The value of COVID-19 tests in Latin America.

Testing is a crucial strategy to control the spread of a pandemic. Voluntary participation in this strategy will depend on individual preferences towards and willingness-to-pay (WTP) for test results. We distributed a web-based, contingent valuation survey to social-media users in 16 Latin American countries to evaluate regional attitudes towards the emerging COVID-19 outbreak and WTP for COVID-19 testing. We observe that the cost of the test and household income are important determinants of testing intentions. We find higher WTP among those reporting greater concern relative to the average respondent. Accounting for uncertainty, our results indicate a WTP of approximately $45 dollars or 4.2 % of monthly income among respondents. These results, paired with our predicted participation rate of between 84-94 % for a $1 test, suggest that local officials will be able to effectively recruit participation in this mitigation strategy given the appropriate subsidization structure.

Introduction

Widespread testing is deemed essential for mitigating global health and economic crises resulting from a pandemic. Yet, little is known about individual preferences (i.e. willingness to pay and intended uptake) for diagnostic tests during these times of crisis. An improved understanding of those preferences may help design public policies aimed at increasing testing rates. We conducted a contingent valuation study to estimate the value that individuals assign to testing for COVID-19 in 16 Latin American countries. We also designed a split-sample treatment to investigate time preferences for test results. To the extent of our knowledge, this is the first study on individual preferences for COVID-19 testing in Latin America and elsewhere. Our study is policy relevant for a region that is in initial stages of the pandemic, and where testing strategies can help flatten the epidemic curve and prevent economic collapse. Our findings indicate that there is a latent demand for COVID-19 tests.

Materials and methods

Latin America provided a unique survey site, as one of the last regions in the world to have a confirmed case of coronavirus ([COVID-19 Data Repository by the Center for Systems Science and Engineering (CSSE) at Johns Hopkins University, 2020). Within our sample, the earliest detected cases were in Brazil (2/26/2020), Mexico (2/28/2020) and Ecuador (2/29/2020), and the latest, Nicaragua and El Salvador (3/18/2020). As shown in Table 1 , over the course of two weeks when we collected data, the number of confirmed cases rose from a minimum of 5 additional cases reported in Nicaragua (a 500 % increase relative to 3/19 baseline) to 10,509 (1692 % increase) in Brazil.1 Detection strategies varied across countries. By mid-June 2020, Chile was conducting about 30,000 tests per million people, and Brazil 4000 tests per million people. Peru followed a different strategy based on rapid testing to improve detection (Ponchner, 2020). As of July 2020, out of the 16 countries in our sample, only Ecuador and El Salvador had an open public testing policy. Eight countries tested individuals with symptoms who met specific criteria (e.g. essential workers and returned from overseas), and six countries tested anyone with symptoms. Across the region, more testing is required to prevent the rapid spread of the virus in the region.

Table 1

Evolution on COVID-19 in Selected Countries and Sample Share by Country.

	Index Case Datea	Confirmed Cases Over Sampling Perioda		Testing Policy as of July 2020b	Unweighted Share of the Sample	Weighted Share of the Sample
		Beginning (3/19/2020)	End (4/5/2020)
Argentina	3−3-20	97	1451	Symptoms & Key Groups	0.213	0.087
Bolivia	3−10-20	12	157	Symptoms & Key Groups	0.039	0.021
Brazil	2−26-20	621	11,130	Anyone with Symptoms	0.070	0.316
Chile	3−3-20	238	4471	Anyone with Symptoms	0.066	0.039
Colombia	3−6-20	102	1485	Anyone with Symptoms	0.141	0.100
Costa Rica	3−6-20	69	454	Symptoms & Key Groups	0.020	0.011
Dominican Rep.	3−1-20	34	1745	Anyone with Symptoms	0.031	0.021
Ecuador	2−29-20	199	3646	Open public testing	0.041	0.031
El Salvador	3−18-20	1	62	Open public testing	0.054	0.012
Guatemala	3−13-20	9	61	Anyone with Symptoms	0.037	0.031
Honduras	3−11-20	12	268	Symptoms & Key Groups	0.048	0.018
Mexico	2−28-20	164	2143	Symptoms & Key Groups	0.086	0.220
Nicaragua	3−18-20	1	6	Symptoms & Key Groups	0.076	0.012
Panama	3−10-20	109	1801	Anyone with Symptoms	0.011	0.008
Peru	3−6-20	234	2281	Symptoms & Key Groups	0.033	0.066
Uruguay	3−13-20	94	406	Symptoms & Key Groups	0.034	0.007

Sources:

CSSE COVID-19 Data Repository.

b/See Roser et al., 2020https://ourworldindata.org/coronavirus-testing#testing-and-contact-tracing-policy (last accessed on July 11, 2020).

