Journal of Health Economics 35 (2014) 147–161

Contents lists available at ScienceDirect

Journal of Health Economics

journal homepage: www.elsevier.com/locate/econbase

Preventing dengue through mobile phones: Evidence from a field

experiment in Peru6
Ana C. Dammert a,b,∗ , Jose C. Galdo a,b , Virgilio Galdo c
a
Carleton University, Canada
b
IZA, Germany
c
World Bank, United States

a r t i cl e i nf o ab s trac t

Article history: Dengue is the most rapidly spreading mosquito-borne viral disease in the world (WHO, 2009). During
Received 24 August 2012 the last two decades, the dramatic rise in the number of dengue infections has been particularly evi -
Received in revised form 7 February 2014
dent in Latin American and the Caribbean countries. This paper examines the experimental evidence
Accepted 7 February 2014
of the effectiveness of mobile phone technology in improving households’ health preventive behavior in
Available online 5 March 2014
dengue-endemic areas. The main results suggest that repeated exposure to health information encourages
households’ uptake of preventive measures against dengue. As a result, the Breteau Index in treatment
JEL classification:
households, an objective measure of dengue risk transmission, is 0.10 standard deviations below the
I10
O12 mean of the control group, which shows a reduction in the number of containers per household that test
positive for dengue larvae.
Keywords: The estimates also show marginally significant effects of the intervention on self-reported dengue
Dengue symptoms. Moreover, we use a multiple treatment framework that randomly assigns households to one
e-Health
of the four treatment groups in order to analyze the impacts of framing on health behavior. Different
Peru
variants emphasized information on monetary and non-monetary benefits and costs. The main results
Framing
Experimental design show no statistical differences among treatment groups.
© 2014 Elsevier B.V. All rights reserved.

1. Introduction Health Organization, dengue represents an enormous global health

Dengue is one of the most serious mosquito-borne viral diseases
affecting humans and is a leading cause of illness in the tropics
and subtropics. It is transmitted by the bite of Aedes mosquitoes
infected with any of four dengue serotypes. According to the World
6 This research has benefited from comments and suggestions received by the edi-
tor, three anonymous referees as well as seminar participants at the 2012 LACEA,
2013 Canadian Economics Association Meeting, and 2013 RECODE meeting. We
thank SASE Asociacion Civil, Centro IDEAS, and the Ministry of Health-DIRESA Sul-
lana for their logistic and institutional support. Special thanks to Cecilia Bustamante,
Teodoro Saez, Dr. Walter Vegas, and Mary Villavicencio for their dedication and
care in the project implementation, field work, and data collection. Minoru Higa
and Rene Castro provided excellent research assistance. This work was supported by
a research grant from the Inter-American Development Bank (Proyecto Red de
Centros de Investigacion - ATN/SF-11298-RG). The standard disclaimer applies.
∗ Corresponding author at: Department of Economics and Norman Paterson
School of International Affairs, 1125 Colonel by Drive, Ottawa, ON K1S5B6, Canada.
Tel.: +1 613 5202600.
E-mail addresses: ana dammert@carleton.ca (A.C. Dammert),
jose galdo@carleton.ca (J.C. Galdo), vgaldo@worldbank.org (V. Galdo).
burden, with 2.5 billion people worldwide at risk of contracting the
disease (WHO, 2009). Most alarming is the fact that, in the two last
decades, humanity has experienced a global emergence of dengue
as a major public health problem due to large demographic shifts,
lack of effective mosquito control, inadequate water and sewer
management systems, and weak public health infrastructure.
This paper presents evidence from the first large-scale, clus-
tered, randomized control trial that evaluates the effectiveness of
mobile phone technologies in enhancing households’ health pre-
ventive behavior in dengue-endemic areas. In recent years, mobile
phone service has become the most rapidly adopted technology in
developing countries, as the costs of installing mobile phone towers
are low relative to those of landlines (Jensen, 2010). Mobile phone
service could facilitate the diffusion of knowledge and best prac-
tices, reduce transaction costs, and improve the delivery of public
services (Aker and Mbiti, 2010; Chong, 2011).
Indeed, a small number of non-experimental microeconomic
studies have investigated the role of mobile phone technologies
in fostering economic development, particularly in rural agricul-
tural markets. Studies of fishermen in India (Jensen, 2007) and of

