Search

Progress toward the United Nations Sustainable Development Goals (SDGs) has been hindered by a lack of data on key environmental and socioeconomic indicators, which historically have come from ground surveys with sparse temporal and spatial coverage. Recent advances in machine learning have made it possible to utilize abundant, frequently-updated, and globally available data, such as from satellites or social media, to provide insights into progress toward SDGs. Despite promising early results, approaches to using such data for SDG measurement thus far have largely evaluated on different datasets or used inconsistent evaluation metrics, making it hard to understand whether performance is improving and where additional research would be most fruitful. Furthermore, processing satellite and ground survey data requires domain knowledge that many in the machine learning community lack. In this paper, we introduce SustainBench, a collection of 15 benchmark tasks across 7 SDGs, including tasks related to economic development, agriculture, health, education, water and sanitation, climate action, and life on land. Datasets for 11 of the 15 tasks are released publicly for the first time. Our goals for SustainBench are to (1) lower the barriers to entry for the machine learning community to contribute to measuring and achieving the SDGs; (2) provide standard benchmarks for evaluating machine learning models on tasks across a variety of SDGs; and (3) encourage the development of novel machine learning methods where improved model performance facilitates progress towards the SDGs., Comment: NeurIPS 2021 (Track on Datasets and Benchmarks)

Accurate and comprehensive measurements of a range of sustainable development outcomes are fundamental inputs into both research and policy. We synthesize the growing literature that uses satellite imagery to understand these outcomes, with a focus on approaches that combine imagery with machine learning. We quantify the paucity of ground data on key human-related outcomes and the growing abundance and resolution (spatial, temporal, and spectral) of satellite imagery. We then review recent machine learning approaches to model-building in the context of scarce and noisy training data, highlighting how this noise often leads to incorrect assessment of models' predictive performance. We quantify recent model performance across multiple sustainable development domains, discuss research and policy applications, explore constraints to future progress, and highlight key research directions for the field.

It has been well established that the high schools students attend contribute significantly to how well they do in college, and even if they choose to attend college at all (Lee & Frank, 1990). Low performing high schools tend to under-prepare their students for post-secondary education and contribute to the very high percentages of students who arrive at college without the requisite skills to succeed in college level courses. The authors have chosen to study those community colleges that disproportionately transfer African American and Latino students from low-performing high schools because they believe there are important lessons to be learned from colleges that have successfully confronted this significant challenge, and because this is a rapidly growing sector of the school-age population across the nation and in California. The present study's design consists of several phases. Phase 1 examined the academic fates of students who entered the California community college system from low performing high schools (the lowest quintile on a composite measure of academic performance and SES) between 1996 and 1998 and who did so with the intention of transferring to a four-year college or university to earn a college degree. Transfer outcomes were assessed 6 years after they entered, between 2002 and 2004. Phase 2 of the study then consisted of conducting case studies of those 5 colleges that showed disproportionate success in transfer of the targeted students. The case studies consisted of multi-day visits to campuses over a period of time, interviews with each President, Vice-Presidents for Student and Academic Affairs, Counseling and Program Staffs, Transfer Center Directors and Financial Aid personnel, and selected faculty who were named as playing a significant role in the transfer function. Phase 3 of the study design involved surveying students, usually through electronic surveys, but sometimes in classrooms, at each of the five campuses, using whatever means were available to oversample Latinos and African Americans. Phase 4 of the study design then involved interviewing key individuals at feeder high schools and receiving universities, this generally included head counselors at the high schools, and outreach and counseling staff at the universities. In the case of the high schools, data collection was limited by the willingness of "low performing schools" to participate. Because this is a "moving picture" the authors are unable to capture all that was occurring at the time that these campuses experienced their successes with their cohorts, and they are unable to know the outcomes of newer innovations that they observed on these campuses in recent times. They are, however, able to make some observations that they think can help move the field forward in significant ways. Across the colleges they studied, five things were common and important to the success they observed. These are: (1) The colleges that showed disproportionate success in transferring African Americans and Latinos from low performing/high need high schools were not necessarily those with strong reputations for transfer; (2) Community college outreach was in many cases the reason that students came to the college in the first place, and connected with appropriate services once there; (3) Strong transfer counseling is the sine quo non of community college transfer, yet it is wholly inadequate and this is not always just because of resource limitations; (4) Every campus immediately pointed to its special support programs for underrepresented students as key to increasing its transfer rate for these students; and (5) Developmental education is the elephant in the living room for transfer of minority students from low performing/high need high schools. (Contains 11 tables, 25 figures and 18 footnotes.) [For the policy brief, "Building Pathways to Transfer: Community Colleges That Break the Chain of Failure for Students of Color. Policy Brief," see ED529494.]

