Search

Abstract: Claiming causal inferences in network settings necessitates careful consideration of the often complex dependency between outcomes for actors. Of particular importance are treatment spillover or outcome interference effects. We consider causal inference when the actors are connected via an underlying network structure. Our key contribution is a model for causality when the underlying network is endogenous; where the ties between actors and the actor covariates are statistically dependent. We develop a joint model for the relational and covariate generating process that avoids restrictive separability and fixed network assumptions, as these rarely hold in realistic social settings. While our framework can be used with general models, we develop the highly expressive class of Exponential-family Random Network models (ERNM) of which Markov random fields and Exponential-family Random Graph models are special cases. We present potential outcome-based inference within a Bayesian framework and propose a modification to the exchange algorithm to allow for sampling from ERNM posteriors. We present results of a simulation study demonstrating the validity of the approach. Finally, we demonstrate the value of the framework in a case study of smoking in the context of adolescent friendship networks.

Networked populations consist of inhomogeneous individuals connected via relational ties. The individuals typically vary in multivariate attributes. In some cases primary interest focuses on individual attributes and in others the understanding of the social structure of the ties. In many circumstances both are of interest, as is their relationship. In this paper we consider this last, most general, case. We model the joint distribution of social ties and individual attributes when the population is only partially observed. Of central interest is when the population is surveyed using a network sampling design. A second situation is when data about a subset of the ties and/or the individual attributes is unintentionally missing. Exponential-family random network models (ERNM)s are capable of specifying a joint statistical representation of both the ties of a network and individual attributes. This class of models allow the nodal attributes to be modeled as stochastic processes, expanding the range and realism of exponential-family approaches to network modeling. In this paper we develop a theory of inference for ERNMs when only part of the network is observed, as well as specific methodology for partially observed networks, including non-ignorable mechanisms for network-based sampling designs. In particular, we consider data collected via contact tracing, of considerable importance to infectious disease epidemiology and public health.

Abstract: Many demographic problems require models for partnership formation. We consider a model for matchings within a bipartite population where individuals have utility for people based on observed and unobserved characteristics. It represents both the availability of potential partners of different types and the preferences of individuals for such people. We develop an estimator for the preference parameters based on sample survey data on partnerships and population composition. We conduct simulation studies based on the Survey of Income and Program Participation showing that the estimator recovers preference parameters that are invariant under different population availabilities and has the correct confidence coverage.

ProblemQuantifying mortality from coronavirus disease (COVID-19) is difficult, especially in countries with limited resources. Comparing mortality data between countries is also challenging, owing to differences in methods for reporting mortality.ContextTracking all-cause mortality (ACM) and comparing it with expected ACM from pre-pandemic data can provide an estimate of the overall burden of mortality related to the COVID-19 pandemic and support public health decision-making. This study validated an ACM calculator to estimate excess mortality during the COVID-19 pandemic.ActionThe ACM calculator was developed as a tool for computing expected ACM and excess mortality at national and subnational levels. It was developed using R statistical software, was based on a previously described model that used non-parametric negative binomial regression and was piloted in several countries. Goodness-of-fit was validated by forecasting 2019 mortality from 2015-2018 data.OutcomeThree key lessons were identified from piloting the tool: using the calculator to compare reported provisional ACM with expected ACM can avoid potential false conclusions from comparing with historical averages alone; using disaggregated data at the subnational level can detect excess mortality by avoiding dilution of total numbers at the national level; and interpretation of results should consider system-related performance indicators.DiscussionTimely tracking of ACM to estimate excess mortality is important for the response to COVID-19. The calculator can provide countries with a way to analyse and visualize ACM and excess mortality at national and subnational levels.

Exponential-family Random Graph models (ERGM) are widely used in social network analysis when modelling data on the relations between actors. ERGMs are typically interpreted as a snapshot of a network at a given point in time or in a final state. The recently proposed Latent Order Logistic model (LOLOG) directly allows for a latent network formation process. We assess the real-world performance of these models when applied to typical networks modelled by researchers. Specifically, we model data from an ensemble of articles in the journal Social Networks with published ERGM fits, and compare the ERGM fit to a comparable LOLOG fit. We demonstrate that the LOLOG models are, in general, in qualitative agreement with the ERGM models, and provide at least as good a model fit. In addition they are typically faster and easier to fit to data, without the tendency for degeneracy that plagues ERGMs. Our results support the general use of LOLOG models in circumstances where ERGMs are considered.

