BIGDATA: Collaborative Research: IA: Population Reproduction of Poverty at Birth from Surveys, Censuses, and Birth Registrations

一代一代向上流动的潜力，对一个民主社会的福祉以及该社会的长期经济生产力至关重要。超过四分之一的美国儿童受出生时贫困的影响。它是儿童今后发展和生活机会的一个关键障碍，因此也是打破代际不利循环的一个关键障碍。母亲和父亲的特点都会影响孩子出生在贫困家庭的几率。因此，重要的是将母亲和父亲都纳入从一代到下一代出生时的贫困经历的科学模型。现有的数据和方法不允许在美国人口的代表性样本中进行这种建模，特别是当它发展成为一个日益多民族的社会时。关于美国每一个新生儿的特征收集的数据有限。人口普查每10年收集一次关于所有个人及其家庭的有限额外数据。每年从代表美国人口的随机样本中收集大量的个人和家庭数据。这些样本包括大规模的美国社区调查和众多的中小型样本调查。本研究开发并评估了统计估计和模拟方法，以结合所有这些来源的数据来回答有关代际流动的问题。这项研究将回答以下问题：在不同的种族和民族群体中，从一代到下一代出生时贫困的持续程度，以及教育和家庭形成在创造向上流动与持续劣势方面的作用。该项目开发的统计方法的开源、用户友好的软件将提供给研究人员。该项目还培养研究生在统计学和社会科学方面的大数据方法技能。该项目的目标是发展一种变革性的大数据方法，利用丰富的“传统”数据源，以统计严谨和经验全面的方式建立社会科学理论。在没有一个具有全国代表性的单一数据来源中观察到母亲-父亲-孩子三位一体的出生时贫困。相反，在一个模拟四个相互关联过程的模型中，三个人在每个人出生时的贫困状况是联系在一起的：(1)根据家庭贫困的出生条件和父母的教育、种族、民族和移民身份预测的教育进步；(2)夫妻的成立和解除；（三）夫妻生育和未婚妇女生育；(4)自己的孩子出生后的家庭贫困。该项目的“大数据”包括超过1亿的出生数据，结合人口普查、微观人口普查、大规模横断面调查以及中小规模纵向调查数据源，这些数据源包括数百万年的学校教育、伴侣配对以及伴侣和（共同居住）非伴侣生育。与许多大数据应用不同，该研究的这些多个数据源要么是完整的枚举，要么是一个明确规定的群体的概率样本。该项目开发了联合调查和人口与调查相结合的估计方法，以提高对过程中个体行为参数的精确估计，并开发了一种模拟建模方法，以产生关于四个相连过程中出现的因果关系的推断，整合了每个组成过程的多个不确定性来源。结果和方法上的进步将通过演讲和同行评议的期刊文章传播给学术界。此外，还将通过项目调查员与新闻媒体和其他论坛的接触，使更广泛的政策团体了解项目结果及其意义，从而进行更广泛的传播。

The potential for upward mobility from one generation to the next is fundamental to the wellbeing of a democratic society, and to that society's long-term economic productivity. Poverty at birth is a condition affecting more than 1 in 4 American children. It is a critical barrier to children's subsequent development and life chances, and therefore to breaking intergenerational cycles of disadvantage. Both mother's and father's characteristics contribute to the chance that a child will be born into poverty. Therefore it is important to incorporate both the mother and father into the scientific modeling of the experience of poverty at birth from one generation to the next. Existing data and methods do not allow for this modeling in representative samples of the U.S. population, especially as it evolves as an increasingly multi-ethnic society. Limited data are collected on the characteristics of every birth in the United States. Limited additional data on all individuals and their households are collected every 10 years in the Census. Larger amounts of data per individual and household are collected every year from random samples that are representative of the U.S. population. These samples include the large-scale American Community Survey and numerous medium- and small-scale sample surveys. This research develops and evaluates statistical estimation and simulation methods to combine data from all these sources to answer questions about intergenerational mobility. The research will answer questions about the degree of persistence of poverty at birth from one generation to the next across different race and ethnic groups, and about the roles of education and family formation in creating upward mobility versus persistence of disadvantage. Open-source, user-friendly software for the statistical methods developed in the project will be made available to researchers. The project also develops graduate students' skills in Big Data methods in statistics and the social sciences. The project has as its goal the development of a transformative, Big Data approach to exploiting the rich "traditional" data sources to build social-scientific theory in a statistically-rigorous and empirically-comprehensive way. In no single nationally-representative data source is poverty at birth observed for the mother-father-child triad. The triad's poverty statuses at each one's birth are instead linked in a model that simulates four connected processes: (1) educational progress predicted from the birth conditions of household poverty and parents' education, race, ethnicity, and immigrant statuses; (2) couple formation and dissolution; (3) couple fertility and unpartnered women's fertility; and (4) household poverty when their own children are born. The "Big Data" of this project consist of more than 100 million births, combined with census, microcensus, large-scale cross-sectional survey, and medium-scale and smaller-scale longitudinal survey data sources that together include millions of years of exposure to schooling, to partner-matching, and to partnered and (co-residentially) unpartnered births. These multiple sources of data of the study, unlike in many Big Data applications, are all either complete enumerations or probability samples of a well-specified population. The project develops combined-survey and combined population-and-survey estimation methods to estimate with enhanced precision the individual behavioral parameters of the process, and develops a simulation-modeling approach to generating inference about causal associations that emerge from the four connected processes, integrating the multiple sources of uncertainty about each component process. Results and methodological advances will be disseminated to the scholarly community through presentations and peer-reviewed journal articles. Additional, broader dissemination will occur through project investigator contact with the news media and other forums for engagement with the broader policy community about the project results and their significance.