ESRA 2025 Preliminary Program

Monday 14 July    Tuesday 15 July    Wednesday 16 July    Thursday 17 July    Friday 18 July   

All time references are in CEST

Methods of including and boosting under-represented population subgroups
Session Organisers	Dr Olga Maslovskaya (University of Southampton) Dr Carina Cornesse (DIW Berlin) Mr Curtis Jessop (NatCen) Dr Mariel Leonard (DIW)
Time	Thursday 17 July, 15:30 - 17:00
Room	Ruppert 002

Some population subgroups are consistently under-represented or excluded from survey samples or may be present in too small frequencies for robust subgroup analysis. While approaches to targeting these population subgroups exist, they may not provide sufficient level of quality for social surveys (e.g., use of convenience samples) or may be prohibitively expensive or unfeasible in a self-completion mode (e.g., over-sampling for the Postcode Address File (PAF) and/or screening). As survey research globally aims to become more inclusive, it is crucial to explore effective methods of including and boosting under-represented population subgroups.

In this session we would like to investigate different effective methods of including and boosting under-represented population subgroups including respondent-driven sampling approach.

We encourage papers from researchers with a variety of backgrounds and across different sectors, including academia, national statistics, and research agencies. We particularly welcome contributions that use experimental designs, and/or other designs that can inform future strategies for including and boosting under-represented subgroups in large scale high-quality probability-based surveys.

The session is proposed by Research Strand 1 of the Survey Futures project, “Enhanced Sampling Frames and Procedures”. Survey Futures is a UKRI-ESRC funded research programme focused on ensuring large-scale social surveys in the UK can innovate and adapt in a changing environment. https://surveyfutures.net/

Keywords: under-represented subgroups, respondent-driven sampling, high quality surveys

Papers

Implementing respondent-driven sampling to increase the diversity of a general population sample

Dr Mariel Leonard (DIW-Berlin) - Presenting Author
Ms Julia Witton (DIW-Berlin)
Dr Carina Cornesse (GESIS)
Dr Julian Axenfeld (DIW-Berlin)
Dr Jean-Yves Gerlitz (University of Bremen)
Dr Olaf Groh-Samberg (University of Bremen)
Dr Sabine Zinn (DIW-Berlin)

Respondent-driven sampling (RDS) is a network sampling technique for surveying complex populations in the absence of sampling frames. The idea is simple: identify some people (“seeds”) who belong or have access to the target population, encourage them to start a survey invitation chain-referral process in their community, ensure that every respondent can be traced back along the referral chain. Due to the reliance on respondent referral, RDS is frequently implemented with hidden or rare target populations, where members are assumed to know each other and thus have a higher degree of access than researchers.

We conducted a pilot study in 2023 where we invited 5,000 panel study members to a general population multi-topic online survey. During the survey, we asked respondents whether they would be willing to recruit up to three of their network members. Willing respondents then received personalized links with which to recruit their network members.

We found that younger individuals, individuals with higher incomes, and individuals with migration backgrounds were all (1) more likely to recruit, and (2) they typically recruited individuals similar to themselves. We additionally found that those recruits also had a higher propensity to participate, thereby potentially increasing the overall diversity of the survey. In this paper, we present a detailed overview of our results, along with relevant methodological findings such as method of recruitment. Additionally, we discuss 2025 fielding of RDS in the Social Cohesion Panel which builds upon the findings from our pilot study.

Respondent-Driven Sampling (RDS) Methods: Evidence review

Dr Olga Maslovskaya (University of Southampton) - Presenting Author
Dr Carina Cornesse (GESIS)
Mr Curtis Jessop (NatCen)
Mr Mark Todd (NatCen)
Mr Toby Li (NatCen)
Mr Stephen Milner (ONS)

Ensuring inclusivity in general population surveys is paramount for robust survey research. However, some population subgroups are consistently under-represented or excluded, or are included in insufficient numbers for meaningful subgroup analysis. While methods to targeting these population subgroups exist, they may not provide sufficient level of quality for social surveys (e.g., use of convenience samples) or may be prohibitively expensive or unfeasible in a self-completion mode (e.g., over-sampling from the Postcode Address File (PAF) in the UK and/or screening).