We designed a web-based survey to gather information regarding experiences, risk perceptions, and preferences for protective measures in the context of the COVID-19 pandemic. The survey had 28 questions, two of which had conditional follow up questions. We used a snowball sampling strategy to recruit respondents from 16 Latin American countries (see Table 1), with initial recruitment based on social media advertising.2 We administered the survey between March 19 and April 5, 2020. After applying data quality controls (e.g. outliers in sociodemographic variables and inconsistencies in household composition indicating survey inattentiveness or manipulation), our sampling procedure yielded 5504 completed surveys.

Following best practices (Boyle et al., 2017; Johnston et al., 2017), we designed a contingent valuation question to elicit the respondent’s uptake intentions and willingness to pay for a COVID-19 test. The time it would take to get test results could vary among one, two, and three days. The price of the proposed vaccine could take a value of $10 to $50 in increments of $10 (see our experimental design in Table 2 ).3 Before responding whether or not they would pay that amount, respondents were reminded of their budget constraint to imprint realism to their choice. The (translated) CV question presented in the survey read as follows:

Table 2

Observations by Experimental Design.

	Waiting Time for Test Results
Cost	1 day	2 days	3 days	Total
$10	370	345	370	1085
$20	394	365	367	1126
$30	337	358	385	1080
$40	393	376	352	1121
$50	354	362	376	1092
Total	1848	1806	1850	5504

For the following question, assume that you suspect that you or a member of your household is infected with COVID-19. Also assume that there is a test available in any clinic or laboratory that can detect in [1 day, 2 days, 3 days] whether someone is infected with COVID-19. That test would cost [$10, $20, $30, $40, $50]. Keep in mind that the money you spend on the test will not be available for other needs at your home (e.g. food, clothes). Would you pay for the test? (Yes/No)

Using a follow-up question, we elicited the certainty level of respondents regarding their answer to the contingent valuation question. The certainty question was based on a four-point scale varying from very unsure to very sure. We used these replies to reduce hypothetical biases related to the uncertainty of respondents regarding their decision to be tested (Blumenschein et al., 2008), which may be particularly important in the initial stages of a disease outbreak. A number of hypothetical versus real comparisons using certainty corrections show that hypothetical bias can be mitigated by recoding favorable responses as negative ones when certainty levels are relatively low (Vossler et al., 2003). Recent contingent valuation studies have utilized similar certainty scales to reduce hypothetical biases (e.g. Ryan et al., 2017; van den Berg et al., 2017v; Vásquez and Rezende, 2019).

Theoretical framework and econometric modeling

The CV method is based on a utility-theoretic framework (Boyle et al., 2017). In the context of this study, the indirect utility of a household (V) is a function of household income (Y), information provided by the COVID-19 test (T), and prices of other goods (P), i.e. V(Y, T, P). The framework assumes that individuals derive utility from income and from knowing whether they are infected with COVID-19, as that information will allow them to proceed accordingly. The maximum amount that a person will pay to be tested (T → T) is equivalent to the income loss that would return their utility to the original level after the test [i.e. V = V)]. This implies that WTP is related to the individual’s income, health information, and prices of other goods.

Given the binary format of our CV question (Yes/No), we indirectly identify the maximum WTP using the equivalence between the probability of being willing to pay for the proposed test and the probability that the respondent’s willingness to pay exceeds the cost of the test presented in the contingent scenario [i.e., Pr(Yes) = Pr(WTP > COST) = Pr(LNWTP > LNCOST)]. Consequently, the probability of a positive response (Pr) can be modeled using the following logit specification:where X is the vector of covariates, α and δ are coefficients to be estimated, and the stochastic error term e is assumed to follow a logistic distribution.

