Search

Statistical profiling of job seekers is an attractive option to guide the activities of public employment services. Many hope that algorithms will improve both efficiency and effectiveness of employment services’ activities that are so far often based on human judgment. Against this backdrop, we evaluate regression and machine-learning models for predicting job-seekers’ risk of becoming long-term unemployed using German administrative labor market data. While our models achieve competitive predictive performance, we show that training an accurate prediction model is just one element in a series of design and modeling decisions, each having notable effects that span beyond predictive accuracy. We observe considerable variation in the cases flagged as high risk across models, highlighting the need for systematic evaluation and transparency of the full prediction pipeline if statistical profiling techniques are to be implemented by employment agencies.

Abstract Background Guidelines and recommendations from public health authorities related to face masks have been essential in containing the COVID-19 pandemic. We assessed the prevalence and correlates of mask usage during the pandemic. Methods We examined a total of 13,723,810 responses to a daily cross-sectional online survey in 38 countries of people who completed from April 23, 2020 to October 31, 2020 and reported having been in public at least once during the last 7 days. The outcome was individual face mask usage in public settings, and the predictors were country fixed effects, country-level mask policy stringency, calendar time, individual sociodemographic factors, and health prevention behaviors. Associations were modeled using survey-weighted multivariable logistic regression. Results Mask-wearing varied over time and across the 38 countries. While some countries consistently showed high prevalence throughout, in other countries mask usage increased gradually, and a few other countries remained at low prevalence. Controlling for time and country fixed effects, sociodemographic factors (older age, female gender, education, urbanicity) and stricter mask-related policies were significantly associated with higher mask usage in public settings. Crucially, social behaviors considered risky in the context of the pandemic (going out to large events, restaurants, shopping centers, and socializing outside of the household) were associated with lower mask use. Conclusion The decision to wear a face mask in public settings is significantly associated with sociodemographic factors, risky social behaviors, and mask policies. This has important implications for health prevention policies and messaging, including the potential need for more targeted policy and messaging design.

Prediction algorithms are regularly used to support and automate high-stakes policy decisions about the allocation of scarce public resources. However, data-driven decision-making raises problems of algorithmic fairness and justice. So far, fairness and justice are frequently conflated, with the consequence that distributive justice concerns are not addressed explicitly. In this paper, we approach this issue by distinguishing (a) fairness as a property of the algorithm used for the prediction task from (b) justice as a property of the allocation principle used for the decision task in data-driven decision-making. The distinction highlights the different logic underlying concerns about fairness and justice and permits a more systematic investigation of the interrelations between the two concepts. We propose a new notion of algorithmic fairness called error fairness which requires prediction errors to not differ systematically across individuals. Drawing on sociological and philosophical discourse on local justice, we present a principled way to include distributive justice concerns into data-driven decision-making. We propose that allocation principles are just if they adhere to well-justified distributive justice principles. Moving beyond the one-sided focus on algorithmic fairness, we thereby make a first step toward the explicit implementation of distributive justice into data-driven decision-making.

Objectives: Real-time data analysis during a pandemic is crucial. This paper aims to introduce a novel interactive tool called Covid-Predictor-Tracker using several sources of COVID-19 data, which allows examining developments over time and across countries. Exemplified here by investigating relative effects of vaccination to non-pharmaceutical interventions on COVID-19 spread.Methods: We combine >100 indicators from the Global COVID-19 Trends and Impact Survey, Johns Hopkins University, Our World in Data, European Centre for Disease Prevention and Control, National Centers for Environmental Information, and Eurostat using random forests, hierarchical clustering, and rank correlation to predict COVID-19 cases.Results: Between 2/2020 and 1/2022, we found among the non-pharmaceutical interventions “mask usage” to have strong effects after the percentage of people vaccinated at least once, followed by country-specific measures such as lock-downs. Countries with similar characteristics share ranks of infection predictors. Gender and age distribution, healthcare expenditures and cultural participation interact with restriction measures.Conclusion: Including time-aware machine learning models in COVID-19 infection dashboards allows to disentangle and rank predictors of COVID-19 cases per country to support policy evaluation. Our open-source tool can be updated daily with continuous data streams, and expanded as the pandemic evolves.

Objectives: To examine the association of non-pharmaceutical interventions (NPIs) with anxiety and depressive symptoms among adults and determine if these associations varied by gender and age.Methods: We combined survey data from 16,177,184 adults from 43 countries who participated in the daily COVID-19 Trends and Impact Survey via Facebook with time-varying NPI data from the Oxford COVID-19 Government Response Tracker between 24 April 2020 and 20 December 2020. Using logistic regression models, we examined the association of [1] overall NPI stringency and [2] seven individual NPIs (school closures, workplace closures, cancellation of public events, restrictions on the size of gatherings, stay-at-home requirements, restrictions on internal movement, and international travel controls) with anxiety and depressive symptoms.Results: More stringent implementation of NPIs was associated with a higher odds of anxiety and depressive symptoms, albeit with very small effect sizes. Individual NPIs had heterogeneous associations with anxiety and depressive symptoms by gender and age.Conclusion: Governments worldwide should be prepared to address the possible mental health consequences of stringent NPI implementation with both universal and targeted interventions for vulnerable groups.

