What to Do with Those Open-Ended Questions? Analyzing Textual Data Using Structural Topic Modeling to Understand How Parents Cope during Coronavirus(COVID-19) Stay-Home Orders in Singapore

Background and Purpose.
Open-ended questions are useful because they provide a view into a respondent’s thinking and unlike close-ended questions, they do not cue respondents to think of particular causes or processes. Yet, analyses of open-ended questions are relatively rare and when conducted are almost exclusively done through human coding. We used an alternative approach, the structural topic modeling (STM) that draws on developments in textual data analysis based on machine learning. STM allows researchers to (a) discover topics from the data, (b) examine how the prevalence of topics changes with respondents’ characteristics (e.g., age), and (c) quickly analyze voluminous data. We applied STM to survey data collected from an open-ended question on parenting during Coronavirus (COVID-19).

Due to stay-home orders, most parents in Singapore have to work and care for their children at home without relief from schools for a month. To understand the challenges of parenting, we implemented a survey to understand how parents cope with the stress of parenting. Since we aimed to quickly get the results out to inform practice, we used STM to analyze parents’ answers to an open-ended question on their coping.

Methods:
We collected survey data from parents who are age 18 and older, living in Singapore, and have children age 12 or younger. We disseminated the surveys on social media groups and to organizations related to families and parenting. The 12-minutes online survey was hosted on Qualtrics and has 50 self-reported questions about the impact of COVID-19, parenting, and demographics. Using STM, we analyzed 198 respondents’ answers on an open-ended question that ask respondents to describe what has helped them to cope with parenting in the last week. Model diagnostics using semantic coherence and exclusivity calculations indicate that a 5-topic model fits the data well. To get a more complete knowledge of the topics, two covariates - relationship to children (mothers or fathers) and the number of children - were used to examine associations with topic prevalence.

Results:
The 5 topics that emerged from the data relating to how respondents cope with parenting include: (a) Managing expectations of self, (b) Having a daily plan/schedule, (c) Communicating with spouse, (d) Relying on faith, and a positive mindset, and (e) Flexibility. The covariate “number of children” was positively associated with the topic prevalence of “Having a daily plan/schedule” (b = .06, p < .001). Mothers were more likely than fathers to use “managing expectations of self” to cope with parenting (b = .01, p < .05).

Conclusions and Implications:
COVID-19 and stay-home orders are both unprecedented events in Singapore and researchers know little or nothing about parenting during such a crisis. Using STM, we found five themes of how parents had cope with the stressors of parenting during “stay-home” orders. STM had allowed us to analyze a large set of responses in a short time to produce results that can quickly inform practice during a crisis. However, the results are exploratory and further analysis using human coders would be important to validate these findings.