Pe

Aa

b

c

a

ARRAA

JIO

KDePFE

1

aai

t2tlTcabC

ST

j

h0

Journal of Health Economics 35 (2014) 147–161

Contents lists available at ScienceDirect

Journal of Health Economics

jou rn al hom epage: www.elsev ier .com/ locate /econbase

reventing dengue through mobile phones: Evidence from a fieldxperiment in Peru�

na C. Dammerta,b,∗, Jose C. Galdoa,b, Virgilio Galdoc

Carleton University, CanadaIZA, GermanyWorld Bank, United States

r t i c l e i n f o

rticle history:eceived 24 August 2012eceived in revised form 7 February 2014ccepted 7 February 2014vailable online 5 March 2014

EL classification:1012

a b s t r a c t

Dengue is the most rapidly spreading mosquito-borne viral disease in the world (WHO, 2009). Duringthe last two decades, the dramatic rise in the number of dengue infections has been particularly evi-dent in Latin American and the Caribbean countries. This paper examines the experimental evidenceof the effectiveness of mobile phone technology in improving households’ health preventive behavior indengue-endemic areas. The main results suggest that repeated exposure to health information encourageshouseholds’ uptake of preventive measures against dengue. As a result, the Breteau Index in treatmenthouseholds, an objective measure of dengue risk transmission, is 0.10 standard deviations below themean of the control group, which shows a reduction in the number of containers per household that testpositive for dengue larvae.

eywords:engue-Healtheruramingxperimental design

The estimates also show marginally significant effects of the intervention on self-reported denguesymptoms. Moreover, we use a multiple treatment framework that randomly assigns households to oneof the four treatment groups in order to analyze the impacts of framing on health behavior. Differentvariants emphasized information on monetary and non-monetary benefits and costs. The main resultsshow no statistical differences among treatment groups.

© 2014 Elsevier B.V. All rights reserved.

Hbd

. Introduction

Dengue is one of the most serious mosquito-borne viral diseases

ffecting humans and is a leading cause of illness in the tropicsnd subtropics. It is transmitted by the bite of Aedes mosquitoesnfected with any of four dengue serotypes. According to the World

� This research has benefited from comments and suggestions received by the edi-or, three anonymous referees as well as seminar participants at the 2012 LACEA,013 Canadian Economics Association Meeting, and 2013 RECODE meeting. Wehank SASE Asociacion Civil, Centro IDEAS, and the Ministry of Health-DIRESA Sul-ana for their logistic and institutional support. Special thanks to Cecilia Bustamante,eodoro Saez, Dr. Walter Vegas, and Mary Villavicencio for their dedication andare in the project implementation, field work, and data collection. Minoru Higand Rene Castro provided excellent research assistance. This work was supportedy a research grant from the Inter-American Development Bank (Proyecto Red deentros de Investigacion - ATN/SF-11298-RG). The standard disclaimer applies.∗ Corresponding author at: Department of Economics and Norman Patersonchool of International Affairs, 1125 Colonel by Drive, Ottawa, ON K1S5B6, Canada.el.: +1 613 5202600.

E-mail addresses: ana dammert@carleton.ca (A.C. Dammert),ose galdo@carleton.ca (J.C. Galdo), vgaldo@worldbank.org (V. Galdo).

dalm

tmvpdasts

sit

ttp://dx.doi.org/10.1016/j.jhealeco.2014.02.002167-6296/© 2014 Elsevier B.V. All rights reserved.

ealth Organization, dengue represents an enormous global healthurden, with 2.5 billion people worldwide at risk of contracting theisease (WHO, 2009). Most alarming is the fact that, in the two lastecades, humanity has experienced a global emergence of dengues a major public health problem due to large demographic shifts,ack of effective mosquito control, inadequate water and sewer

anagement systems, and weak public health infrastructure.This paper presents evidence from the first large-scale, clus-

ered, randomized control trial that evaluates the effectiveness ofobile phone technologies in enhancing households’ health pre-

entive behavior in dengue-endemic areas. In recent years, mobilehone service has become the most rapidly adopted technology ineveloping countries, as the costs of installing mobile phone towersre low relative to those of landlines (Jensen, 2010). Mobile phoneervice could facilitate the diffusion of knowledge and best prac-ices, reduce transaction costs, and improve the delivery of publicervices (Aker and Mbiti, 2010; Chong, 2011).

Indeed, a small number of non-experimental microeconomictudies have investigated the role of mobile phone technologiesn fostering economic development, particularly in rural agricul-ural markets. Studies of fishermen in India (Jensen, 2007) and of

1 ealth

f2waatuccttth

dnelaorAvatttn

atIuptuvaehdthGpediromcb2

pawlts

b

vi

p(dclgtesoeis

wipfioitaiqe

pBsetTaisbttvd(

iiostadcscr

48 A.C. Dammert et al. / Journal of H

armers in Niger (Aker, 2010) and Uganda (Mutu and Yamano,009) have shown that access to mobile phone service is associatedith significant increases in arbitrage, declines in price dispersion,

nd increases in the number of markets over which farmers trade,ll of which have led to improvements in farmers’ welfare. Despitehese promising results, mobile phone service has not yet beensed extensively to advance preventive health care in developingountries (Blaya et al., 2010). Given that households in developingountries invest little in preventive health care, and consideringhat the treatment of one single case of dengue ranges from US$10o US$25 (WHO, 2009), the provision of preventive informationhrough text messages has the potential to be a cost-effectiveealth care intervention.

We also contribute to the literature regarding mosquito-borneiseases, which has focused on malaria due to the significantumber of lives worldwide that have been claimed by this dis-ase. Not much has been written on dengue in the economicsiterature. Dengue can cause recurring and debilitating infectionsnd without adequate treatment, it increases vulnerability tother diseases, affects educational performance of children, andeduces labor market productivity of adults (Beatty et al., 2011;nderson et al., 2007). There is no specific antiviral medicine oraccine against dengue; thus policy makers have turned theirttention to prevention policies which can be classified intohose that kill adult mosquitoes (indoor residual sprays), thosehat inhibit mosquito breeding (larval habitat management), andhose that reduce dengue infection in humans (mosquito-treatedets).1

Previous research in the medical literature has shown that therere large private returns from the prevention of dengue, althoughhe adoption of preventive measures is low (Anders and Hay, 2012).n the standard model of investments in human capital, individ-als invest in health products if the expected benefits from thereventive product outweigh its costs. Low adoption of preven-ive products could be due to a lack of knowledge or people’snderestimation of the importance of controlling the mosquitoector, since dengue may present as a mild illness episode (Eldernd Lloyd, 2006). Given that the Aedes mosquito thrives in urbannvironments with limited water supply and ensuing shortage, itas been shown that one of the most effective ways to controlengue is to provide households with preventive information sohat they can eliminate the breeding places of the mosquito throughouse maintenance and disposal of tires and plastics (Espinoza-omez et al., 2002). Thus, the goal of our intervention was torovide repetitive access to information in order to improve knowl-dge of preventive practices, which may lead to reductions inengue infestation risk. In contrast to recent experimental stud-

es on health information, where households received a one-timeandomly assigned message, we sent 30 messages over a periodf 3 months before the peak of the dengue season. Reminders canitigate ‘attentional failure’ and thus change intertemporal allo-

ations, and improve consumer welfare by providing associationsetween future opportunities and today’s choices (Karlan et al.,010).

We conducted the field experiment in 100 urban localities in therovince of Talara in the department of Piura in northern Peru. Ourrea of study is considered an endemic dengue area because of itseather conditions, proximity to the Equatorial tropical area, and

ow development of water supply and sanitation facilities. In fact,he Peruvian Ministry of Health has declared the area as endemicince a dengue outbreak in 2001 infected 11,703 people with this

1 We do not focus on indoor residual spray, since the costs of this intervention areorne by the government, which is in charge of conducting the spraying campaigns.

deaas

t

Economics 35 (2014) 147–161

irus in Piura. Since then, more than 20,000 people have been clin-cally diagnosed with the virus (Ramírez, 2011).

Following the World Health Organization guidelines on denguerevention, we sent information regarding the mosquito’s life cyclee.g., eggs laid on the wet walls of containers of water), the con-itions that allow dengue to spread, and several strategies forontrolling the spread of the disease. To be effective, messages wereocally relevant by customizing them based on local uses of lan-uage and local illness classifications. We measure the impact ofhis informational exposure on health-preventive behavior (cov-ring of water reservoirs, the use of mosquito nets and windowcreenings, among others), self-reported dengue symptoms, casesf diarrhea, and the presence of dengue larvae in water contain-rs three months later. Importantly, the presence of externalitiess addressed by using GIS coordinates for all households in theample.

Based on several prior experiments on health prevention thatere done mainly in developed countries, we are also interested

n analyzing whether the perspective in which the information isresented affects preventive behavior. It has been shown that theraming of messages could shape attitudes toward risk and therebynfluence behavior and choices (Kahneman and Tversky, 1979). Inrder to analyze the impacts of message framing on health behav-or, we randomly assigned localities within the treatment groupo one of four treatment groups. Each framing group highlighted

piece of general information; some messages provided generalnformation by highlighting the positive (and negative) conse-uences of adopting a preventive behavior while other messagesmphasized the monetary cost.

Several results emerged. First, exposure to repeated healthreventive information affects households’ health behavior. Thereteau Index, an objective measure of dengue risk transmission,hows that households exposed to preventive information experi-nced a decrease in the number of water containers per householdesting positive for dengue with respect to the control group.his is explained by changes in household behavior since therere statistically significant increases in the probability of cover-ng water reservoirs, cleaning of water reservoirs, consumption ofafe water, and the use of screens in windows and/or mosquitoed nets. These findings contribute to the literature on preven-ive behavior, the area in which most of the experimental studieshat find small effects provided information during a one-timeisit (Dupas, 2009). Studies that find positive effects were con-ucted over many months by providing repetitive informationCairncross et al., 2005; Luby et al., 2004; Pop-Eleches et al., 2011;Lester et al., 2010).

