Recipient Organization
UNIV OF CONNECTICUT
438 WHITNEY RD EXTENSION UNIT 1133
STORRS,CT 06269
Performing Department
Allied Health Sciences
Non Technical Summary
Mothers can influence childhood obesity risk through what and how they feed their children. Many mothers turn to social media for health information and support around parenting issues, including childhood nutrition. Nutrition misinformation is common on social media, and much is unknown about how mothers vet child nutrition information they encounter. In the proposed mixed method study, we will explore how mothers evaluate the veracity of child nutrition information they encounter on social media and examine factors influencing the spread of child nutrition misinformation in online social networks. First, we will engage an expert panel of health care professionals and researchers to identify common child nutrition misconceptions. Second, we will survey mothers to explore how they evaluate the veracity of child nutrition information they see on social media. Finally, using publicly available social media posts, we will characterize content related to child nutrition and then examine characteristics associated with greater dissemination through online social networks using qualitative content analysis, natural language processing, and social network analyses. Findings will inform the development of public health campaigns and educational curricula that counter-messages common misconceptions related to child nutrition.
Animal Health Component
10%
Research Effort Categories
Basic
80%
Applied
10%
Developmental
10%
Goals / Objectives
The goals of this mixed method study are to explore how mothers evaluate the veracity of child nutrition information they encounter on social media and to examine factors influencing the spread of child nutrition misinformation in online social networks. First, we will engage an expert panel of health care professionals and researchers to identify common child nutrition misconceptions. Second, we will survey mothers to explore how they evaluate the veracity of child nutrition information they see on social media. Finally, using publicly available social media posts, we will characterize content related to child nutrition and then examine characteristics associated with greater dissemination through online social networks using qualitative content analysis, natural language processing, and social network analyses. Analyses will be descriptive and/or exploratory.
Project Methods
Step 1: Expert panel.Efforts: We will ask an expert panel of health care professionals and researchers to identify common child nutrition misconceptions. We will recruit experts (n=20) through our professional networks. Eligible experts must work clinically with mothers of infants/children (0-12 years) or conduct child nutrition research; comfortable completing the survey in English; able and willing to provide informed consent. Experts will complete a 10-minute online survey. We will email a $20 gift card to experts who choose to provide an email address. The survey will ask experts to list common child nutrition misconceptions they encounter in their work and on social media. To describe our expert panel, experts will also report their education, professional credentials, current position/employment, years of experience in the field, use of social media, and demographics.Evaluation: We summarize characteristics of experts using descriptive statistics. We will enumerate child nutrition misconceptions and rank them by frequency.Step 2: Survey of mothers.Efforts: We will survey mothers to explore what types of child nutrition information they encounter on social media, how they evaluate the veracity of information, and factors associated with higher perceived information credibility. We will recruit participants (n=200) via social media. Recruitment postings will direct users to an online eligibility screener. Eligible participants will be aged 18 years or older; is a mother of at least one child 0-12 years old; has an account on at least one major social media platform (Twitter, Facebook, Instagram, Pinterest) and has engaged on the platform in the past 7 days; currently lives in the United States; comfortable completing the survey in English; able and willing to provide informed consent. Eligible women will be presented an information sheet and asked to indicate informed consent before proceeding to a 30-minute survey. We will email a $20 gift card to participants who provide an email address. To minimize potential for duplicate or fraudulent responses, we will include quality and attention checks in the survey, and limit incentives to valid responses from unique email addresses. The survey will include both closed- and open-ended questions about how mothers evaluate the veracity of child nutrition-related information they encounter on social media. We will prepare examples of social media posts based on the child nutrition misconceptions identified by our expert panel, additional items based on national child nutrition guidelines. Example posts will include both child nutrition misinformation and evidence-based recommendations. For both posts containing evidence-based recommendations and misinformation, we include both positively- and negatively-framed posts. Mothers will rate the perceived credibility of each post and provide open-ended comments on how they determined credibility. We will also ask mothers how they would likely respond to the post if they saw it in their social media feed. We ask mothers from what sources they see child nutrition information, including specific accounts. Mothers will rate the credibility of child nutrition information they see on social media generally, and by source, the size of the poster's following, post format, and frequency of exposure.Evaluation: We will conduct a directed content analysisof responses to open-ended questions about how mothers evaluate health information encountered on social media. We will develop an initial codebook based on initial review of responses. We will revise the codebook to incorporate additional themes emerging during the initial review. The team will discuss initial results and finalize themes before final independent coding and consensus. We will calculate inter-rater reliabilityand will reach consensus on any discrepant coding through discussion. We will use linear regression models to examine post characteristics (e.g., source, format) associated with higher perceived credibility (measured on 0-10 scale anchored with "not at all" and "a lot"). To estimate multivariate associations, we will add variables to the model one at a time, in order of largest crude relationship. The most informative model for each outcome will be chosen after considering estimated betas, 95% CIs, and AIC for the model. Results from these qualitative and quantitative analyses will provide insights onto how mother evaluate the veracity of child nutrition information they see on social media.Step 3: Analysis of child nutrition misconceptions on Twitter.Efforts: We will characterize child nutrition tweets and examine which characteristics are associated with greater spread through online social networks. We will finalize our sampling strategy based on findings from the survey of mothers (Step 2). We will identify tweets based on search terms, hashtags, or specific accounts (e.g., mommy bloggers), as we have done previously.71We will only sample publicly-posted tweets. We have tools for downloading publicly-posted tweets, including the NCapture add-on for Chrome and NVivo (QSR International, Melbourne, Australia) and previously-developed Python scripts. We will download tweet data, including tweet URL, username, tweet text, hashtags, tweet date, replies, and likes. We will review tweets to exclude those not written in English and any not relevant to the current project, and will prospectively collect tweets until we achieve a sample of 1,000 tweets.Evaluation. We will use a combination of traditional qualitative content analysis methods and natural language processing techniques. We will conduct a directed content analysis to categorize tweet characteristics. We will evaluate whether the information included in the tweets are consistent with current recommendations for child nutrition, and categorize tweets and replies as consistent with recommendations, contradicts recommendations, or neutral. In addition, we will perform several natural language processing analyses to validate the results from qualitative content analysis and provide additional information: (1) we will perform topic modeling to uncover the hidden thematic structure of the tweets and extract relevant topics; (2) we will perform sentiment analysis to identify the emotion of tweets; (3) we will perform additional content analysis to identify the extent of cognitive and social processes of the tweets; (4) we will build a machine learning model (e.g., neural network) to predict the likelihood of a tweet to be misinformation. These analyses and resulting machine learning model will provide insights into post characteristics associated with greater likelihood of being misinformation, and can be used to identify childhood nutrition misinformation on social media and to inform counter-messages and educational materials to help mothers vet child nutrition information they encounter on social media. We will characterize the spread of misinformation using social network analyses. Specifically, we will construct the retweet network of each post by tracking how it is retweeted, and characterize the dynamics of the spread of misinformation over time through visualization and network modeling. Furthermore, we will derive dissemination metrics from the constructed retweet networks, such as the scale (number of retweets), structural virality (average distance between all pair of nodes in the network), and number and variance of the connected components (subnetwork in which any two nodes are connected to each other by path). Then, we will examine the associations between post characteristics and dissemination metrics using crude and multivariable-adjusted regression models. We will add post characteristics bivariately associated with dissemination metrics at p<0.10 to the adjusted model one at a time, in order of largest to smallest association; variables associated with the outcome at p<0.05 will be retained in final adjusted models.