http://dx.doi.org/10.1016/j.jhealeco.2014.02.002
0167-6296/© 2014 Elsevier B.V. All rights reserved.
148 A.C. Dammert et al. / Journal of Health Economics 35 (2014) 147–161
farmers in Niger (Aker, 2010) and Uganda (Mutu and Yamano,
2009) have shown that access to mobile phone service is associated
with significant increases in arbitrage, declines in price dispersion,
and increases in the number of markets over which farmers trade,
all of which have led to improvements in farmers’ welfare. Despite
these promising results, mobile phone service has not yet been
used extensively to advance preventive health care in developing
countries (Blaya et al., 2010). Given that households in developing
countries invest little in preventive health care, and considering
that the treatment of one single case of dengue ranges from US$10
to US$25 (WHO, 2009), the provision of preventive information
through text messages has the potential to be a cost-effective
health care intervention.
We also contribute to the literature regarding mosquito-borne
diseases, which has focused on malaria due to the significant
number of lives worldwide that have been claimed by this dis-
ease. Not much has been written on dengue in the economics
literature. Dengue can cause recurring and debilitating infections
and without adequate treatment, it increases vulnerability to
other diseases, affects educational performance of children, and
reduces labor market productivity of adults (Beatty et al., 2011;
Anderson et al., 2007). There is no specific antiviral medicine or
vaccine against dengue; thus policy makers have turned their
attention to prevention policies which can be classified into
those that kill adult mosquitoes (indoor residual sprays), those
that inhibit mosquito breeding (larval habitat management), and
those that reduce dengue infection in humans (mosquito-treated
nets).1
Previous research in the medical literature has shown that there
are large private returns from the prevention of dengue, although
the adoption of preventive measures is low (Anders and Hay, 2012).
In the standard model of investments in human capital, individ-
uals invest in health products if the expected benefits from the
preventive product outweigh its costs. Low adoption of preven-
tive products could be due to a lack of knowledge or people’s
underestimation of the importance of controlling the mosquito
vector, since dengue may present as a mild illness episode (Elder
and Lloyd, 2006). Given that the Aedes mosquito thrives in urban
environments with limited water supply and ensuing shortage, it
has been shown that one of the most effective ways to control
dengue is to provide households with preventive information so
that they can eliminate the breeding places of the mosquito through
house maintenance and disposal of tires and plastics (Espinoza-
Gomez et al., 2002). Thus, the goal of our intervention was to
provide repetitive access to information in order to improve knowl-
edge of preventive practices, which may lead to reductions in
dengue infestation risk. In contrast to recent experimental stud-
ies on health information, where households received a one-time
randomly assigned message, we sent 30 messages over a period
of 3 months before the peak of the dengue season. Reminders can
mitigate ‘attentional failure’ and thus change intertemporal allo-
cations, and improve consumer welfare by providing associations
between future opportunities and today’s choices (Karlan et al.,
2010).
We conducted the field experiment in 100 urban localities in the
province of Talara in the department of Piura in northern Peru. Our
area of study is considered an endemic dengue area because of its
weather conditions, proximity to the Equatorial tropical area, and
low development of water supply and sanitation facilities. In fact,
the Peruvian Ministry of Health has declared the area as endemic
since 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 are
borne by the government, which is in charge of conducting the spraying campaigns.
virus in Piura. Since then, more than 20,000 people have been clin-
ically diagnosed with the virus (Ramírez, 2011).
Following the World Health Organization guidelines on dengue
prevention, we sent information regarding the mosquito’s life cycle
(e.g., eggs laid on the wet walls of containers of water), the con-
ditions that allow dengue to spread, and several strategies for
controlling the spread of the disease. To be effective, messages were
locally relevant by customizing them based on local uses of lan-
guage and local illness classifications. We measure the impact of
this informational exposure on health-preventive behavior (cov-
ering of water reservoirs, the use of mosquito nets and window
screenings, among others), self-reported dengue symptoms, cases
of diarrhea, and the presence of dengue larvae in water contain-
ers three months later. Importantly, the presence of externalities
is addressed by using GIS coordinates for all households in the
sample.
Based on several prior experiments on health prevention that
were done mainly in developed countries, we are also interested
in analyzing whether the perspective in which the information is
presented affects preventive behavior. It has been shown that the
framing of messages could shape attitudes toward risk and thereby
influence behavior and choices (Kahneman and Tversky, 1979). In
order to analyze the impacts of message framing on health behav-
ior, we randomly assigned localities within the treatment group
to one of four treatment groups. Each framing group highlighted
a piece of general information; some messages provided general
information by highlighting the positive (and negative) conse-
quences of adopting a preventive behavior while other messages
emphasized the monetary cost.
Several results emerged. First, exposure to repeated health
preventive information affects households’ health behavior. The
Breteau Index, an objective measure of dengue risk transmission,
shows that households exposed to preventive information experi-
enced a decrease in the number of water containers per household
testing positive for dengue with respect to the control group.
This is explained by changes in household behavior since there
are statistically significant increases in the probability of cover-
ing water reservoirs, cleaning of water reservoirs, consumption of
safe water, and the use of screens in windows and/or mosquito
bed nets. These findings contribute to the literature on preven-
tive behavior, the area in which most of the experimental studies
that find small effects provided information during a one-time
visit (Dupas, 2009). Studies that find positive effects were con-
ducted over many months by providing repetitive information
(Cairncross et al., 2005; Luby et al., 2004; Pop-Eleches et al., 2011;
Lester et al., 2010).
Second, we evaluated the impact of exposure to repetitive
information on different dengue indicators. The follow-up data
included self-reported information on dengue incidence and illness
of household members during the treatment period. The results
show that behavioral changes associated with the intervention
translated into a reduction in dengue symptoms (fever, headache,
and eyes pain) but did not affect the number of clinical diagnosed
dengue cases. Mild episodes of dengue may not lead to significant
costs to households, but they affect people’s perception toward the
severity of the disease. We also analyzed whether the behavioral
changes are big enough to have an effect on preventable water-
related diseases. The results show a decline in the incidence of
diarrhea in treatment households, although the null of no differ-
ence between the treatment and control areas cannot be rejected
at standard levels. These indicators, however, were self-reported
and thus are likely to be affected by recall error and misdiagno-
sis.
Third, the point estimates suggest that households exposed
to non-monetary loss messages experienced a slightly higher
 A.C. Dammert et al. / Journal of Health Economics 35 (2014) 147–161
increase in almost all preventive outcomes relative to other types
of messages. However, the null of equality of coefficients between
treatment groups is not rejected at standard levels in most cases.
This result is similar to other experimental studies that find no
framing effects on health behavior (e.g., Dupas, 2009). Moreover,
there is some evidence of the heterogeneous effects as the increase
in preventive measures against dengue was concentrated among
the poorest households. We do not find, however, differences
according to the gender of the cell-phone owner or by previous
exposure to dengue in the house. Finally, we do not find evidence
of geographical spillovers except in one outcome, the Container
Index. This finding shows that, as we increase the number of
treated households in the neighborhood, households in the treat-
ment group experienced a decrease in the number of containers
testing positive for dengue.
The remainder of the paper is organized as follows. Section 2
reviews the literature. Section 3 describes the study area, inter-
vention, and dataset. Section 4 presents our estimates. Section 5
concludes and outlines the envisaged main policy implications.
2. Related literature
In recent years, the use of text messages to deliver health ser-
vices has been studied mostly in developed countries with the
aimed at encouraging people to change their health behavior. For
instance, reminders for outpatient appointments were associated
with a reduction in the likelihood of patients missing their appoint-
ments (Koshy et al., 2008), promotion of weight-loss behavior
among overweight people (Patrick et al., 2009; Joo and Kim, 2007),
smoking cessation (Rodgers et al., 2005) and adherence to dia-
betes treatment among children and adolescents (Franklin et al.,
2006). Nonetheless, after a careful meta-evaluation of more than 33
studies from the medical literature in developed countries, Fjelsoe
et al. (2009) reported that only fourteen studies satisfied qual-
ity standards, four studies targeted preventive health behaviors,
and ten studies targeted clinical care. The authors conclude that
SMS-delivered interventions have positive short-term behavioral
outcomes, but further research is required to evaluate preventive
health behaviors.
Text messaging has not been used to spread awareness and
information about infectious diseases in endemic regions despite
recent evidence suggesting that households are responsive to infor-
mation on the health risks they face (Dupas, 2011a; Cairncross et al.,
2005; Rhee et al., 2005). Madajewicz et al. (2007), for example,
show that informing households about the arsenic concentration
in their well water increased the probability that they would switch
to a safer well. Similarly, Jalan and Somanathan (2008) report
that households that receive information about the concentration
of fecal bacteria in drinking water improved water-purification
behavior. Prevention is a key component of dengue control and can
be achieved through vector control and personal protection meas-
ures (WHO, 2009). Not all information, however, can be effective.
Dupas’ (2011b) review of the literature shows that the provision
of information can influence people’s behavior when they are not
fully informed about the health situation they face, when the source
of information is credible, and when they are able to process the
new information.
It is also important to analyze how informational campaigns
affect the adoption of preventive measures against the disease.
Prior work in psychology has shown that message framing has
an effect on motivating people to change their behavior (for
a review of framing on health outcomes, see Rothman and
Salovey, 1997; Salovey and Williams-Piehota, 2004; Rothman
et al., 2006). In particular, framing might have differential effects
149
on health prevention, detection, and treatment, depending on
whether the outcome is perceived to involve some risk or uncer-
tainty. For example, preventive behavior could be viewed as
relatively safe (i.e., it provides a relatively certain outcome), while
detection could be viewed as the act of taking a risk of find-
ing the presence of a health problem (Rothman and Salovey,
1997). Therefore, the main implication of the prospect theory
on health behavior is that gain-framed messages are more per-
suasive than loss-framed messages in encouraging preventive
behavior, while loss-framed messages are more persuasive than
gain-framed messages in encouraging detection behavior. Alter-
natively, independent of the perceived risk of the outcome, loss
framing might have greater effects due to negativity bias in pro-
cessing the information wherein people are influenced more by
negative information than with comparable positive information
(Meyerowitz and Chaiken, 1987). Thus, people are more motivated
to avoid a loss than to attain a gain of equal magnitude or loss
aversion.
We have scant empirical evidence on how framing shapes eco-
nomic behavior in developing countries. Bertrand et al. (2010)
show that demand for credit can be manipulated through letters
sent to potential borrowers containing randomly assigned psycho-
logical features motivated by specific types of frames and cues
such as profile pictures in South Africa. Chong et al. (2013), on the
contrary, show that informational messages emphasizing environ-
mental, social, or authority content were not effective in increasing
recycling behavior in Peru.
In a more related paper, Dupas (2009) analyzed an experimental
design in Western Kenya, where households receiving vouchers for
insecticide-treated bed nets were exposed to a one-time-only ran-
domly assigned message (flyer). Participants in the health-framing
group received a marketing message in which morbidity and mor-
tality were emphasized, while participants in the financial group
received a marketing message where financial gains from pre-
venting malaria were emphasized. The main results suggest that
neither of the two framing options affected the uptake of bed nets.
One possible explanation is that liquidity constraints are the main
barriers to investments in malaria prevention.
Our paper follows this line of inquiry, although it differs in
important ways. First, we sent multiple messages over a period of
three months, since repetition of information affects the processing
of the information into the memory system (encoding), which in
