Longitudinal social networks impacts on weight and weight-related behaviors assessed using mobile-based ecological momentary assessments: Study Protocols for the SPARC study

The purpose of the SPARC study is to determine mechanisms by which friendship networks impact eating, physical activity and weight among diverse college freshmen. The SPARC study is grounded in a socioecological framework with an emphasis on the interpersonal level by tracking changes in friends’ relationships (perceived, direct report, and social network analysis) with nutrition and PA behaviors, and weight status over a single academic year (9 months). Data will be collected in waves throughout participants’ first year at a large southwestern university from freshmen students living in residence halls. Instruments include web-based surveys, mobile-based ecological momentary assessments (mEMAs), and student card data (time-dated data on food purchasing, use of physical activity venues, and geographical information system (GIS) location of these activities relative to other students in their social networks) from freshmen living in residence halls. The web-based surveys and anthropometrics are scheduled for four time points (the beginning and end of each semester). The m EMAs are scheduled to be administered over four waves. A wave consists of a 1-week period during each of the four target months during the academic year (see Table 1). The student card data are secondary data (times entered dining halls and recreation centers) that will be collected at the conclusion of the study. Each dataset provides unique information and will help us understand the mechanisms and the contextual factors related to friendship networks’ role in students’ eating and PA behaviors, and weight over time (see Table 2 for specific measures).

We aim to saturate the residence halls (i.e., recruit all freshmen) to have as complete a friendship network as possible (close friends, roommates, friend groups, and networks at the residence hall floor and residence hall level will also be included). Residence Life and Resident Assistants (Community Mentors and Community Assistants) from each residence hall helped to facilitate recruitment.

For each completed assessment, participants earn incremental monetary awards (up to $110) and additional earned incentives (“swag”: e.g., study branded water bottles, t-shirts, Frisbees, ear buds, tote bags). In order to not impact the social network of the study, floor-level incentives (e.g., pizza party) are not offered. When floors reach 60 % or higher enrollment, individual participants are offered additional swag of their choice. Community Mentors and Assistants are offered $15 gift cards if their floor reaches 60 % participation and $40 in gift cards if their floor reaches 80 % participation at each data collection point.

Web-based surveys and anthropometric measurements: The web-based surveys address personal, interpersonal, and environmental factors related to participants’ weight and weight related behaviors and takes 20–30 min to complete. Included in the survey are validated measures of eating, PA, and weight status, as well as a series of questions about participants’ relationships with friends. Baseline demographic characteristics (e.g., age, gender, race/ethnicity, Pell Grant status, and parental education) are collected at the first data collection point.

The validated 26-item Dietary Screener Questionnaire (DSQ) used in the 2010 National Health Interview Survey Cancer Control Supplement, a free tool developed by the National Cancer Institute of NIH [25, 26] assesses the frequency of consumption of key food items and groups. While this tool does not estimate individual’s caloric intake, it allows for tracking of the consumption of major food groups (e.g., fruits and vegetables, high fat foods, sugar sweetened beverages) related to weight. In addition, participants are asked to report how often they ate breakfast [27], evening meals [28], and fast food [27]. The USDA six-item food security short form is included to examine changes in food security status [29].

PA is examined with the Godin-Shepard PA assessment [30, 31], which assesses usual vigorous, moderate, and light PA: “In a usual week, how many hours do you spend doing the following activities: Strenuous exercise (heart beats rapidly)?; Moderate exercise (not exhausting)?; Mild exercise (little effort)?” Response options ranged from none to more than 6 h per week. A sum of the time spent in PA and moderate-to-vigorous PA will be created. Sedentary activities are assessed with the question: “Yesterday, how much time did you spend in front of a screen (excluding time in class and being physically active)?” Response options ranged from zero to more than 6 h and will be summed to create a total time spend in sedentary behaviors [32, 33].

Participants complete friendship network questions on each web-based survey for which they list their 5 closest male and female friends [34‐37] and report the time spent with them eating, being physically active, or sedentary. Participants also respond to how long they have been friends with that person and whether or not the friend is their best friend, roommate, suitemate, or significant other. Participants are asked to indicate their level of closeness with each nominated friend [38] and how they maintain their friendship with each person [39]. In addition, participants are asked to reflect on their openness to new friends while on campus [40‐42]. These survey data will be linked among nominated friends and roommates to assess associations over time with questions about weight-related behaviors and outcomes at each of the four time points.

