Editorial to the Special Issue on “New challenges for sample surveys: innovation through tradition”

Monitoring Round Goby ( Neogobius melanostomus) Population Expansion in Eastern and Northern Lake Michigan

Workshop Synthesis: Survey Methods for Hard-to-Reach Groups and Modes

Quasi-randomization approaches estimate latent participation probabilities for units from a nonprobability / convenience sample. Estimation of participation probabilities for convenience units allows their combination with units from the randomized survey sample to form a survey-weighted domain estimate. One leverages convenience units for domain estimation under the expectation that estimation precision and bias will improve relative to solely using the survey sample; however, convenience sample units that are very different in their covariate support from the survey sample units may inflate estimation bias or variance. This paper develops a method to threshold or exclude convenience units to minimize the variance of the resulting survey-weighted domain estimator. We compare our thresholding method with other thresholding constructions in a simulation study for two classes of datasets based on the degree of overlap between survey and convenience samples on covariate support. We reveal that excluding convenience units that each express a low probability of appearing in both reference and convenience samples reduces estimation error.

The COVID-19 pandemic has posed a substantial threat to people’s lives and aroused health concerns. This study aims at exploring the following questions. First, how does the COVID-19 pandemic affect people’s willingness to pay for health (WPH) in the short and long term? Second, what is the psychological mechanism underlying such an effect? Finally, what are the boundary conditions for this effect? To answer these questions, we conducted three longitudinal surveys. The first survey was launched in February 2020—the time of the most serious outbreak of COVID-19 in China. Data were obtained from 1548 participants through questionnaires on an online survey platform. The sample covered 297 prefecture-level cities in 31 provincial administrative regions. Subsequently, we conducted two follow-up surveys in August 2020 and July 2021. The samples of these surveys were randomly selected from the sample of the first survey. The findings showed that the pandemic promoted people’s WPH in the outbreak period. The fear of death and self-esteem mediated and moderated this effect, respectively. Moreover, the effect persisted for six months after the COVID-19 pandemic had been brought under control (August 2020). However, the effect disappeared after a year and a half (July 2021). These results indicate that the COVID-19 pandemic promoted people’s WPH and that this effect was sustained in the short term after the pandemic had been brought under control but not in the long term.

Concerns about the cost-effectiveness of conservation monitoring prompt a reassessment of systematic aerial sample surveys, which have been widely applied to census wildlife and livestock in African savannas for more than 30 years. First, we use results from high-resolution sample surveys in Laikipia District (northern Kenya) to directly compare results from aerial total and sample surveys, showing few systematic differences in their estimates of wildlife abundance but great differences in cost and scope. Second, we quantify how the precision of population estimates is affected by survey resolution and species density. Results suggest that lower survey resolutions widely used in the past to census wildlife and livestock resources have been insufficient to reliably estimate all but the most abundant species. Third, we describe how sample survey data can be used to map the potential for human–wildlife conflict across large landscapes. High-resolution sample surveys in Laikipia have revealed causes and consequences of ecosystem change, advanced our understanding of ungulate population dynamics, guided wildlife management and conservation action, and increased confidence in sample survey methodology. However, further refinements in sample survey methods are needed to improve cost-effectiveness.