turn affects comprehension and long-term retention (PAHO, 2004).
From a theoretical economic standpoint, this might be explained
by models of ‘attentional failure’ (Karlan et al., 2010) or ‘rational
inattention’ (Reis, 2006). Reminders might change intertemporal
allocations, and improve consumer welfare, by providing asso-
ciations between future opportunities and today’s choices that
mitigate the attention failure (Karlan et al., 2010). Second, we are
interested in the effects of the provision of preventive informa-
tion through text messages. A number of recent empirical studies
in developing countries have shown that text reminders increase
adherence to HIV antiretroviral treatment (Pop-Eleches et al., 2011;
Lester et al., 2010) and health workers’ adherence to malaria
treatment (Zurovac et al., 2011). Third, our experimental design
considers that households might be willing to change their behav-
ior but may not be able to do so if the income constraint is binding
or if they are unable to borrow to purchase mosquito nets or
screens. For that reason, we also provided preventive information
that does not involve any direct cost, such as cleaning and covering
water reservoirs or discarding water-holding solid waste where
mosquitoes breed outside the house (e.g., tires, bottles, or other
small water reservoirs). Finally, we focus on a Latin American coun-
try where experimental interventions addressing health practices
are scant.
150 A.C. Dammert et al. / Journal of Health Economics 35 (2014) 147–161
Fig. 1. Study areas. Note: Study areas included 100 localities in Parinas, Piura. Red dots represent localities in the treatment group, while blue dots represent those in the
control group. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)
3. Experimental design
We implemented a randomized intervention in the district of
Pariñas, Department of Piura, located in the northern region of Peru
(Fig. 1). Most households in Pariñas have access to piped water
through house connections; however, safe water is only provided
a few days a week due to problems with the distribution pipe and
inadequate energy sources. Households report receiving water on
average 3.4 times a week, with an average of 7 h per day. Thus,
households use different reservoirs to store water: plastic storage
tanks or cisterns, drums, discarded buckets, or ground-level cement
water storage tanks, among others.
The Ministry of Health has divided the district of Pariñas into
19 health zones containing 100 localities based on the availabil-
ity of safe water, dengue propagation, and roads, among others
(see Appendix A). A household eligible to participate in the exper-
imental design is one where the head of household or spouse is a
cell phone user and is literate enough to read a simple message.
According to the 2007 Peruvian Census, 55.8% of households in
Pariñas had a cell phone in 2007. From that year to the year of
the intervention, the percentage of households with mobile phones
has increased significantly (INEI, 2012). Appendix B shows that the
average cell phone sample household looks similar to the average
household that owns a cell phone in Pariñas, according to the 2007
Peruvian Census data. It is important to note that our results are
informative for those households who had cell phones and there-
fore are somewhat richer and more educated than households with
no cell phone are. In this regard, one should assess with caution the
external validity of the results.
Two independent sets of randomizations at the locality level
were conducted based on the administrative information from
the Ministry of Health, complemented with a customized base-
line survey data. In the first step, a stratified random assignment
between treatment and control localities was implemented fol-
lowing a ratio of 2:1. The stratification was based on two key
characteristics: poverty index and dengue index. The former was
constructed from the principal component analysis of 13 house-
hold assets and 6 dwelling characteristics, including access to safe
water and toilet facilities. The latter was constructed using health
administrative data regarding the incidence of dengue at the local-
ity level. The dengue index is based on various indicators measuring
the percentage of houses where dengue larvae was found, the
percentage of containers where the larvae was found, and the
percentage of infected containers per 100 inspected houses. We
complemented this analysis with baseline information on the avail-
ability of mosquito nets, window screenings, and indoor residual
spraying at the locality level. Random stratification proceeded after
we split the data into four categories according to the distribution of
the poverty and dengue indices: high/high, high/low, low/high, and
low/low. Thresholds were determined according to the median of
the distribution for each index. Within each category, we conducted
random assignment between treatment and control localities fol-
lowing a ratio of 2:1. As a result, we randomly assigned 64 localities
to the treatment group and 36 localities to the control group.
In the second step, within treatment localities, we randomly
assigned localities to two orthogonal information sets: gain/loss
messages and monetary/non-monetary messages. This framework
allows us to make causal comparisons within each domain (i.e.,
gain versus loss groups and monetary versus nonmonetary groups)
as well as across four different combinations (i.e., monetary/gain,
monetary/loss, nonmonetary/gain, and nonmonetary/loss treat-
ment groups).
Approximately 1 out of 5 households that own a mobile phone
in Pariñas was included in the final sample. This represents 2021
households, of which 1350 belong to the treatment localities and
671 to control localities.2 Of the initial 2021 households with com-
plete baseline information, 1784 (88%) were re-interviewed in early
March 2010, three months after the intervention started and dur-
ing the month at which temperatures in the region are at their
peak. The main reasons for attrition were migration (3.9%), the
inability to find eligible respondents despite repeated visits (5.7%),
2
Within the treatment localities, 309 households were assigned to the mone-
tary/gain framing, 383 to monetary/loss framing, 362 to nonmonetary/gain framing,
and 296 to nonmonetary/loss framing.
 A.C. Dammert et al. / Journal of Health Economics 35 (2014) 147–161
and refusal (2.4%). Appendix C shows similar statistical significant
attrition rates across all experimental groups.
3.1. Treatment and SMS framing
Based on PAHO’s (2004) framework, we created 21 messages
targeting two different goals: (i) preventing production of adult
mosquitoes; i.e., water storage, refuse disposal, water-holding solid
waste in the yard, among others, and (ii) preventing exposure to the
bites of Aedes mosquitoes; i.e., screening of windows and doors,
use of bed nets, and use of repellents. We also created nine neutral
patient-care text messages or disease recognition messages, but
we did not frame them. In total, households received over 30 SMS
regarding dengue prevention, detection, and control over a period
of 3 months.3 Every effort was made to ensure that participants
knew how to send and receive text messages. It is important to note
that incoming SMS messages are free of charge, so households did
not incur any cost when they received a text message.
Households in the randomly assigned treatment group were
exposed to different framed messages according to their assigned
framing group. In the monetary framing group, households
received messages about the financial gains or losses they would
realize (e.g., working days lost to illness, cost of medical care). In
the gain/loss groups, some households received messages about
the benefits of taking preventive measures, while others received
messages about the losses from not taking such measures. Follow-
ing Levin et al. (1998), the negative frames were written using a
simple negation (i.e., not adopting the behavior), to avoid incorpo-
rating biases due to differences in linguistic connotation by using
alternative terminology. Table 1 provides some examples.
3.2. Baseline household survey
The baseline survey was completed in July–September 2009.
The head of household provided information on demographic
characteristics of each household member (i.e., age, educational
attainment, gender, siblings) and dengue episodes of any house-
hold 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-
tant to tailor the messages to the uses and representations of the local people. For
this reason, we conducted a focus group in the area of intervention during July
2009. The use of focus groups was particularly important to gather information
on language uses in terms of dengue’s knowledge and preventive practices. Rep-
resentatives from the Asociación de Mujeres de Negritos, an NGO actively working
on health interventions, participated in this first focus group. Their feedback was
complemented with detailed interviews with personnel from the public health sys-
tem, who provided useful information about current policies implemented by the
Ministry of Health aimed at facilitating communication with the population in the
area of intervention. Once the content of text messages was defined, a second focus
group was carried out in Piura with the participation of communication specialists
in epidemiological health care. The aim was to check the relevance of the infor-
mation, identify the messages that needed to be adjusted (e.g., better wording),
identify additional information that should be included, and validate the health
content based on local illness classifications. Finally, we received feedback from the
local enumerators after testing their comprehension of the messages.
4
The NGO IDEAS in collaboration with FORO Salud-Piura administered the pre-
intervention baseline survey. All local enumerators, supervisors, and monitors
participated in an intensive weekend training program, including discussion of
interview methods, questionnaire content, survey protocol, data quality checks,
and mock interviews. Survey manuals developed specifically for the baseline survey
were provided to all enumerators, supervisors, and monitors. Numerators assisted
in reviewing the survey instruments for inconsistencies and problems before imple-
menting the survey. In addition, we provided each enumerator with an information
pamphlet with local phone numbers in case the members of the household had ques-
tions about the veracity of the survey. In the baseline survey, the surveyor gathered
verbal consent to participate in this study and provided information regarding the
possibility of receiving text messages.
151
the dataset provided detailed information on dwelling characteris-
tics including toilet facilities, house infrastructure, and household
density, which proxy for households’ long-run economic status
(Filmer and Pritchett, 2001).
Table 2 shows the p-values for the mean differences in rele-
vant baseline covariates across treatment and control localities. The
results do not reject the null hypothesis of equality of means for
most covariates, as the differences are not statistically significantly
different from zero.5 Likewise, Table 2 shows no statistically signif-
icant differences across all treatment groups except for household
size, the rate of computer ownership, and the proportion of house-
holds reporting that vaccinations prevent dengue. The differences
are very small for the last two variables though, and they are only
statistically significant due to the size of the standard deviations.
Nonetheless, we control for these differences in the econometric
estimations. Overall, the result shows that the stratified ran-
domization was effective in balancing the covariate mean values
between treatment and control households.
From Table 2, the average head of household is a 50-year-old
male, with 9 years of completed schooling, living with 4.9 family
members in the household. About 96% of cell phone users have a
pre-paid phone, which is used mostly to communicate with fam-
ily members (85% among heads of households and 96% among
spouses). The average amount spent on pre-paid phone cards was
18 soles (about US$6.50) during the month before the survey. About
5% of the sample did not send and/or receive text messages, the
majority of them report not knowing how to do so.
Most respondents in the sample had heard about dengue; 89% of
the sample knew that it is transmitted by mosquito bite. However,
23% reported that dengue could be prevented by using the chemi-
cal treatment temephos (distributed under the name Abate) to kill
the larvae in the container. Abate is distributed free of charge by
the Ministry of Health and is one of the cores of the current public
educational campaign. In addition, 3% incorrectly report that vac-
cinations can prevent dengue. Concerning dengue incidence, about
25% of the sample self-reported being diagnosed with dengue dur-
ing the previous two years: 64% were diagnosed by a doctor or
nurse, 13% were diagnosed by a family member or self-diagnosed,
and 7% were diagnosed by a pharmacist. On average, the illness
lasted for over a week. It is important to note that the head of the
household or the spouse provided this information for all members
of the household; therefore, recall error might have affected these
self-reported data.
4. Results
4.1. Overall effects of text messaging
This section examines whether providing any type of infor-
mation about preventive health affected household behavior. The
relevant parameter of interest is the intent-to-treat (ITT) that meas-
ures the average impact of being exposed to preventive information
via text messages. For household i, the estimated regression for
each preventive outcome Yi (buying nets or screens, treating stor-
age water, cleaning reservoirs, among others), is given by:
Yi = ˛ + ˇ1 Ti + X ∗ μ + εi , (1)
where Ti is an indicator that equals 1 if the household was randomly
exposed to the preventive information and X is a vector of base-
line characteristics, including head of household characteristics,
5
Unreported results show that there are no differences in means and significance
of the tests of equality if the means are computed over the sample of households
present at follow-up only. Results are available upon request.
152 A.C. Dammert et al. / Journal of Health Economics 35 (2014) 147–161

