Preferences and Demand for Mosquito Control among Dengue-Endemic Communities in Peñuelas, Puerto Rico: An Application of the Best-Worst Choice Model

ABSTRACT.

Aedes aegypti is the primary vector of dengue virus and threatens 3.9 billion people living in many tropical and subtropical countries. Prevention and reduction of dengue and other Aedes-borne viruses, including Zika and chikungunya, requires control of mosquito populations. Community mobilization and input are essential components of vector control efforts. Many vector control campaigns do not engage communities prior to implementation, leading to program failure. Those that do often conduct basic knowledge, attitude, and practice surveys that are not designed to explicitly elicit preferences. Here, we applied a novel stated preference elicitation tool, best-worst choice, to understand preferences, willingness to participate, and willingness to pay for mosquito control in dengue-endemic communities of Peñuelas, Puerto Rico. Findings revealed that the community preferred mosquito control programs that are 1) applied at the neighborhood level, 2) implemented by the local government, and 3) focused specifically on reducing disease transmission (e.g., dengue) instead of mosquito nuisance. Programs targeting the reduction of disease transmission and higher educational level of participants increased willingness to participate. Participants were willing to pay an average of $72 annually to have a program targeting the reduction of diseases such as dengue. This study serves as a model to engage communities in the design of mosquito control programs and improve stakeholders’ decision-making.

INTRODUCTION

Aedes aegypti is one of the most important disease-transmitting arthropods in the world. Its preferences for feeding on human blood and breeding in human-made water-holding containers in and around houses facilitates the transmission of arboviruses, including dengue virus (DENV), chikungunya, and Zika.^1–6 An estimated 3.9 billion people are at risk of DENV, with about 390 million infections occurring every year.^7,8 The WHO reported that the incidence of dengue has increased by 30-fold over the last 50 years.⁹ The occurrence of DENV has expanded globally, but its burden is highest in tropical and subtropical countries in Africa, the Americas, the Asia-Pacific region, and the Middle East.^8,10

The control of Ae. aegypti using conventional vector control methods (e.g., insecticide spraying) has been difficult to achieve and maintain, resulting in failure to prevent arbovirus epidemics and the rapid geographic expansion of Ae. aegypti.⁴ Community concerns around safety and environmental impacts can influence a community’s willingness to participate, support, and pay for such efforts.^11,12 Public perception of mosquito control can be affected by limited knowledge about the specific control measures targeting different mosquito species and by sudden increases in mosquito control activities only during outbreaks and epidemics of mosquito-borne diseases.¹³

Community investment and engagement in mosquito control is critical. Aedes aegypti is a human commensal that prefers to colonize peri-domestic environments. The presence and density of mosquito populations depend on human behavior and immediate household environments. Any private or public benefits conferred by vector control activities will directly affect the communities.^14–21 Adopting vector control strategies that are sympathetic to the community-level landscape and integrating community-led efforts into routine vector control programs can improve sustainability.^{15,18,22–24} Community-engaged vector control programs have proven sustainable and cost-effective in dengue-endemic countries, including Cuba, Singapore, Nicaragua, and Mexico.^16,25–27

The process of obtaining community input prior to implementing mosquito control interventions must be capable of eliciting preferences around key aspects of the interventions and developing an understanding of the value people place on mosquito control. If seen as a public good, acceptance and adoption will increase.²⁸ Stated preferences tools (SPTs) are used in economics for the non-market valuation of goods and services.^29,30 Stated preferences tools allow the estimation of the demand and preferences for programs or interventions yet to be implemented (i.e., new community-based mosquito control programs).^29,31–33 Several studies have been conducted to measure value and preferences for mosquito control, but the scope has been limited to adopting contingent valuation methods (CVMs) to study demand and preferences for insecticide-treated nets (ITNs).^19,34–36 Innovative SPTs have been applied to mosquito control to a lesser extent (i.e., discrete choice experiments, or DCEs).^11,37

