Search

Introduction & Background Linking social media and survey data at the individual level has the potential to add evidence to a variety of research questions. To make this data openly available to others, social media data need to be converted into useful metrics that minimise issues of disclosure while maximising utility. This research explores linkages of Twitter data and survey data in the Understanding Society Innovation panel, focusing on the usage of non-disclosive metrics created from Twitter data alongside the similarly anonymised survey data. Objectives & Approach The Innovation Panel asked for consent to link Twitter data to survey responses and data has been collected from the Twitter API. However, Twitter’s unstructured nature necessitates creating measures that can be used jointly with linked survey data. We have developed a framework to create social media metrics that can be combined with survey data that also remove any disclosive data, so these data can be widely shared for maximum utility. The current research analyses these data to understand what the metrics look like through presentation of descriptive statistics. We also begin to show these data may be used in combination with survey data through inclusion of a set of metrics in logistic regression models predicting attrition and measurement of mental health. Relevance to Digital Footprints Social media is a prevalent aspect of social life and leaves a substantial digital footprint. However, there are a number of limitations to these data, including a lack of understanding of who is producing the data, and having the ability to relate these to a variety of specific (and possibly higher quality) measures for a representative sample of the population. Linkage to surveys address these problems and can lead to new research opportunities using digital footprint data. Results While small sample sizes impact the power of some analyses, the methods developed are illustrative of ways to use this novel data source. Results show that there is high variation in the created metrics, and initial analysis shows that the inclusion of a set of user-level Twitter data is not significantly related to attrition. However, more accounts followed on Twitter and the number of user retweets are significantly related to higher levels of mental distress on the GHQ scale. Conclusions & Implications Overall, there is some evidence that social media helps to understand survey outcomes, perhaps more so on measurement outcomes. This study provides an initial start on how to use these curated linked social media and survey data, and we note there are other social media networks that we can apply this strategy to; for example, LinkedIn, particularly with changes made to Twitter (X).

Introduction & Background LinkedIn, with its extensive global network of over 900 million members across more than 200 countries, presents a unique repository for examining labour market dynamics, professional development, and the impact of social networking on employment opportunities. Despite its potential, LinkedIn's wealth of data on professional trajectories, skills, and labour market outcomes remains largely untapped in survey research due to challenges in data collection. Objectives & Approach This paper introduces a novel methodology for integrating LinkedIn data with survey responses using data from the fourteenth wave of the Innovation Panel (IP14) of Understanding Society: The UK Household Longitudinal Study (UKHLS), conducted in 2021. In IP14, we probed the extent of LinkedIn usage among the UK population and assessed users' willingness to link their LinkedIn profiles with their survey responses. Those consenting to link their accounts were asked for specific details — namely their first and last names, employer, and job title — to enable profile identification on LinkedIn. Faced with the unavailability of a unique platform identifier and the cessation of LinkedIn’s API, this information was crucial for matching profiles accurately. We crafted a framework using PhantomBuster for ethical data extraction and a probabilistic string-matching technique to ensure precise linkage between survey responses and LinkedIn profiles. PhantomBuster, a cloud-based tool, efficiently scrapes dynamic content using JavaScript in a headless browser environment, sidestepping IP-related restrictions while adhering to website terms of service. It streamlines the data collection process. Identified profiles were subjected to an iterative probabilistic string matching, using respondent-provided metadata alongside supplementary data, to maximize the accuracy of matching the profiles to our survey participants. Relevance to Digital Footprints The described method advances digital footprint research in data collection and linkage. It automates the retrieval of vast online data sets; compiles information efficiently in an organized format; saves time and labour by mechanizing monotonous tasks; circumvents platform-imposed IP restrictions; and imposes fewer barriers to entry as it requires less technical skill than other scraping tools like Selenium. Conclusions & Implications This approach not only facilitates the precise identification and collection of LinkedIn profile data but also sets a precedent for ethical considerations in web scraping practices. By documenting this methodology, we aim to equip researchers with a scalable and replicable tool for future studies, enriching the analysis of labour market outcomes and the interplay between formal education, informal training, and professional success through the integration of LinkedIn and survey data.

Introduction & Background Uses and gratification (U&G) theory posits individuals’ engagement with social media is a deliberate effort to fulfill various needs, like information seeking, entertainment, and networking. However, prior studies predominantly addressed whether individuals use social media to satisfy their needs, leaving a gap in understanding how individuals behave online to satisfy needs. This study fills this gap by merging survey responses with actual Twitter activity, to investigate how individuals behave online to satisfy distinctive motivations, including (a) self-expression, (b) seeking entertainment, (c) business and working, (d) staying informed with news, and (e) networking. We also investigated how these online behaviors vary among individuals with different demographic features, including socio-economic classes, gender, and age. Objectives & Approach Our research addressed questions by linking survey responses with actual Twitter activities within the U.K. Participants were asked to provide survey responses surrounding age, gender, socio-economic class, and motivations for using social media. They were also queried about the existence of Twitter account, willingness to disclose Twitter username, and, if agreeable, the username itself. The survey continued until a total of 2,195 individuals shared Twitter handles. Following the removal of accounts that were either suspended or nonexistent, the study proceeded with a final count of 1,915. We collected each user’s Twitter metadata with Twitter API, including tweet count, follower count, following count, and bio information, and linked each user’s metadata with survey responses. To ensure respondents’ anonymity, survey, Twitter and linked data are stored separately, and can only be accessed by designated researcher. Relevance to Digital Footprints The study's approach of linking survey responses with actual Twitter activity offers a detailed insight into the digital footprints left by users as they engage with social media to satisfy their diverse needs. By analyzing the behaviors associated with motivations, this research illuminates the specific ways individuals curate their digital presence. Results Regression analysis indicated that individuals motivated by self-expression tend to tweet (b = .28, SE = .06, p < .001), follow account (b = .38, SE = .06, p < .001), gain followers (b = .13, SE = .06, p = .035), and post bio details (b = .89, SE = .13, p < .001). Work and business motivation leads to post bio information (b = .38, SE = .15, p = .012), while networking leads to follow more accounts (b = .28, SE = .06, p < .001). Social-economic class moderated associations between networking motivation and tweet count (b = -.25, SE = .09, p = .004), and between self-expression and tweet count (b = .20, SE = .08, p = .009). For individuals with higher socio-economic, self-expression has a higher effect on tweet count, whereas networking motivation has a less effect on tweet count. Additionally, we found gender moderated the association between self-expression and tweet count (b = .25, SE = .12, p = .04) and between keeping updated with news and tweet count (b = .11, SE = .05, p = .03). Conclusions & Implications These findings offer a nuanced understanding of social media usage, highlighting how different motivations influence specific online behaviors. The novel approach of linking surveys with actual social media activity provides a more accurate representation of user behavior, contributing insights for academic and practical social media strategy and design.