Academic and public debates are increasingly concerned with the question whether and how algorithmic decision-making (ADM) may reinforce social inequality. Most previous research on this topic originates from computer science. The social sciences, however, have huge potentials to contribute to research on social consequences of ADM. Based on a process model of ADM systems, we demonstrate how social sciences may advance the literature on the impacts of ADM on social inequality by uncovering and mitigating biases in training data, by understanding data processing and analysis, as well as by studying social contexts of algorithms in practice. Furthermore, we show that fairness notions need to be evaluated with respect to specific outcomes of ADM systems and with respect to concrete social contexts. Social sciences may evaluate how individuals handle algorithmic decisions in practice and how single decisions aggregate to macro social outcomes. In this overview, we highlight how social sciences can apply their knowledge on social stratification and on substantive domains of ADM applications to advance the understanding of social impacts of ADM.

The COVID-19 pandemic and associated government lockdown restrictions have fueled a high demand for survey data on how individuals and establishments are coping with the restrictions. However, the pandemic has also dramatically affected surveys themselves, forcing research institutes to adapt their fieldwork operations to the uncertain and evolving conditions. This paper documents the specific impacts of the pandemic on several ongoing surveys at the Institute for Employment Research in Germany, and describes how these impacts have been addressed. In addition, this paper summarizes efforts to rapidly develop new Covid-19 surveys to help meet the increasing demand for timely data about the effects of the Corona-Crisis on the labor market.

This paper describes a partnership between Facebook and academic institutions to create a global COVID-19 symptom survey. The survey is available in 56 languages. A representative sample of Facebook users is invited on a daily basis to report on symptoms, social distancing behavior, mental health issues, and financial constraints. Facebook provides weights to reduce nonresponse and coverage bias. Privacy protection and disclosure avoidance mechanisms are implemented by both partners to meet global policy and industry requirements. Country and region-level statistics are published daily via dashboards, and microdata are available for researchers via data use agreements. Over 1 million responses are collected weekly.

Survey researchers are increasingly seeking opportunities to link interview data with administrative records. However, obtaining consent from all survey respondents (or certain subgroups) remains a barrier to performing record linkage in many studies. We experimentally investigated whether emphasizing different benefits of record linkage to respondents in a telephone survey of employee working conditions improves respondents’ willingness to consent to linkage of employment administrative records relative to a neutral consent request. We found that emphasizing linkage benefits related to “time savings” yielded a small, albeit statistically significant, improvement in the overall linkage consent rate (86.0) relative to the neutral consent request (83.8 percent). The time savings argument was particularly effective among “busy” respondents. A second benefit argument related to “improved study value” did not yield a statistically significant improvement in the linkage consent rate (84.4 percent) relative to the neutral request. This benefit argument was also ineffective among the subgroup of respondents considered to be most likely to have a self-interest in the study outcomes. The article concludes with a brief discussion of the practical implications of these findings and offers suggestions for possible research extensions.

Predictive modeling methods from the field of machine learning have become a popular tool across various disciplines for exploring and analyzing diverse data. These methods often do not require specific prior knowledge about the functional form of the relationship under study and are able to adapt to complex non-linear and non-additive interrelations between the outcome and its predictors while focusing specifically on prediction performance. This modeling perspective is beginning to be adopted by survey researchers in order to adjust or improve various aspects of data collection and/or survey management. To facilitate this strand of research, this paper (1) provides an introduction to prominent tree-based machine learning methods, (2) reviews and discusses previous and (potential) prospective applications of tree-based supervised learning in survey research, and (3) exemplifies the usage of these techniques in the context of modeling and predicting nonresponse in panel surveys.

Trust is praised by many social scientists as the foundation of functioning social systems owing to its assumed connection to cooperative behavior. The existence of such a link is still subject to debate. In the present study, we first highlight important conceptual issues within this debate. Second, we examine previous evidence, highlighting several issues. Third, we present findings from an original experiment, in which we tried to identify a "real" situation that allowed us to measure both trust and cooperation. People's expectations and behavior when they decide to share (or not) their data represents such a situation, and we make use of corresponding data. We found that there is no relationship between trust and cooperation. This non-relationship may be rationalized in different ways which, in turn, provides important lessons for the study of the trust-behavior nexus beyond the particular situation we study empirically.