This paper outlines the utility of statistical methods for sample surveys in analysing clinical trials data. Sample survey statisticians face a variety of complex data analysis issues deriving from the use of multi-stage probability sampling from finite populations. One such issue is that of clustering of observations at the various stages of sampling. Survey data analysis approaches developed to accommodate clustering in the sample design have more general application to clinical studies in which repeated measures structures are encountered. Situations where these methods are of interest include multi-visit studies where responses are observed at two or more time points for each patient, multi-period cross-over studies, and epidemiological studies for repeated occurrences of adverse events or illnesses. We describe statistical procedures for fitting multiple regression models to sample survey data that are more effective for repeated measures studies with complicated data structures than the more traditional approaches of multivariate repeated measures analysis. In this setting, one can specify a primary sampling unit within which repeated measures have intraclass correlation. This intraclass correlation is taken into account by sample survey regression methods through robust estimates of the standard errors of the regression coefficients. Regression estimates are obtained from model fitting estimation equations which ignore the correlation structure of the data (that is, computing procedures which assume that all observational units are independent or are from simple random samples). The analytic approach is straightforward to apply with logistic models for dichotomous data, proportional odds models for ordinal data, and linear models for continuously scaled data, and results are interpretable in terms of population average parameters. Through the features summarized here, the sample survey regression methods have many similarities to the broader family of methods based on generalized estimating equations (GEE). Sample survey methods for the analysis of time-to-event data have more recently been developed and implemented in the context of finite probability sampling. Given the importance of survival endpoints in late phase studies for drug development, these methods have clear utility in the area of clinical trials data analysis. A brief overview of methods for sample survey data analysis is first provided, followed by motivation for applying these methods to clinical trials data. Examples drawn from three clinical studies are provided to illustrate survey methods for logistic regression, proportional odds regression and proportional hazards regression. Potential problems with the proposed methods and ways of addressing them are discussed.

THE methodology reported in this chapter is of recent origin. Literature on the application of sampling theory to practical survey problems as we know it today was virtually unavailable as recently as 1940. Ten years ago the basic principles of the theory and the major operating technics of survey sampling were known only by a few experts. Theory and practice in the field covered by this chapter have been developed largely outside education in opinion polling, in market research, and in census operations. It is not surprising, therefore, that there is considerably more literature about survey sampling than there is about survey sampling in education. The wealth of understandable and technically adequate sources on survey sampling which accumulated during the period covered in this issue of the REVIEW justifies the devotion of a separate chapter to it. This chapter supplements a section of the Johnson and Moonan chapter (27) of a previous issue which covered survey sampling thru the middle of 1951. Survey sampling methods apply to problems of enumeration (determining how many or how much), particularly with reference to finite populations. As such they differ considerably from common theory and methods in educational statistics, which deal with infinite populations and with sample methods in experimental design. Altho in recent years the statistical literature has included many items on sampling concerned primarily with sampling inspection (acceptance sampling), this area is excluded from the review of this chapter except as immediately applicable to survey sampling. Sampling surveys involving measurement, psychometric problems of item design, the measurement of attitudes and opinions, and scaling are not covered in this chapter.

In assembling our guidelines, we examined ideas in the area of population estimation and also in two areas related to population estimation: (1) sample (statistical) surveys; and (2) population forecasting. In regard to the two related fields, we found ideas that can be applied directly as guidelines for developing and presenting estimates and others that can be easily adapted for this purpose. While, as this book attests, there are substantial areas of overlap among sample surveys, population forecasting, and population estimation, there are important distinctions. And these distinctions lead to different guidelines.

IntroductionAcross sub‐Saharan Africa, selling sex puts young women at high risk of HIV. Some young women who sell sex (YWSS) may self‐identify as sex workers, while others may not, having implications for how to reach them with HIV prevention. We describe characteristics, sexual behaviours and health service use of YWSS in Zimbabwe, comparing women who identified as female sex workers (FSW) and women who did not (non‐identifying‐YWSS), and explore factors associated with HIV infection.MethodsWe analysed data from respondent‐driven sampling (RDS) surveys among YWSS aged 18 to 24 implemented in six sites in Zimbabwe from April to July 2017. RDS was used to enrol YWSS into an impact evaluation of the multi‐country DREAMS (Determined, Resilient, Empowered, AIDS‐free, Mentored and Safe) Partnership, which provides comprehensive HIV prevention programming to adolescent girls and young women. Women completed an interviewer‐administered questionnaire and were offered HIV testing services. We used logistic regression (RDS‐II‐weighted, normalized by site) to identify factors associated with prevalent HIV infection.ResultsForty‐four seeds recruited 2387 YWSS. RDS‐adjusted HIV prevalence was 24%; 67% of women identified as FSW. FSW were older and had lower educational attainment than non‐identifying‐YWSS. While 40% of FSW reported 10+ clients in the previous month, 9% of non‐identifying‐YWSS did so. FSW were more likely to have accessed HIV‐related services, including HIV testing in the last six months (FSW: 70%; non‐identifying‐YWSS: 60%). Over half of all YWSS described selling sex as their main financial support (FSW: 88%; non‐identifying YWSS: 54%). Increasing age, lower educational attainment, younger age of first selling sex and higher number of clients in the previous month were associated with prevalent HIV.ConclusionsYWSS in Zimbabwe have a high prevalence of HIV, reported high numbers of sexual partners and depend financially on selling sex. Non‐identifying‐YWSS differed socio‐demographically to FSW, yet factors associated with HIV risk were similar for all women. Women not identifying as FSW were less likely to access services, suggesting they should be prioritized for HIV prevention. Network‐based recruitment may enhance their inclusion in programmes, like DREAMS, which aim to reach young women at highest‐risk with comprehensive health, HIV prevention and social protection services.

