploring various combinations of the physical-symbolic properties ol environments. A brief plea is made for the second strategy.     

Noisy Winter: The DDT Controversy in the Years before Silent Spring

Abstract In this paper, we examine three unanticipated findings from a social constructionist analysis of popular media coverage of the pesticide DDT from the years 1944 to 1961. The first unanticipated finding was the early (1945) appearance of negative or cautionary claims in the media source examined, the New York Times. Second, while negative or cautionary claims about the pesticide did constitute a minority voice during this time period, it was nonetheless a persistent voice. The third unanticipated finding was the predominance of the U.S. Department of Agriculture, and the State Agricultural Experiment Stations among those claimsmakers initially cautioning potential users about unintended and potentially deleterious impacts. The concept of "routine monitoring mechanisms" is introduced to explain this third finding. We conclude by considering the potential impact of this coverage on the subsequent development of the controversy.     

Race,SES, and Obesity Among Men

Over the last decade, obesity has increased significantly among men but few national studies have empirically examined racial and socioeconomic differences in obesity among men. In this paper, we utilized logistic regression to evaluate the potential associations that race and socioeconomic status may have with obesity among men in the National Survey of American Life: an in-person household survey of non-institutionalized U.S. blacks and whites who lived in communities where at least 10% of the community residents were black Americans. A greater proportion of black men were likely to be obese than white men, but no interaction among race, SES, and obesity was detected when potential confounding variables were included. There was not a relationship between SES and obesity for white men, but there was an apparent positive relationship between SES and obesity for black men that did not remain significant in adjusted models. No relationship was found between age and obesity among black men, though white men who were 55 and older were more likely than those 18–34 to be obese in confounder adjusted models. Among white men, no relationships were found between obesity and education, household income, or marital status. Black men in the lowest income category were less likely to be obese than those in the highest income category, in bivariate but not adjusted models. These findings suggest that the way racial, economic, stress and behavioral factors combine to affect obesity in black and white men may be different.     

Log-ratios and the logarithmic mean

Logarithmically transformed ratios have an important role to play in some branches of statistics and economics. Focus of this paper is on an unfamiliar interpretation of these logratios, which are in fact absolute differences being divided by the logarithmic mean. It is demonstrated that familiarity with the logarithmic mean and its relation to more common mean values leads to view a problem from a new angle whenever logratios are involved. Examples investigated are Fisher’s z-transformation, the likelihood-ratio test of goodness-of-fit, Theil’s measure of inequality, and the log-ratio itself as one important device for measuring relative change within timeseries.