In this study, we used best-worst choice (BWC), a novel SPT that combines best-worst scaling (BWS) with binary choice (BC; derived from DCEs) to estimate the preferences or utility and demand for a hypothetical good or service.^29,38–42 Best-worst choice instructs participants to perform two tasks. First, for the BWS, participants were asked to choose a “most important” and a “least important” attribute level (i.e., mosquito control programmatic feature) from a hypothetical mosquito control program, which elicited preferences. Second, for the BC task, participants were asked if they would participate (“yes”) or not participate (“no”) in that specific hypothetical mosquito control program, which elicited willingness to participate. The BC model allows for the estimation of willingness to pay (WTP), defined as the marginal rate at which a person substitutes a given attribute or attribute level for money.⁴³ Both BWS and BC have a theoretical foundation on the random utility theory, which assumes that individuals will choose the alternatives yielding the maximum perceived value of a utility from the choice sets.²⁹ To our knowledge, BWC has not been applied to assess mosquito control preferences and demand.

Puerto Rico has the highest risk of dengue in the United States. With regular seasonal transmission and periodic epidemics of dengue reported, Ae. aegypti is a major public health concern in Puerto Rico. Entomological and epidemiological surveillance has been used to guide control efforts led by the government and the private sector. Surveillance efforts have been concentrated in major metropolitan areas such as San Juan, Caguas, and Salinas, but not on smaller towns at risk of dengue.^44–49 Traditional control strategies for Ae. aegypti in Puerto Rico have included pyrethroid insecticide spraying (with ultralow-volume truck-mounted sprayers),⁵⁰ government-led source reduction (management or elimination of breeding sites),⁵¹ and application of larvicides,⁵¹ which have shown a minimal degree of success over the last 50 years. These approaches have limited coverage as deployment requires residents’ availability and acceptance and material and personnel resources that increase operational costs.^51,52 Given the pervasive nature of the Ae. aegypti, vector control strategies that achieve low coverage will not result in community-wide reductions in adult mosquito populations.⁵¹ To date, community-based mosquito control strategies have been tested but not successfully adopted in Puerto Rico.^53,54

The objective of this study was to understand preferences for a mosquito control program for Ae. aegypti with residents of Peñuelas, Puerto Rico, using a BWC approach. More specifically, this project developed a strategy to 1) determine what aspects of a mosquito control program are most/least important to residents; 2) assess their willingness/unwillingness to participate in a mosquito control program; and 3) estimate how much residents are willing to pay when considering a hypothetical mosquito control program. Peñuelas, a rural municipality of Puerto Rico, serves as an example with no prior history of mosquito surveillance or organized control, where public choice for a potential mosquito control program had not yet been taken into consideration. Results from the BWC model can provide critical information to guide resource allocation and improve satisfaction among consumers (i.e., community members) and may help overcome acceptance and engagement challenges in traditional mosquito control programming.

MATERIALS AND METHODS

Study area.

Our study area was Peñuelas, a small municipality located on the southern coast of Puerto Rico. There are approximately 6,900 households located across 13 neighborhoods.⁵⁵ In Peñuelas, as in many other municipalities of Puerto Rico, the population has a habit of storing water for drinking, bathing, cooking, watering plants, etc.⁴⁵ Peñuelas was greatly affected by Hurricane Maria in September 2017 and by an earthquake that measured 6.4 on the Richter Scale in January 2020, causing massive structural damages and population displacement.⁵⁶ Abandoned houses⁴⁷ and boats (in the coastal neighborhoods) and insecure water sources may facilitate Ae. aegypti proliferation throughout the communities. Currently, there are no consistent mosquito control activities, nor is there mosquito surveillance in Peñuelas.

Survey design and implementation.

We selected and refined the attributes and varying levels for the BWC questions based on literature, feedback from experts, and input obtained during key informant interviews with community leaders, government officials, and representatives of nonprofit organizations of Peñuelas. We also pretested the survey with 15 residents of Peñuelas (via Qualtrics) to limit the number of attributes and levels presented, selecting comprehensive and meaningful scenarios, and to reduce the tasks’ complexity.^{29,42,57–59} Those individuals in the pretesting phase were excluded if they lived within the areas to be sampled. The attributes included in our survey were implementation, location of application, focus, and annual costs of the program (Table 1). The inclusion of annual cost as an attribute allowed for the estimation of WTP for a hypothetical mosquito control program. The levels of the annual costs attributed were based on a previous study conducted in Key West, Florida, that estimated that people were willing to pay between $25 and $200 for the expansion of mosquito control.¹¹

Table 1.