The New York State Water Resources Institute at Cornell University undertook a two‐year sampling survey of pesticides in ground water beginning in 1985. The survey focused on areas where combinations of agricultural pesticide use, soil texture, and ground water occurrence seemed likely to lead to leaching. The sampling survey included samples from four types of sampling points: (1) monitoring wells; (2) existing water supply wells; (3) test holes; and (4) tile drains. The monitoring wells were sampled repeatedly throughout the project to attempt to characterize temporal changes in water quality corresponding with seasonal changes in ground water levels. The pesticides studied for this project were atrazine, alachlor, cyanazine, metolachlor, carbaryl, carbofuran (and a metabolite, 3‐hydroxy carbofuran), and simazine. All, except for carbaryl, have been found in ground water in other sampling surveys in the United States.The results of the sampling survey did not reflect the careful choices of enviromental characteristics and pesticide use that seemed likely to lead to leaching. Residues of three pesticides were detected in six single samples from separate sources at four of the 30 sites tested. Three of the six samples came from shallow test holes that were used to sample the shallowest possible saturated soils beneath fields. The three pesticides detected were atrazine, simazine, and 3‐hydroxy carbofuran. Of the six samples, a single sample from a test hole contained atrazine concentrations equal to the current federal health advisory for long‐term exposure to atrazine (3 ppb).The remaining detections were between the limit of detection for analytical methods and the federal health advisory for each pesticide. The federal health advisories were formulated after the end of the project. Analytical methods may have been insensitive with respect to these advisories. Sampling results from other surveys suggest that many detections of the same pesticides lie below the limits of detection used for this sampling survey.A possible explanation for the lack of detections, given the design of the sampling survey, may lie in the agricultural practices noted at sampled sites. Most of the farm managers rotated their crops and pesticides on many small fields. Although the environmental conditions chosen for sampling sites were expected to lead to contamination, reported pesticide applications varied from year to year and field to field according to rotational schedules. The inconsistency of applications from year to year may explain the lack of detections (at the limits of quantification used for analyses) noted in this sampling survey.

For many sample surveys, researchers, policymakers, and other stakeholders are interested in obtaining estimates for various domains, such as for geographic levels, for demographic groups, or a cross‐classification of both. Often, the demand for estimates at a disaggregated level exceeds what the sample size and survey design can support when estimation is done by traditional design‐based estimation methods. Small area estimation involves exploiting relationships among domains and borrowing strength from multiple sources of information to improve inference relative to direct survey methods. This typically involves the use of models whose success depends heavily on the quality and predictive ability of the sources of information used. Possible sources of auxiliary information include administrative records, Censuses, big data such as traffic or cell phone data, or previous vintages of the same survey. One rich source of information is that of other surveys, especially in countries like the United States, where multiple surveys exist that cover related topics. We will provide an introduction to the topic of combining information from multiple surveys in small area estimation using area‐level models, including practical advice and a technical introduction, and illustrating with applications. We will discuss reasons to combine surveys and give an overview of some of the most common types of models.This article is categorized under: Statistical Models > Multivariate Models

Effects of sample survey design on the accuracy of classification tree models in species distribution models