Second, we evaluated the impact of exposure to repetitivenformation on different dengue indicators. The follow-up datancluded self-reported information on dengue incidence and illnessf household members during the treatment period. The resultshow that behavioral changes associated with the interventionranslated into a reduction in dengue symptoms (fever, headache,nd eyes pain) but did not affect the number of clinical diagnosedengue cases. Mild episodes of dengue may not lead to significantosts to households, but they affect people’s perception toward theeverity of the disease. We also analyzed whether the behavioralhanges are big enough to have an effect on preventable water-elated diseases. The results show a decline in the incidence ofiarrhea in treatment households, although the null of no differ-nce between the treatment and control areas cannot be rejectedt standard levels. These indicators, however, were self-reported

nd thus are likely to be affected by recall error and misdiagno-is.

Third, the point estimates suggest that households exposedo non-monetary loss messages experienced a slightly higher

ealth

iotTftitaeoItmt

rvc

2

vaiwmasb2seiaSoh

irm2sittobbuDofon

aPaaSe

owtrdi1osbgnfcn(ta

nsslscmr

didgtrvnOb

itotFbiacmitiaLtciostw

A.C. Dammert et al. / Journal of H

ncrease in almost all preventive outcomes relative to other typesf messages. However, the null of equality of coefficients betweenreatment groups is not rejected at standard levels in most cases.his result is similar to other experimental studies that find noraming effects on health behavior (e.g., Dupas, 2009). Moreover,here is some evidence of the heterogeneous effects as the increasen preventive measures against dengue was concentrated amonghe poorest households. We do not find, however, differencesccording to the gender of the cell-phone owner or by previousxposure to dengue in the house. Finally, we do not find evidencef geographical spillovers except in one outcome, the Containerndex. This finding shows that, as we increase the number ofreated households in the neighborhood, households in the treat-

ent group experienced a decrease in the number of containersesting positive for dengue.

The remainder of the paper is organized as follows. Section 2eviews the literature. Section 3 describes the study area, inter-ention, and dataset. Section 4 presents our estimates. Section 5oncludes and outlines the envisaged main policy implications.

. Related literature

In recent years, the use of text messages to deliver health ser-ices has been studied mostly in developed countries with theimed at encouraging people to change their health behavior. Fornstance, reminders for outpatient appointments were associated

ith a reduction in the likelihood of patients missing their appoint-ents (Koshy et al., 2008), promotion of weight-loss behavior

mong overweight people (Patrick et al., 2009; Joo and Kim, 2007),moking cessation (Rodgers et al., 2005) and adherence to dia-etes treatment among children and adolescents (Franklin et al.,006). Nonetheless, after a careful meta-evaluation of more than 33tudies from the medical literature in developed countries, Fjelsoet al. (2009) reported that only fourteen studies satisfied qual-ty standards, four studies targeted preventive health behaviors,nd ten studies targeted clinical care. The authors conclude thatMS-delivered interventions have positive short-term behavioralutcomes, but further research is required to evaluate preventiveealth behaviors.

Text messaging has not been used to spread awareness andnformation about infectious diseases in endemic regions despiteecent evidence suggesting that households are responsive to infor-ation on the health risks they face (Dupas, 2011a; Cairncross et al.,

005; Rhee et al., 2005). Madajewicz et al. (2007), for example,how that informing households about the arsenic concentrationn their well water increased the probability that they would switcho a safer well. Similarly, Jalan and Somanathan (2008) reporthat households that receive information about the concentrationf fecal bacteria in drinking water improved water-purificationehavior. Prevention is a key component of dengue control and cane achieved through vector control and personal protection meas-res (WHO, 2009). Not all information, however, can be effective.upas’ (2011b) review of the literature shows that the provisionf information can influence people’s behavior when they are notully informed about the health situation they face, when the sourcef information is credible, and when they are able to process theew information.

It is also important to analyze how informational campaignsffect the adoption of preventive measures against the disease.rior work in psychology has shown that message framing has

n effect on motivating people to change their behavior (for

review of framing on health outcomes, see Rothman andalovey, 1997; Salovey and Williams-Piehota, 2004; Rothmant al., 2006). In particular, framing might have differential effects

msta

Economics 35 (2014) 147–161 149

n health prevention, detection, and treatment, depending onhether the outcome is perceived to involve some risk or uncer-

ainty. For example, preventive behavior could be viewed aselatively safe (i.e., it provides a relatively certain outcome), whileetection could be viewed as the act of taking a risk of find-

ng the presence of a health problem (Rothman and Salovey,997). Therefore, the main implication of the prospect theoryn health behavior is that gain-framed messages are more per-uasive than loss-framed messages in encouraging preventiveehavior, while loss-framed messages are more persuasive thanain-framed messages in encouraging detection behavior. Alter-atively, independent of the perceived risk of the outcome, loss

raming might have greater effects due to negativity bias in pro-essing the information wherein people are influenced more byegative information than with comparable positive informationMeyerowitz and Chaiken, 1987). Thus, people are more motivatedo avoid a loss than to attain a gain of equal magnitude or lossversion.

We have scant empirical evidence on how framing shapes eco-omic behavior in developing countries. Bertrand et al. (2010)how that demand for credit can be manipulated through lettersent to potential borrowers containing randomly assigned psycho-ogical features motivated by specific types of frames and cuesuch as profile pictures in South Africa. Chong et al. (2013), on theontrary, show that informational messages emphasizing environ-ental, social, or authority content were not effective in increasing

ecycling behavior in Peru.In a more related paper, Dupas (2009) analyzed an experimental

esign in Western Kenya, where households receiving vouchers fornsecticide-treated bed nets were exposed to a one-time-only ran-omly assigned message (flyer). Participants in the health-framingroup received a marketing message in which morbidity and mor-ality were emphasized, while participants in the financial groupeceived a marketing message where financial gains from pre-enting malaria were emphasized. The main results suggest thateither of the two framing options affected the uptake of bed nets.ne possible explanation is that liquidity constraints are the mainarriers to investments in malaria prevention.

Our paper follows this line of inquiry, although it differs inmportant ways. First, we sent multiple messages over a period ofhree months, since repetition of information affects the processingf the information into the memory system (encoding), which inurn affects comprehension and long-term retention (PAHO, 2004).rom a theoretical economic standpoint, this might be explainedy models of ‘attentional failure’ (Karlan et al., 2010) or ‘rational

nattention’ (Reis, 2006). Reminders might change intertemporalllocations, and improve consumer welfare, by providing asso-iations between future opportunities and today’s choices thatitigate the attention failure (Karlan et al., 2010). Second, we are

nterested in the effects of the provision of preventive informa-ion through text messages. A number of recent empirical studiesn developing countries have shown that text reminders increasedherence to HIV antiretroviral treatment (Pop-Eleches et al., 2011;ester et al., 2010) and health workers’ adherence to malariareatment (Zurovac et al., 2011). Third, our experimental designonsiders that households might be willing to change their behav-or but may not be able to do so if the income constraint is bindingr if they are unable to borrow to purchase mosquito nets orcreens. For that reason, we also provided preventive informationhat does not involve any direct cost, such as cleaning and coveringater reservoirs or discarding water-holding solid waste whereosquitoes breed outside the house (e.g., tires, bottles, or other

mall water reservoirs). Finally, we focus on a Latin American coun-

ry where experimental interventions addressing health practicesre scant.

150 A.C. Dammert et al. / Journal of Health Economics 35 (2014) 147–161

F dots rc e read

3

P(taiahtw

1i(icAPthahPifne

wtlblcch

waitppcaswtltrlt

amagamm

ih6pMing the month at which temperatures in the region are at theirpeak. The main reasons for attrition were migration (3.9%), theinability to find eligible respondents despite repeated visits (5.7%),

ig. 1. Study areas. Note: Study areas included 100 localities in Parinas, Piura. Redontrol group. (For interpretation of the references to color in this figure legend, th

. Experimental design

We implemented a randomized intervention in the district ofarinas, Department of Piura, located in the northern region of PeruFig. 1). Most households in Parinas have access to piped waterhrough house connections; however, safe water is only provided

few days a week due to problems with the distribution pipe andnadequate energy sources. Households report receiving water onverage 3.4 times a week, with an average of 7 h per day. Thus,ouseholds use different reservoirs to store water: plastic storageanks or cisterns, drums, discarded buckets, or ground-level cementater storage tanks, among others.

The Ministry of Health has divided the district of Parinas into9 health zones containing 100 localities based on the availabil-

ty of safe water, dengue propagation, and roads, among otherssee Appendix A). A household eligible to participate in the exper-mental design is one where the head of household or spouse is aell phone user and is literate enough to read a simple message.ccording to the 2007 Peruvian Census, 55.8% of households inarinas had a cell phone in 2007. From that year to the year ofhe intervention, the percentage of households with mobile phonesas increased significantly (INEI, 2012). Appendix B shows that theverage cell phone sample household looks similar to the averageousehold that owns a cell phone in Parinas, according to the 2007eruvian Census data. It is important to note that our results arenformative for those households who had cell phones and there-ore are somewhat richer and more educated than households witho cell phone are. In this regard, one should assess with caution thexternal validity of the results.

Two independent sets of randomizations at the locality levelere conducted based on the administrative information from

he Ministry of Health, complemented with a customized base-ine survey data. In the first step, a stratified random assignmentetween treatment and control localities was implemented fol-

owing a ratio of 2:1. The stratification was based on two keyharacteristics: poverty index and dengue index. The former wasonstructed from the principal component analysis of 13 house-old assets and 6 dwelling characteristics, including access to safe

ta

epresent localities in the treatment group, while blue dots represent those in theer is referred to the web version of the article.)

ater and toilet facilities. The latter was constructed using healthdministrative data regarding the incidence of dengue at the local-ty level. The dengue index is based on various indicators measuringhe percentage of houses where dengue larvae was found, theercentage of containers where the larvae was found, and theercentage of infected containers per 100 inspected houses. Weomplemented this analysis with baseline information on the avail-bility of mosquito nets, window screenings, and indoor residualpraying at the locality level. Random stratification proceeded aftere split the data into four categories according to the distribution of

he poverty and dengue indices: high/high, high/low, low/high, andow/low. Thresholds were determined according to the median ofhe distribution for each index. Within each category, we conductedandom assignment between treatment and control localities fol-owing a ratio of 2:1. As a result, we randomly assigned 64 localitieso the treatment group and 36 localities to the control group.

