Program Evaluation and Causal Inference With High‐Dimensional Data

In this paper, we provide efficient estimators and honest confidence bands for a variety of treatment effects including local average (LATE) and local quantile treatment effects (LQTE) in data‐rich environments. We can handle very many control variables, endogenous receipt of treatment, heterogeneous treatment effects, and function‐valued outcomes. Our framework covers the special case of exogenous receipt of treatment, either conditional on controls or unconditionally as in randomized control trials. In the latter case, our approach produces efficient estimators and honest bands for (functional) average treatment effects (ATE) and quantile treatment effects (QTE). To make informative inference possible, we assume that key reduced‐form predictive relationships are approximately sparse. This assumption allows the use of regularization and selection methods to estimate those relations, and we provide methods for post‐regularization and post‐selection inference that are uniformly valid (honest) across a wide range of models. We show that a key ingredient enabling honest inference is the use of orthogonal or doubly robust moment conditions in estimating certain reduced‐form functional parameters. We illustrate the use of the proposed methods with an application to estimating the effect of 401(k) eligibility and participation on accumulated assets. The results on program evaluation are obtained as a consequence of more general results on honest inference in a general moment‐condition framework, which arises from structural equation models in econometrics. Here, too, the crucial ingredient is the use of orthogonal moment conditions, which can be constructed from the initial moment conditions. We provide results on honest inference for (function‐valued) parameters within this general framework where any high‐quality, machine learning methods (e.g., boosted trees, deep neural networks, random forest, and their aggregated and hybrid versions) can be used to learn the nonparametric/high‐dimensional components of the model. These include a number of supporting auxiliary results that are of major independent interest: namely, we (1) prove uniform validity of a multiplier bootstrap, (2) offer a uniformly valid functional delta method, and (3) provide results for sparsity‐based estimation of regression functions for function‐valued outcomes.     

The Possibility of Mixed-Mode Surveys in Sociological Studies

Abstract:  This paper explores the possibility of the mixed-mode survey in Japan in order to ensure a wider range of respondents, focusing on the specific type of mixture that is designed to make contact with a respondent by more than one mode and in which a respondent can choose an alternative mode convenient for him or her. The dataset mainly used comes from the survey Study on the Social Composition of the Urban Local Community conducted in Shinagawa Ward of Tokyo in 1999. Three other studies are introduced for additional information. As an indicator of the effectiveness of a mixed-mode design, the response rates of four studies are examined and some improvement in the response rates is observed, compared to that of a single-mode design. Another indicator is the representativeness of a sample and it is evaluated using the data of the Shinagawa survey. When the discrepancy between valid respondents and the population on some demographic characteristics is regarded as representation bias, a mixed-mode design is found to reduce the bias. An inherent problem in a mixed-mode design is "mode effect." In order to control for the mode effect in analyses, the inclusion of an additional variable indicating "mode" in the model is recommended. It is further suggested that the interaction effects between the "mode" and the respondents' traits should be checked, on the assumption that the survey mode affects respondents differently. The attempts are made and discussed in this paper using two variables drawn from the Shinagawa survey.     

The consequences of unobserved heterogeneity in a sequential logit model

Cameron and Heckman (1998) established that a sequential logit model is more sensitive than many other models to the possible biasing influence of unobserved heterogeneity. This article proposes a method which allows researchers to find out how large this potential problem is for their data, their model, and their hypothesis of interest. This is done by proposing a set of scenarios for this unobserved heterogeneity, and showing how to estimate the effects of interest given these scenarios. The set of results from these scenarios give an indication of how sensitive the results are to assumptions regarding unobserved heterogeneity. This sensitivity analysis has been applied to a study of educational attainment in the Netherlands, and it showed that that the finding that the effect of father's education declined over transitions is quite sensitive to the assumptions made about unobserved heterogeneity, but that the finding that the effect of father's education declined over birth cohorts is more robust than is often feared.     

A Weighted Composite Dose – Response Model For Human Salmonellosis

This article describes the development of a weighted composite dose – response model for human salmonellosis. Data from previously reported human challenge studies were categorized into two different groups representing low and moderately virulent/pathogenic Salmonella strains based on a disease end point. Because epidemiological data indicate that some Salmonella strains are particularly pathogenic, and in the absence of human feeding study data for such strains, Shigella dysenteriae was used as a proxy for highly virulent strains. Three single-hit dose – response models were applied to the human feeding study data and evaluated for best fit using maximum likelihood estimation: (1) the exponential (E-1pop), (2) the two-subpopulation exponential (E-2pop), and (3) the Beta-Poisson (BP). Based on the goodness-of-fit test, the E-1pop and BP were the best-fit models for low and moderately virulent/pathogenic Salmonella strains, and the E-2pop and BP models were better for highly virulent/pathogenic strains. Epistemic analysis was conducted by determining the degree of confidence associated with the selected models, which was found to be 50%/50% (E-1pop/BP) for low and moderately pathogenic Salmonella strains, and 9.8%/90.2% (E-2pop/BP) for highly virulent strains. The degree of confidence for each component model and variations in the proportion of strains within each virulence/pathogenicity category were incorporated into the overall composite model. This study describes the influence of variation in strain virulence and host susceptibility on the shape of the population dose – response relationship.     

