Relevance & Research Question: Although online probability-based panels aim for accuracy, they can exhibit a left-leaning bias in public opinion research due to the overrepresentation of politically and civically engaged individuals. Researchers employ weighting techniques to correct sample imbalances relative to the population. This study aims to assess the extent to which diverse adjustment variables and weighting techniques can mitigate this left-leaning bias and enhance the accuracy of estimates from probability-based panels.
Methods & Data: This research examines samples from KnowledgePanel Europe (KP Europe), Ipsos' random probability online panel, which has been operational in the US since 1999 and the UK since 2020, with expansion across Europe since 2022. The evaluation focuses on the effectiveness of different weighting techniques, including raking, propensity score adjustments and generalized regression estimation (GRE). These methods are implemented using two sets of adjustment variables: basic demographics (age, sex, education and geographic region) and a more comprehensive set that includes demographics and variables linked to political attitudes and engagement.

To gauge the relative advantages of various adjustment procedures and variables, each was evaluated based on its success in reducing bias for different benchmarks from high-quality, "gold-standard" surveys. These benchmarks cover a range of topics, like civic engagement, living situation and technology use. Besides biases, the variance or precision of estimates is crucial. The "margin of error" (MOE) describes the expected variance in survey estimates if repeated multiple times under identical circumstances. The MOE is calculated for estimates from all benchmark variables to see how different weighting procedures and variables affect variability.

Results: Initial findings reveal variability in left-leaning bias in KP Europe samples. While various weighting methods effectively reduce bias and align results with population distributions, the choice of adjustment variables significantly affects the accuracy of the estimates. Additionally, incorporating political variables alongside basic demographics has a different impact on the MOE across KP Europe's countries.

Added Value: This study highlights the critical role of adjustment variables in improving the accuracy of estimates and provides valuable insights into the effectiveness of weighting techniques for reducing bias in political and public opinion research across diverse European contexts.

Relevance & Research Question
Gaining insights into how different sampling methods may affect the generalizability and validity of survey-embedded experiments is essential for online survey actors. Previous studies have found that social psychological experiments appear to generate comparable experimental treatment effects across sampling (probability and nonprobability) and stratification schemas (prestratified or not). This study replicates and expands such research to survey embedded experiments from political science (Krishnarajan, 2023) and media and communication studies (Van Erkel & Soontjens, 2022) and assesses whether effect sizes vary over types of sampling and stratification strategies.
Methods & Data
Two survey-embedded experiments were administered to participants of the Swedish Citizen Panel (SCP), an academically run non-commercial online access panel. Participants were 2,000 panelists recruited to the SCP using probability sampling and 2,000 panelists who self-selected into the panel. These panelists were then selected either through prestratified random sampling or simple random sampling, yielding a total of four independent groups of invited panelists: (1) a nonprobability-based sample stratified on age, education, and gender, (2) a nonprobability-based sample that was not selected using prestratification criteria, (3) a probability-based sample stratified on age, education, and gender, and (4) a probability-based sample that was not selected using prestratification criteria.
Results
The results are not yet available; the present abstract will be revised with the results amended in January 2026. The data will be analyzed by comparing the outcomes of the political and media and communication survey embedded experiments in the four sample groups.
Added Value
This study adds to the research on survey methodology and the impact sampling has in treatment effects. A better understanding of how nonprobability and probability samples might affect data quality is crucial to be able to conduct more precise and valuable research in the future.

Relevance & Research Question
The expansion of internet access in many countries raises the question of whether it is now feasible to conduct web-only surveys of the general population without compromising representativeness. This presentation will examine the representativeness of web-only surveys of the general population in the United Kingdom. The analysis addresses the following three research questions:

• RQ1: How have internet exclusion and intensity of internet use changed over time?

• RQ2: What are the characteristics of different types of internet users and non-users? How representative are these groups? How has this changed over time?

• RQ3: How does the representativeness of web respondents compare to the representativeness of different groups of internet users? How has this changed over time?

Methods & Data
We use data from the United Kingdom Household Longitudinal Study (UKHLS). The main survey (2009-2022), a probability-based sample of the UK household population, is used to explore the evolution of internet exclusion over time (RQ1 and RQ2). Moreover, we benefit from a mixed-mode experiment embedded in the Innovation Panel (2012-2023), a probability-based sample of the Great Britain household population. In the experiment, a random sample of households was assigned to a web-first and CAPI sequential design. This enables us to evaluate the representativeness of web respondents and compare it with that of internet users (RQ3).

We use coefficients of variation of the response propensities to estimate the representativeness of internet users and web respondents with regard to a set of auxiliary variables.
Results
The results show a significant decrease in internet exclusion. Internet users are increasingly representative of the general population, although gaps still remain among older adults, those with lower education levels, and other disadvantaged groups. Web survey respondents have also become more representative of the general population over the past decade, but they remain less representative than internet users.
Added Value

The results offer valuable empirical evidence about the quality of web-only surveys in the past and present, which will assist survey practitioners in understanding the opportunities and risks of conducting web-only surveys now and in the near future.