In the second step, within treatment localities, we randomlyssigned localities to two orthogonal information sets: gain/lossessages and monetary/non-monetary messages. This framework

llows us to make causal comparisons within each domain (i.e.,ain versus loss groups and monetary versus nonmonetary groups)s well as across four different combinations (i.e., monetary/gain,onetary/loss, nonmonetary/gain, and nonmonetary/loss treat-ent groups).Approximately 1 out of 5 households that own a mobile phone

n Parinas was included in the final sample. This represents 2021ouseholds, of which 1350 belong to the treatment localities and71 to control localities.2 Of the initial 2021 households with com-lete baseline information, 1784 (88%) were re-interviewed in earlyarch 2010, three months after the intervention started and dur-

2 Within the treatment localities, 309 households were assigned to the mone-ary/gain framing, 383 to monetary/loss framing, 362 to nonmonetary/gain framing,nd 296 to nonmonetary/loss framing.

ealth

aa

3

tmwbupwroktn

efrrttmisra

3

Tcah

tt2oroctMagimicl

ipiawimptvp

ttd(

vrmdishasNedb

mmpis15m

t2cttec2

A.C. Dammert et al. / Journal of H

nd refusal (2.4%). Appendix C shows similar statistical significantttrition rates across all experimental groups.

.1. Treatment and SMS framing

Based on PAHO’s (2004) framework, we created 21 messagesargeting two different goals: (i) preventing production of adult

osquitoes; i.e., water storage, refuse disposal, water-holding solidaste in the yard, among others, and (ii) preventing exposure to the

ites of Aedes mosquitoes; i.e., screening of windows and doors,se of bed nets, and use of repellents. We also created nine neutralatient-care text messages or disease recognition messages, bute did not frame them. In total, households received over 30 SMS

egarding dengue prevention, detection, and control over a periodf 3 months.3 Every effort was made to ensure that participantsnew how to send and receive text messages. It is important to notehat incoming SMS messages are free of charge, so households didot incur any cost when they received a text message.

Households in the randomly assigned treatment group werexposed to different framed messages according to their assignedraming group. In the monetary framing group, householdseceived messages about the financial gains or losses they wouldealize (e.g., working days lost to illness, cost of medical care). Inhe gain/loss groups, some households received messages abouthe benefits of taking preventive measures, while others received

essages about the losses from not taking such measures. Follow-ng Levin et al. (1998), the negative frames were written using aimple negation (i.e., not adopting the behavior), to avoid incorpo-ating biases due to differences in linguistic connotation by usinglternative terminology. Table 1 provides some examples.

.2. Baseline household survey

The baseline survey was completed in July–September 2009.he head of household provided information on demographic

haracteristics of each household member (i.e., age, educationalttainment, gender, siblings) and dengue episodes of any house-old member over the two years prior to the survey.4 In addition,

3 Given that text messages must contain up to 143 characters, it was quite impor-ant to tailor the messages to the uses and representations of the local people. Forhis reason, we conducted a focus group in the area of intervention during July009. The use of focus groups was particularly important to gather informationn language uses in terms of dengue’s knowledge and preventive practices. Rep-esentatives from the Asociación de Mujeres de Negritos, an NGO actively workingn health interventions, participated in this first focus group. Their feedback wasomplemented with detailed interviews with personnel from the public health sys-em, who provided useful information about current policies implemented by the

inistry of Health aimed at facilitating communication with the population in therea of intervention. Once the content of text messages was defined, a second focusroup was carried out in Piura with the participation of communication specialistsn epidemiological health care. The aim was to check the relevance of the infor-

ation, identify the messages that needed to be adjusted (e.g., better wording),dentify additional information that should be included, and validate the healthontent based on local illness classifications. Finally, we received feedback from theocal enumerators after testing their comprehension of the messages.

4 The NGO IDEAS in collaboration with FORO Salud-Piura administered the pre-ntervention baseline survey. All local enumerators, supervisors, and monitorsarticipated in an intensive weekend training program, including discussion of

nterview methods, questionnaire content, survey protocol, data quality checks,nd mock interviews. Survey manuals developed specifically for the baseline surveyere provided to all enumerators, supervisors, and monitors. Numerators assisted

n reviewing the survey instruments for inconsistencies and problems before imple-enting the survey. In addition, we provided each enumerator with an information

amphlet with local phone numbers in case the members of the household had ques-ions about the veracity of the survey. In the baseline survey, the surveyor gatherederbal consent to participate in this study and provided information regarding theossibility of receiving text messages.

inalhos

4

4

mruvea

Y

wel

op

Economics 35 (2014) 147–161 151

he dataset provided detailed information on dwelling characteris-ics including toilet facilities, house infrastructure, and householdensity, which proxy for households’ long-run economic statusFilmer and Pritchett, 2001).

Table 2 shows the p-values for the mean differences in rele-ant baseline covariates across treatment and control localities. Theesults do not reject the null hypothesis of equality of means forost covariates, as the differences are not statistically significantly

ifferent from zero.5 Likewise, Table 2 shows no statistically signif-cant differences across all treatment groups except for householdize, the rate of computer ownership, and the proportion of house-olds reporting that vaccinations prevent dengue. The differencesre very small for the last two variables though, and they are onlytatistically significant due to the size of the standard deviations.onetheless, we control for these differences in the econometricstimations. Overall, the result shows that the stratified ran-omization was effective in balancing the covariate mean valuesetween treatment and control households.

From Table 2, the average head of household is a 50-year-oldale, with 9 years of completed schooling, living with 4.9 familyembers in the household. About 96% of cell phone users have a

re-paid phone, which is used mostly to communicate with fam-ly members (85% among heads of households and 96% amongpouses). The average amount spent on pre-paid phone cards was8 soles (about US$6.50) during the month before the survey. About% of the sample did not send and/or receive text messages, theajority of them report not knowing how to do so.Most respondents in the sample had heard about dengue; 89% of

he sample knew that it is transmitted by mosquito bite. However,3% reported that dengue could be prevented by using the chemi-al treatment temephos (distributed under the name Abate) to killhe larvae in the container. Abate is distributed free of charge byhe Ministry of Health and is one of the cores of the current publicducational campaign. In addition, 3% incorrectly report that vac-inations can prevent dengue. Concerning dengue incidence, about5% of the sample self-reported being diagnosed with dengue dur-

ng the previous two years: 64% were diagnosed by a doctor orurse, 13% were diagnosed by a family member or self-diagnosed,nd 7% were diagnosed by a pharmacist. On average, the illnessasted for over a week. It is important to note that the head of theousehold or the spouse provided this information for all membersf the household; therefore, recall error might have affected theseelf-reported data.

. Results

.1. Overall effects of text messaging

This section examines whether providing any type of infor-ation about preventive health affected household behavior. The

elevant parameter of interest is the intent-to-treat (ITT) that meas-res the average impact of being exposed to preventive informationia text messages. For household i, the estimated regression forach preventive outcome Yi (buying nets or screens, treating stor-ge water, cleaning reservoirs, among others), is given by:

i = + ˇ1Ti + X ′� + εi, (1)

here Ti is an indicator that equals 1 if the household was randomlyxposed to the preventive information and X is a vector of base-ine characteristics, including head of household characteristics,

5 Unreported results show that there are no differences in means and significancef the tests of equality if the means are computed over the sample of householdsresent at follow-up only. Results are available upon request.

152 A.C. Dammert et al. / Journal of Health Economics 35 (2014) 147–161

Table 1Message treatment (examples).

Frame Messages

Water Window screens Water-holding solidwaste

Symptoms (same forall)

Monetary Gain You will not spendmoney eradicatingdengue in your house ifyou cover the waterused for personalconsumption

Use window screens toprevent dengue in yourhouse. You will notspend money intreatment!

You will not spendmoney in denguetreatment if youdiscard old tires,broken waterreservoirs, and bottlesfrom your yard

Go to the nearesthealth center if you ora family member hasfever, headache, painbelow the eyes, or arash. It could bedengue!

Loss You will spend moneyeradicating dengue inyour house if you don’tcover the water usedfor personalconsumption

If you don’t usewindow screens,dengue will spread inyour house and youwill spend money intreatment!

You will spend moneyin dengue treatment ifyou don’t discard oldtires, broken waterreservoirs, and bottlesfrom your yard

Go to the nearesthealth center if you ora family member hasfever, headache, painbelow the eyes, or arash. It could bedengue!

Non-monetary Gain You will not getdengue in your house ifyou cover the waterused for personalconsumption

Use window screens toprevent dengue in yourhouse!

You will not getdengue in your house ifyou discard old tires,broken waterreservoirs, and bottlesfrom your yard

Go to the nearesthealth center if you ora family member hasfever, headache, painbelow the eyes, or arash. It could bedengue!

Loss You will get dengue inyour house if you don’tcover the water usedfor personalconsumption

You will get dengue inyour house if you don’tuse window screens!

You will get dengue inyour house if you don’tdiscard old tires,broken waterreservoirs, and bottlesfrom your yard

Go to the nearesthealth center if you ora family member hasfever, headache, painbelow the eyes, or arash. It could be

ptit

imi

Y

womtup

sairbirwuasr

i

brpFtiepv

(drtptb(etK

nPtegin

overty index, and dengue index. Standard errors are clustered athe locality level to account for random assignment across local-ties. We also report the p-value from a test of equality of thereatment indicator for the different groups.

Since we have multiple measures of dengue preventive behaviorn the survey, following Kling et al. (2007), we created a summary

easure (Y*) defined as the unweighted average of the differentndicators as follows:

∗ = 1k

K∑

k=1

YK − �K

�K

here each indicator k is standardized by the mean and variancef the control group at baseline. Thus, the magnitude of the esti-ated index shows where the mean of the treatment group is in

he distribution of the control group in terms of standard deviationnits. This allows us to test whether the treatment had an overallositive or negative ITT effect.