Table 1
Message treatment (examples).

Frame Messages

Water Window screens Water-holding solid Symptoms (same for

waste all)

Monetary Gain You will not spend
money eradicating
dengue in your house if
you cover the water
used for personal
consumption
Loss You will spend money
eradicating dengue in
your house if you don’t
cover the water used
for personal
consumption
Use window screens to
prevent dengue in your
house. You will not
spend money in
treatment!
If you don’t use
window screens,
dengue will spread in
your house and you
will spend money in
treatment!
You will not spend Go to the nearest
money in dengue health center if you or
treatment if you a family member has
discard old tires, fever, headache, pain
broken water below the eyes, or a
reservoirs, and bottles rash. It could be
from your yard dengue!
You will spend money Go to the nearest
in dengue treatment if health center if you or
you don’t discard old a family member has
tires, broken water fever, headache, pain
reservoirs, and bottles below the eyes, or a
from your yard rash. It could be
dengue!

Non-monetary Gain You will not get
dengue in your house if
you cover the water
used for personal
consumption
Loss You will get dengue in
your house if you don’t
cover the water used
for personal
consumption
Use window screens to
prevent dengue in your
house!
You will get dengue in
your house if you don’t
use window screens!
You will not get Go to the nearest
dengue in your house if health center if you or
you discard old tires, a family member has
broken water fever, headache, pain
reservoirs, and bottles below the eyes, or a
from your yard rash. It could be
dengue!
You will get dengue in Go to the nearest
your house if you don’t health center if you or
discard old tires, a family member has
broken water fever, headache, pain
reservoirs, and bottles below the eyes, or a
from your yard rash. It could be
dengue!

poverty index, and dengue index. Standard errors are clustered at
the locality level to account for random assignment across local-
ities. We also report the p-value from a test of equality of the
treatment indicator for the different groups.
Since we have multiple measures of dengue preventive behavior
in the survey, following Kling et al. (2007), we created a summary
measure (Y*) defined as the unweighted average of the different
indicators as follows:
K vention.
1 . YK − µ K
Y∗= oK
k k=1
where each indicator k is standardized by the mean and variance
of the control group at baseline. Thus, the magnitude of the esti-
mated index shows where the mean of the treatment group is in
the distribution of the control group in terms of standard deviation
units. This allows us to test whether the treatment had an overall
positive or negative ITT effect.
Panel A of Table 3 shows that households receiving a text mes-
sage reported a change in their health preventive behavior. Looking
at the overall index, the average impact is a statistically signif-
icant 0.12 standard deviation of the control group. Households
reported an increase in the probability of covering water reservoirs
by 3.4 percentage points (column 2), a change that represents a 4.6%
improvement relative to the control group. Moreover, households
receiving a text message reported an increase in the frequency in
which they clean the water reservoirs by 8.4 percentage points (col-
umn 3) and in the consumption of safe water (either boiling it or
adding chlorine) by 3.3 percentage points. These numbers repre-
sent a 9.6% and 3.8% improvement relative to the control group,
respectively.
Households exposed to the preventive information reported an
increase in the usage of screening in the windows and/or mosquito
bednets of about 4.5 percentage points, or a 4.5% improvement
relative to the control group and a reduction of 2.6 percentage
points in solid waste (tires or bottles) left outside the house.
Furthermore, there is no effect of exposure to preventive informa-
tion on the usage of chemical larvicida (Abate) to kill the larvae
in the container. The information sent by text message did not
emphasize the usage of Abate given that the Ministry of Health
provides Abate free of cost to all households in the area of inter-
Our findings are related to Cairncross et al. (2005) and Luby et al.
(2004), who find a positive effect of information campaigns con-
ducted over many months on hygiene. In addition, our results are
related to recent studies that focus on preventive e-health interven-
tions in developing countries. Pop-Eleches et al. (2011) show that
participants exposed to weekly text message reminders increase
their adherence to HIV antiretroviral treatment (of at least 90%)
by 13–16% relative to the control group. Similarly, Lester et al.
(2010) show an increase in HIV antiretroviral treatment in patients
exposed to weekly text messages inquiring about their health sta-
tus and reminders about the availability of phone-based support in
Kenya.
With regard to framing effects, we do not find statistically sig-
nificant differences between groups, except for one outcome. As
Panel B of Table 3 shows, the monetary frame is equally effec-
tive as the non-monetary frame. We find statistically significant
effects of the non-monetary message compared to the control
group when the change in behavior does not involve a direct cost
incurred by the household. For example, households exposed to
non-monetary messages report an increase in the frequency in
which they clean the water reservoirs by 9.6 percentage points,
consumption of safe water by 4.1 percentage points, and having less
water-holding solid waste outside the household by −3.6 percent-
age 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 2
Descriptive statistics – 2009 baseline health survey.