Respondent Driven Sampling (RDS) has emerged as a successful strategy for recruiting hard-to-reach groups, such as LGBTQ+ communities, migrants, or ethnic minorities (Heckathorn, 1997). There have also been attempts to apply RDS in the context of boosting population subgroups in probability-based surveys.

Developed as a network sampling approach, RDS combines snowball sampling with the mathematical properties of probability-based sampling techniques to approximate a probability sample under Markov Chain assumptions (e.g., Thompson, 2014).

An evidence review is essential to consolidate the existing literature, grey literature, and reports, summarising various approaches used for RDS. We created a detailed coding scheme to capture all indicators of RDS methodologies. This review assesses strengths and limitations of these approaches in terms of effectiveness, costs of implementation, and speed of implementation. Additionally, we identify gaps in the evidence, which will provide valuable directions for future research in this area.

In this presentation, we summarise all findings from the evidence review and discuss recommendations for survey practice.

Suitability of the Chosen Variables for Stratification in Large Scale Survey: An Exploration from Indian Version of DHS

Mr SOMNATH JANA (PHD RESEARCH SCHOLAR) - Presenting Author

Survey accuracy is an important consideration in large-scale investigations, particularly when attempting to get precise and dependable estimates from varied populations. Stratification, which includes separating the population into discrete subgroups depending on important traits, has long been seen to be an effective way to enhance survey accuracy. By reducing sample variability among these homogenous groupings, stratified sampling has the potential to lower standard errors while increasing data representativeness.
The study aimed to evaluate the use of stratification in improving the precision of survey estimates in large-scale surveys by comparing the standard errors obtained with and without stratification. This study investigates the effectiveness of different stratification approaches in large-scale surveys, focusing on health indicators in India. Comparing 2x stratification (urban-rural division) with 4x stratification (three rural segments and one urban segment), the research analyzes data from districts in Uttar Pradesh and Kerala (States of India). Key indicators include stunting, immunization, antenatal care visits, and access to sanitation facilities.
Results show that 2x stratification generally outperforms 4x stratification across most districts and indicators. In Uttar Pradesh, 2x stratification performed better in 64% of districts for stunting, 54.67% for immunization, 58.67% for antenatal care, and 46.67% for sanitation. Kerala showed even more pronounced advantages for 2x stratification.
The study concludes that 2x stratification may be more effective at capturing local contexts, leading to more representative sampling and precise estimates. These findings have significant implications for survey design and implementation, suggesting that simpler stratification approaches can often yield more accurate and cost-effective results in health and development surveys in India.

Collecting Aggregate Relational Data from People Who Use Drugs using Social Media Advertisement

Mr ihsan kahveci (University of Washington) - Presenting Author
Dr McKenna Parnes (University of Washington)
Dr Brittany Blanchard (University of Washington)
Dr Tyler McCormick (University of Washington)

The U.S. overdose crisis claimed over 107,000 lives last year, with nearly 20 million Americans at risk due to fentanyl-contaminated drug supplies. Proven public health interventions, such as overdose education, naloxone distribution, and drug testing, can prevent overdose deaths. Unfortunately, access to these programs is often hindered by social stigma and systemic inequities. A promising solution lies in leveraging the social networks of people who use drugs (PWUD). Research shows that trusted peers can effectively disseminate information and resources, improve access to services, promote safer behaviors, and reduce stigma. However, traditional network analyses require complete data, often infeasible due to recruitment challenges and high costs.

This study leverages Aggregate Relational Data (ARD) to explore the social connections of PWUD and estimate key network characteristics, such as degree and centrality. Through social media advertisements on Meta, 3,600 participants were recruited to provide relational data by answering questions such as, “How many people do you know with a given trait X?” Data were collected on both close friendships and broader acquaintance networks, providing a comprehensive view of social connections within this population. The Network Scale-Up Method (NSUM) was applied to ARD, estimating an average degree of 2.7 for close friendships and 35.6 for acquaintances. Additionally, ARD methods were used to model network features and identify key actors positioned to facilitate harm reduction interventions, such as naloxone distribution and overdose prevention education. To address selection bias and ensure the sample accurately represents the target population, we applied post-stratification and design weights, calibrated using demographic data from the U.S. Census and Meta Audience Insights for Washington State.

These findings highlight the potential of ARD to collect behavioral and network data from hard-to-reach populations for designing effective public health interventions.