Table 3 presents the variables included in vector X. The variable LNCOST is included to investigate how the cost affects the likelihood of being tested. According to our theoretical framework, we expect this effect to be negative because such payment would reduce income available for other goods and services. Two indicators, WAITING2 and WAITING3, depict differentials among our split-sample treatments. Both indicators are expected to have negative effects assuming that individuals would prefer to have information sooner rather than later. We also included sociodemographic variables to control for the heterogeneity of respondents. Based on our theoretical framework, the likelihood of paying for the test increases with household income. The effect of other individual and household characteristics remains to be empirically estimated. All models include country fixed effects and a day-specific time trend to account for heterogeneity across nations and over time as disease burden and anxiety grew.

Table 3

Variables Definition and Descriptive Statistics.

Variables	Definition	Unweighted Mean	Weighted Mean
LNCOST	Natural logarithm of the out-of-pocket cost of the test presented in the contingent scenario	3.262	3.262
WAITING2	If the time to get results is 2 days in the contingent scenario (1=Yes; 0=Otherwise)	0.328	0.333
WAITING3	If the time to get results is 3 days in the contingent scenario (1=Yes; 0=Otherwise)	0.336	0.334
CONCERNED	If the respondent is very concerned about COVID-19 (1=Yes; 0=Otherwise)	0.578	0.546
FEMALE	Respondent’s sex (1=female; 0=male)	0.765	0.584
AGE	Respondent’s age (in years)	36.209	39.992
EDUCATION	Respondent’s education (in schooling years)	15.363	16.382
HOUSEHOLD	Number of household members	4.567	4.175
INCOME	Monthly household income (in 1000s US$)	0.777	1.074

The amount that the median household is willing to pay for a COVID-19 test (median WTP) can be computed as the exponential of the linear combination of the average of covariates other than LNCOST and corresponding estimated logit coefficients scaled by the negative reciprocal of the logit coefficient of LNCOST:where is a vector that includes the weighted sample means of covariates (as reported in Table 3) and the sample share of each country (as reported in Table 1), is the coefficient of LNCOST, and is a vector of logit coefficients of the remaining covariates.4

In addition to using the original responses of the full sample of respondents to estimate Eqs. 1 and 2 (Models 1 and 2 in Table 4 ), we used the follow-up certainty question to recode positive responses as negative if the respondent was uncertain about her decision to be tested for COVID-19 (Models 3 and 4). This approach produces more conservative and arguably more precise WTP estimates (Blumenschein et al., 2008), relevant during a time of substantial uncertainty. We also corrected for potential coverage biases that are typical in web-based surveys,5 following an iterative proportional fitting (raking) procedure to generate weights that estimate the total population by country of residence, sex, and age (Kolenikov, 2014). In total, we divided the population in the selected Latin American countries in 192 groups (16 countries x 2 sex groups x 6 age groups: 18–24, 25–34, 35–44, 45–54, 55–64, and 65+). In the next section, we present unweighted (Models 1 and 3) and weighted results (Models 2 and 4).

Table 4

Logit Models of Willingness to Pay for a COVID-19 Test (Marginal Effects).

	Model 1:Raw Data	Model 2:Raw Data	Model 3:Corrected for Uncertainty	Model 4:Corrected for Uncertainty
LNCOST	−0.109(0.008)***	−0.106(0.016)***	−0.103(0.012)***	−0.117(0.017)***
WAITING2	0.013(0.011)	0.039(0.012)***	−0.001(0.013)	0.015(0.010)
WAITING3	0.007(0.009)	0.030(0.013)**	−0.026(0.012)**	−0.025(0.017)
CONCERNED	0.051(0.012)***	0.058(0.013)***	0.048(0.011)***	0.043(0.012)***
FEMALE	0.038(0.012)***	0.027(0.016)*	0.014(0.014)	0.020(0.017)
AGE	0.000(0.000)	0.000(0.000)	0.001(0.000)*	0.001(0.001)
EDUCATION	−0.001(0.002)	0.001(0.002)	−0.001(0.002)	0.001(0.002)
HOUSEHOLD	−0.000(0.002)	−0.000(0.003)	−0.002(0.002)	0.001(0.003)
INCOME	0.149(0.025)***	0.086(0.022)***	0.141(0.018)***	0.091(0.019)***
Pseudo R2	0.0645	0.0596	0.0733	0.0756
Weights	No	Yes	No	Yes

Notes: Observations = 5,504. Standard errors, clustered by country, are reported in parentheses. ***, ** and * imply statistical significance at 1%, 5% and 10% level, respectively. All models include country-fixed effects, and a variable representing the day the survey was completed to depict effects of the growth in reported cases.