Treatment groups Control Difference p-Value p-Value of joint

group (T − C) F-test
[T1 = T2 = T3 = T4]

Any Monetary Monetary Non- Non-

gain (T1) loss (T2) monetary monetary
gain (T3) loss (T4)

Cement wall 0.825 0.847 0.809 0.869 0.776 0.796 0.029 0.544 0.481
Electricity 0.988 0.994 0.969 0.998 0.990 0.979 0.009 0.331 0.499
Radio 0.963 0.963 0.974 0.967 0.950 0.968 −0.004 0.687 0.630
TV 0.975 0.977 0.977 0.970 0.976 0.980 −0.005 0.573 0.965
Refrigerator 0.593 0.530 0.589 0.659 0.594 0.602 −0.010 0.837 0.452
Computer 0.196 0.150 0.247 0.249 0.139 0.214 −0.018 0.565 0.019
Piped water 0.914 0.910 0.915 0.947 0.885 0.908 0.006 0.827 0.419
Piped sewage 0.943 0.962 0.948 0.969 0.894 0.914 0.029 0.346 0.648
Wealth index 0.677 0.675 0.695 0.712 0.627 0.685 −0.008 0.828 0.426
Dengue index 0.372 0.346 0.347 0.405 0.389 0.356 0.016 0.710 0.791
Head: age 50.334 49.534 50.987 51.522 48.875 49.307 1.027 0.366 0.589
Head: male 0.736 0.761 0.738 0.732 0.713 0.773 −0.037 0.235 0.837
Head has health 0.584 0.602 0.581 0.608 0.541 0.539 0.045 0.247 0.634
insurance
Head: years of 9.486 9.828 9.291 9.398 9.488 9.787 −0.301 0.394 0.832
completed
schooling
Head: level of education
No schooling 0.025 0.032 0.029 0.025 0.014 0.019 0.006 0.444 0.437
Primary 0.296 0.249 0.291 0.326 0.314 0.267 0.029 0.360 0.498
Secondary 0.444 0.476 0.476 0.403 0.422 0.477 −0.032 0.274 0.126
Post-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 0.014
Dengue is 0.888 0.913 0.885 0.854 0.909 0.903 −0.015 0.424 0.240
transmitted by a
mosquito bite
Dengue could be avoided by:
Getting
vaccinated 0.032 0.010 0.039 0.047 0.027 0.021 0.011 0.212 0.026
Using abate in water 0.213 0.184 0.259 0.221 0.176 0.232 −0.019 0.738 0.683
reservoirs
A member of the 0.239 0.208 0.238 0.249 0.258 0.251 −0.012 0.687 0.483
household was
diagnosed with
dengue during the
past two years
By a doctor/nurse 0.641 0.652 0.688 0.583 0.640 0.632 0.009 0.858 0.316
By family member/self 0.139 0.184 0.112 0.140 0.137 0.198 −0.057 0.317 0.731
diagnosed
Pharmacist 0.067 0.046 0.047 0.088 0.085 0.044 0.023 0.359 0.578
Number of days ill 6.924 7.045 7.013 6.724 6.949 6.837 0.088 0.754 0.558
Monthly money 20.843 22.972 21.113 19.047 20.740 20.994 −0.151 0.923 0.143
spend on pre-paid
cell phone cards
(soles): head
Spouse 16.850 18.463 16.438 15.914 16.455 17.483 −0.634 0.499 0.399
Don’t know how to 0.053 0.045 0.047 0.058 0.061 0.067 −0.014 0.476 0.913
send/read a text
message: head
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 the
means 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.

that the gain frame is equally effective as the loss frame except
for cleaning water reservoirs daily or weekly. Households receiv-
ing loss messages experienced a statistically significant increase in
the probability of cleaning the water reservoirs by 14 percentage
points. From a health policy standpoint, these results suggest that
SMS framing for health preventive behavior is not as important as
providing the information itself. This lack of statistical significance
for framing effects is consistent with the evidence found in Dupas
(2009).
We are also interested in learning whether there are het-
erogeneous effects, since households’ investment decisions on
preventive health care depend on factors such as access to financial
markets and ex ante information on the burden associated with
the specific disease. We then consider variation in the effects of
text messaging as a function of observable characteristics through
the interaction of the treatment indicator with baseline indicators.
First, households are classified in three categories: high if their
baseline 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 3
Effects of text messaging.

Overall Cover water Clean water Drink boiled Change abate Use mosquito Have
standardized reservoirs reservoirs daily water or treat every 3 months nets for water-holding
index or weekly water with sleeping and/or solid waste
chlorine screens in (tires, bottles,
windows or others)
(1) (2) (3) (4) (5) (6) (7)

Panel A
Any 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 B
Monetary 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 0.829 0.597 0.576 0.358 0.228 0.917 0.216
[M = NM]

Panel C
Gain 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 0.380 0.260 0.008 0.852 0.119 0.129 0.654
[G = L]
Mean control 0.927 0.639 0.895 0.685 0.734 0.065
group
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%.
*** 1%.

distribution (richest), medium if their index belongs to the 2nd and estimates between households who had experienced dengue
3rd quartiles, and low if their index belongs to the lowest quartile before and those who did not.
(poorest) (Panel A of Table 4). Panel A shows that a higher frac-
tion of low assets index households (poorest) exposed to the text 4.2. Impacts on dengue indicators
messages report treating their water for consumption, having less
solid waste outside the household, and using mosquito nets and/or In order to compare the objective measures of dengue risk
screens in windows compared to medium- and high-wealth index transmission with self-reported measures that might suffer from
households. On the contrary, the results show no differential effects
non-random measurement error, a week after the follow-up, sur-
of messages on the probability of covering water reservoirs. vey health personnel from the Ministry of Health collected larvae
Moreover, we analyzed whether the gender of the cell phone
samples by inspecting water containers and other water-holding
owner had a differential effect on the outcomes of interest, since
solid waste (e.g., tires, bottles or other small water reservoirs) in
women are usually the ones responsible for the activities related to
each household of our sample.7 Households were not aware of this
dengue prevention inside the house and they tend to invest more
visit.
in goods improving child’s well-being and health (Thomas, 1990).
Following WHO (2009), containers were examined for the pres-
At baseline, the proportion of male cell ownership is similar across
ence of Aedes larvae and pupae. Based on this information, we
groups, 44% in the treatment group and 46% in the control group,
created three commonly used indices of the abundance of Aedes
and the difference is not statistically significant different from zero
mosquitoes: (i) the House Index, defined as the percentage of
(p-value = 0.468).6 As Panel B of Table 4 shows, we do not find evi- houses testing positive for Aedes larvae; (ii) the Container Index,
dence of gender differential effects at follow-up. Note, however, defined as the percentage of water-holding containers testing pos-
that we did not randomly select the head of household or spouse itive for Aedes larvae; and (iii) the Breteau Index, defined as the
for participation in the program to examine whether the adoption
number of positive water-holding containers per household. Each
of preventive action varied by gender.
index was standardized by the mean and variance of the control
Finally, one might wonder if households are more responsive to
group.
SMS preventive health information if a family member had been
As Table 5 shows, our objective measures of dengue infestation
diagnosed with dengue in the past. For instance, Dupas’ (2011b)
show significant point estimates of exposure to preventive infor-
meta-analysis shows that provision of information is less effec-
mation on the Container Index and the Breteau Index (Panel A).
tive in influencing people’s preventive behavior when they are well
The Container Index shows that households in the treatment group
informed about the health situation they face. As Panel C of Table 4
shows, 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 head
of household and the spouse owned a cell phone, (95 cases, or 5% of the sample).
7
In addition, our previous self-reported estimates could be affected by the
Hawthorne effect, as study participants having received messages stressing preven-
tion measures might be more likely to report having taken those measures, even if
not 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 4
Heterogeneous effects of text messaging.

Overall Cover water Clean water Drink boiled water Use mosquito nets Have
standardized reservoirs reservoirs daily or or treat water with for sleeping and/or water-holding
index weekly chlorine screens in windows solid waste (tires,
bottles, or others)
(1) (2) (3) (4) (5) (6)

Panel A: wealth index

T∗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 0.0280 0.1326 0.2539 0.0633 0.0677 0.0777
T∗Low = T∗Med = T∗High

Panel B: gender of the cellphone owner

T 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 dengue

T 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 and
dengue index. Observations are clustered at the locality level. Robust standard errors in parenthesis.
*
10%.
**
5%.