Panel A of Table 3 shows that households receiving a text mes-age reported a change in their health preventive behavior. Lookingt the overall index, the average impact is a statistically signif-cant 0.12 standard deviation of the control group. Householdseported an increase in the probability of covering water reservoirsy 3.4 percentage points (column 2), a change that represents a 4.6%

mprovement relative to the control group. Moreover, householdseceiving a text message reported an increase in the frequency inhich they clean the water reservoirs by 8.4 percentage points (col-mn 3) and in the consumption of safe water (either boiling it ordding chlorine) by 3.3 percentage points. These numbers repre-

ent a 9.6% and 3.8% improvement relative to the control group,espectively.

Households exposed to the preventive information reported anncrease in the usage of screening in the windows and/or mosquito

wcwa

dengue!

ednets of about 4.5 percentage points, or a 4.5% improvementelative to the control group and a reduction of 2.6 percentageoints in solid waste (tires or bottles) left outside the house.urthermore, there is no effect of exposure to preventive informa-ion on the usage of chemical larvicida (Abate) to kill the larvaen the container. The information sent by text message did notmphasize the usage of Abate given that the Ministry of Healthrovides Abate free of cost to all households in the area of inter-ention.

Our findings are related to Cairncross et al. (2005) and Luby et al.2004), who find a positive effect of information campaigns con-ucted over many months on hygiene. In addition, our results areelated to recent studies that focus on preventive e-health interven-ions in developing countries. Pop-Eleches et al. (2011) show thatarticipants exposed to weekly text message reminders increaseheir adherence to HIV antiretroviral treatment (of at least 90%)y 13–16% relative to the control group. Similarly, Lester et al.2010) show an increase in HIV antiretroviral treatment in patientsxposed to weekly text messages inquiring about their health sta-us and reminders about the availability of phone-based support inenya.

With regard to framing effects, we do not find statistically sig-ificant differences between groups, except for one outcome. Asanel B of Table 3 shows, the monetary frame is equally effec-ive as the non-monetary frame. We find statistically significantffects of the non-monetary message compared to the controlroup when the change in behavior does not involve a direct costncurred by the household. For example, households exposed toon-monetary messages report an increase in the frequency in

hich they clean the water reservoirs by 9.6 percentage points,

onsumption of safe water by 4.1 percentage points, and having lessater-holding solid waste outside the household by −3.6 percent-

ge points, compared to the control group. Panel C of Table 3 shows

A.C. Dammert et al. / Journal of Health Economics 35 (2014) 147–161 153

Table 2Descriptive statistics – 2009 baseline health survey.

Treatment groups Controlgroup

Difference(T − C)

p-Value p-Value of jointF-test[T1 = T2 = T3 = T4]

Any Monetarygain (T1)

Monetaryloss (T2)

Non-monetarygain (T3)

Non-monetaryloss (T4)

Cement wall 0.825 0.847 0.809 0.869 0.776 0.796 0.029 0.544 0.481Electricity 0.988 0.994 0.969 0.998 0.990 0.979 0.009 0.331 0.499Radio 0.963 0.963 0.974 0.967 0.950 0.968 −0.004 0.687 0.630TV 0.975 0.977 0.977 0.970 0.976 0.980 −0.005 0.573 0.965Refrigerator 0.593 0.530 0.589 0.659 0.594 0.602 −0.010 0.837 0.452Computer 0.196 0.150 0.247 0.249 0.139 0.214 −0.018 0.565 0.019Piped water 0.914 0.910 0.915 0.947 0.885 0.908 0.006 0.827 0.419Piped sewage 0.943 0.962 0.948 0.969 0.894 0.914 0.029 0.346 0.648Wealth index 0.677 0.675 0.695 0.712 0.627 0.685 −0.008 0.828 0.426Dengue index 0.372 0.346 0.347 0.405 0.389 0.356 0.016 0.710 0.791Head: age 50.334 49.534 50.987 51.522 48.875 49.307 1.027 0.366 0.589Head: male 0.736 0.761 0.738 0.732 0.713 0.773 −0.037 0.235 0.837Head has healthinsurance

0.584 0.602 0.581 0.608 0.541 0.539 0.045 0.247 0.634

Head: years ofcompletedschooling

9.486 9.828 9.291 9.398 9.488 9.787 −0.301 0.394 0.832

Head: level of educationNo schooling 0.025 0.032 0.029 0.025 0.014 0.019 0.006 0.444 0.437Primary 0.296 0.249 0.291 0.326 0.314 0.267 0.029 0.360 0.498Secondary 0.444 0.476 0.476 0.403 0.422 0.477 −0.032 0.274 0.126Post-secondary 0.235 0.243 0.204 0.246 0.250 0.237 −0.002 0.955 0.849

Household size 4.973 4.770 5.228 4.867 4.986 4.867 0.106 0.347 0.014Dengue istransmitted by amosquito bite

0.888 0.913 0.885 0.854 0.909 0.903 −0.015 0.424 0.240

Dengue could be avoided by:Getting

vaccinated 0.032 0.010 0.039 0.047 0.027 0.021 0.011 0.212 0.026Using abate in water

reservoirs0.213 0.184 0.259 0.221 0.176 0.232 −0.019 0.738 0.683

A member of thehousehold wasdiagnosed withdengue during thepast two years

0.239 0.208 0.238 0.249 0.258 0.251 −0.012 0.687 0.483

By a doctor/nurse 0.641 0.652 0.688 0.583 0.640 0.632 0.009 0.858 0.316By family member/self

diagnosed0.139 0.184 0.112 0.140 0.137 0.198 −0.057 0.317 0.731

Pharmacist 0.067 0.046 0.047 0.088 0.085 0.044 0.023 0.359 0.578Number of days ill 6.924 7.045 7.013 6.724 6.949 6.837 0.088 0.754 0.558

Monthly moneyspend on pre-paidcell phone cards(soles): head

20.843 22.972 21.113 19.047 20.740 20.994 −0.151 0.923 0.143

Spouse 16.850 18.463 16.438 15.914 16.455 17.483 −0.634 0.499 0.399Don’t know how tosend/read a textmessage: head

0.053 0.045 0.047 0.058 0.061 0.067 −0.014 0.476 0.913

Spouse 0.043 0.045 0.039 0.039 0.051 0.063 −0.020 0.163 0.933

Note: Data from 2021 households included in the pre-intervention baseline survey. For each variable, the p-values report the values for a test of the null hypothesis that themeans are identical in treatment and control groups and the values for a test of the null hypothesis that the means are identical across the experimental groups.

tfitpSpf(

epmt

hat the gain frame is equally effective as the loss frame exceptor cleaning water reservoirs daily or weekly. Households receiv-ng loss messages experienced a statistically significant increase inhe probability of cleaning the water reservoirs by 14 percentageoints. From a health policy standpoint, these results suggest that

MS framing for health preventive behavior is not as important asroviding the information itself. This lack of statistical significanceor framing effects is consistent with the evidence found in Dupas2009).

ttFb

We are also interested in learning whether there are het-rogeneous effects, since households’ investment decisions onreventive health care depend on factors such as access to financialarkets and ex ante information on the burden associated with

he specific disease. We then consider variation in the effects of

ext messaging as a function of observable characteristics throughhe interaction of the treatment indicator with baseline indicators.irst, households are classified in three categories: high if theiraseline wealth index belongs to the highest quartile of the index

154 A.C. Dammert et al. / Journal of Health Economics 35 (2014) 147–161

Table 3Effects of text messaging.

Overallstandardizedindex

Cover waterreservoirs

Clean waterreservoirs dailyor weekly

Drink boiledwater or treatwater withchlorine

Change abateevery 3 months

Use mosquitonets forsleeping and/orscreens inwindows

Havewater-holdingsolid waste(tires, bottles,or others)

(1) (2) (3) (4) (5) (6) (7)

Panel AAny message 0.120*** 0.034** 0.084** 0.033** 0.017 0.046** −0.026*

(0.034) (0.014) (0.036) (0.015) (0.020) (0.022) (0.014)

Panel BMonetary 0.115** 0.030* 0.072 0.025 0.030 0.047* −0.015

(0.040) (0.017) (0.040) (0.018) (0.024) (0.025) (0.014)Non-monetary 0.124** 0.038** 0.096** 0.040** 0.003 0.044* −0.037**

(0.038) (0.016) (0.042) (0.017) (0.022) (0.026) (0.020)p-Value of

[M = NM]0.829 0.597 0.576 0.358 0.228 0.917 0.216

Panel CGain 0.102** 0.025 0.027 0.034* −0.001 0.065** −0.030**

(0.041) (0.021) (0.048) (0.026) (0.028) (0.030) (0.015)Loss 0.137*** 0.042** 0.140** 0.031* 0.034 0.026 −0.022

(0.037) (0.018) (0.053) (0.018) (0.025) (0.032) (0.027)p-Value of

[G = L]0.380 0.260 0.008 0.852 0.119 0.129 0.654

Mean controlgroup

0.927 0.639 0.895 0.685 0.734 0.065

Note: N = 1754 households. Controls in all regressions include head of household characteristics at baseline (age, gender, schooling, employment, and health insurance),assets index and dengue index. Observations are clustered at the locality level. Robust standard errors in parenthesis.

* 10%.** 5%.

d3(tmssho

owdiAga(dtfo

Sdmtis

o

eb

4

tnvssev

ecmhdinig

show significant point estimates of exposure to preventive infor-mation on the Container Index and the Breteau Index (Panel A).

*** 1%.

istribution (richest), medium if their index belongs to the 2nd andrd quartiles, and low if their index belongs to the lowest quartilepoorest) (Panel A of Table 4). Panel A shows that a higher frac-ion of low assets index households (poorest) exposed to the text

essages report treating their water for consumption, having lessolid waste outside the household, and using mosquito nets and/orcreens in windows compared to medium- and high-wealth indexouseholds. On the contrary, the results show no differential effectsf messages on the probability of covering water reservoirs.

