The World Bank’s Indonesia Covid-19 Observatory is a Indonesia Brief multi-GP partnership that aims to generate usable and Covid–19 Observatory №2 near-real-time insights on the impact of the pandemic to inform the Government of Indonesia’s policy response to the crisis. P LE ASE DIRECT INQ UIRIE S TO L AURA RALSTON — LR A LSTO N@WO R LDBA NK .O R G 26.06.2020 Page 1/4 RAB IA ALI — R A LI 1 @WO R LDBA NK .O R G Social Media Monitoring A P R I L – M AY 2 0 2 0 ❶ Key Messages Discussion on COVID-19 in social media There is a high level of knowledge on spiked around the time large-scale COVID-19, with most respondents aware social restrictions were introduced and of common symptoms and following common topics of conversation revolved While the overall tone in the discussions preventative health measures. However, around health care, food access, and job on social media and in online news citizens face challenges maintaining loss. Complementary survey data finds sources has been moderating, the social distancing recommendations when that these are citizens’ biggest concerns, government handling of the situation is they need to purchase food and in their over half indicated working fewer days, widely discussed, and survey respondents jobs. These challenges are more prevalent and about a quarter facing some level of indicate concerns regarding social unrest among lower income groups food insecurity. and a wish for further government action. ❷ Findings 2.1 On awareness and behavior change Figure 1 Challenges in following social distancing and public health measures T here is a high level of knowl- Even though most people try to follow edge on COVID-19. For exam- social distancing, only 41 percent report Can’t avoid crowded ple, over 75 percent of online having no challenges in adhering to it. spaces when buying 37% survey respondents correctly Needing to go out to work, working in in- food identify the three main symp- teractive environments and being unable Can’t stay at home toms of COVID-19 as shortness of breath, to avoid crowded places when purchasing as I need to work outside 33% fever, and cough.1 This high level of aware- food are the main obstacles reported (Fig- Can’t avoid people in ness is followed by changes in behavior to ure 1). All these challenges are more severe my daily work 24% follow recommended preventative health for lower income groups. measures. For instance, 95 percent report- Important to social- ed washing their hands with soap more About 7 percent of respondents report ize in my community 17% often, 75 percent reported avoiding touch- moving back to their villages because of Need to visit mosque/ ing their eyes, nose, and mouth, 92 percent unemployment. This result is more mod- church 8% avoid social gatherings, and 91 percent est than other survey findings, where up to cancelled their travel plans for Idul Fitri. 38 percent of men and 34 percent of wom- 1 Data was collected online via social media from 3,375 These changes in behavior can also be ob- en report moving since the crisis began.2 survey respondents between 20th April to 20th May served through conversations on social Migratory movements are more common 2020 - see methods section for more details. media where #dirumahaja (stay at home) within districts or cities than across dis- 2 JPAL survey data reported at 10th May from data col- lected since 29th March 2020, covering 2,500 respon- is one of the most shared hashtags. tricts or cities, however. dents Indonesia Covid–19 Observatory 26.06.2020 Page 2/4 2.2 On concerns and vulnerabilities Figure 2 Distribution of topics A discussed on Twitter and Instagram ttention to COVID-19, as On the government handling of COVID-19, measured through online Indonesia’s national police and soldiers conversations, has peak- are widely discussed subtopics as they Twitter ed around important pol- are primary actors in ensuring that LSSR Confidence in government 27% icy announcements. For are implemented. Discussion around lo- Food access 14% example, the introduction of large-scale cal government and central government Travel restrictions 14% social restrictions (LSSR) or the ban on actions also emerges as a subtopic given the Idul Fitri mudik, led to large spikes their differing policies to help recovery Job loss 11% in the volume of tweets.3 Geographically, from pandemic. In late March, wet markets Health care 10% cities and districts with higher COVID-19 were a major topic under food access as Mask for all 9% caseloads, also have higher volumes of they changed the opening hours. On trav- Social distancing 6% tweets. In both Instagram and Twitter, el restriction, ‘mudik’ has been a trending Stigma 0.3% the government handling of the situation topic as many people are still discussing and food access are the top two most sa- whether to go home despite the govern- lient topics discussed (Figure 2). However, ment’s travel advisories. Instagram across time and across platforms there are variations in the attention to different top- The concerns raised through social media Food access 50% ics (Figure 3). For example, attention to job are also reflected in the online survey (Fig- Confidence in government 38% loss is more prevalent on Twitter and has ure 4). After getting sick and caring for the Health care 33% grown recently, while attention to health- sick, the next highest concerns are job loss Social distancing 27% care, food access and social distancing is and food access. Having enough to eat is a greater on Instagram. This may reflect the more pronounced concern among the poor- Travel restrictions 20% gender differences in user bases with more est, where 21% report it as their biggest worry. Job loss 15% females using Instagram and more males These results align with other crowdsourced Mask for all 14% using Twitter. survey efforts which find that getting infect- Stigma 4% ed is a primary concern, followed by the neg- ative short-term economic effects.4 3 This is true even after normalizing by total number of tweets per location to account for population density effects. 