Effect of perceived cost on judgments regarding the efficacy of investment

Three experiments examined the effect of costs associated with investments on the judgments of the causal effectiveness of those investments, and on the type of information utilised when making those judgements. Using operant conditioning technology, in an experimental analogue of making economic investments, it was found that greater costs associated with an investment reduced the judgement about the effectiveness of the response (Experiment 1). Greater investment costs tended to bias the subject toward using information integrated over a longer period of time; whereas low costs tended to produce judgements made on the basis of the conditions prevailing immediately prior to the return being acquired (Experiments 1–3). The effect of cost was modulated by the subjects' own financial background. The same absolute cost being treated as “greater” by those from less wealthy backgrounds compared to those from more affluent backgrounds (Experiments 2 and 3). These results suggest that as well as the nature of the schedule relating investment to return, the perceived cost of the investment determines the manner in which it is judged with respect to its efficacy.     

材料采购内部控制缺失风险及应对

从采购合同和采购成本两个方面分析公司材料采购内部控制缺失引致的风险及成因，提出材料采购风险的应对措施。     

Logistic与分类树模型变量筛选的比较——基于信用卡邮寄业务响应率分析

基于信用卡邮寄业务响应率分析来讨论Logistic模型和分类树模型在变量选取上的区别,并尝试从几个不同角度去解释两类模型变量筛选差异的原因。笔者认为没有绝对占优势的方法,需要结合具体场景和模型的特点来选择合适的模型。分类树模型在训练集上容易过度拟合,对单个变量的影响很敏感,在进行危险因素分析时结果更能强调危险因素,对孤立点的识别率很高。Logistic模型容易受到解释变量依存关系的影响,加上分类变量的影响容易过多地选入变量或者因子,对孤立点敏感,对噪点不敏感。判别函数的差异是变量筛选差异的关键因素。     

Book Reviews

Books reviewed:
Office International Des Epizooties (World Organization for Animal Health) World Animal Health in 1998, Part 1: Reports on the Animal Health Status and Disease Control Methods and Tables on Incidence of List A Diseases and Part 2: Tables on the Animal Health Status and Disease Control Methods
Peter Sedlmeier. Lawrence Erlbaum Associates, Mahwah Improving Statistical Reasoning: Theoretical Models and Practical Implications
Andrew Ford. Island Press Modeling the Environment. An Introduction to System Dynamics Modeling of Environmental Systems
Jacob I. Bregman. Lewis Publishers, Boca Raton Environmental Impact Statements, Second Edition     

Challenging passive partnerships with parents and children in the core group forum: a framework for a proactive approach

This paper looks at two concepts central to effective child protection practice: partnerships and the core group system, which need to work together if we are to improve the worrying, ad hoc approaches to child protection work once a child's name has been added to the child protection register. These concepts emerged at around the same point in time, but little has been written about the core group, even though it has been given the key tasks of formulating, implementing and reviewing the child protection plan, and forging effective working partnerships with other professionals, parents/carers and their children ( Calder 1991, 1995 b; Calder & Horwath 1999 ). Whilst the concept of partnership has received considerable media and publishing attention, no previous paper has explored partnerships with families within the core group phase. This reflects the low priority currently afforded to activity following the initial child protection conference, and arguably sets the core group up to fail, even before it has started ( Calder 1996a, 1998 ). The authors draw on their national research into core group practice to explore the notion of partnerships between the core group and parents, carers and children, concluding with some recommendations on how Area Child Protection Committees can create the environment in which we can move from passive to proactive partnerships within a higher profile for this stage and focus of the work. The authors draw on some suggestions from Messages from Research ( Department of Health 1995a ) and from several small-scale research studies into core group practice.     

Cellular Automata‐Based Systematic Risk Analysis Approach for Emergency Response

Emergency response is directly related to the allocation of emergency rescue resources. Efficient emergency response can reduce loss of life and property, limit damage from the primary impact, and minimize damage from derivative impacts. An appropriate risk analysis approach in the event of accidents is one rational way to assist emergency response. In this article, a cellular automata‐based systematic approach for conducting risk analysis in emergency response is presented. Three general rules, i.e., diffusive effect, transporting effect, and dissipative effect, are developed to implement cellular automata transition function. The approach takes multiple social factors such as population density and population sensitivity into consideration and it also considers risk of domino accidents that are increasing due to increasing congestion in industrial complexes of a city and increasing density of human population. In addition, two risk indices, i.e., individual risk and aggregated weighted risk, are proposed to assist decision making for emergency managers during emergency response. Individual risk can be useful to plan evacuation strategies, while aggregated weighted risk can help emergency managers to allocate rescue resources rationally according to the degree of danger in each vulnerable area and optimize emergency response programs.