Attributes and levels used to create the best-worst choice experiment questions for a hypothetical mosquito control program in Peñuelas, Puerto Rico

Attribute	Description	Levels
Implementation	Describes who is responsible for the implementation of a mosquito control strategy	Me or household member
		Private company
		Local government
Location	Options for where a mosquito control method can be applied	Private property only (outside of household)
		Neighborhood
Focus	Focus of a given program	Reducing disease transmission
		Reducing discomfort caused by mosquito bites
Annual cost	Annual fee per household for the implementation of a given program	U.S. $20
		U.S. $80
		U.S. $140
		U.S. $200

A full factorial design would have resulted in 48 possible combinations (i.e., 3 × 2 × 2 × 4 = 48) of hypothetical mosquito control programs based on the attributes and levels listed in Table 1. Therefore, we adopted a D-efficiency main effects design to reduce the total BWC questions, thus shortening the survey and reducing respondent fatigue.^29,42,60 The final design had a D-efficiency score of 93.8. A total of 12 scenarios with varying levels of the attributes listed in Table 1 were separated into two blocks of six questions each (see Supplemental Table 1). Each participant was randomly assigned to a block, using Qualtrics when the survey was administered with a tablet or by the principal investigator (PI) when paper surveys were administered. A total of 87 and 88 participants were randomized into Block A and Block B, respectively. The BWC tasks were developed using the guidelines for stated preferences choice modeling and Soto et al.^29,41

In the survey, participants were given a detailed explanation of all attributes and their levels. This was followed by a brief script or “cheap talk” to address hypothetical bias (arising from asking people to answer hypothetical questions), which made subjects aware of their hypothetical bias (e.g., be mindful and try not to overstate your WTP or willingness to participate).^61–64 Participants were asked to consider each of the BWC tasks separately, followed by asking participants the certainty of their response using a 10-point Likert scale (i.e., 1 was least certain, and 10 was most certain). This certainty scale was used to address hypothetical bias once again.⁶⁴ Supplemental Figure 1 shows an example of a BWC question included in the survey.

Sampling and recruitment.

The study population comprised adults aged 18 years or older who were residents of Peñuelas for at least a year. Participants were recruited following a modified two-stage sampling methodology based on the CDC’s Community Assessment for Public Health Emergency Response Toolkit.⁶⁵ This method was selected because of its potential to be implemented in low-resource communities that need mosquito control in dengue-endemic countries. First, 30 census blocks were randomly selected with a probability proportional to the number of households within each census block. The total number of households by census block was obtained from the 2010 U.S. Census occupancy status tables (Tables H1/H3).⁶⁶ Two census blocks had fewer than 10 households; therefore, we spatially joined them to their closest census block to increase the number of households available for sampling.⁶⁵ Second, we used a systematic random sampling to select seven households within each selected census block.⁶⁵ The household count (or the number of households in between the selected households) was calculated by dividing the total number of households within each selected cluster by 7. Because a cluster was selected twice, we selected 14 households. For our study, the target sample size was 210 households within 29 randomly selected census blocks.

The first household to be recruited was selected by creating a random point within each census block using ArcGIS Pro software. The selection of houses followed the serpentine method, which consisted of passing through the front or entrance of every house in the selected census block. We went door-to-door to recruit seven household heads to participate in the survey. If no one was home, a flyer containing information about the study, an announcement that another visit would take place, and the contact information of the principal investigator was left at the front door or gate of the house. Household heads approached at their residence had the following options: 1) answering the questions orally at the time of visit; 2) making an appointment for the interview; 3) completing the survey independently by either using a Qualtrics link or scanning a QR code from Qualtrics on their phones; or 4) not participating. Households were replaced if members declined, the house was vacant, or there was no response at the house after three visits.

Data collection.

Trained interviewers administered questionnaires from July 1, 2021 through July 21, 2021. The questionnaires were administered in Spanish, the local language, by trained field assistants. The questionnaire was administered via Qualtrics using a tablet or a paper version when no Internet connection was available in remote areas. The PI checked all the questionnaires for completeness. Prior to field mobilization, local field assistants were trained on recruitment, consent, and survey administration by the PI following the field training guide created by the Johns Hopkins Bloomberg School of Public Health.⁶⁷ The training also involved best practices for questionnaire administration and data quality enhancement procedures.