Table 5 experienced a decrease of 0.108 standard deviations in the percent-

Dengue risk transmission indices.
age of water-holding containers testing positive for dengue, while
Standardized Standardized Standardized the Breteau Index shows that households in treatment localities
Container House Breteau experienced a decrease in the number of positive containers per
Index Index Index
household by 0.098 standard deviations of the mean of the control
(1) (2) (3)
group. We do not find, however, statistically significant effects on
Panel A the Standardized House Index. These results imply that households
Any message −0.108** −0.045 −0.098**
(0.046) (0.064) (0.047)
in the treatment group experienced a decreased in the percent-
age of water-holding containers testing positive for dengue larvae
Panel B
(1.44% vs. 2.47% in the treatment and control groups) as well as the
Monetary −0.106** −0.047 −0.089
(0.051) (0.079) (0.054)
number of positive water-holding containers per household (10.66
Non-monetary −0.109** −0.044 −0.107** vs. 18.91 in the treatment and control groups). These effects were
(0.049) (0.069) (0.048) not big enough, however, to change the percentage of houses with
Panel C at least one positive water-holding container.
Gain −0.081* 0.006 −0.073 It is important to note that, even though there is a corre-
(0.052) (0.076) (0.052) lation between the critical levels of the mosquito breeding and
Loss −0.133** −0.093 −0.121** the intensity of virus transmission, the interpretation of the var-
(0.049) (0.073) (0.050)
ious indices in relation to human infection and disease can be
Panel D difficult to assess (Focks, 2004). For example, larvae might be clus-
Monetary gain −0.047 0.050 −0.041
tered in a small number of containers so that the Container Index
(0.062) (0.104) (0.068)
Monetary loss −0.152** −0.122 −0.127**
will be low, although the number of infectious bites per person
(0.053) (0.093) (0.057) may be high. The relationship between dengue risk measures and
Non-monetary gain −0.112** −0.036 −0.102* dengue infestation gets more complicated when one realizes that
(0.056) (0.089) (0.053)
the latter is affected by mosquito longevity, the immunological
Non-monetary loss −0.106* −0.054 −0.113**
(0.054) (0.080) (0.052)
status of the human population, and temperature (Silver, 2008).
International evidence, for instance, suggests that the majority of
Notes: ***1%. N = 1496 households. Controls in all regressions include head of house-
dengue outbreaks occurred in localities with low Aedes indices, and
hold characteristics at baseline (age, gender, schooling, employment, and health
insurance), assets index and dengue index. Observations are clustered at the locality more importantly, even when there is some statistical relationship
level. Robust standard errors in parenthesis. between the indices, these were inconsistent across years within
*
10%. the same localities (Shah et al., 2012). Indeed, it is widely argued
**
5%. the need for better methods to quantify the relationship between
dengue risk and dengue endemicity to ensure better future
156 A.C. Dammert et al. / Journal of Health Economics 35 (2014) 147–161

vaccination and control strategies (Anders and Hay, 2012). Table 6

Effects of text messages on dengue symptoms and diarrhea.
Nonetheless, the Pan American Health Organization defines three
levels of risk for dengue transmission as low (HI < 0.1%), medium Had a member of the Had a member of the
(HI 0.1–5%), and high (HI > 5%) (MINSA 2001). Our data estimate household show household had
dengue symptoms diarrhea during the
that the House Index was 8.15% in the treatment group and 9.26%
(fever, headache, eye previous month?
in the control group at follow-up. In both areas, the level of risk pain) since November (2)
transmission was high.8 of 2009?
The follow-up survey also collected information on dengue inci- (1)
dence and illness of household members during the treatment Panel A
period. In our sample, less than 1% of households self-reported Any message −0.031* −0.021
having a member clinically diagnosed with dengue at the time of (0.018) (0.019)

the follow-up survey.9 This low number may have been accrued Panel B
because the follow-up occurred too soon after the intervention Monetary −0.032* −0.053**
to capture a significant number of dengue episodes or because (0.019) (0.023)
Non-monetary −0.031 0.013
self-reported measures may be affected by the underreporting (0.020) (0.024)
of dengue. In fact, reporting an objective measure of ‘incidence p-Value of 0.9360 0.0167
of dengue’ is not simple. First, only health officials can provide [M = NM]
information of dengue incidence after performing laboratory tests Panel C
on individuals. Routine checkups or self-observation of symp- Gain −0.029 −0.021
toms is not enough for dengue diagnosis. Moreover, even if this (0.021) (0.023)
*
information would be available from public health centers, it will Loss −0.033 −0.020
(0.019) (0.025)
be severely affected by undercoverage, as it is a common prac- p-Value of [G = L] 0.8670 0.9831
tice among many individuals affected by this virus to be treated Mean control 0.066 0.159
by means of domestic health practices at home (Beatty et al., group
2011).10 N 1730 1754
Given that undercoverage and misreported error may bias self- Note: ***1%. Controls in all regressions include head of household characteristics at
reported measures of dengue incidence, as an alternative mea- baseline (age, gender, schooling, employment, and health insurance), assets index
sure, we also recorded whether a household member had dengue and dengue index. Observations are clustered at the locality level. Robust standard
errors in parenthesis.
symptoms (fever, headaches, eye pain, and diarrhea) during the *
10%.
treatment period. At follow-up, 3.4% of households in the treat- **
5%.
ment group reported having a household member with dengue
symptoms, while 6.6% in the control group did so. Column 1 of
4.3. Spillover effects
Table 6 shows that the difference of 3 percentage points is statisti-
cally significant at the 10 percent level. Yet it is worth mentioning
Several health studies have shown significant spillover effects.
that most of the dengue symptoms can be easily confused with
Depending on the characteristics of the preventive product, these
other common health problems, which makes difficult to connect
effects can either increase adoption of preventive measures (e.g.,
symptoms or dengue risk measures to morbidity and dengue infes-
bednets as discussed in Dupas, 2013) or decrease it (e.g., deworm-
tation.
ing drugs as discussed in Miguel and Kremer, 2004). In our setting,
We also analyzed whether any member of the household expe-
there are possible channels through which externalities may occur:
rienced diarrhea during the month prior to the follow-up. Since
people may share the information they got from the text message
households report cleaning their water reservoirs and treating
or people who change their preventive behavior may discuss the
water for consumption, one might wonder, whether these changes
benefits with friends and family. A standard approach to assess
in behavior are big enough to have an effect on preventable water-
individual externalities would be to rely on the information on
related diseases. Column 2 of Table 6 shows that households
the number and type of social links for both participant and non-
exposed to text messages self-report a decline in the incidence
participant households, as in Tontarawongsa et al. (2011) when
of diarrhea by 2 percentage points, which translates into a 13%
studying bednet adoption in India.
improvement relative to the control group (albeit not statistically
In terms of mosquito density, residential proximity is also
significant). Looking into the different groups, as Panel B of Table 6
important since increasing uptake of preventive measures among
shows, the reduction in the incidence of diarrhea is higher and sta- those exposed to the repetitive information may change the
tistically significant for households receiving monetary messages mosquito vector population in the area under study. We follow this
(−5.3 percentage points). line of inquiry by providing evidence on spatial spillovers using GPS
location to examine the extent to which outcomes are influenced
by treatment density (Dupas, 2013; Tarozzi et al., 2013).11
We conducted a complete mapping of the study area and
recorded longitude and latitude coordinates for 1754 households
8
Malaria endemicity is commonly measured by the proportion of a population
(households with complete address information and that are
with detectable malaria parasite in their blood. As Anders and Hay (2012) showed, present at both baseline and follow-up). Based on the GIS coor-
however, no comparable measure exists to quantify the endemic level of dengue dinates, we constructed measures of treatment density within a
where most epidemiological data rely on clinical reports. given radius of our sample household. On average, households have
9
Reports from the Ministry of Health also show low levels of clinical dengue cases
in the area during the same months in 2010.
10
An alternative explanation of the low dengue infection prevalence in our sample
11
population could be due to the suppression of the mosquito vector population in the It is important to note that, in our paper, as in Tarozzi et al. (2013), there is no
area under study. This might not be the case, however, given that Talara reported treatment variation within localities. This is different from Dupas (2013), where the
about 52% of the total number of clinical cases in the region during the summer author is able to exploit the exogenous variation given by the randomization of the
months of 2013. fraction of beneficiary households within each locality.
 A.C. Dammert et al. / Journal of Health Economics 35 (2014) 147–161 157

Table 7
Testing for spillover effects.

