How the UN worked with big data to understand violence against women during COVID-19

COVID-19 has dominated search engines and social media over the past year. But for thousands of women in eight countries in the Asia-Pacific, among the most searched keywords were how to “cover bruises on face,” what are “signs of domestic abuse,” and whether a “one-stop crisis centre” was available. One of the biggest concerns amid COVID-19 is the increase in violence against women and girls. Early on in the pandemic, women’s rights organizations had voiced out the challenges in collecting information and measuring violence against women and girls. Population-based surveys, which require face-to-face interviews in safe, private spaces, are difficult to carry out given mobility restrictions in many countries. Many women and girls are stuck at home and are unable to speak out in the presence of their abusers. To address this issue, UN Women, in partnership with the U.N. Population Fund, International Labour Organization, and Quilt.AI, a company using big data and artificial intelligence to provide insights into human behavior, found a way to get a picture of what’s going on in select countries. “Detailed research into violence against women that has been safely and ethically conducted before the pandemic remains relevant for understanding the patterns and consequences of various forms of violence against women.” --— Sujata Tuladhar, technical specialist on gender-based violence, UNFPA's Asia-Pacific Regional Office A proxy measure of violence They analyzed data from online searches such as Google Trends, and public posts on social media channels including Facebook, Twitter, Instagram, YouTube, and Sharechat between September 2019 and November 2020. A total of 3,502 keywords were used, and over 20 million unique online searches were employed across eight countries: Bangladesh, India, Indonesia, Malaysia, Nepal, Singapore, Thailand, and the Philippines. The countries were chosen in consultation with UN Women and UNFPA, said Priyanka Bhalla, social impact director for Quilt.AI. These are where the U.N. agencies and their partners can directly follow up with survivors on the ground, including women that are migrating. The keywords used were based on questionnaires from demographic and household-level surveys that looked at different acts of violence in each country, as well as literature review, which included academic reports, insights from household and demographic surveys, documents published by international organizations and NGOs, news reports, blogs, and public Facebook discussions groups. The keywords were also reviewed by UN Women country offices to ensure local terms and contexts were included. These then were run in a search analysis tool, and historical search data was extracted to understand when the terms were increasingly used. Quilt.AI used Google mobility reports to identify peak periods of lockdown in each country, which they found to be in the month of April 2020, with the exception of May 2020 for Singapore. This helped them compare patterns of violence against women before, during, and after COVID-19 lockdown. “Before the Google mobility report, we also used month-on-month search data that gave us information on search patterns before, during, and after COVID-19 [lockdown],” Bhalla said. Sujata Tuladhar, technical specialist on gender-based violence at UNFPA's Asia-Pacific Regional Office, told Devex in an email that the online searches were used as a proxy measure of violence that could be indirectly linked to increases or decreases in women's experiences with violence, but that should not be taken as a direct measure of the level of violence happening in countries. “This is not about measurement of prevalence of VAW [violence against women],” she said. “This is a study that offers signposts to GBV [gender-based violence and] VAW by hinting strongly at the true picture, but not quantifying it,” she said. What the data says A key limitation of the study is people’s use of and access to the Internet. The data that was available was only from people who have access to the Internet and were able to use it amid the pandemic. This is why UNFPA advises that it’s not possible to compare results between countries. Another key limitation is that data contained little or no information on the individuals and their circumstances. “Data about online search terms and social media data, while being accessible and providing insights into people's interests, tell us little about the person searching/posting and their reasons for doing so or the context,” Tuladhar said. But the data showed a trend of increase in violence against women during COVID-19, particularly domestic violence, although help-seeking behavior online, based on help-seeking keywords such as “domestic violence hotline,” “sexual abuse lawyers,” and “sexual abuse counseling,” decreased in Indonesia and India. Bhalla said this could be due to a number of factors, such as service providers not being able to effectively share and advertise their information and resources online, making it difficult for survivors to find information. Mistrust in the services provided by the government and police could have also played a role, as well as survivors’ limited access to information during lockdown. An abusive relationship or abusive husband or partner was a dominant search term in India, Bangladesh, Thailand, Singapore, Philippines, and Indonesia. In Nepal, there was high interest in survivor or victim information. Key similarities between countries however include distrust in national security and justice institutions, as well as discourse on victim shaming. There was a rise too in online campaigns and support groups for women and girls who were victims of violence. However, there was little discussion on violence against members of the LGBTQI+ community. Opportunities and challenges Tuladhar said all data sources have strengths and limitations. But nontraditional means of collecting information, such as what they’ve done with big data from online searches and social media posts, have become an important data source for producing official statistics. “They provide new and cost-effective sources of information, especially when combined with traditional survey and administrative data and qualitative research,” she said. But while they provide useful insight, they only reflect a small portion of the data landscape. “Detailed research into violence against women that has been safely and ethically conducted before the pandemic remains relevant for understanding the patterns and consequences of various forms of violence against women,” Tuladhar said. Nevertheless, they plan to use the evidence from this survey to reinforce their call for strengthened policies and enhanced investments in survivor-centered services that address the needs of women survivors of violence.