Econometric analysis.

A paired estimation approach, the conditional logit, was used to analyze the BWS data, as it accounts for correlation among observations for all 12 scenarios (six per person given the block design) for any given individual.^39,41,68,69 The BC data were analyzed using a random effects logit to adjust for heterogeneity in individuals’ responses (e.g., socioeconomic characteristics of participants, potential survey fatigue).^38,41 All categorical variables were entered using dummy coding. Two distinct model specifications were used for the BC data. Model 1 was analyzed using the stated responses to the binary questions as entered by participants, whereas Model 2 adjusted for hypothetical bias using the certainty scale calibration as the dependent variable (i.e., switching yes to no if certainty was less than 7). Both Model 1 and Model 2 controlled for sex (female or male), age (18 to 34 years, 35 to 54 years, or 55 years or more), and education level (less than a high school diploma, high school diploma, some college or an associate degree, bachelor’s degree or more). In addition, the attribute of “Cost” was quantitatively coded to obtain estimates of WTP. Here, we calculated WTP by dividing the coefficient of a given attribute by the coefficient of the cost attribute.^43,70 Results of the econometric analysis were interpreted following Louviere et al.,²⁹ as well as standard practice, with statistical significance denoted by P-values of < 0.01, < 0.05, and < 0.10.^{38,41,42,68,71} Best-worst choice data produce measurements for attribute impact for each attribute and level scale values for each level.³⁹ The attribute impact variable relates to the mean importance across all levels of an attribute, whereas the level scale values represent the location of each attribute level on an underlying scale of importance (e.g., number line).^39,72 One of the level scale values was omitted in the regression to serve as the base level or the zero in the number line.³⁹ All analyses were performed using Stata 14 (StataCorp, College Station, TX). For additional details on the econometric analysis, refer to Supplemental Methods 1.

Ethics statement.

Consent to participate was obtained verbally from participating household heads. The study was approved by the University of Arizona Institutional Review Board and was deemed to be no more than minimal risk (Protocol number: 2106904326).

RESULTS

Sample characteristics.

Household questionnaires were completed at 175 distinct households in 29 census blocks out of the 210 households intended to be interviewed. The response rate was 81% (based on all in-person contacts made), whereas the refusal rate was 18% (see Supplemental Figure 2). Most respondents were female (61.7%) and high school graduates (32.0%), and they earned a household income less than $25,000 annually (68.0%) (Table 2). Most participants were 55 years or older (57.1%). Our sample mostly consisted of people who had lived in Peñuelas for more than 10 years (78.3%), who owned a house (85.7%) and resided in Quebrada Ceiba (24.0%), Santo Domingo (20.0%), and Tallaboa Alta (17.7%) neighborhoods. A total of 45.1% and 29.7% of participants were retired and unemployed, respectively.

Table 2.

Demographic characteristics from survey participants with best-worst choice experiment questions for a hypothetical mosquito control program in Peñuelas, Puerto Rico, 2021 (N = 175)

Characteristic	n (%)
Sex
Female	108 (61.7)
Male	67 (38.3)
Age
18–34 years	19 (10.9)
35–54 years	56 (32.0)
55 years or older	100 (57.1)
Education of household head
Less than high school	36 (20.6)
High school graduate	56 (32.0)
Some college or associate’s degree	44 (25.1)
Bachelor’s degree or more	39 (22.3)
Annual household income
Less than $25,000	119 (68.0)
$25,000–$49,999	27 (15.4)
$50,000–$99,999	6 (3.4)
Prefer not to answer	23 (13.1)
Time residing in Peñuelas
1–9 years	38 (21.7)
10 years or more	137 (78.3)
Employment status
Employed	38 (21.7)
Unemployed	50 (29.7)
Retired	79 (45.1)
Student	5 (2.9)
Prefer not to answer	1 (0.6)
Type of residence
Own a house	150 (85.7)
Rent a house	17 (9.7)
Government/nonprofit subsidized housing	7 (4.0)
Prefer not to answer	1 (0.6)
Neighborhood
Coto	11 (6.3)
Cuebas	4 (2.3)
Jaguas	28 (16.0)
Pueblo	7 (4.0)
Quebrada Ceiba	42 (24.0)
Rucio	2 (1.1)
Santo Domingo	35 (20.0)
Tallaboa Alta	31 (17.7)
Tallaboa Encarnación	7 (4.0)
Tallaboa Poniente	8 (4.6)

Preferences for mosquito control.