Abstract Linked social media and survey data have the potential to be a unique source of information for social research. While the potential usefulness of this methodology is widely acknowledged, very few studies have explored methodological aspects of such linkage. Respondents produce planned amounts of survey data, but highly variant amounts of social media data. This study explores this asymmetry by examining the amount of social media data available to link to surveys. The extent of variation in the amount of data collected from social media could affect the ability to derive meaningful linked indicators and could introduce possible biases. Linked Twitter data from respondents to two longitudinal surveys representative of Great Britain, the Innovation Panel and the NatCen Panel, show that there is indeed substantial variation in the number of tweets posted and the number of followers and friends respondents have. Multivariate analyses of both data sources show that only a few respondent characteristics have a statistically significant effect on the number of tweets posted, with the number of followers being the strongest predictor of posting in both panels, women posting less than men, and some evidence that people with higher education post less, but only in the Innovation Panel. We use sentiment analyses of tweets to provide an example of how the amount of Twitter data collected can impact outcomes using these linked data sources. Results show that more negatively coded tweets are related to general happiness, but not the number of positive tweets. Taken together, the findings suggest that the amount of data collected from social media which can be linked to surveys is an important factor to consider and indicate the potential for such linked data sources in social research.

Front-of-pack labels (FOPLs) provide simplified nutritional information that aims to inform consumer choice and encourage reformulation. We conducted an online randomised controlled experiment on a representative British sample to test the effectiveness of FOPLs across a range of outcomes. The primary outcomes have been published; here, we present the secondary outcomes: the ability to rank the healthiest product and the time to complete the rankings by comparing the FOPL groups and a no-label control, as well as a descriptive analysis of the perceptions. Participants from the NatCen panel were randomised to one of five experimental groups (Multiple Traffic Lights; Nutri-Score; Warning Label; Positive Choice tick; no-label control). Six food/drink categories were selected (pizza, drinks, cakes, crisps, yoghurts, breakfast cereals), and three products were created with varying healthiness. The participants (analytic sample = 4530) were asked to rank the products in order of healthiness twice (baseline: no label; follow-up: experimental group label). Compared to the control, the probability of correctly ranking the healthiest product at follow-up was significantly greater for the N-S, MTL and WL across all products. The time to correctly complete the ranking was fastest for the N-S, PC and no-label control. The descriptive analysis showed that the FOPLs were perceived favourably, and especially N-S and MTL. The findings were supportive of the primary analyses, with those results suggesting that N-S performed the best, and then MTL.

Front of pack food labels (FOPLs) provide accessible nutritional information to guide consumer choice. Using an online experiment with a large representative British sample, we aimed to examine whether FOPLs improve participants’ ability to identify the healthiness of foods and drinks. The primary aim was to compare ability to rank between FOPL groups and a no label control. Adults (≥18 years), recruited from the NatCen panel, were randomised to one of five experimental groups (Multiple Traffic Light, MTL; Nutri-Score, N-S; Warning Label, WL; Positive Choice tick, PC; no label control). Stratification variables were year of recruitment to panel, sex, age, government office region, and household income. Packaging images were created for three versions, varying in healthiness, of six food and drink products (pizza, drinks, cakes, crisps, yoghurts, breakfast cereals). Participants were asked to rank the three product images in order of healthiness. Ranking was completed on a single occasion and comprised a baseline measure (with no FOPL), and a follow-up measure including the FOPL as per each participant’s experimental group. The primary outcome was the ability to accurately rank product healthiness (all products ranked correctly vs. any incorrect). In 2020, 4504 participants had complete data and were included in the analysis. The probability of correct ranking at follow-up, and improving between baseline and follow-up, was significantly greater across all products for the N-S, MTL and WL groups, compared to control. This was seen for only some of the products for the PC group. The largest effects were seen for N-S, followed by MTL. These analyses were adjusted for stratification variables, ethnicity, education, household composition, food shopping responsibility, and current FOPL use. Exploratory analyses showed a tendency for participants with higher compared to lower education to rank products more accurately. Conclusions: All FOPLs were effective at improving participants’ ability to correctly rank products according to healthiness in this large representative British sample, with the largest effects seen for N-S, followed by MTL.