Height and weight are measured privately by trained research staff at the same times as the web-based survey using Seca scales and stadiometers to track changes in body mass index. Waist and hip circumference are collected using flexible tape measurers. Participants are asked to self-report on their height and weight, as well as whether they are trying to change their weight in any way [43].

mEMAs: During each wave of mEMA data collection, students are prompted via SMS text messages to complete the mEMA eight times per day on four of the 7 days in their assessment period, with at least 1 weekend day per wave. As such, a potential for 128 repeated measures for each participant is possible with the mEMA. A random, interval-contingent schedule is used for the mEMA prompts. Twice during each of the four established time periods per day (9 am–12 pm, 12–3 pm, 3–7 pm, and 7–10 pm) the system randomly prompts participants to complete a brief survey. In order to ensure the momentary nature of the mEMA, participants are allotted 35 min to respond to the prompt by completing a 1-min survey, with the survey being available for 5 min prior to, and 30 min after, the text message prompt. Outside of these times, the mEMA surveys are not available to complete on the app. Because of Apple restrictions, the devilSPARC app was developed through Apple Enterprise and is available only through a study website. Trained research assistants assist with downloading the devilSPARC mEMA app to each participant’s smartphone and provide demonstrations at each data collection period on how to use the devilSPARC app. For participants who do not have android or iOS operating systems on their smart phones (n = 27), a data-enabled smart phone is loaned to them for the 1 week duration of each wave of mEMA data collection, a method that previous studies found successful [44, 45]. If participants have a new phone or need to update the app, trained research assistants assist with downloading the devilSPARC mEMA app. In addition, each participant receives paper instructions on how to download the app and detailed instructions are emailed and posted on the study website.

To minimize participant burden and increase response rates, each mEMA is limited to 5–8 questions as has been suggested in previous research [46, 47] addressing different weight behaviors and contextual factors about the behavior (e.g., who the participant is with and how they are feeling). Depending on the behaviors that the participant reports, a skip pattern is enabled (e.g., if participants are not being physically active, then questions about their physical activity are not viewable). Eating behavior measures in the mEMA are based on common foods related to weight among college students as identified by Laska and colleagues [48] and also foods reported in 24 h recalls in validation testing of the app [49]. See Fig. 2 and Table 2 for detailed descriptions of the mEMA items.

Student card data: Students use their student card to purchase food, attend university activities, and access facilities such as dining halls and recreation centers. University databases will provide student card data for students’ purchases and activities related to eating and PA, which will be linked to their friendship networks’ activities. The mEMA data will also be matched with participants’ student card activity. With this information (e.g., student access to campus recreation centers, purchases at certain food outlets and the dining halls), we will capture additional behavioral, temporal (time stamp), and geospatial data on what friends are doing together.

To analyze the collected data, we will use mixed model regression techniques to develop and test egocentric models (analysis derived from an “index” participant) over the course of the year [50]. Egocentric models will include associations between index student behaviors and behavioral measures of corresponding friendship networks (residence hall-level and floor-level) and friends (e.g., roommate, close friends and friend group). We will also examine moderation of these effects by contextual factors, such as location, time of day, and individual differences (e.g., gender, race/ethnicity). Friend effects require special attention in egocentric models. If two individuals in the study are friends with each other, then each person’s data could be used as both an outcome and a predictor in the same analysis, thus violating the generalized linear model’s assumption of independence among observations. Accordingly, we will use a generalized linear mixed (or multilevel) model framework. Due to the sampling design of this study, empirical clustering among students’ responses within residence halls and within floors of residence halls is likely. If warranted, we will account for these additional sources of non-independence among observations by including random intercept effects for different levels of nesting (e.g., index participants within floors, floors within residence halls). Effects of age, gender, race/ethnicity, and socioeconomic status (SES) will be adjusted for as necessary. Assumptions regarding linearity of associations and homoscedasticity will be checked via analysis and visual inspection of residuals. Further, we will test for temporal autocorrelation in the data and, as necessary, account for it (e.g., by specifying autoregressive error structures in our models). The choice of link function (e.g., identity, logit) and error distributions (e.g., Gaussian, binomial, Poisson) will correspond to the nature of the outcome measure being modeled. Though some continuous outcomes may not be normally distributed, based on the central limit theorem, we expect the large sample size proposed here to ensure that model estimates will be asymptotically unbiased [51]. Conservative power analyses indicate that the projected sample size of 1100 cases should afford power of .80 to detect small effects (ds < .25), assuming residence hall-level ICCs of .01 and alpha of .05.

Stochastic actor-oriented models (SAOMs) [52, 53] will be used to analyze influence on key outcomes while controlling for selection into friendships. The SAOM is a longitudinal model with separate functions to estimate change in behavior and change in friendship networks due to varied selection processes. This model form allows both behavior and friend selection to be modeled endogenously, such that we can untangle issues related to homophily (i.e., students choosing similar friends) and contagion (i.e., transmission of ideas from friend to friend) [21]. The SAOMs will also distinguish between the influence of friends, best friends, and roommates as predictors of individual behavior. The SAOMs will account for the embedded clustering in the sampling design due to students being sampled from floors within residence halls, constituting a multilevel network (i.e., floor networks nested within residence halls). SAOMs require “complete” network data that includes most individuals within a bounded setting, hence the saturated sampling design. Social network analyses is possible with the high saturation rates. For this study only floors with response rates in the neighborhood of 75 % saturation or greater will be included in the SAOM analysis. SAOMs will be used in addition to the egocentric models to address the study aims.