Looping questions are used to collect data about several similar events, such as employment spells, retirement accounts, or marriages. The loops gather information about the number of events experienced as well as details about each one. The questions require respondents to think hard to recall each event and are often lengthy and repetitive. Looping questions can be asked in two formats, and which format a survey uses may aect the quality of the data collected. We develop hypotheses about the eects of format on measurement error in looping questions and test the hypotheses using experimental data from a web survey with a link to administrative records. Results show that one format collects more accurate event reports, but the other format provides higher quality data to the follow up questions. We conclude with guidance for those who write survey questions as well as those who rely on survey data for substantive analyses.

Abstracts not available for BookReviews

Abstracts not available for BookReviews

Record linkage is becoming more important as survey budgets are tightening while at the same time demands for more statistical information are rising. Not all respondents consent to linking their survey answers to administrative records, threatening inferences made from linked data sets. So far, several studies have identified respondent-level attributes that are correlated with the likelihood of providing consent (e.g., age, education), but these factors are outside the control of the survey designer. In the present study three factors that are under the control of the survey designer are evaluated to assess whether they impact respondents' likelihood of linkage consent: 1) the wording of the consent question; 2) the placement of the consent question and; 3) interviewer attributes (e.g., attitudes toward data sharing and consent, experience, expectations). Data from an experiment were used to assess the impact of the first two and data from an interviewer survey that was administered prior to the start of data collection are used to examine the third. The results show that in a telephone setting: 1) indicating time savings in the wording of the consent question had no effect on the consent rate; 2) placement of the consent question at the beginning of the questionnaire achieved a higher consent rate than at the end and; 3) interviewers' who themselves would be willing to consent to data linkage requests were more likely to obtain linkage consent from respondents.

Administrative records are increasingly being linked to survey records to highten the utility of the survey data. Respondent consent is usually needed to perform exact record linkage; however, not all respondents agree to this request and several studies have found significant differences between consenting and non-consenting respondents on the survey variables. To the extent that these survey variables are related to variables in the administrative data, the resulting administrative estimates can be biased due to non-consent. Estimating non-consent biases for linked administrative estimates is complicated by the fact that administrative records are typically not available for the non-consenting respondents. The present study can overcome this limitation by utilizing a unique data source, the German Panel Study "Labour Market and Social Security" (PASS), and linking the consent indicator to the administrative records (available for the entire sample). This situation permits the estimation of non-consent biases for administrative variables and avoids the need to link the survey responses. The impact of non-consent bias can be assessed relative to other sources of bias (nonresponse, measurement) for several administrative estimates. The results show that non-consent biases are present for few estimates, but are generally small relative to other sources of bias.

Education researchers who study the effect of family social background on student achievement often use students’ survey reports of parental education to investigate these effects. However, past research has demonstrated that students misreport their parents’ education levels. We expand upon this research in two ways. First we use cognitive theories about the response process to develop and test hypotheses about reporting inconsistencies across these variables. Second we evaluate the impact of student misreporting on estimates of the relationship between parental education levels and student math achievement. Using data from the German administration of PISA 2000 (OECD Programme for International Student Assessment) in which both students and parents were asked to report parental variables, we show that reporting inconsistencies are a function of student achievement: students with higher math scores tend to provide reports that are more consistent with their parents' reports. This interesting case of differential measurement error has consequences for comparisons of the effects of parental background on student achievement across different subgroups of the population and across countries (a common use of PISA data and other international studies similar to PISA).

Abstracts not available for BookReviews

Data used in nationwide face-to-face surveys are always collected in multistage cluster samples. The relative homogeneity of the clusters selected in this way can lead to design effects at the sampling stage. Interviewers can further homogenize answers within the sampling points. The study presented here was designed to separate between interviewer effects and sampling-point effects using interpenetrated samples for conducting a nationwide survey on fear of crime. Even though one might, given the homogeneity of neighborhoods, assume that sampling-point effects would be especially strong for questions related to fear of crime in one.

Objectives. To assess the impact of the COVID-19 pandemic on mental distress in US adults. Methods. Participants were 5065 adults from the Understanding America Study, a probability-based Internet panel representative of the US adult population. The main exposure was survey completion date (March 10-16, 2020). The outcome was mental distress measured via the 4-item version of the Patient Health Questionnaire. Results. Among states with 50 or more COVID-19 cases as of March 10, each additional day was significantly associated with an 11% increase in the odds of moving up a category of distress (odds ratio = 1.11; 95% confidence interval = 1.01, 1.21; P  = .02). Perceptions about the likelihood of getting infected, death from the virus, and steps taken to avoid infecting others were associated with increased mental distress in the model that included all states. Individuals with higher consumption of alcohol or cannabis or with history of depressive symptoms were at significantly higher risk for mental distress. Conclusions. These data suggest that as the COVID-19 pandemic continues, mental distress may continue to increase and should be regularly monitored. Specific populations are at high risk for mental distress, particularly those with preexisting depressive symptoms.