Results

Table 4 shows the marginal effects of covariates on the likelihood of paying for the COVID-19 test based on four logit models. Estimation results are relatively consistent across all models. The coefficients of LNCOST suggest that the likelihood of paying for the COVID-19 test decreases with each additional dollar of the cost of the test. Our findings also indicate that, depending on the model, individuals who are very concerned about COVID-19 would be four to six percentage points more likely to get the test than those with lower levels of concern.6 Similarly, estimated coefficients for INCOME are positive and statistically significant, suggesting that the likelihood of testing increases with household income.

The indicators for the time it takes to get test results (WAITING2 and WAITING3) were statistically insignificant in the unweighted model (Model 1), but significant in the weighted model (Model 2), in which we observe a four and three percentage point increase in participation with longer wait times, a counterintuitive result. However, once we correct for uncertainty, the sign changes for WAITING3, and each coefficient becomes insignificant. These results could indicate that respondents perceive longer wait times as correlated with accuracy, and that individuals understand that, amid a pandemic, it takes time to receive test results. Individual characteristics other than household income were statistically insignificant with the exception of FEMALES in Models 1 and 2. That coefficient suggests that females would be between three and four percentage points more likely to get tested than males, although that effect is not significant after correcting for respondents’ uncertainty.

Fig. 1 shows predicted uptake rates across different costs for a test that would yield results in one day. Model 3 (unweighted, corrected for response uncertainty) provided the most conservative estimates. Depending on the model specification, the predicted uptake rate would be between 84 % and 94 % for an out-of-pocket test cost of $1. In contrast, the predicted uptake rate could be as low as 35 % for a test cost of $200.

Fig. 1

Test Uptake Rates.

Notes: Results based on the raw reported data (certainty-adjusted data) are presented in gray (black), and the weighted estimates are shown with a dashed line. Model 1 – Raw Data, Unweighted; Model 2 – Raw Data, Weighted; Model 3 – Data Corrected for Uncertainty, Unweighted; Model 4 – Data Corrected for Uncertainty, Weighted.

Table 5 shows median WTP estimates for each model, by the time it would take to receive test results. Based on the weighted model with original responses (Model 2), the median person would pay between 142 and 204 US dollars for a COVID-19 test. Conservative estimates indicate that the median person would pay at least 45 US dollars for a COVID-19 test, even of the longest wait time (see Model 3, waiting time = 3 days).

Table 5

Median Willingness to Pay Estimates (95 % Confidence Intervals).

	Model 1: Raw Data	Model 2: Raw Data	Model 3: Corrected for Uncertainty	Model 4: Corrected for Uncertainty
Waiting = 1 day
Median WTP	166.47	142.31	57.72	70.56
Lower Bound	114.24	70.78	45.81	51.56
Upper Bound	218.69	213.85	69.63	89.56
Waiting = 2 day
Median WTP	187.30	204.23	57.09	80.10
Lower Bound	126.81	62.30	46.35	50.60
Upper Bound	247.80	346.16	67.84	109.60
Waiting = 3 day
Median WTP	177.35	188.05	45.04	57.20
Lower Bound	121.88	53.02	33.58	36.05
Upper Bound	232.82	323.09	56.50	78.36

Note: WTP estimates reported are evaluated at the averages listed in Table 2.

Finally, we consider whether the price elasticity of testing intentions varies with any individual and household-level controls by incorporating interaction terms with LNCOST. Table 6 reports the marginal effects of LNCOST by AGE and CONCERN.7 We find that concerned and younger respondents’ testing intentions are less responsive to higher fees. For example, we observe a decrease of 6.7 (8.5) percentage points in the probability of testing associated with a 1% increase in cost for concerned (unconcerned) 20-year-olds (Model 6), consistent with the belief that concerned individuals likely want confirmation on disease-status. If we compare by age, we find that while older respondents are more likely to test, they are more price sensitive, i.e. compare the decrease in the probability of testing of 6.7, 11.1, and 15 percentage points for concerned 20, 40 and 60-year-olds, respectively. In contrast, we find no significant differential effects by education, household income, or household size.

Table 6

Marginal Effects on LNCOST on the Probability of Being Willing to Pay for a COVID-19 Test by Age and Level of Concern.

