DieTryin: An R package for data collection, automated data entry, and post-processing of network-structured economic games, social networks, and other roster-based dyadic data

Abstract

Researchers studying social networks and inter-personal sentiments in bounded or small-scale communities face a trade-off between the use of roster-based and free-recall/name-generator-based survey tools. Roster-based methods scale poorly with sample size, and can more easily lead to respondent fatigue; however, they generally yield higher quality data that are less susceptible to recall bias and that require less post-processing. Name-generator-based methods, in contrast, scale well with sample size and are less likely to lead to respondent fatigue. However, they may be more sensitive to recall bias, and they entail a large amount of highly error-prone post-processing after data collection in order to link elicited names to unique identifiers. Here, we introduce an R package, DieTryin, that allows for roster-based dyadic data to be collected and entered as rapidly as name-generator-based data; DieTryin can be used to run network-structured economic games, as well as collect and process standard social network data and round-robin Likert-scale peer ratings. DieTryin automates photograph standardization, survey tool compilation, and data entry. We present a complete methodological workflow using DieTryin to teach end-users its full functionality.

Friendship and partner choice in rural Colombia

“…RICH economic games involve tasks where participants (often called deciders or focals) have a chance to: (i) allocate money to, (ii) take money from, and (iii) at a cost to themselves, reduce the payouts of other individuals (a.k.a., alters). In these games, a photograph roster is used to allow each focal individual to make decisions with respect to each other person in the study (see Ross and Redhead, 2021, for details and software tools). The two RICH games used here-the allocation and exploitation games-have important differences in framing.…”

Social hierarchies and social networks in humans

“…Advanced tools for data collection (e.g. [ 208 ]) and network inference are continually being developed and refined. Generative network models, such as exponential random graph models [ 209 ], stochastic actor-oriented models [ 210 ], and latent network frameworks [ 211 ] hold particular promise, as they allow for the simultaneous consideration of multiple mechanisms operating across scales.…”

Reliable network inference from unreliable data: A tutorial on latent network modeling using STRAND.

“…It is common for social networks to be measured using free-list self-report nominations (i.e., in “name-generator” designs; Marin & Hampton, 2007), or self-reports of the relationship between every potential dyad in which the respondent could appear (i.e., in roster-based designs; Marsden, 2005). While social relationships can be captured using several different measurement instruments (see Marsden, 2005; Ross & Redhead, 2021 for reviews), the most popular questionnaire-based approaches fall into two broad categories: event-recall questions and perceptual questions.…”

“…That is, rather than solely analyzing self-reports of a given social network, we simulate 12 time points of sparse transfer data (in which transfers were accurately documented), and analyze these data in addition to the standard self-report data. These sparse networks of “ground-truth” data might come from field observations (e.g., “scan sampling” or “spot-checks”; Borgerhoff Mulder et al, 1985), diary methods (Paolisso & Hames, 2010), experimental games (Ross & Redhead, 2021), video recordings (DeTroy et al, 2021), GPS tracking (Davis et al, 2018), proximity detection (e.g., through cell phone data; Urban, 2021), or a variety of other methods. Even with sparse, incomplete “ground-truth” data, network reconstruction can be improved, and covariate effects on network structure and reporting parameters disambiguated.…”