In stark contrast to the Arctic, there have been statistically significant positive trends in total Antarctic sea ice extent since 1979. However, the short and highly variable nature of observed Antarctic sea ice extent limits the ability to fully understand the historical context of these recent changes. To meet this challenge, we have created robust, observation-based reconstruction ensembles of seasonal Antarctic sea ice extent since 1905. Using these reconstructions, here we show that the observed period since 1979 is the only time all four seasons demonstrate significant increases in total Antarctic sea ice in the context of the twentieth century and that the observed increases are juxtaposed against statistically significant decreases throughout much of the early and middle twentieth century. These reconstructions provide reliable estimates of seasonally resolved total Antarctic sea ice extent and are skilful enough to better understand aspects of air–sea–ice interactions within the Antarctic climate system.

Abstract. Landfast sea ice (fast ice) is an important though poorly-understood component of the cryosphere on the Antarctic continental shelf, where it plays a key role in atmosphere-ocean-ice sheet interaction and coupled ecological and biogeochemical processes. Here, we present a first in-depth baseline analysis of variability and change in circum-Antarctic fast-ice distribution (including its relationship to bathymetry), based on a new high-resolution satellite-derived time series for the period 2000 to 2018. This reveals a) an overall trend of −882 ± 824 km²/y (−0.19 ± 0.18 %/y); and b) eight distinct regions in terms of fast-ice coverage and modes of formation. Of these, four exhibit positive trends over the 18 y period and four negative. Positive trends are seen in East Antarctica and in the Bellingshausen sea, with this region claiming the largest positive trend of +1,198 ± 359 km²/y (+1.10 ± 0.35 %/y). The four negative trends predominantly occur in West Antarctica, with the largest negative trend of −1,206 ± 277 km²/y (−1.78 ± 0.41 %/y) occurring in the Victoria and Oates Lands region in the eastern Ross Sea. All trends are significant. This new baseline analysis represents a significant advance in our knowledge of the current state of both the global cryosphere and the complex Antarctic coastal system that is vulnerable to climate variability and change. It will also inform a wide range of other studies.

Abstract. Landfast sea ice (fast ice) is an important though poorly understood component of the cryosphere on the Antarctic continental shelf, where it plays a key role in atmosphere–ocean–ice-sheet interaction and coupled ecological and biogeochemical processes. Here, we present a first in-depth baseline analysis of variability and change in circum-Antarctic fast-ice distribution (including its relationship to bathymetry), based on a new high-resolution satellite-derived time series for the period 2000 to 2018. This reveals (a) an overall trend of -882±824 km2 yr−1 (-0.19±0.18 % yr−1) and (b) eight distinct regions in terms of fast-ice coverage and modes of formation. Of these, four exhibit positive trends over the 18-year period and four negative. Positive trends are seen in East Antarctica and in the Bellingshausen Sea, with this region claiming the largest positive trend of +1198±359 km2 yr−1 (+1.10±0.35 % yr−1). The four negative trends predominantly occur in West Antarctica, with the largest negative trend of -1206±277 km2 yr−1 (-1.78±0.41 % yr−1) occurring in the Victoria and Oates Land region in the western Ross Sea. All trends are significant. This new baseline analysis represents a significant advance in our knowledge of the current state of both the global cryosphere and the complex Antarctic coastal system, which are vulnerable to climate variability and change. It will also inform a wide range of other studies.

We revisit a novel causal model published in Demography by Hicks et al. (2018), designed to assess whether exposure to neighborhood disadvantage over time affects children's reading and math skills. Here, we provide corrected and new results. Reconsideration of the model in the original article raised concerns about bias due to exposure-induced confounding (i.e., past exposures directly affecting future exposures) and true state dependence (i.e., past exposures affecting confounders of future exposures). Through simulation, we show that our originally proposed propensity function approach displays modest bias due to exposure-induced confounding but no bias from true state dependence. We suggest a correction based on residualized values and show that this new approach corrects for the observed bias. We contrast this revised method with other causal modeling approaches using simulation. Finally, we reproduce the substantive models from Hicks et al. (2018) using the new residuals-based adjustment procedure. With the correction, our findings are essentially identical to those reported originally. We end with some conclusions regarding approaches to causal modeling.