Moreover, we analyzed whether the gender of the cell phonewner had a differential effect on the outcomes of interest, sinceomen are usually the ones responsible for the activities related toengue prevention inside the house and they tend to invest more

n goods improving child’s well-being and health (Thomas, 1990).t baseline, the proportion of male cell ownership is similar acrossroups, 44% in the treatment group and 46% in the control group,nd the difference is not statistically significant different from zerop-value = 0.468).6 As Panel B of Table 4 shows, we do not find evi-ence of gender differential effects at follow-up. Note, however,hat we did not randomly select the head of household or spouseor participation in the program to examine whether the adoptionf preventive action varied by gender.

Finally, one might wonder if households are more responsive toMS preventive health information if a family member had beeniagnosed with dengue in the past. For instance, Dupas’ (2011b)eta-analysis shows that provision of information is less effec-

ive in influencing people’s preventive behavior when they are wellnformed about the health situation they face. As Panel C of Table 4hows, there are no statistical differences in the intent-to-treat

6 We inputted the gender of the head of household in cases where both the headf household and the spouse owned a cell phone, (95 cases, or 5% of the sample).

T

Htn

stimates between households who had experienced dengueefore and those who did not.

.2. Impacts on dengue indicators

In order to compare the objective measures of dengue riskransmission with self-reported measures that might suffer fromon-random measurement error, a week after the follow-up, sur-ey health personnel from the Ministry of Health collected larvaeamples by inspecting water containers and other water-holdingolid waste (e.g., tires, bottles or other small water reservoirs) inach household of our sample.7 Households were not aware of thisisit.

Following WHO (2009), containers were examined for the pres-nce of Aedes larvae and pupae. Based on this information, wereated three commonly used indices of the abundance of Aedesosquitoes: (i) the House Index, defined as the percentage of

ouses testing positive for Aedes larvae; (ii) the Container Index,efined as the percentage of water-holding containers testing pos-

tive for Aedes larvae; and (iii) the Breteau Index, defined as theumber of positive water-holding containers per household. Each

ndex was standardized by the mean and variance of the controlroup.

As Table 5 shows, our objective measures of dengue infestation

he Container Index shows that households in the treatment group

7 In addition, our previous self-reported estimates could be affected by theawthorne effect, as study participants having received messages stressing preven-

ion measures might be more likely to report having taken those measures, even ifot true. We thank one of the referees for pointing this out.

A.C. Dammert et al. / Journal of Health Economics 35 (2014) 147–161 155

Table 4Heterogeneous effects of text messaging.

Overallstandardizedindex

Cover waterreservoirs

Clean waterreservoirs daily orweekly

Drink boiled wateror treat water withchlorine

Use mosquito netsfor sleeping and/orscreens in windows

Havewater-holdingsolid waste (tires,bottles, or others)

(1) (2) (3) (4) (5) (6)

Panel A: wealth indexT∗High 0.076 0.037* 0.015 −0.006 0.071 −0.026

(0.069) (0.020) (0.061) (0.033) (0.045) (0.018)T∗Medium 0.087* 0.027 0.042 0.025 0.015 −0.016

(0.051) (0.022) (0.053) (0.019) (0.035) (0.029)T∗Low 0.153** 0.027 0.129* 0.069** 0.084** −0.052**

(0.066) (0.037) (0.070) (0.029) (0.039) (0.024)Medium 0.091 −0.005 0.118* 0.010 0.000 0.014

(0.066) (0.027) (0.067) (0.027) (0.047) (0.029)Low 0.039 −0.048 0.043 −0.006 −0.023 0.035

(0.067) (0.031) (0.075) (0.033) (0.046) (0.026)p-Value

T∗Low = T∗Med = T∗High0.0280 0.1326 0.2539 0.0633 0.0677 0.0777

Panel B: gender of the cellphone ownerT 0.120** 0.037* 0.070* 0.020 0.078** −0.004

(0.039) (0.020) (0.039) (0.018) (0.027) (0.016)T∗Male −0.003 −0.008 0.031 0.028 −0.073 −0.046

(0.042) (0.023) (0.051) (0.026) (0.046) (0.028)

Panel C: household experienced previous dengueT 0.124** 0.040* 0.078 0.025 0.052 −0.025

(0.042) (0.024) (0.048) (0.024) (0.032) (0.018)T∗Yes −0.009 −0.013 0.012 0.014 −0.012 −0.001

(0.045) (0.025) (0.050) (0.030) (0.043) (0.022)

Note: ***1%. Controls in all regressions include head of household characteristics at baseline (age, gender, schooling, employment, and health insurance), assets index anddengue index. Observations are clustered at the locality level. Robust standard errors in parenthesis.

* 10%.** 5%.

Table 5Dengue risk transmission indices.

StandardizedContainerIndex(1)

StandardizedHouseIndex(2)

StandardizedBreteauIndex(3)

Panel AAny message −0.108** −0.045 −0.098**

(0.046) (0.064) (0.047)

Panel BMonetary −0.106** −0.047 −0.089

(0.051) (0.079) (0.054)Non-monetary −0.109** −0.044 −0.107**

(0.049) (0.069) (0.048)

Panel CGain −0.081* 0.006 −0.073

(0.052) (0.076) (0.052)Loss −0.133** −0.093 −0.121**

(0.049) (0.073) (0.050)

Panel DMonetary gain −0.047 0.050 −0.041

(0.062) (0.104) (0.068)Monetary loss −0.152** −0.122 −0.127**

(0.053) (0.093) (0.057)Non-monetary gain −0.112** −0.036 −0.102*

(0.056) (0.089) (0.053)Non-monetary loss −0.106* −0.054 −0.113**

(0.054) (0.080) (0.052)

Notes: ***1%. N = 1496 households. Controls in all regressions include head of house-hold characteristics at baseline (age, gender, schooling, employment, and healthinsurance), assets index and dengue index. Observations are clustered at the localitylevel. Robust standard errors in parenthesis.

* 10%.** 5%.

eatehgtia(nvna

ltidtwmdtsIdmbttd

xperienced a decrease of 0.108 standard deviations in the percent-ge of water-holding containers testing positive for dengue, whilehe Breteau Index shows that households in treatment localitiesxperienced a decrease in the number of positive containers perousehold by 0.098 standard deviations of the mean of the controlroup. We do not find, however, statistically significant effects onhe Standardized House Index. These results imply that householdsn the treatment group experienced a decreased in the percent-ge of water-holding containers testing positive for dengue larvae1.44% vs. 2.47% in the treatment and control groups) as well as theumber of positive water-holding containers per household (10.66s. 18.91 in the treatment and control groups). These effects wereot big enough, however, to change the percentage of houses witht least one positive water-holding container.

It is important to note that, even though there is a corre-ation between the critical levels of the mosquito breeding andhe intensity of virus transmission, the interpretation of the var-ous indices in relation to human infection and disease can beifficult to assess (Focks, 2004). For example, larvae might be clus-ered in a small number of containers so that the Container Indexill be low, although the number of infectious bites per personay be high. The relationship between dengue risk measures and

engue infestation gets more complicated when one realizes thathe latter is affected by mosquito longevity, the immunologicaltatus of the human population, and temperature (Silver, 2008).nternational evidence, for instance, suggests that the majority ofengue outbreaks occurred in localities with low Aedes indices, andore importantly, even when there is some statistical relationship

etween the indices, these were inconsistent across years withinhe same localities (Shah et al., 2012). Indeed, it is widely arguedhe need for better methods to quantify the relationship betweenengue risk and dengue endemicity to ensure better future

1 ealth Economics 35 (2014) 147–161

vNl(tit

dphtbtsooiotibtb2

ssstmsTctost

rhwireoisst(

whw

i

paam

Table 6Effects of text messages on dengue symptoms and diarrhea.

Had a member of thehousehold showdengue symptoms(fever, headache, eyepain) since Novemberof 2009?(1)

Had a member of thehousehold haddiarrhea during theprevious month?(2)

Panel AAny message −0.031* −0.021

(0.018) (0.019)

Panel BMonetary −0.032* −0.053**

(0.019) (0.023)Non-monetary −0.031 0.013

(0.020) (0.024)p-Value of

[M = NM]0.9360 0.0167

Panel CGain −0.029 −0.021

(0.021) (0.023)Loss −0.033* −0.020

(0.019) (0.025)p-Value of [G = L] 0.8670 0.9831Mean control

group0.066 0.159

N 1730 1754

Note: ***1%. Controls in all regressions include head of household characteristics atbaseline (age, gender, schooling, employment, and health insurance), assets indexand dengue index. Observations are clustered at the locality level. Robust standarde

4

Debitpobitps

itm

56 A.C. Dammert et al. / Journal of H

accination and control strategies (Anders and Hay, 2012).onetheless, the Pan American Health Organization defines three

evels of risk for dengue transmission as low (HI < 0.1%), mediumHI 0.1–5%), and high (HI > 5%) (MINSA 2001). Our data estimatehat the House Index was 8.15% in the treatment group and 9.26%n the control group at follow-up. In both areas, the level of riskransmission was high.8

The follow-up survey also collected information on dengue inci-ence and illness of household members during the treatmenteriod. In our sample, less than 1% of households self-reportedaving a member clinically diagnosed with dengue at the time ofhe follow-up survey.9 This low number may have been accruedecause the follow-up occurred too soon after the interventiono capture a significant number of dengue episodes or becauseelf-reported measures may be affected by the underreportingf dengue. In fact, reporting an objective measure of ‘incidencef dengue’ is not simple. First, only health officials can providenformation of dengue incidence after performing laboratory testsn individuals. Routine checkups or self-observation of symp-oms is not enough for dengue diagnosis. Moreover, even if thisnformation would be available from public health centers, it wille severely affected by undercoverage, as it is a common prac-ice among many individuals affected by this virus to be treatedy means of domestic health practices at home (Beatty et al.,011).10

Given that undercoverage and misreported error may biaself-reported measures of dengue incidence, as an alternative mea-ure, we also recorded whether a household member had dengueymptoms (fever, headaches, eye pain, and diarrhea) during thereatment period. At follow-up, 3.4% of households in the treat-

ent group reported having a household member with dengueymptoms, while 6.6% in the control group did so. Column 1 ofable 6 shows that the difference of 3 percentage points is statisti-ally significant at the 10 percent level. Yet it is worth mentioninghat most of the dengue symptoms can be easily confused withther common health problems, which makes difficult to connectymptoms or dengue risk measures to morbidity and dengue infes-ation.

