key: cord-0927975-t1eui4xz authors: Bansal, Prateek; Raj, Alok; Mani Shukla, Dhirendra; Sunder, Naveen title: COVID-19 vaccine preferences in India() date: 2022-03-07 journal: Vaccine DOI: 10.1016/j.vaccine.2022.02.077 sha: 2353951e1d89b31145ca5c3bdfe562a3a1bc6546 doc_id: 927975 cord_uid: t1eui4xz India’s mass vaccination efforts have been slow due to high levels of vaccine hesitancy. This study uses data from an online discrete choice experiment with 1371 respondents to rigorously examine the factors shaping vaccine preference in the country. We find that vaccine efficacy, presence of side effects, protection duration, distance to vaccination centre and vaccination rates within social network play a critical role in determining vaccine demand. We apply a non-parametric model to uncover heterogeneity in the effects of these factors. We derive two novel insights from this analysis. First, even though, on average, domestically developed vaccines are preferred, around 30 percent of the sample favours foreign-developed vaccines. Second, vaccine preference of around 15 percent of the sample is highly sensitive to the presence of side effects and vaccination uptake among their peer group. These results provide insights for the ongoing policy debate around vaccine adoption in India. India, the context that we investigate in the current study, is the country with the secondhighest incidence of COVID-19 with more than 34 million cases and over 480,000 deaths as of early January 2022 (World Health Organization, 2022) . The Indian government has tried to prevent the spread of the virus through measures such as mask mandates, lockdowns, and mass vaccination. There have been four major national lockdowns, in addition to numerous other state and local ones (Soni, 2021) . These lockdowns have controlled virus transmission, but they have also been accompanied by economic losses (World Bank, 2021) , social distress (Taylor et al., 2021) and increase in domestic violence (Ravindran & Shah, 2020) . While a combination of interventions are needed to curb the spread of the coronavirus, mass vaccination efforts hold the greatest promise for bringing an end to the current pandemic (Schwarzinger et al., 2021; Moor et al., 2021) . However, vaccination rates in India remain low in comparison to other countries across the globe, with only around 44 percent of Indians being fully vaccinated (Bloomberg, 2022) . In addition to supply side challenges related to vaccine procurement, prioritization, and distribution , there are several demand-side factors that have hindered vaccination efforts (Pal et al., 2021; Tarfe, 2021) . Recent studies suggest that vaccine hesitancy is one of the main hurdles across the globe (Corse, 2021; Kessels et al., 2021; Qin et al., 2021; Schwarzinger et al., 2021) . Vaccine hesitancy, estimated to be between 29 and 42 percent (Lazarus et al., 2021) , is potentially the primary driver of low vaccination rates in India. Hence, there is an urgent need to understand characteristics shaping preferences for the COVID-19 vaccine in the Indian context. A few recent studies have explored the preferences of Indians towards the safety and effectiveness of COVID-19 vaccine (Jain et al., 2021; Panda et al., 2021; Sharun et al., 2020) , but they do not quantify the effect of that various vaccine attributes have on preferences related to vaccines. Our analysis plugs this gap in the literature and is one of the first detailed investigations of the factors affecting COVID-19 vaccine preferences in India. Using a Discrete Choice Experiment (DCE), we quantify the sensitivity of consumers' vaccine preference relative to changes in various attributes such as efficacy, protection duration, side effects, price and administration location. Finally, we explore heterogeneities in the effectiveness of drivers of vaccine preferences using a non-parametric empirical model. Such a demand-side analysis is timely and critical because countries need to ramp up vaccination efforts in the face of the emergence of new variants. The findings of our study have the potential to shape ongoing policy discussions in India and other developing countries. The data for this study comes from an online discrete choice experiment (DCE) conducted between May and June 2021 2 . The respondents were recruited from a panel enlisted by MarketXcel, a market research agency in India. The sample consists of individuals who are i) above 18 years of age, ii) have not yet received COVID-19 vaccine, and iii) residing in one of the following five states -Maharashtra, Tamil Nadu, Uttar Pradesh, West Bengal, and Delhi. These states were chosen because they represent approximately 41 percent of the Indian population (as per Census 2011), and accounted for a disproportionately high number of COVID-19 cases/deaths in the country 3 . We use a quota sampling approach, where respondents were stratified based on gender, marital status, and age. In total, we received 1675 responses, but the final analysis sample consists of 1371 observations -we removed 304 observations because the survey completion time was less than 3.5 minutes (that is, half of the median survey completion time). The summary statistics are presented in Table 1 male, single, and younger age groups are overrepresented in our sample. DCEs are widely applied in health economics to study the trade-offs between attributes in individual-level preferences for health services (Clark et al., 2014; van Egmond et al., 2021) . More recently, DCEs have also been used to study the factors affecting the uptake of COVID-19 vaccine (McPhedran and Toombs., 2021, Dong et al., 2020) . In this study, we also use a DCE design, where the selection of attributes is guided by a thorough literature review. The review suggests that the prominent factors affecting consumer preferences for COVID-19 vaccine are -prevalence side effects (Borriello et al., 2021; El-Elima et al., 