For our BWS model, the attribute impact variable “location” was the most important, closely followed by “focus” (Table 3). The “annual cost” was the least important attribute, and it was omitted because it served as the reference case. For the level scale values, the coefficients of both “me or household member” and “private company” were negative and significant (P < 0.05 and P < 0.01, respectively). This indicates that having a mosquito control program implemented by the local government was significantly more important than one implemented by respondents (or a household member) or a private company. The least important level scale value was “private property,” whereas “neighborhood” was the level scale value with the highest level of importance. A mosquito control program focused on “reducing disease transmission” was significantly more important than a program focused on “reducing discomfort caused by mosquito bites” (P < 0.01). The highest level of importance among the annual costs attribute was “$200” (P < 0.05) in comparison to “$20.” It is worth noting that the coefficients of annual cost of “$140” and $200 were close in magnitude, which may suggest that respondents viewed both level scale values as similarly important or a potential ceiling effect for annual cost.

Table 3.

Results from best-worst scaling questions from survey participants in Peñuelas, Puerto Rico, using conditional logit analysis

Attribute	Coefficient (SE)*	P-value†
Attribute impact variables‡
Implementation	1.12 (0.17)	< 0.01
Location	1.58 (0.16)	< 0.01
Focus	1.54 (0.15)	< 0.01
Annual cost	0 [Ref]	–
Level scale values§
Implementation
Me or household member	−0.24 (0.12)	< 0.05
Private company	−0.82 (0.13)	< 0.01
Local government	1.06‖	–
Location
Private property	−1.18 (0.12)	< 0.01
Neighborhood	1.18‖	–
Focus
Reducing disease transmission	0.28 (0.09)	< 0.01
Reducing discomfort caused by mosquito bites	−0.28‖	–
Cost
U.S. $20	−0.54‖	–
U.S. $80	0.04 (0.12)	0.707
U.S. $140	0.24 (0.13)	< 0.10
U.S. $200	0.26 (0.11)	< 0.05
Number of respondents	175	–
Number of choices	12,600	–
Log likelihood	−2196.8784	–
χ2 Statistic¶	0.0000	–

Statistically significant results are in bold.

^*Numbers in parenthesis are standard errors.

^†Statistical significance denoted by P < 0.10, P < 0.05, and P < 0.01.

^‡Attribute impact variable relates to the mean importance of the attribute regardless of the levels.^39,72

^§Level scale values refer to the additional importance given or taken away from an attribute based on its levels.^39,72

^‖Effects coded: negative sum of the above level scale values corresponding to this attribute.

^¶Chi-square statistic of hypothesis that all model parameters are zero.

Willingness to participate.

For our BC data, two distinct model specifications were used. The certainty calibration improved our model estimates as more variables were significant in Model 2. As shown in Table 4, the coefficients of me or household member, private company, and private property were insignificant, suggesting the absence of an association between these program features and residents’ participation. As anticipated, reducing disease transmission was positive and significant (P < 0.05). This indicates that a program focusing on reducing disease transmission increases the participation decision. We also found evidence that education level was significantly associated with the likelihood of participating in a mosquito control program. Having a bachelor’s degree or higher significantly increased the likelihood of participating in a mosquito control program. Age and sex were not statistically significant.

Table 4.