We also analyzed whether any member of the household expe-ienced diarrhea during the month prior to the follow-up. Sinceouseholds report cleaning their water reservoirs and treatingater for consumption, one might wonder, whether these changes

n behavior are big enough to have an effect on preventable water-elated diseases. Column 2 of Table 6 shows that householdsxposed to text messages self-report a decline in the incidencef diarrhea by 2 percentage points, which translates into a 13%mprovement relative to the control group (albeit not statisticallyignificant). Looking into the different groups, as Panel B of Table 6hows, the reduction in the incidence of diarrhea is higher and sta-istically significant for households receiving monetary messages−5.3 percentage points).

8 Malaria endemicity is commonly measured by the proportion of a populationith detectable malaria parasite in their blood. As Anders and Hay (2012) showed,owever, no comparable measure exists to quantify the endemic level of denguehere most epidemiological data rely on clinical reports.9 Reports from the Ministry of Health also show low levels of clinical dengue cases

n the area during the same months in 2010.10 An alternative explanation of the low dengue infection prevalence in our sampleopulation could be due to the suppression of the mosquito vector population in therea under study. This might not be the case, however, given that Talara reportedbout 52% of the total number of clinical cases in the region during the summeronths of 2013.

llb

r(pdg

taf

rrors in parenthesis.* 10%.

** 5%.

.3. Spillover effects

Several health studies have shown significant spillover effects.epending on the characteristics of the preventive product, theseffects can either increase adoption of preventive measures (e.g.,ednets as discussed in Dupas, 2013) or decrease it (e.g., deworm-

ng drugs as discussed in Miguel and Kremer, 2004). In our setting,here are possible channels through which externalities may occur:eople may share the information they got from the text messager people who change their preventive behavior may discuss theenefits with friends and family. A standard approach to assess

ndividual externalities would be to rely on the information onhe number and type of social links for both participant and non-articipant households, as in Tontarawongsa et al. (2011) whentudying bednet adoption in India.

In terms of mosquito density, residential proximity is alsomportant since increasing uptake of preventive measures amonghose exposed to the repetitive information may change the

osquito vector population in the area under study. We follow thisine of inquiry by providing evidence on spatial spillovers using GPSocation to examine the extent to which outcomes are influencedy treatment density (Dupas, 2013; Tarozzi et al., 2013).11

We conducted a complete mapping of the study area andecorded longitude and latitude coordinates for 1754 householdshouseholds with complete address information and that are

resent at both baseline and follow-up). Based on the GIS coor-inates, we constructed measures of treatment density within aiven radius of our sample household. On average, households have

11 It is important to note that, in our paper, as in Tarozzi et al. (2013), there is noreatment variation within localities. This is different from Dupas (2013), where theuthor is able to exploit the exogenous variation given by the randomization of theraction of beneficiary households within each locality.

A.C. Dammert et al. / Journal of Health Economics 35 (2014) 147–161 157

Table 7Testing for spillover effects.

Cover waterreservoirs

Clean waterreservoirsdaily orweekly

Drink boiledwater ortreat waterwith chlorine

Use mosquitonets forsleepingand/orscreens inwindows

Have water-holding solidwaste (tires,bottles, orothers)

StandardizedContainerIndex

StandardizedHouse Index

StandardizedBreteauIndex

(1) (2) (3) (4) (5) (6) (7) (8)

Within 50 m of householdNumber of

households0.000 −0.006 0.003 −0.004 0.002 0.008 0.000 −0.010

(0.003) (0.006) (0.004) (0.006) (0.003) (0.011) (0.014) (0.010)Number of

treatedhouseholds

0.006 0.007 0.001 0.008 0.003 0.028 −0.017 0.008

(0.006) (0.013) (0.008) (0.012) (0.006) (0.024) (0.030) (0.022)Treatment∗# of

treatedhouseholds

−0.004 0.006 0.003 −0.007 −0.002 −0.043* 0.018 −0.003

(0.006) (0.013) (0.008) (0.013) (0.006) (0.025) (0.031) (0.023)

Within 100 m of householdNumber of

households−0.002 −0.005 0.003 −0.002 0.003** 0.004 0.001 0.005

(0.001) (0.003) (0.002) (0.003) (0.001) (0.005) (0.007) (0.005)Number of

treatedhouseholds

0.003 0.011** −0.006* −0.004 −0.003 0.016* 0.015 0.004

(0.002) (0.005) (0.003) (0.004) (0.002) (0.009) (0.011) (0.008)Treatment∗# of

treatedhouseholds

−0.001 −0.003 0.004 0.005 0.001 −0.023** −0.017* −0.010

(0.002) (0.005) (0.003) (0.004) (0.002) (0.008) (0.011) (0.008)

Within 150 m of householdNumber of

households−0.002** −0.001 0.003** −0.001 0.003** 0.002 −0.002 0.003

(0.001) (0.002) (0.001) (0.002) (0.001) (0.004) (0.005) (0.003)Number of

treatedhouseholds

0.004** 0.004 −0.003 −0.002 −0.004** 0.008 0.011 0.002

(0.002) (0.003) (0.002) (0.003) (0.002) (0.006) (0.007) (0.005)Treatment∗# of

treatedhouseholds

−0.001 −0.002 0.000 0.003 0.001 −0.012** −0.007 −0.006

(0.001) (0.003) (0.002) (0.003) (0.001) (0.005) (0.007) (0.005

N 1754 1754 1754 1754 1754 1487 1487 1487

Note: ***1%. The estimation includes locality fixed effects and controls for head of household characteristics at baseline (age, gender, schooling, employment, and healthinsurance) Robust standard errors in parenthesis.

35estfsp

tso

oei

tpmshmIt

* 10%.** 5%.

.8 neighbors within a 50-meter radius (10.3 neighbors within0–100 m radius, and 12.5 within 100–150 m radius) who werexposed to text messages. We also constructed a measure of totalample population density within a given radius. Table 7 presentshe coefficient estimates controlling for locality fixed effects. Weocus on a radius of less than 150 meters, since epidemiologicaltudies of dengue have shown that the Aedes mosquito has a dis-ersal of approximately 100 m (Anders and Hay, 2012).12

As Table 7 shows, the coefficient estimates on the interaction

erms between the treatment group and treatment intensity are nottatistically significantly different from zero for all self-reportedutcomes (columns 1–5). These results suggest that households in

12 We have assessed the presence of externalities by testing whether the numberf control households living within a 150 m radius of treatment localities experi-nced changes in behavior and health outcomes. The point estimates, however, aremprecisely estimated. We thank one of the referees for pointing this out.

Iitmbaina

reatment and control localities experienced similar effects fromroximity to treated households. Columns 6–8 show the objectiveeasures of dengue risk transmission; however, the results suggest

ome positive spillover effects. If we increase the number of treatedouseholds within 50 m, 100 m, or 150 m, households in the treat-ent group experienced a decrease in the standardized Container

ndex (column 6). As expected, this effect declines as distance fromhe index household increases.

Why are no spatial spillovers observed in the self-reported data?t might be possible that the follow-up occurred too soon after thentervention and there was not enough time for the information toravel in the neighborhood. An alternative explanation is that infor-

ation on preventive behavior may have been spread out, but itsenefits are not immediate; thus, people in endemic dengue areas

re waiting for the peak of the dengue season to change their behav-or and/or for the Ministry of Health to spray the area once a certainumber of clinical episodes are observed. A study conducted over

longer time horizon and identifying social networks may be able

1 ealth Economics 35 (2014) 147–161

tac

5

taedwtipipp

ttthmasatadgidpa2c

mtttmfd

Table A1Urban Parinas – Ministry of Health Division.

Sector Localities

1 J.CHAVEZ,M.SULLON, A.H.ALGARROBOS, AH.VENECEDORES,LASGARDENIAS Y PILAR NORES, MARIO AGUIRRE

2 QUINONES,9OCTUBRE,SANSEBASTIAN,STA.ROSA,B.AIRES,S.CERRO,J.BASADRE

3 7JUNIO,A.GARCIA,L.A.SANCHEZ,M.CHAVEZ,A.FUJIMORI,LOSGERANIOS,AH.LOS ROBLES,AH.MIRAFLORES,BELLOHORIZONTE,H.CARLIN,2FEBRERO,LOS FICUS,JOSE.C.MARIATEGUI

4 TALARA ALTA,LAS PALMERAS,URB.LOS ROBLES,CRISTO REY CRUZDE MAYO,02 MAYO,URB.MARIA AUXILIADORA

5 AV. F PAR, URB. LOS VENCEDORES, JAMES STORE, SUDAMERICA,ENAPU, V.TALARA, L.ALVA CASTRO, M.CABREDO, LUCIANOCASTILLO.

6 APROVISER,FONAVI,LAS MERCEDES7 URB. LOS PINOS. URBA, VILLA FAP CORPAC8 PARQUES DEL 43–57, AV. H PAR; AV. E, F IMPAR Y G.9 PARQUE DEL 35–42 y 58–62; AV. H IMPAR AV. D E y G.10 PARQUES DEL 63–72, LA PARADA, AV. E, AV. A; URB LOS JAZMINES,

CENTRO CIVICO Y MERCADO CENTRAL.11 PARQUE DEL 8– 22, AV A y B, URB. A.TABOADA12 PARQUES 1–7 y 22–27 AV. A, B y C; PARQUE 78; UNIDAD VECINAL;

CALLE 413 PARQUE 28–34 AV. B Y BARRIO PARTICULAR, BLOCK MILITAR,

BLOCK PROFESORES Y URB. STA.ROSA14 URB. MUNICIPAL, L.VILLANUEVA, SAN JUDAS TADEO, J.OLAYA, B.EL

PESCADOR, SAN MARTIN, LAS MERCEDES, CEMENTERIO, TALLERESUNIDOS, MUELLE, B.ESPECIALISTAS

15 SAN PEDRO, JESUS MARIA, STA. RITA, B. ESTIBADORES, LASPENITAS.

16 PT. ARENAS17 SACOBSA18 NEGREIROS19 ENACE

ritiata

TC

SNa

58 A.C. Dammert et al. / Journal of H

o identify some of these effects more accurately. Moreover, vari-tion in treatment density may be correlated with unobservableharacteristics of observed location characteristics.