4 Premise Global Impact Survey. Indonesia results, N=2,749, as of 14th May 2020. Data collection started 2nd April 2020. Figure 3 Weekly topic trends on Twitter and Instagram Confidence in Twitter Instagram government 100 Food access 80 Health care 60 Job loss Mask for all 40 Social 20 distancing Stigma 0 MAR MAR APR APR APR APR MAY MAY MAR MAR APR APR APR APR MAY MAY 23-29 30 6-12 13-19 20-26 27 4–10 11–17 23-29 30 6-12 13-19 20-26 27 4–10 11–17 Travel –APR –MAY –APR –MAY restrictions 5 3 5 3 Figure 4 Top Concerns of Survey Respondents Caring for sick/elderly Getting sick Having enough to eat Losing job/not having No concerns household members enough work Total Indonesia Age group 18-34 35-50 50+ Gender Female Male Income Group < IDR 2 million IDR 2-5 million IDR 5-10 million IDR 10-20 million > IDR 20 million 0 50 100 Indonesia Covid–19 Observatory 26.06.2020 Page 3/4 2.3 Sentiment analysis Figure 5 Sentiments regarding government handling of the situation in T he overall sentiment tone governments. In relation to this up to 73 %, mid-April to mid-May 2020 on social media and online percent of respondents are worried about news is mildly negative, and social disorder and unrest. There is also a Q: Do you think the reaction by the in online news it has fur- wide perception (by 65 percent) that it will central government to the current ther moderated since the take more than 2 months for things to re- coronavirus outbreak is appropriate, end of April. However, the online survey turn to normal. too extreme or not sufficient? shows stronger negative citizen perspec- tives, as more than half of the respondents Comparison surveys confirm the dissatis- Too extreme felt that the government handling of the faction with government actions, although Appropriate situation has been insufficient. This was different surveys report different levels of with respect to both central and provincial magnitude.5 30% 5 Consortium of international researchers via covid19-survey.org find 84% Indonesians felt government response was insufficient (N=1,618; data collected between March 20-May 1, 2020), while Premise Global Impact Survey find 22% felt government response was insufficient (N=2,749; April 2- May 14, 2020). 68% Not 2.4 Sample representativeness and survey comparisons sufficient C ompared to the population of IDR 2-5 million group), and higher rep- adult Indonesians with inter- resentativeness in the Java-Bali region. Q: Do you think the reaction by your net access, the online survey Skewness to the younger age group reflects provincial government to the current sample is skewed to a young- the demographics of social media users as coronavirus outbreak is appropriate, er age group (18-34), lower in- both Twitter and Instagram users are heav- too extreme or not sufficient? come group (less than IDR 2 million and ily distributed in 19-29 age group. Too extreme Appropriate 40% 57% Not sufficient Figure 6 How well does the SM survey Figure 7 How well does the SM Survey capture capture Indonesia’s online population Indonesia's online population by income group? on age distribution? SM Survey Less than IDR 2-5 IDR 5-10 IDR 10-20 More than IDR 2 million million million million IDR 20 million Susenas 78% SM SURVEY 61% 22% 45% 15% 9% 9% 30% SUSENAS 14% 9% 9% 7% 32% 39% 12% 9% 18-34 35-50 50+ Figure 8 How well does the SM Survey capture Indonesia's online population by region? 80 SM Survey Susenas 60 40 20 0 Jakarta Java-Bali Kalimantan Maluku Nusa Papua Sulawesi Sumatera Tenggara Indonesia Covid–19 Observatory 26.06.2020 Page 4/4 ❸ Methodology Figure 9 Twitter Users Distribution Online conversations about COVID-19 ticles and 98 different sources. Data collec- Age are monitored through collecting Twitter tion for all sources began on in the last week 51% tweets and Instagram posts with COVID-re- of March 2020. lated hashtags and geotag information. This amounted to over 3 million tweets from To compare online conversation with om-the- 33% 410,000 unique Twitter users and 5.1 million ground situation, an online survey collected Instagram posts from over 1 million unique additional data from 3,375 respondents be- Instagram users.6 The content of these posts tween April 20th to May 20th. This survey was 8% 8% is analyzed using two approaches: (i) unsu- distributed through Facebook to targeted au- <=18 19 - 29 30 - 39 >= 40 pervised Latent Dirichlet Allocation (LDA) to diences and through the networks of social review trending topics, and (ii) word embed- media influencers. Ex-post survey weights ding methods to identify the volume of con- were applied to improve representativeness of versation on pre-identified topics. To com- sample. In order to provide a holistic view of the Gender plement this social media data, online news on-the ground situation, these survey results media data is accessed from GDELT reposi- were compared to other COVID-19 surveys tory of headlines covering over 200,000 ar- conducted in Indonesia since March 2020. 41% Female 59% Male 6 This user base has a young demographic with the majority being aged below 30 years old. Twitter has a stronger male user base, while Instagram has a stronger female userbase. Figure 9 Instagram Users Distribution Age 64% 27% 8% 0.3% <=18 19 - 29 30 - 39 >= 40 Gender 37% Male 63% Female