Results from binary choice questions from survey participants in Peñuelas, Puerto Rico, using the random effects logit model

Attribute	Model 1		Model 2*		WTP Model
	Coefficient (SE)†	P-value‡	Coefficient (SE)†	P-value‡
Implementation
Me or household member	0.19 (0.40)	0.634	−0.38 (0.41)	0.356	–
Private company	0.08 (0.39)	0.843	−0.23 (0.41)	0.571	–
Local government	Ref	–	Ref	–	–
Location
Private property	−0.49 (0.35)	0.161	−0.04 (0.36)	0.908	–
Neighborhood	Ref	–	Ref	–	–
Focus
Reducing disease transmission	−0.08 (0.33)	0.816	0.72 (0.34)	< 0.05	$72
Reducing discomfort caused by mosquito bites	Ref	–	Ref	–	–
Annual cost (quantitative)	−0.01 (0.003)	< 0.01	−0.01 (0.003)	< 0.01	–
Sex
Female	−0.47 (1.23)	0.703	−0.93 (0.99)	0.347	–
Male	Ref	–	Ref	–	–
Age
18–34 years	Ref	–	Ref	–	–
35–54 years	−0.94 (1.94)	0.627	1.07 (1.63)	0.511	–
55 years or more	−2.84 (1.83)	0.120	−0.62 (1.55)	0.689	–
Education of household head
Less than a high school diploma	Ref	–	Ref	–	–
High school graduate	4.93 (2.14)	< 0.05	10.18 (1.44)	< 0.01	–
Some college or associate’s degree	5.21 (2.21)	< 0.05	12.06 (1.56)	< 0.01	–
Bachelor’s degree or more	5.29 (2.19)	< 0.05	12.44 (1.59)	< 0.01	–
Constant	6.62 (2.84)	< 0.05	−4.83 (2.04)	< 0.05	–
Number of respondents	175	–	175	–	–
Number of choices	1,050	–	1,050	–	–
Log likelihood	−261.81	–	−300.24	–	–
χ2 Statistic§	0.0028	–	0.000	–	–

WTP = willingness-to-pay.

Statistically significant results are in bold.

^*Model 2 includes the certainty calibration for binary choice.

^†Numbers in parenthesis are standard errors.

^‡Statistical significance denoted by P < 0.10, P < 0.05, and P < 0.01.

^§Chi-square statistic of hypothesis that all model parameters are zero.

Willingness to pay.

The coefficient of annual cost was quantitatively coded to allow for the estimation of WTP for those attribute levels that were significant. In our study, residents were willing to pay an average of $72 annually for a mosquito control program focused on reducing disease transmission (Table 4).

DISCUSSION

In this study, we implemented a novel stated preferences methodology using BWC to examine residents’ preferences and demand regarding different components of a mosquito control program. The BWS results from our survey showed a higher preference for a neighborhood-level application of the hypothetical mosquito control program. This was followed by having a program implemented by the local government while being focused on reducing disease transmission (i.e., dengue). The preference for a government-led mosquito control program was unexpected, as participants expressed their beliefs that the local government struggles to offer basic services to the residents and their mistrust toward the local government and its capacity to execute a program across Peñuelas. Moreover, participants preferred that the program focus on disease-transmitting mosquitoes over nuisance mosquitoes, which enables considering mosquito control strategies targeting Ae. aegypti mosquitoes. The annual cost attribute had the lowest overall importance, which coincided with the residents expressing that they prioritized other aspects of the program over the annual cost. Furthermore, during the interviews, participants stated that they are willing to pay for a mosquito control program that aims to protect their health, regardless of the cost.

In our study, we were able to assess whether the features of a mosquito control program affected the likelihood of a resident participating in the program with the BC questions. The analysis showed that the level scale values (or the additional importance given or taken away from an attribute) related to implementation and location did not influence people’s likelihood to participate in a mosquito control program. These findings differ from the BWS results, indicating that even though there is a preference for a government-led mosquito control program, this program attribute does not influence their decision of participating in the program. Likewise, a mosquito control program implemented at the household level or neighborhood level had no effect on their willingness to participate. Similar to the results of BWS, having a program targeting the reduction of health risks, in the form of disease transmission, had a positive effect on hypothetical program participation and its importance. Hence, this finding reinforces that a mosquito control program can specifically target Ae. aegypti mosquitoes, which transmit dengue in the study area, instead of all mosquito species (namely, nuisance mosquito species) across Peñuelas. In this study, those who had a higher education level were more likely to participate in a mosquito control program.