Cover water Clean water Drink boiled Use mosquito Have water- Standardized Standardized Standardized
reservoirs reservoirs water or nets for holding solid Container House Index Breteau
daily or treat water sleeping waste (tires, Index Index
weekly with chlorine and/or bottles, or
screens in others)
windows
(1) (2) (3) (4) (5) (6) (7) (8)

Within 50 m of household
Number of 0.000 −0.006 0.003 −0.004 0.002 0.008 0.000 −0.010
households
(0.003) (0.006) (0.004) (0.006) (0.003) (0.011) (0.014) (0.010)
Number of 0.006 0.007 0.001 0.008 0.003 0.028 −0.017 0.008
treated
households
(0.006) (0.013) (0.008) (0.012) (0.006) (0.024) (0.030) (0.022)
Treatment∗# of −0.004 0.006 0.003 −0.007 −0.002 −0.043* 0.018 −0.003
treated
households
(0.006) (0.013) (0.008) (0.013) (0.006) (0.025) (0.031) (0.023)

Within 100 m of household

Number of −0.002 −0.005 0.003 −0.002 0.003** 0.004 0.001 0.005
households
(0.001) (0.003) (0.002) (0.003) (0.001) (0.005) (0.007) (0.005)
Number of 0.003 0.011** −0.006* −0.004 −0.003 0.016* 0.015 0.004
treated
households
(0.002) (0.005) (0.003) (0.004) (0.002) (0.009) (0.011) (0.008)
Treatment∗# of −0.001 −0.003 0.004 0.005 0.001 −0.023** −0.017* −0.010
treated
households
(0.002) (0.005) (0.003) (0.004) (0.002) (0.008) (0.011) (0.008)

Within 150 m of household

Number of −0.002** −0.001 0.003** −0.001 0.003** 0.002 −0.002 0.003
households
(0.001) (0.002) (0.001) (0.002) (0.001) (0.004) (0.005) (0.003)
Number of 0.004** 0.004 −0.003 −0.002 −0.004** 0.008 0.011 0.002
treated
households
(0.002) (0.003) (0.002) (0.003) (0.002) (0.006) (0.007) (0.005)
Treatment∗# of −0.001 −0.002 0.000 0.003 0.001 −0.012** −0.007 −0.006
treated
households
(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 health
insurance) Robust standard errors in parenthesis.
*
10%.
**
5%.

3.8 neighbors within a 50-meter radius (10.3 neighbors within 50–
100 m radius, and 12.5 within 100–150 m radius) who were
exposed to text messages. We also constructed a measure of total
sample population density within a given radius. Table 7 presents
the coefficient estimates controlling for locality fixed effects. We
focus on a radius of less than 150 meters, since epidemiological
studies of dengue have shown that the Aedes mosquito has a dis-
persal of approximately 100 m (Anders and Hay, 2012).12
As Table 7 shows, the coefficient estimates on the interaction
terms between the treatment group and treatment intensity are not
statistically significantly different from zero for all self-reported
outcomes (columns 1–5). These results suggest that households in
12
We have assessed the presence of externalities by testing whether the number
of control households living within a 150 m radius of treatment localities experi-
enced changes in behavior and health outcomes. The point estimates, however, are
imprecisely estimated. We thank one of the referees for pointing this out.
treatment and control localities experienced similar effects from
proximity to treated households. Columns 6–8 show the objective
measures of dengue risk transmission; however, the results suggest
some positive spillover effects. If we increase the number of treated
households within 50 m, 100 m, or 150 m, households in the treat-
ment group experienced a decrease in the standardized Container
Index (column 6). As expected, this effect declines as distance from
the index household increases.
Why are no spatial spillovers observed in the self-reported data?
It might be possible that the follow-up occurred too soon after the
intervention and there was not enough time for the information to
travel in the neighborhood. An alternative explanation is that infor-
mation on preventive behavior may have been spread out, but its
benefits are not immediate; thus, people in endemic dengue areas
are waiting for the peak of the dengue season to change their behav-
ior and/or for the Ministry of Health to spray the area once a certain
number of clinical episodes are observed. A study conducted over
a longer time horizon and identifying social networks may be able
158 A.C. Dammert et al. / Journal of Health Economics 35 (2014) 147–161

to identify some of these effects more accurately. Moreover, vari- Table A1

Urban Pariñas – Ministry of Health Division.
ation in treatment density may be correlated with unobservable
characteristics of observed location characteristics. Sector Localities

1 J.CHAVEZ,M.SULLON, A.H.ALGARROBOS, AH.VENECEDORES,LAS

5. Conclusions GARDENIAS Y PILAR NORES, MARIO AGUIRRE
2 QUIÑONES,9OCTUBRE,SANSEBASTIAN,STA.ROSA,B.AIRES,S.CERRO,
J.BASADRE
The dengue burden is high for developing countries that face 3 7JUNIO,A.GARCIA,L.A.SANCHEZ,M.CHAVEZ,A.FUJIMORI,LOS
tight health budget constraints. Several countries in Latin American GERANIOS,AH.LOS ROBLES,AH.MIRAFLORES,BELLO
and the Caribbean, including Peru, have implemented strategies to HORIZONTE,H.CARLIN,2FEBRERO,LOS FICUS,JOSE.C.MARIATEGUI
eradicate the disease, including indoor residual spraying and the 4 TALARA ALTA,LAS PALMERAS,URB.LOS ROBLES,CRISTO REY CRUZ
DE MAYO,02 MAYO,URB.MARIA AUXILIADORA
distribution of temephos (distributed as Abate) to kill the larvae in 5 AV. F PAR, URB. LOS VENCEDORES, JAMES STORE, SUDAMERICA,
water containers. Dengue infections, however, continue to rise over ENAPU, V.TALARA, L.ALVA CASTRO, M.CABREDO, LUCIANO
the last years (PAHO, 2007). A key stylized fact on health behav- CASTILLO.
ior in developing countries is that households invest little in the 6 APROVISER,FONAVI,LAS MERCEDES
7 URB. LOS PINOS. URBA, VILLA FAP CORPAC
preventive health course. A plausible explanation for this behavior
8 PARQUES DEL 43–57, AV. H PAR; AV. E, F IMPAR Y G.
is the lack of information on illness prevention. Our intervention 9 PARQUE DEL 35–42 y 58–62; AV. H IMPAR AV. D E y G.
provides evidence on the link between information campaigns and 10 PARQUES DEL 63–72, LA PARADA, AV. E, AV. A; URB LOS JAZMINES,
preventive behavior. CENTRO CIVICO Y MERCADO CENTRAL.
11 PARQUE DEL 8– 22, AV A y B, URB. A.TABOADA
We randomly assigned over 2000 households in northern Peru
12 PARQUES 1–7 y 22–27 AV. A, B y C; PARQUE 78; UNIDAD VECINAL;
to information treatment and control groups to evaluate the effec- CALLE 4
tiveness of providing repetitive information via mobile phone 13 PARQUE 28–34 AV. B Y BARRIO PARTICULAR, BLOCK MILITAR,
technologies on enhancing households’ health behavior. House- BLOCK PROFESORES Y URB. STA.ROSA
holds in the treatment group received more than 30 tailored text 14 URB. MUNICIPAL, L.VILLANUEVA, SAN JUDAS TADEO, J.OLAYA, B.EL
PESCADOR, SAN MARTIN, LAS MERCEDES, CEMENTERIO, TALLERES
messages during a three-month period designed to increase the
UNIDOS, MUELLE, B.ESPECIALISTAS
adoption of preventive measures against dengue. The main results 15 SAN PEDRO, JESUS MARIA, STA. RITA, B. ESTIBADORES, LAS
suggest that exposure to repeated health preventive information PEÑITAS.
affected household behavior. Overall, the Breteau Index, an objec- 16 PT. ARENAS
17 SACOBSA
tive measure of dengue infestation, shows that households that
18 NEGREIROS
are repeatedly exposed to preventive information experienced a 19 ENACE
decrease of −0.10 standard deviations of the mean of the control
group, which is explained by positive changes in household behav-
ior and is translated into a reduction of self-reported incidence of
dengue symptoms. Given that mobile phone service offers a large
pool of individual instant and inexpensive access to information, results. From a policy standpoint, this result suggests that the fram-
and since it costs US$10 to US$25 to treat one case of dengue (WHO, ing of SMS health messages is not as important as the provision of
2009), this health preventive care intervention can be considered the information itself. Finally, this study showed that the largest
cost-effective. impacts on preventive behavior against dengue were concentrated
Further evidence shows positive externalities for the objective among the poorest households, a pattern that could be related to
measures of risk transmission, as the number of water-holding con- the fact that these households in general are the least informed
tainers testing positive for dengue decreases in treatment areas as about the health situation they face.
the number of treated households increase. This result suggests These findings provide valuable evidence for other developing
that changes in household behavior may affect the density of the countries interested in incorporating digital technologies on health
mosquito vector, however further research is needed to analyze programs. Although text messages considered in this evaluation
further the spillover effects in the prevention of mosquito borne were related to dengue prevention, this cost-effective approach
diseases. Moreover, we do not find statistically significant framing has the potential for use in the prevention of other types of

Table A2
Census 2007.