2021; Mouter et al., 2022; Reiter et al., 2020) , the origin of developer/manufacturer (Dong et al., 2020; Schwarzinger et al., 2021) , place of administration (McPhedran and Toombs., 2021; Kreps et al., 2020) , efficacy against the virus (Reiter et al., 2020; Leng et al., 2021) , out-of-pocket costs (Borriello et al., 2021; McPhedran and Toombs, 2021) , duration of protection (Dong et al., 2020) , number of doses required (Kreps et al., 2020) , vaccine adoption in social networks (Reiter et al., 2020) , source of COVID-19 related information (Reiter et al., 2020; Goruntla et al., 2020) . We include a bulk of these attributes in our experiment 4 . The levels of the attributes are chosen such that i) they span the entire attribute support, and ii) vaccine profiles are comparable to the ones that were available at the time of the experiment in India 5 . Table 2 presents details on the attributes that were included in our experiment. In the experiment, respondents were asked to choose their preferred vaccine between two alternatives based on vaccine attributes. Given the number of alternatives and the possible levels of different attributes, a full-factorial design would have meant a total of over 1.3 million choice scenarios (2 3 × 4 2 × 3 2 ) 2 . Since it is infeasible to present all these choice scenarios, we generated a DCE design using the D-efficient experimental design approach with zero priors (Kessels et al., 2006; Rose and Bliemer, 2009 ). The aim of the D-efficient design is to create a subset of all possible choice scenarios that optimize a function (e.g., minimises the determinant) of the asymptotic variance-covariance matrix, i.e. generate choice data that could result in the highest possible confidence in the parameter estimates for a given sample size. Our experiment consists of six blocks with six choice scenarios per block, and each respondent was shown a randomly selected block 6 . In each scenario, a respondent was 4 The number of doses and information about vaccines are not included because these attributes are not as relevant in the Indian context. 5 At the time of the survey, there were three vaccines that had been approved for use and had been deployed in India -Covishield, Covaxin and Sputnik. 6 There is no consensus in the literature against including or excluding the opt-out alternative (Ryan and Skåtun, 2004) . To circumvent potential misinterpretation of the opt-out alternative and avoid modelling challenges arising from a zero-level alternative, we do not present it to respondents. If we were interested in welfare estimation, an opt-out alternative could have been included. asked that "based on this information, which COVID-19 vaccine would you prefer to uptake?" An example choice scenario is presented in Table 3 . The proportion of friends and family members who has taken the vaccine 10% 90% Based on the above information, which COVID-19 vaccine would you prefer to uptake? Vaccine 1 Vaccine 2 We use a conditional logit (CL) model to estimate the effect of attributes on the preference for COVID-19 vaccine. The indirect utility under this specification can be expressed as: . (1) = ′ + In Equation (1) (2) = exp ( ) ∑ ∀ exp ( ) Additionally, we quantify the unobserved heterogeneity in main effects across different parts of the population using a non-parametric logit mixed logit (LML) model (Bansal et al., 2018; Train, 2016) . The indirect utility in the LML specification is: . ( A key point of departure from the CL model is that the LML specification contains random parameters , which are assumed to have a discrete mixing distribution over their ( ) finite support set (or multi-dimensional grid). The joint probability mass function of random parameters in LML is specified as follows: (4) In Equation (4), is assumed to be a spline function, and is the corresponding ( ) vector of parameters. The LML model is estimated using maximum simulated likelihood estimator, which is described in Bansal et al. (2018) . Standard errors are obtained using bootstrapping. The results from the CL model (columns 1-2, We show that Indian consumers have a lower probability of picking a vaccine developed outside of India (odds ratio = 0.93), while their odds of getting the vaccine when all of their peer group has been fully vaccinated is, on average, 1.41 times higher than those with an entirely unvaccinated peer group. We further probe for heterogeneities in the observed results using a non-parametric LML model. Instead of a single coefficient (CL estimates), the LML model provides a probability (cumulative) distribution function of the odds ratios -the results for each attribute are presented in Figure 1 . The graphs in Figure 1 demonstrate that the distribution of odds ratio varies significantly from the CL estimate, although the mean odds ratio from the LML model are similar in magnitude to the CL estimates (see columns 2 and 5 of Table 1 ). The heterogeneity in effects, as visible in Figure 1 , are statistically confirmed by the significant estimates of standard deviations (column 4 in Table 4 ). We uncover some important patterns from the LML estimates. Although the coefficient estimate of the CL model (column 1 of Table 4 ) suggests that there is a preference for vaccination at home, the cumulative distribution function (CDF) from LML in Figure 1 , however, indicates that 29.7 percent of individuals preferred vaccination in private/government hospitals. Note that the right vertical axis and blue curve in Figure 1 represent the CDF of the odds ratios from LML. The CDF value at the odds ratio of 1 for vaccination at home is 0.297, which implies that 29.7 percent of individuals have an odds ratio below one for vaccination at home (i.e., they preferred vaccination in private/government hospitals over home). Similarly, CL results suggest a higher inclination of individuals towards domestically developed vaccines, but the CDF of odds ratios from LML