Understanding the variability of Antarctic sea ice is an ongoing challenge given the limitations of observed data. Coupled climate model simulations present the opportunity to examine this variability in Antarctic sea ice. Here, the daily sea ice extent simulated by the newly released National Center for Atmospheric Research (NCAR) Community Eart h System Model Version 2 (CESM2) for the historical period (1979–2014) is compared to the satellite-observed daily sea ice extent for the same period. The comparisons are made using a newly developed suite of statistical metrics that estimates the variability of the sea ice extent on timescales ranging from the long-term decadal to the short term, intraday scales. Assessed are the annual cycle, trend, day-to-day change, and the volatility, a new statistic that estimates the variability at the daily scale. Results show that the trend in observed daily sea ice is dominated by subdecadal variability with a weak positive linear trend superimposed. The CESM2 simulates comparable subdecadal variability but with a strong negative linear trend superimposed. The CESM2's annual cycle is similar in amplitude to the observed, key differences being the timing of ice advance and retreat. The sea ice begins its advance later, reaches its maximum later and begins retreat later in the CESM2. This is confirmed by the day-to-day change. Apparent in all of the sea ice regions, this behavior suggests the influence of the semiannual oscillation of the circumpolar trough. The volatility, which is associated with smaller scale dynamics such as storms, is smaller in the CESM2 than observed.

Abstract. The total Antarctic sea ice extent (SIE) experiences a distinct annual cycle, peaking in September and reaching its minimum in February. In thispaper we propose a mathematical and statistical decomposition of this temporal variation inSIE. Each component is interpretable and, when combined,gives a complete picture of the variation in the sea ice. We consider timescales varying from the instantaneous and not previously defined to themulti-decadal curvilinear trend, the longest. Because our representation is daily, these timescales of variability give precise information about thetiming and rates of advance and retreat of the ice and may be used to diagnose physical contributors to variability in the sea ice. We definea number of annual cycles each capturing different components of variation, especially the yearly amplitude and phase that are major contributors toSIE variation. Using daily sea ice concentration data, we show that our proposed invariant annual cycle explains 29 % more of the variation indaily SIE than the traditional method. The proposed annual cycle that incorporates amplitude and phase variation explains 77 % more variation thanthe traditional method. The variation in phase explains more of the variability in SIE than the amplitude. Using our methodology, we show that theanomalous decay of sea ice in 2016 was associated largely with a change of phase rather than amplitude. We show that the long term trend inAntarctic sea ice extent is strongly curvilinear and the reported positive linear trend is small and dependent strongly on a positive trend thatbegan around 2011 and continued until 2016.

Abstract. The total Antarctic sea ice extent (SIE) experiences a distinct annual cycle, peaking in September and troughing in March. In this paper we propose a mathematical and statistical decomposition of this temporal variation in SIE. Each component is interpretable and, when combined, give a complete picture of the variation of the sea ice. We consider time scales varying from the instantaneous, and not previously defined, to the multidecadal curvilinear trend, the longest. Because our representation is daily, these timescales of variability give precise information about the timing and rates of advance and retreat of the ice and may be used to diagnose physical contributors to variability in the sea ice. We define a number of annual cycles each capturing different components of variation, especially the yearly amplitude and phase that are major contributors to SIE variation. Using daily sea ice concentration data, we show that our proposed invariant annual cycle explains 29 % more of the variation of daily SIE than the traditional method. The proposed annual cycle that incorporates amplitude and phase variation explains 77 % more variation than the traditional method. The variation in phase explains more of the variability in SIE than the amplitude. Using our methodology, we show that the anomalous decay of sea ice in 2016 was associated largely with a change of phase rather than amplitude. We show that the long term trend in Antarctic sea ice extent is strongly curvilinear and the reported positive linear trend is small and dependent strongly on a positive trend that began around 2011 and continued until 2016.