An important result of our study is that participants prefer mosquito control programs that focus on reducing disease risk over nuisance reduction. Bithas et al.⁵⁸ also showed that people preferred controlling mosquitoes that have a health impact over controlling for mosquito nuisance (especially nuisance-causing native mosquitoes) in Greece. However, it should be noted that the framing for health and nuisance impact by Bithas et al.⁵⁸ differed from ours in that they used two separate attributes to measure nuisance (i.e., night versus day nuisance), and health impact was linked to West Nile virus risk reduction. A recent study in Ponce, Puerto Rico, revealed that there was community-level support toward multiple mosquito control strategies specifically targeting Ae. aegypti to reduce disease transmission, including the release of Wolbachia-infected mosquitoes.⁷³ The qualitative and quantitative data collected for this study speak to the community-level support for mosquito control efforts, but they did not explicitly elicit preferences or measure willingness to participate or WTP for these efforts.⁷³ In our study, participants were willing to pay at least $72 for some form of mosquito control that targets disease-transmitting mosquitoes. Our WTP results are consistent with the findings from Dickinson and Paskewitz,⁷⁴ who showed that people were willing to pay at least $10 for mosquito control and up to $160 to control West Nile virus-transmitting mosquitoes in Madison, WI.

The present study had multiple strengths based on how it was designed and executed and on application of the BWC model. To our knowledge, this is the first time that BWC has been applied to measure preferences regarding mosquito control. Our paper adds to the limited literature on the preferences for control among communities at risk of mosquito-borne diseases. A scarce number of studies have assessed preferences for mosquito control using non-market valuation techniques, such as choice experiments and CVMs (often used to estimate WTP).^11,28,74,75 Among these, the overwhelming majority have focused on investigating preferences regarding ITNs in malaria-endemic countries, with the estimation of WTP being the most used method of CVM.^{34,36,76–79} Another strength of the current study was the systematic semi-random methodology for sampling and recruitment of participants. This sampling strategy and high participation rate allowed us to obtain a sample representative of Peñuelas with less potential for bias.⁶⁵ The strategy was implemented by a small team over the course of a 3-week period, which demonstrates the feasibility of implementing this type of assessment even in communities with limited resources. Community engagement was strengthened by having community leaders accompanying the field team and the principal investigator’s personal knowledge of the communities; both strategies facilitated navigation through the communities and outreach to household heads. The survey design integrated feedback from different people in key positions and community members in Peñuelas through key informant interviews and questionnaire pretesting. This resulted in a comprehensive and meaningful set of questions to be asked of the residents of Peñuelas and allowed us to have insight into the aspects of mosquito control programs that needed to be included in the BWC sets.

The BWC method has several advantages that result from implementing BWS and BC in combination. Best-worst scaling forces participants to choose two program features, a “most important” and a “least important,” which produces conditional demand data (as the information is conditional on a respondent inevitably choosing two alternatives) and also results in the estimation of parameters under a common utility scale that allows for comparisons.^39,41,42,72 The BC component enables the estimation of unconditional demand by allowing participants to opt out or pick a “not participate” option.^29,39,42 Furthermore, BWC measures utility directly with the BWS component by observing trade-offs among attribute levels and indirectly through BC by inferring the utility of attribute levels by measuring the “acceptance” or “rejection” of an entire choice set (which is consistent with traditional demand theory).^29,41–43 The BWS component of the BWC emphasizes the citizen behavior aspect of consumption and demand and characterizes the heterogeneity of preferences of an individual,⁷² although the BC results align with consumer behavior, as it reflects the decision-making process while purchasing or not purchasing a good or service. The integration of BWS enables the estimation of the importance or impact of a given attribute, as it is not embedded (or confounded) within the level scale values.^38,72 In other words, parameter estimates do not combine the relative impact or importance of a given attribute with its levels.^39,72,80 The BC model produces estimates for WTP that are not obtained from BWS.⁴³ Lastly, the utilization of BWC could help reduce survey exhaustion and obtain both measures of preferences and importance by having BWS and BC combined into a single choice task instead of having two separate survey questions to elicit preferences.^29,41,72

There were several limitations of the study. One of the major criticized aspects of stated choice methods is that the results produced may suffer from hypothetical bias and can cause an overestimation of the model parameters.⁶⁴ In our study, we applied two different but well-established methods, cheap talk and certainty scales, to reduce said hypothetical bias. The availability of current ArcGIS maps limited our ability to identify households prior to entering a community. Google Maps and community leaders’ knowledge were used to assist in navigation. Data collection was carried out during the SARS-CoV-2, commonly known as COVID-19, pandemic, which could have affected participation. The study team adhered to best practices of the CDC to prevent the spread of COVID-19 (e.g., wearing face masks). Another limitation of recruitment of households was the high number of abandoned and inaccessible houses located within the selected census blocks. The extent of the displacement caused by Hurricane Maria and the earthquake was observed in one census block where house replacement was not possible as the number of abandoned houses surpassed the number of occupied houses.