AIM-Back is an embedded pragmatic clinical trial (ePCT) with cluster randomization designed to increase access and compare the effectiveness of two different non-pharmacological care pathways for low back pain (LBP) delivered within the Veteran Administration Health Care System (VAHCS). This manuscript describes baseline characteristics of AIM-Back participants as well as the representativeness of those referred to the AIM-Back program by sex, age, race, and ethnicity, relative to Veterans with low back pain at participating clinics. To be eligible for AIM-Back, Veterans were referred to the randomized pathway at their clinic by trained primary care providers (Referral cohort). Veterans from the Referral cohort that participated in the study included: 1) an Electronic Health Record (EHR) sample of Veterans enrolled in the program (i.e., attended initial AIM-Back visit with no consent required) and a Survey sample of Veterans that were consented for further study. Descriptive statistics for age, race, ethnicity, sex, high-impact chronic pain (HICP), a comorbidity measure, post-traumatic stress diagnosis (PTSD) and opioid exposure were reported for the Referral cohort and by sample; mean baseline PROMIS pain interference, physical function and sleep disturbance scores were reported by sample. Additional measures of pain, mental health and social risk were reported on the Survey sample. Participation to prevalence ratios (PPRs) were calculated for sex, age, race, and ethnicity by clinic to describe representativeness of the Referral cohort. Across 17 randomized primary care clinics, the Referral cohort included 2767 unique Veterans with n=1817 in the EHR sample, n=996 in the Survey sample and n=799 of the EHR sample (44%) were also in the Survey sample. High rates of HICP were observed in the EHR and Survey samples (>59%). Mean scores (SD) based on self-reported PROMIS Pain Interference (63.2 (6.8), 63.1 (6.6)) and PROMIS Physical Function (37.1 (5.3), 38.1 (5.8)) indicated moderate impairment in the EHR sample and Survey sample respectively. Approximately 10% of the EHR sample had documented opioid use in the year leading up to the AIM-Back referral. At most clinics, older Veterans (>=65 years) were underrepresented in the Referral cohort compared to those with LBP visits at clinics (PPRs < 0.8). The AIM-Back trial will conduct analysis to examine the comparative effectiveness of the two care pathways and identify individual characteristics that may improve responses to each pathway. The trial is expected to complete 12-month follow-up data collection by December 2024, with subsequent analyses and publications providing insights into optimizing non-pharmacological care for Veterans with LBP. NCT04411420 (clinicaltrials.gov).

A significant share of transportation forecasting models are based on household travel survey data, frequently gathered through phone-based travel surveys. Over the last decade, these surveys have faced a variety of issues, namely a decrease in response rates and increases in potential sampling biases since their sampling frames are typically constructed using a list of land-line numbers. Although the weighting process can manage some of these issues, their scale is becoming a problem. In recent years, several experiments have been conducted in order to integrate multiple survey methods and data sources to obtain better quality databases. This paper reports on an experimentation conducted in 2012 in the region of Sherbrooke, in the province of Quebec, Canada. A web based travel survey was conducted among the regions’ student population—in parallel to the large-scale household phone travel survey- in order to increase both the representativeness of the sampling frame and the response rate among 20–29 year-olds. This paper presents the methodology used to integrate both survey samples in order to create a more representative view of typical travel behaviors in the region of Sherbrooke. The paper demonstrates that by combining the samples, we succeed in reducing heterogeneity in sampling rates among population segments which translates into an increase in trip rates (global and transit) for the target population. Challenges related to the combination of survey methods and samples are discussed.

The paper presents the quality problem of fishery statistics produced by the conducted land-based and sea-going, large scale sample surveys of the survey programme of the Institute of Marine Biological Resources of the Hellenic Centre for Marine Research (IMBR/HCMR) in Greece, and the optimality strategies developed in their sampling designs for the maximization of precision of the calculated sample estimates for a given cost of sampling.The optimality problems of the sampling designs of the individual large scale sample surveys are described, and the optimality solutions developed under the sampling variance structure are explained.The paper deals with the optimization of the following three large scale sample surveys: biological sample survey of commercial landings (BSCL), experimental fishing sample survey (EFSS), and commercial landings and effort sample survey (CLES).