Conference Agenda

Aggregate relational data (ARD) on relationships between individuals and groups in a population can shed light on acquaintanceship network size, the level of segregation in contact with subpopulations, and the size of unlisted groups. Despite their wide range of applications, ARD questions can be cognitively demanding for respondents, as they require reporting the number of acquaintances in various subpopulations within the constrained time frames of typical survey settings. However, research on the susceptibility of ARD instruments to measurement error remains scarce. This study leverages the panel design of the Chilean Longitudinal Social Survey (ELSOC) to examine the reliability of individual ARD items and the network size measure obtained by combining them. To estimate reliability, I employ structural equation and multilevel modeling approaches commonly used in survey methodology research, adapting them to accommodate the unique structure of ARD. Individual items have reliability scores ranging from 0.41 to 0.51, depending on the reliability assessment method used. The reliability of the log-transformed network size is 0.59, indicating that combining the items reduces measurement error. Nevertheless, the reliability estimates of both individual items and the composite measure fall short of the commonly accepted 0.70 reliability threshold, raising concerns about bias and precision in ARD-based estimates of network size, segregation, and unlisted group sizes. These findings highlight the need for methodological refinements to improve the quality of ARD-derived data.

Feld and McGail (2020) explain how social networks distort perceptions of the social world. Highly connected individuals (high-degree nodes) are overrepresented in personal networks, leading to misperceptions where their traits are seen as more typical than is actually the case (Feld, 1991). The extent of this distortion depends on variation in individuals’ visibility, or indegree, which is difficult to assess when indegrees are unknown.

We propose a method using population surveys to estimate target indegree variance by counting how often the same targets are named by multiple informants. A high number of repeated mentions indicates high variation. We show how the ratio of repeated mentions to the square of total mentions provides a quantitative measure of the distorted experience of these targets.

We illustrate our method in academic citation networks, where readers may get inflated impressions of the value of academic articles in a journal, because the papers they find through citations are more highly cited than average for that journal. We measure this distortion in each journal by counting repeat citations in a sample of papers.

High variation in visibility among stigmatized individuals could lead to public perception being shaped by a few highly visible extroverts, while most others remain overlooked. Our method quantifies this variation in visibility. We illustrate its application and recognize challenges in implementation, addressing potential limitations.

This paper highlights the importance of measuring indegree variation to better understand distorted perceptions and provides a practical method for assessing distortions in many situations.