shows that 68.8 percent of individuals preferred domestic vaccine and the remaining 31.2 percent preferred foreign-developed vaccines. The LML results also demonstrate that around 12 percent of the sample have odds ratio of the vaccinated social network above five, while around 17 percent have a strong preference for a vaccine with no side effects (odds ratios > 6), which are very different from the CL estimates. Loglikelihood -5171.5 -5013.6 ***p-value < 0.01, **0.01 < p-value < 0.05, *0.05 < p-value < 0.15. The results from this study are largely congruent with recent evidence from other contextsvaccine efficacy, ease of inoculation, social networks and presence of side effects are important determinants of preferences for COVID-19 vaccine (see Dong et al., 2020; Schwarzinger et al., 2021) . An encouraging result from our analysis is that individuals are much more likely to choose a vaccine with more than 90 percent efficacy -at the time of the survey, two out of the three vaccines approved for use in India met this criterion. Although the aggregate effects favour a domestically developed vaccine, we also demonstrate that there is a sizable part of the population (31.2 percent) that has a higher likelihood of selecting a foreign-developed vaccine. This suggests that giving individuals a choice regarding vaccine type can potentially increase uptake. This is important from a policy perspective since at the moment most facilities in India do not provide such a choice. Our findings also suggest that providing vaccines in people's homes might further increase uptake, which suggests that the government could plausibly increase door-to-door vaccination efforts, especially in locations with low health facility density. It is important that these efforts should complement, and not replace, the current strategy of vaccination at health centres -this is because our data suggest that a considerable proportion (30 percent estimates the effect of only quantitative attributes on the individual's preference for COVID-19, but several attitudinal (e.g., trust in vaccine technology) and pandemic-related factors (e.g., hospitalization/death due to COVID-19 in one's social network) are not considered here. These omitted variables might induce biases in the parameters estimates. Future investigations might consider including questions to capture this kind of information to address omitted variable biases. Third, the data for the study comes from information collected using an internet-based survey-the usual caveats of such a data collection procedure applies to our study, especially that vulnerable demographic groups (e.g., old age group) with a lack of internet access might be underrepresented in this study. Despite these limitations, the present study provides timely quantitative evidence on factors that shape vaccine preferences in India, one of the nations that is in dire need of ramping up vaccination efforts. It is also one of the first detailed studies of determinants of preferences for COVID-19 vaccines in the South Asian context and will contribute to policy discussions on ways to expedite COVID-19 vaccine delivery in the region. Extending the logit-mixed logit model for a combination of random and fixed parameters 26 Billion Shots Given: Covid-19 Tracker Preferences for a COVID-19 vaccine in Australia Discrete choice experiments in health economics: a review of the literature Public preference for COVID-19 vaccines in China: A discrete choice experiment Acceptance and attitudes toward COVID-19 vaccines: A cross-sectional study from Jordan Predictors of acceptance and willingness to pay for the COVID-19 vaccine in the general public of India: A health belief model approach COVID-19 vaccine hesitancy among undergraduate medical students: results from a nationwide survey in India Public attitudes toward COVID-19 vaccination: The role of vaccine attributes, incentives, and misinformation A comparison of criteria to design efficient choice experiments Willingness to get vaccinated against Covid-19 and attitudes toward vaccination in general A global survey of potential acceptance of a COVID-19 vaccine Individual preferences for COVID-19 vaccination in China Rationing of a scarce life-saving resource: Public preferences for prioritizing COVID-19 vaccination Efficacy or delivery? An online Discrete Choice Experiment to explore preferences for COVID-19 vaccines in the UK Development of a discrete choice experiment to understand patient preferences for diabetes and hypertension management in rural Uganda. The Lancet Global Health 8 Please, you go first!" preferences for a COVID-19 vaccine among adults in the Netherlands Can states bring down vaccine prices? Yes, by coming together as a single buyer. The Indian Express Covid-19 vaccine, acceptance, and concern of safety from public perspective in the state of Odisha Chinese consumers' willingness to get a COVID-19 vaccine and willingness to pay for it Unintended consequences of lockdowns: covid-19 and the shadow pandemic (No. w27562) Acceptability of a COVID-19 vaccine among adults in the United States: How many people would get vaccinated? Constructing efficient stated choice experimental designs Modelling non-demanders in choice experiments COVID-19 vaccine hesitancy in a representative working-age population in France: a survey experiment based on vaccine characteristics Covid-19 vaccine acceptance: Beliefs and barriers associated with vaccination among the general population in India Effects of COVID-19 lockdown phases in India: an atmospheric perspective How India's Covid-19 communication strategy is failing to combat vaccine hesitancy Substance use and abuse, COVID-19-related distress, and disregard for social distancing: A network analysis Mixed logit with a flexible mixing distribution What do patients and dermatologists prefer regarding lowrisk basal cell carcinoma follow-up care? A discrete choice experiment