. Conclusions

The dengue burden is high for developing countries that faceight health budget constraints. Several countries in Latin Americannd the Caribbean, including Peru, have implemented strategies toradicate the disease, including indoor residual spraying and theistribution of temephos (distributed as Abate) to kill the larvae inater containers. Dengue infections, however, continue to rise over

he last years (PAHO, 2007). A key stylized fact on health behav-or in developing countries is that households invest little in thereventive health course. A plausible explanation for this behavior

s the lack of information on illness prevention. Our interventionrovides evidence on the link between information campaigns andreventive behavior.

We randomly assigned over 2000 households in northern Peruo information treatment and control groups to evaluate the effec-iveness of providing repetitive information via mobile phoneechnologies on enhancing households’ health behavior. House-olds in the treatment group received more than 30 tailored textessages during a three-month period designed to increase the

doption of preventive measures against dengue. The main resultsuggest that exposure to repeated health preventive informationffected household behavior. Overall, the Breteau Index, an objec-ive measure of dengue infestation, shows that households thatre repeatedly exposed to preventive information experienced aecrease of −0.10 standard deviations of the mean of the controlroup, which is explained by positive changes in household behav-or and is translated into a reduction of self-reported incidence ofengue symptoms. Given that mobile phone service offers a largeool of individual instant and inexpensive access to information,nd since it costs US$10 to US$25 to treat one case of dengue (WHO,009), this health preventive care intervention can be consideredost-effective.

Further evidence shows positive externalities for the objectiveeasures of risk transmission, as the number of water-holding con-

ainers testing positive for dengue decreases in treatment areas ashe number of treated households increase. This result suggests

hat changes in household behavior may affect the density of the

osquito vector, however further research is needed to analyzeurther the spillover effects in the prevention of mosquito borneiseases. Moreover, we do not find statistically significant framing

cpwh

able A2ensus 2007.

Baseline survey (2009) Urban

All

Cement wall 0.817 0.751

Electricity 0.986 0.873

Piped water 0.917 0.765

Piped sewage 0.933 0.767

Refrigerator 0.598 0.422

Computer 0.211 0.142

Head: male 0.749 0.740

Head: level of completed schoolingNo schooling 2.53 2.66

Primary 29.29 27.80

Secondary 45.68 37.59

Post-secondary 22.50 31.95

Number of households 2021 20772

ource: 2007 Census (www.inei.gob.pe).ote: Ownership of radio and TV are not comparable with those in our data. The Census qny TV (color or black and white TV). Likewise, our questionnaire asks for a radio while th

esults. From a policy standpoint, this result suggests that the fram-ng of SMS health messages is not as important as the provision ofhe information itself. Finally, this study showed that the largestmpacts on preventive behavior against dengue were concentratedmong the poorest households, a pattern that could be related tohe fact that these households in general are the least informedbout the health situation they face.

These findings provide valuable evidence for other developingountries interested in incorporating digital technologies on health

rograms. Although text messages considered in this evaluationere related to dengue prevention, this cost-effective approachas the potential for use in the prevention of other types of

Parinas: Census 2007

Cell-phone household Non-cell phone household

0.811 0.6960.921 0.8270.821 0.7050.820 0.7150.537 0.2770.220 0.0450.762 0.712

1.83 3.7121.40 35.8437.13 38.1839.64 22.2711570 9202

uestionnaire asks for color TV only while our questionnaire asks for ownership ofe Census asks for radio separately from ownership of any music equipment.

ealth Economics 35 (2014) 147–161 159

ditmcbc

cbatowaru

A

A

fTdaloAmtp

fgthTaitdwtog

A

lfttrpbnc

All

Trea

tmen

t

Mon

etar

y

gain

(T1)

Mon

etar

y

loss

(T2)

Non

-mon

etar

yga

in

(T3)

Non

-mon

etar

ylo

ss

(T4)

Con

trol

Dif

fere

nce

T

−

C[p

-val

ue]

p-V

alu

e

of

F-te

st[T

1

=

T2

=

T3

=

T4]

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

0.12

1

(0.0

89)

0.11

2

(0.0

87)

0.12

4

(0.1

13)

0.08

6

(0.0

73)

0.10

5

(0.0

60)

0.13

5

(0.0

87)

0.13

7

(0.0

92)

−0.0

25

[0.1

74]

0.38

73

0.04

4

(0.0

56)

0.03

7

(0.0

52)

0.04

8

(0.0

56)

0.02

0 (0

.033

)

0.04

0

(0.0

55)

0.04

1

(0.0

59)

0.05

5

(0.0

63)

−0.0

18

[0.1

34]

0.25

60

0.05

6

(0.0

64)

0.05

4

(0.0

66)

0.04

3

(0.0

65)

0.06

0

(0.0

72)

0.03

3

(0.0

38)

0.08

2

(0.0

71)

0.05

8

(0.0

62)

−0.0

04

[0.7

59]

0.08

660.

022

(0.0

35)

0.02

1

(0.0

35)

0.03

4

(0.0

48)

0.00

8

(0.0

15)

0.03

3

(0.0

40)

0.00

9

(0.0

20)

0.02

4

(0.0

37)

−0.0

03

[0.6

66]

0.03

30

s

base

d

on

regr

essi

on

of

attr

itio

n

on

trea

tmen

t

grou

p

wit

h

clu

ster

ed

stan

dar

d

erro

rs

at

the

loca

lity

leve

l.

N

=

2021

incl

ud

ing

1350

hou

seh

old

s

in

trea

tmen

t

grou

p

loca

liti

es

and

671

in

con

trol

grou

pia

tion

s

in

par

enth

esis

.

A.C. Dammert et al. / Journal of H

iseases or risky behaviors. In addition, the findings highlight themportance of considering the dynamic nature of the informationo alter health behavior where the repeated dissemination of infor-

ation is important. Reminders might mitigate attentional failure,hange intertemporal allocations, and improve consumer welfarey providing associations between future opportunities and today’shoices (Karlan et al., 2010).

Indeed, booming mobile phone connectivity in developingountries is shifting policy attention focus and increasing the num-er of initiatives aimed at empowering vulnerable groups withccess to tailored content-value information delivered to themhrough a new technological platform. We believe that, in contextsf poverty, weak institutional structures, and imperfect marketshere information is often costly or incomplete, mobile phone

pplications that provide repetitive information to people aboutisky behaviors could have non-negligible effects that need to benveiled.

ppendix A.

See Table A1.

ppendix B. Census data

The 2007 Peruvian Census provides online information for dif-erent economic and demographic variables at the district level.able A2 presents the descriptive statistics of all localities in theistrict of Parinas, disaggregated by whether the households had

mobile phone or not. The first column corresponds to the base-ine information presented in Table A2. First, we investigate howur sample is similar to cell phone households in the same district.s the table shows, 56% of households in Parinas had at least oneobile phone in 2007. By comparing the first and third columns,

he results show that our sample looks similar to the average cellhone household in Parinas.

Second, we analyze how different are cell-phone householdsrom non-cell phone households in Parinas. Comparing demo-raphic and household characteristics across columns, we can seehat cell phone households are more educated and have betterousehold infrastructure than non-cell phone households have.his is not surprising given that the cost of handsets and servicesre not affordable to all households. Thus, we cannot general-ze the findings to all households in urban areas. Nevertheless,he empirical findings suggest the potential relevance of ICTs andifferent messages on household behavior for cell phone users,ho are an important fraction of Peruvian households. Moreover,

here are potential positive spillovers accrued from the reductionf the Aedes mosquito in cell phone households as Table 7 sug-est.

ppendix C. Attrition

Table A3 presents the attrition that occurred between the base-ine and follow-up surveys by treatment status. Attrition was 12%or the entire sample, 11.2% in the treatment group and 13.7% in

he control group. As column 8 shows, the difference is not sta-istically significant from zero. Likewise, after disaggregating theate of attrition across three different sources (respondent notresent, migration, and refusal) we observe statistical similarityetween the treatment and control groups. The test statistic doesot reject the null hypotheses of equality of means across all threeategories. Ta

ble

A3

Att

riti

on

rate

s.

Att

riti

on

Rat

e(n

=

100

loca

liti

es)

Res

pon

den

tn

ot

pre

sen

tM

igra

tion

Ref

usa

l

Not

e:

Test

of

equ

alit

y

ilo

cali

ties

. Sta

nd

ard

dev

1 ealth

stadcfsd

R

A

A

A

A

B

B

B

C

C

C

D

D

D

D

E

E

F

F

F

F

I

J

J

J

J

K

K

K

K

L

L

L

M

M

M

M

P

PP

P

R

R

R

R

R

R

S

S

S

60 A.C. Dammert et al. / Journal of H

Looking into across the various treatment groups, columns 3–6how that there are no statistically significant differences acrosshe different treatment and control groups except for the percent-ge of refused interviews. The gain group has a 1-percentage pointifference in the percentage of refused interviews compared to theontrol group, while the loss group has a −1.5 percentage point dif-erence with the control group. The overall difference, however, ismall and not statistically significant due to the size of the standardeviations.

eferences

ker, J., 2010. Information markets near and far: mobile phones and agriculturalmarkets in Niger. American Economic Journal: Applied Economics 2 (3), 46–59.

ker, J., Mbiti, I., 2010. Mobile phones and economic development in Africa. Journalof Economic Perspectives 24 (3), 207–232.

nders, K., Hay, S., 2012. Lessons from malaria control to help meet the ris-ing challenge of dengue. The Lancet Infectious Diseases 12 (12), 977–984.

nderson, K., Chunsuttiwat, S., Nisalak, A., Mammen, P., Libraty, D., Rothman, A.,Green, S., Vaughn, D., Ennis, F., Endyl, T., 2007. Burden of symptomatic dengueinfection in children at primary school in Thailand: a prospective study. TheLancet 369 (9571), 1452–1459.