We develop a Bayesian model-based approach to finite population estimation accounting for spatialdependence. Our innovation here is a framework that achieves inference for finite population quantities inspatial process settings. A key distinction from the small area estimation setting is that we analyze finitepopulations referenced by their geographic coordinates (point-referenced data). Specifically, we consider atwo-stage sampling design in which the primary units are geographic regions, the secondary units arepoint-referenced locations, and the measured values are assumed to be a partial realization of a spatialprocess. Traditional geostatistical models do not account for variation attributable to finite populationsampling designs, which can impair inferential performance. On the other hand, design-based estimateswill ignore the spatial dependence in the finite population. This motivates the introduction of geostatisticalprocesses that will enable inference at arbitrary locations in our domain of interest. We demonstrate usingsimulation experiments that process-based finite population sampling models considerably improve modelfit and inference over models that fail to account for spatial correlation. Furthermore, the process basedmodels offer richer inference with spatially interpolated maps over the entire region. We reinforce theseimprovements and demonstrate scaleable inference for groundwater Nitrate levels in the population ofCalifornia Central Valley wells by offering estimates of mean Nitrate levels and their spatially interpolatedmaps.

Respondent-driven sampling (RDS) is a method for sampling from a target population by leveraging social connections. RDS is invaluable to the study of hard-to-reach populations. However, RDS is costly and can be infeasible. RDS is infeasible when RDS point estimators have small effective sample sizes (large design effects) or when RDS interval estimators have large confidence intervals relative to estimates obtained in previous studies or poor coverage. As a result, researchers need tools to assess whether or not estimation of certain characteristics of interest for specific populations is feasible in advance. In this paper, we develop a simulation-based framework for using pilot data-in the form of a convenience sample of aggregated, egocentric data and estimates of subpopulation sizes within the target population-to assess whether or not RDS is feasible for estimating characteristics of a target population. in doing so, we assume that more is known about egos than alters in the pilot data, which is often the case with aggregated, egocentric data in practice. We build on existing methods for estimating the structure of social networks from aggregated, egocentric sample data and estimates of subpopulation sizes within the target population. We apply this framework to assess the feasibility of estimating the proportion male, proportion bisexual, proportion depressed and proportion infected with HIV/AIDS within three spatially distinct target populations of older lesbian, gay and bisexual adults using pilot data from the caring and Aging with Pride Study and the Gallup Daily Tracking Survey. We conclude that using an RDS sample of 300 subjects is infeasible for estimating the proportion male, but feasible for estimating the proportion bisexual, proportion depressed and proportion infected with HIV/AIDS in all three target populations.

Respondent-Driven Sampling (RDS) is a network-based method for sampling hard-to-reach populations that is widely used by public health agencies and researchers worldwide. Estimation of population characteristics from RDS data is challenging due to the unobserved population network, and multiple point and variance estimators have been proposed. Research evaluating these estimators has been limited and largely focused on point estimation; this analysis is the first evaluation of multiple variance estimators currently in use. We evaluated the performance of RDS variance estimators via simulations of RDS on synthetic networked populations constructed from 40 RDS surveys of injection drug users in the United States. In these simulations, average design effects (DEs) were lower and average 95% confidence interval (CI) coverage percentages were higher than suggested in previous work: typical DE range=1-3; average 95% CI coverage=93%. However, DE and CI coverage vary across the 40 sets of simulations, suggesting that the characteristics of a given study should be evaluated to assess estimator performance. We also found that simulation results are sensitive to whether sampling is conducted with replacement and the approach used to create CIs. We conclude that CI coverage rates and DEs are often acceptable but not perfect and that RDS estimates are usually reliable in scenarios where RDS assumptions are met. While RDS estimation performed reasonably well, we found strong evidence that the simple random sample variance estimator and corresponding CIs significantly underestimate variance and should not be used to analyze RDS data.

Prior research has suggested that children living in a disadvantaged neighborhood have lower achievement test scores, but these studies typically have not estimated causal effects that account for neighborhood choice. Recent studies used propensity score methods to account for the endogeneity of neighborhood exposures, comparing disadvantaged and nondisadvantaged neighborhoods. We develop an alternative propensity function approach in which cumulative neighborhood effects are modeled as a continuous treatment variable. This approach offers several advantages. We use our approach to examine the cumulative effects of neighborhood disadvantage on reading and math test scores in Los Angeles. Our substantive results indicate that recency of exposure to disadvantaged neighborhoods may be more important than average exposure for children's test scores. We conclude that studies of child development should consider both average cumulative neighborhood exposure and the timing of this exposure.