This report is an effort to assess what has been accomplished in successfully diversifying and desegregating historically segregated and unequal higher educational institutions across Kentucky, and how this was done. The report discusses the changing legal setting for these initiatives, analyzes the educational achievements and challenges, and suggests how the progress can be continued and expanded in changing circumstances. Much has been accomplished, but educational inequality continues in some important ways. There are still massive gaps in the educational pipeline that threaten realization of the state's goals. Changing circumstances, both in law and in the state's demography, require new approaches that deal both with major vestiges of the historic discrimination against African Americans as well as equity for Latinos, whose numbers are now growing fast. Legal Principles for a Diversity Plan are appended. (Contains 13 tables, 2 figures and 234 footnotes.)

This report provides evidence that the first semester in community college is a pivotal point in students' academic careers. Students whose first academic experience in college is positive and successful are more likely to remain in school, and to sustain their aspiration to transfer to a four-year college. Those whose first experiences are less successful are far less likely to persist towards their goals. The majority of young adult high school graduates who entered community college in fall 1998 aspired to transfer to a four-year college to earn a bachelor's degree, but only a minority of these students made it past the first semester with their goals intact. Of the young adult high school graduates who entered college aspiring to transfer, most either left school after the first semester or lowered their educational expectations. One quarter did not return for the spring semester. Among those who returned, only a slight majority still had transferring to a four-year college as their primary educational goal. Overall, only four in ten returned for the spring semester and maintained their educational aspirations. One in three returned with lowered expectations. Ultimately, only 41.3 percent of this subset of students transferred to a four-year college or university. The report also shows that students who took a full courseload of transfer-eligible courses (four or more) in their first semester were far more likely to transfer than other students, a pattern found for every racial/ethnic group. Similarly, students with high first semester GPAs in transfer-eligible courses were more likely to transfer. (Contains 5 tables and 12 figures.)

This study examined transitions into and out of poverty and welfare across 4-year time periods and their implications for math and reading skills and behavior among 10- and 11-year-olds. Analyses of data from the National Longitudinal Survey of Youth-Child Supplement indicate that even with controls for factors that select families into poverty, children who do not experience poverty or welfare over this time period are advantaged relative to children who experience either. Children who are continuously poor but never receive welfare have more favorable outcomes than poor children who receive welfare. Among children experiencing changing economic circumstances, if the family manages to leave poverty, child outcomes are more positive; children whose families fall into poverty experience more negative outcomes than children living consistently above the poverty line. Fluctuations in family economic circumstances are also associated with poorer child outcomes. (Contains 32 references and 6 tables.) (JPB)

Receipt of welfare is often negatively correlated with children's outcomes. However, because virtually all children who live in households that receive public assistance are poor, the question arises whether poor child outcomes are truly an effect of welfare, have only a spurious relationship to welfare receipt, or are a result of welfare selection factors. This study used data from children in the National Longitudinal Survey of Youth-Child Supplement who were 9 to 14 years of age in 1992, and controlled for poverty and for selection onto welfare. The findings indicated that: (1) controlling for child and maternal characteristics accounted for the majority of bivariate negative associations between welfare and cognitive ability and behavior problems among black children; (2) controlling for poverty did little to change the negative relationship between welfare and measures of academic achievement and behavior problems for either blacks or whites; and (3) controlling for selection onto welfare, through a two-stage selection model, reduces, but does not eliminate the negative relationship between welfare receipt and outcomes among white children and has little discernible effects among black children. (Ten tables detail statistical information. Contains 26 references.) (Author/KB)