eatty, M., Beutels, P., Meltzer, M., Shepard, D., Hombach, J., Hutubessy, R., Dessis,D., Coudeville, L., Dervaux, B., Wichmann, O., Margolis, H., Kuritsky, J., 2011.Health economics of dengue: a systematic literature review and expert panel’sassessment. The American Journal of Tropical Medicine and Hygiene 84 (3),473–488.

ertrand, M., Karlan, D., Mullainathan, S., Shafir, E., Zinman, J., 2010. What’sadvertising content worth? Evidence from a consumer credit marketing fieldexperiment. The Quarterly Journal of Economics 125 (1), 263–305.

laya, J., Fraser, H., Holt, B., 2010. E-health technologies show promise in developingcountries. Health Affairs 29 (2), 244–251.

airncross, S., Shordt, K., Zacharia, S., Govindan, B., 2005. What causes sustainablechanges in hygiene behavior? A cross-sectional study from Kerala, India. SocialScience and Medicine 61 (10), 2212–2220.

hong, A., 2011. Development Connections: Unveiling the Impacts of New Informa-tion Technologies. Palgrave MacMillan, Washington, D.C.

hong, A., Karlan, D., Shapiro, J., Zinman, J., 2013. (Ineffective) messages to encour-age recycling: evidence from a randomized evaluation in Peru. World BankEconomic (forthcoming).

upas, P., 2009. What matters (and what does not) in households’ decision to investin malaria prevention? American Economic Review Papers and Proceedings 99(2), 224–230.

upas, P., 2011a. Do teenagers respond to HIV risk information? Evidence from afield experiment in Kenya. American Economic Journal: Applied Economics 3(1), 1–34.

upas, P., 2011b. Health behavior in developing countries. Annual Review of Eco-nomics 3, 425–449.

upas, P., 2013. Short-Run Subsidies and Long-Run Adoption of New Health Prod-ucts: Evidence from a Field Experiment, Mimeo.

lder, J., Lloyd, L., 2006. Achieving Behaviour Change for Dengue Control: Methods,Scaling-Up, and Sustainability. In: Working Paper for the Scientific WorkingGroup on Dengue Research, Convened by the Special Programme for Researchand Training in Tropical Diseases, 1–5 October 2006, Geneva.

spinoza-Gomez, F., Hernandez-Suarez, M., Coll-Cardenas, R., 2002. Educationalcampaign versus malathion spraying for the control of Aedes aegyptiin Colima, Mexico. Journal of Epidemiology Community Health 56, 148–152.

ilmer, D., Pritchett, L., 2001. Estimating wealth effects without expenditure data –or tears: an application of educational enrolment in states of India. Demography38 (1), 115–132.

jelsoe, B., Marshall, A., Miller, Y., 2009. Behavior change interventions delivered bymobile phone short message service. American Journal of Preventive Medicine36 (2), 165–173.

ranklin, V., Waller, A., Pagliari, C., Greene, S.A., 2006. A randomized controlled trialof Sweet Talk, a text-messaging system to support young people with diabetes.Diabetic Medicine 23, 1332–1338.

ocks, D., 2004. A Review of Entomological Sampling Methods and Indicators forDengue Vectors. WHO.

NEI, 2012. Las Tecnologías de Información y Comunicación en los Hogares. InformeTécnico Nro 2. INEI, Lima.

alan, J., Somanathan, E., 2008. The importance of being informed: experimental evi-dence on demand for environmental quality. Journal of Development Economics

87 (1), 14–28.

ensen, R., 2007. The digital provide: information (technology), market performance,and welfare in the South Indian fisheries sector. The Quarterly Journal of Eco-nomics CXX11 (3), 879–924.

T

T

Economics 35 (2014) 147–161

ensen, R., 2010. Information, efficiency, and welfare in agricultural markets. Agri-cultural Economics 41 (1), 203–216.

oo, N.-S., Kim, B.-T., 2007. Mobile phone short message service for behavior mod-ification in a community-based weight control program in Korea. Journal ofTelemedicine and Telecare 13, 416–420.

ahneman, D., Tversky, A., 1979. Prospect theory: an analysis of decision under risk.Econometrica 47 (2), 263–291.

arlan, D., McConnel, M., Mullainathan, S., Zinman, J., 2010. Getting to the Top ofMind: How Reminders Increase Saving, NBER Working Papers 16205. NationalBureau of Economic Research, Inc.

ling, J., Liebman, J., Katz, L., 2007. Experimental analysis of neighborhood effects.Econometrica 75 (1), 83–119.

oshy, E., Car, J., Majeed, A., 2008. Effectiveness of mobile-phone short messageservice (SMS) reminders for ophthalmology outpatient appointments. BMCOphthalmology 8, 9.

ester, R., Ritvo, P., Mills, E., Kariri, A., Karanja, S., Chung, M., Jack, W., Habyari-mana, J., Sadatsafavi, M., Najafzadeh, M., Marra, C., Estambale, B., Ngugi, E., Ball,B., Thabane, L., Gelmon, L., Kimani, J., Ackers, M., Plummer, F., 2010. Effectsof a mobile phone short message service on antiretroviral treatment adher-ence in Kenya (WelTel Kenya1): a randomised trial. The Lancet 376 (9755),1838–1845.

evin, I.P., Schneider, S.L., Gaeth, G., 1998. All frames are not created equal: a typo-logy and critical analysis of framing effects. Organizational Behavior and HumanDecision Processes 76 (2), 149–188.

uby, S.P., Agboatwalla, M., Painter, J., Altaf, A., Billhimer, W.L., Hoekstra, R.M., 2004.Effect of intensive handwashing promotion on childhood diarrhea in high-riskcommunities in Pakistan: a randomized controlled trial. Journal of the AmericanMedical Association 291 (21), 2547–2554.

adajewicz, M., Pfaff, A., van Geen, A., Graziano, J., Hussein, I., Momotaj, H., Sylvi, R.,Ahsan, H., 2007. Can information alone change behavior? Response to arseniccontamination of groundwater in Bangladesh. Journal of Development Eco-nomics 84 (2), 731–754.

eyerowitz, B., Chaiken, S., 1987. The effect of message framing on breast self-examination attitudes, intentions, and behavior. Journal of Personality andSocial Psychology 52 (3), 500–510.

iguel, E., Kremer, M., 2004. Worms: identifying impacts on education andhealth in the presence of treatment externalities. Econometrica 72 (1),159–217.

utu, M., Yamano, T., 2009. The impact of mobile phone coverage expansion onmarket participation: panel data evidence from Uganda. World Development37 (12), 1887–1896.

AHO, 2004. Planning Social Mobilization and Communication for Dengue FeverPrevention and Control. A Step-by-Step Guide. Pan-American Health Organiza-tion.

AHO, 2007. Health in the Americas. Pan-American Health Organization.atrick, K., Raab, F., Adams, M., Dillon, L., Zabinski, M., Rock, C., Griswold, W., Nor-

man, G., 2009. A text message-based intervention for weigh loss: randomizedcontrolled trial. Journal of International Medical Research 11, 1.

op-Eleches, C., Thirumurthy, H., Habyarimana, J., Zivin, J., Goldstein, M., de Walque,D., Mackeen, L., Haberer, J., Kimaiyo, S., Sidle, J., Ngare, D., Bangsberg, D.,2011. Mobile phone technologies improve adherence to antiretroviral treat-ment in a resource-limited setting: a randomized controlled trial of text messagereminders. AIDS 25 (6), 825–834.

amírez, G., 2011. Situación Epidemiológica de Dengue en el Perú y Definiciones deCaso de Dengue Hemorrágico de la OMS. Reporte preparado para el Ministeriode Salud, Lima.

eis, R., 2006. Inattentive consumers. Journal of Monetary Economics 53 (8),1761–1800.

hee, M., Sissoko, M., Perry, S., McFarland, W., Parsonnet, J., Doumbo, O., 2005. Useof Insecticide-Treated Nets (ITNs) following a malaria education interventionin Piron, Mali: a control trial with systematic allocation of households. MalariaJournal 4, 35.

odgers, A., Corbett, T., Bramley, D., 2005. Do u smoke after txt? Results of a stan-dardized trial of smoking cessation using mobile phone text messaging. TobaccoControl 14, 255–261.

othman, A., Salovey, P., 1997. Shaping perceptions to motivate healthy behavior:the role of message framing. Psychological Bulletin 121 (1), 3–19.

othman, A., Bartels, R., Wlaschin, J., Salovey, P., 2006. The strategic useof gain- and loss-framed messages to promote healthy behavior:how theory can inform practice. Journal of Communication 56, S202–S220.

alovey, P., Williams-Piehota, P., 2004. Field experiments in social psychology:message framing and the promotion of health protective behaviors. AmericanBehavioral Scientist 47, 488–505.

hah, S., Sani, J., Hassan, M., Safian, M., Aizuddin, A., Hod, R., 2012. Relationshipsbetween Aedes indices and dengue outbreaks in Selangor, Malaysia. DengueBulletin 36, 166–174.

ilver, J., 2008. Sampling the larval population. In: Mosquito Ecology Field SamplingMethods.

arozzi, A., Mahajan, A., Blackburn, B., Kopf, D., Krishnan, L., Yoong, J., 2013. Micro-

loans, Insecticide-Treated Bednets and Malaria: Evidence from a RandomizedControlled Trial in Orissa (India), Mimeo.

homas, D., 1990. Intra-household resource allocation: an inferential approach.Journal of Human Resources 25 (4), 635–664.

ealth

T

W

A.C. Dammert et al. / Journal of H

ontarawongsa, C., Mahajan, A., Tarozzi, A., 2011. (Limited) Diffusion of Health-protecting Behaviors: Evidence from Non-beneficiaries of a Public HealthProgram in Orissa, India, Mimeo.

HO, 2009. Dengue Guidelines for Diagnosis, Treatment, Prevention, and Control.World Health Organization.

Z

Economics 35 (2014) 147–161 161

urovac, D., Sudoi, R., Akhwale, W., Ndiritu, M., Hamer, D., Rowe, A., Snow, R., 2011.The effect of mobile phone text-message reminders on Kenyan health work-ers’ adherence to malaria treatment guidelines: a cluster randomised trial. TheLancet 378 (9793), 795–803.