In terms of the BWC application, the concepts of mosquito control were kept as abstract qualities instead of describing a specific technology, which made it difficult to infer the specific methods for Ae. aegypti mosquito control that could be implemented. However, we have identified options of key mosquito control strategies that potentially meet the preferred characteristics. Mosquito control measures such as source reduction (or elimination of breeding sites), mosquito adult trapping, larviciding containers around houses, or more innovative control strategies (i.e., releasing Wolbachia-infected mosquitoes) are options that meet the communities’ preference for a mosquito control program focused on reducing disease transmission. Moreover, there is potential to explore strategies that combine government-led efforts with community-led efforts. For example, the local government can recruit volunteers to host the Gravid Aedes Trap (GAT) in their households. The GAT is a passive and low-cost trap used to capture gravid Ae. aegypti mosquitoes⁸¹ and can be easily managed by community members. The World Mosquito Project (WMP) had people hosting and monitoring mosquito traps and actively participating in local community committees working on progress in monitoring of the WMP Wolbachia Method in Brazil, and this approach proved effective in the implementation of the program.⁸² The WMP aims to reduce mosquito-borne diseases by releasing Ae. aegypti mosquitoes infected with the wMel strain of Wolbachia.⁸² The local government of Peñuelas can also work with the community to distribute and apply larvicides to the most productive breeding sites in the area, including bromeliads (results not shown here). In our survey, participants were also asked to rank four mosquito control strategies (1 = most preferred and 4 = least preferred). The GAT was ranked as the top choice at 32.0%, followed by insecticides with 28.0%, genetically modified mosquitoes with 24.0%, and Wolbachia mosquitoes with 15.4% (results not shown here). The results from our ranking exercise demonstrate that community-led strategies can supplement governments mosquito control activities.

Prevention and reduction of arboviral diseases, including dengue, continues to depend on controlling mosquito populations at the global scale.^14,83,84 Addressing community participation in mosquito control efforts is challenging. Using BWCs can provide knowledge about trade-offs that people make regarding mosquito control and their preferences and intention of participating in and paying for these services. When community preferences are taken into consideration prior to establishing a mosquito control program, it maximizes program success.^14–18 Stakeholders can leverage this methodology to efficiently develop a bi-directional relationship to develop mosquito control programs that suit their community.

Future studies should explore the relative role that income disparities may have on people’s preferences. Further research should be conducted to explore the environmental impacts of mosquito control strategies and how people’s perceptions about potential environmental impacts affect their support, acceptability, and preferences for mosquito control. In our study, participants shared their concerns about the negative effects that applying insecticides or releasing Wolbachia-infected mosquitoes could have on the environment (results not shown here). Moreover, our survey results showed that 85% of respondents believed that mosquito control strategies should prioritize the safety of both humans and the environment (results not shown here). There is a critical need to identify strategies to integrate top-down (government-led) and bottom-up (community-led) strategies. Methods where preferences are identified and then honored in programming could reduce the mistrust the community has in government programs. To reduce the disparity between those who pursue higher education and those who do not, educational programs focused on disease prevention and mosquito control should be integrated into the school’s academic curriculum from elementary school through high school. Lastly, a BWC approach can be adopted to measure people’s preferences for specific mosquito control strategies (i.e., application of pesticides versus at-home mosquito traps) that have the potential to be offered individually or in combination.

In conclusion, engaging communities to express their preferences in mosquito control programming yields information that can help key decision-makers and stakeholders. These results can improve upon the decision-making process by stakeholders, including the local government, public health officials, and community brigades, regarding controlling mosquito populations and the strategies adopted.

Note: Supplemental material appears at www.ajtmh.org.