	AGE
CONCERNED	20	30	40	50	60	70
	Based on Model 5 : Raw Data
No (0)	−0.086 (0.012)***	−0.106 (0.014)***	−0.126 (0.021)***	−0.145 (0.029)***	−0.163 (0.038)***	−0.181 (0.046)***
Yes (1)	−0.058 (0.007)***	0.076 (0.008)***	−0.093 (0.016)***	−0.111 (0.024)***	−0.127 (0.032)***	−0.143 (0.041)***
	Based on Model 6 : Corrected for Uncertainty
No (0)	−0.085 (0.018)***	−0.108 (0.014)***	−0.130 (0.019)***	−0.151 (0.028)***	−0.172 (0.038)***	−0.191 (0.048)***
Yes (1)	−0.067 (0.014)***	−0.089 (0.011)***	−0.111 (0.018)***	−0.131 (0.028)***	−0.150 (0.039)***	−0.169 (0.049)***

Notes: Observations = 5,504. Standard errors, clustered by country, are reported in parentheses. ***, ** and * imply statistical significance at 1 %, 5 % and 10 % level, respectively. These marginal effects were estimated based on Models 5 and 6 reported in Appendix Table A1.

Table A1

Logit Model of Willingness to Pay for a COVID-19 Test with Interaction Terms (Coefficients).

	Model 5:Raw Data	Model 6:Corrected for Uncertainty
LNCOST	−0.216 (0.140)	−0.165 (0.181)
LNCOST x CONCERNED	0.101 (0.021)***	0.070 (0.020)***
AGE	0.037 (0.18)**	0.040 (0.018)**
LNCOST x AGE	−0.011 (0.005)**	−0.011 (0.006)*
WAITING2	0.202 (0.064)***	0.064 (0.044)
WAITING3	0.149 (0.071)**	−0.126 (0.082)
FEMALE	0.140 (0.084)*	0.087 (0.079)
EDUCATION	0.005 (0.012)	0.002 (0.008)
HOUSEHOLD	−0.001 (0.017)	0.005 (0.015)
INCOME	0.466 (0.126)***	0.425 (0.094)***
Pseudo R2	0.0613	0.0756

Notes: Observations = 5,504. Standard errors, clustered by country, are reported in parentheses. ***, ** and * imply statistical significance at 1 %, 5 % and 10 % level, respectively. Both models include country fixed effects, a variable representing the day the survey was completed to depict effects of the growth in reported cases. Sampling weights are used to estimate both models.

Discussion and conclusions

We implemented a web-based survey to investigate individual preferences for COVID-19 tests. Our findings suggest that there is a latent demand for COVID-19 tests in the Latin American countries included in this study. The median person would be willing to pay at least 4.2 % of the average monthly income for a COVID-19 test (or $45). Our WTP estimates are consistent with charges made by private labs at the beginning of the pandemic (March 2020), of up to $70 in Brazil, $140 in Chile, $80 in Colombia, $300 in Ecuador, $420 in Mexico, and $137 in Uruguay.8 Policymakers can use these estimates to design subsidization schemes that allow reaching a targeted level of testing, while preventing rent-seeking behaviors amid the pandemic. Our findings also indicate that constituents would support the extension of testing as a strategy to mitigate health and economic consequences of the COVID-19 pandemic.

While our study provides timely and policy relevant information on preferences for COVID-19 testing, more research on this topic is still needed. Future studies can be conducted with more representative samples of respondents at the country level, and consider new tests for antibodies for those not tested during an active infection. Another logical extension to this study would be the analysis of preferences for test accuracy, mode of specimen gathering (e.g. nasopharyngeal, blood samples), and the time it takes for results beyond 1–3 days used in this study, given that respondents were not impatient to receive test results in that window of time. This could have been associated with the widespread belief (shared by more than 90 % of respondents) that the number of tests available is insufficient to face such an unprecedented pandemic. However, we leave the evolution of such preferences over the course of a pandemic to future investigations.

Funding

This research was supported by funding from , Fairfield, CT and granted approval by Fairfield University Institutional Review Board under protocol no. 3752-2020 and Sacred Heart University’s Institutional Review Board under protocol no. 200316A.

CRediT authorship contribution statement

Jennifer M. Trudeau: Investigation, Data curation, Writing - original draft. Jessica Alicea-Planas: Writing - review & editing. William F. Vásquez: Methodology, Formal analysis, Writing - original draft.

Declaration of Competing Interest

None.