Baseline survey (2009) Urban Pariñas: Census 2007

All Cell-phone household Non-cell phone household

Cement wall 0.817 0.751 0.811 0.696

Electricity 0.986 0.873 0.921 0.827
Piped water 0.917 0.765 0.821 0.705
Piped sewage 0.933 0.767 0.820 0.715
Refrigerator 0.598 0.422 0.537 0.277
Computer 0.211 0.142 0.220 0.045
Head: male 0.749 0.740 0.762 0.712
Head: level of completed schooling
No schooling 2.53 2.66 1.83 3.71
Primary 29.29 27.80 21.40 35.84
Secondary 45.68 37.59 37.13 38.18
Post-secondary 22.50 31.95 39.64 22.27
Number of households 2021 20772 11570 9202

Source: 2007 Census (www.inei.gob.pe).

Note: Ownership of radio and TV are not comparable with those in our data. The Census questionnaire asks for color TV only while our questionnaire asks for ownership of
any TV (color or black and white TV). Likewise, our questionnaire asks for a radio while the Census asks for radio separately from ownership of any music equipment.
 A.C. Dammert et al. / Journal of Health Economics 35 (2014) 147–161 159

diseases or risky behaviors. In addition, the findings highlight the

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

Note: Test of equality is based on regression of attrition on treatment group with clustered standard errors at the locality level. N = 2021 including 1350 households in treatment group localities and 671 in control group
importance of considering the dynamic nature of the information
to alter health behavior where the repeated dissemination of infor-
mation is important. Reminders might mitigate attentional failure,

0.3873

0.2560

0.0866
0.0330
change intertemporal allocations, and improve consumer welfare

(9)
by providing associations between future opportunities and today’s
choices (Karlan et al., 2010).
Indeed, booming mobile phone connectivity in developing

Difference T − C

−0.025 [0.174]

−0.018 [0.134]

−0.004 [0.759]
−0.003 [0.666]
countries is shifting policy attention focus and increasing the num-
ber of initiatives aimed at empowering vulnerable groups with

[p-value]
access to tailored content-value information delivered to them
through a new technological platform. We believe that, in contexts

(8)
of poverty, weak institutional structures, and imperfect markets
where information is often costly or incomplete, mobile phone
applications that provide repetitive information to people about

0.137 (0.092)

0.055 (0.063)

0.058 (0.062)
0.024 (0.037)
risky behaviors could have non-negligible effects that need to be
unveiled.

Control

(7)
Appendix A.

Non-monetary
See Table A1.

0.135 (0.087)

0.041 (0.059)

0.082 (0.071)
0.009 (0.020)
loss (T4)
Appendix B. Census data

(6)
The 2007 Peruvian Census provides online information for dif-
ferent economic and demographic variables at the district level.

Non-monetary

0.040 (0.055)

0.033 (0.038)
0.105 (0.060)

0.033 (0.040)
Table A2 presents the descriptive statistics of all localities in the

gain (T3)
district of Pariñas, disaggregated by whether the households had
a mobile phone or not. The first column corresponds to the base-

(5)
line information presented in Table A2. First, we investigate how
our sample is similar to cell phone households in the same district.
As the table shows, 56% of households in Pariñas had at least one
Monetary loss

mobile phone in 2007. By comparing the first and third columns, 0.086 (0.073)

0.020 (0.033)

0.060 (0.072)
0.008 (0.015)
the results show that our sample looks similar to the average cell
phone household in Pariñas.
(T2)

Second, we analyze how different are cell-phone households

(4)

from non-cell phone households in Pariñas. Comparing demo-

graphic and household characteristics across columns, we can see
that cell phone households are more educated and have better
Monetary gain

0.124 (0.113)

0.048 (0.056)

0.043 (0.065)
0.034 (0.048)

household infrastructure than non-cell phone households have.

This is not surprising given that the cost of handsets and services
are not affordable to all households. Thus, we cannot general-
(T1)
(3)

ize the findings to all households in urban areas. Nevertheless,

the empirical findings suggest the potential relevance of ICTs and
different messages on household behavior for cell phone users,
who are an important fraction of Peruvian households. Moreover,
0.112 (0.087)

0.037 (0.052)

0.054 (0.066)
0.021 (0.035)

there are potential positive spillovers accrued from the reduction

Treatment

of the Aedes mosquito in cell phone households as Table 7 sug-

gest.
(2)

localities. Standard deviations in parenthesis.

Appendix C. Attrition
0.121 (0.089)

0.044 (0.056)

0.056 (0.064)
0.022 (0.035)

Table A3 presents the attrition that occurred between the base-

line and follow-up surveys by treatment status. Attrition was 12%
(1)
All

for the entire sample, 11.2% in the treatment group and 13.7% in
the control group. As column 8 shows, the difference is not sta-
tistically significant from zero. Likewise, after disaggregating the
rate of attrition across three different sources (respondent not
Respondent
not present
Attrition Rate
Attrition rates.

Migration
localities)

present, migration, and refusal) we observe statistical similarity

(n = 100

Refusal

between the treatment and control groups. The test statistic does
Table A3

not reject the null hypotheses of equality of means across all three
categories.
160 A.C. Dammert et al. / Journal of Health Economics 35 (2014) 147–161
Looking into across the various treatment groups, columns 3–6
show that there are no statistically significant differences across
the different treatment and control groups except for the percent-
age of refused interviews. The gain group has a 1-percentage point
difference in the percentage of refused interviews compared to the
control group, while the loss group has a −1.5 percentage point dif-
ference with the control group. The overall difference, however, is
small and not statistically significant due to the size of the standard
deviations.
References
Aker, J., 2010. Information markets near and far: mobile phones and agricultural
markets in Niger. American Economic Journal: Applied Economics 2 (3), 46–59.
Aker, J., Mbiti, I., 2010. Mobile phones and economic development in Africa. Journal
of Economic Perspectives 24 (3), 207–232.
Anders, 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.
Anderson, K., Chunsuttiwat, S., Nisalak, A., Mammen, P., Libraty, D., Rothman, A.,
Green, S., Vaughn, D., Ennis, F., Endyl, T., 2007. Burden of symptomatic dengue
infection in children at primary school in Thailand: a prospective study. The
Lancet 369 (9571), 1452–1459.
Beatty, 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’s
assessment. The American Journal of Tropical Medicine and Hygiene 84 (3),
473–488.
Bertrand, M., Karlan, D., Mullainathan, S., Shafir, E., Zinman, J., 2010. What’s
advertising content worth? Evidence from a consumer credit marketing field
experiment. The Quarterly Journal of Economics 125 (1), 263–305.
Blaya, J., Fraser, H., Holt, B., 2010. E-health technologies show promise in developing
countries. Health Affairs 29 (2), 244–251.
Cairncross, S., Shordt, K., Zacharia, S., Govindan, B., 2005. What causes sustainable
changes in hygiene behavior? A cross-sectional study from Kerala, India. Social
Science and Medicine 61 (10), 2212–2220.
Chong, A., 2011. Development Connections: Unveiling the Impacts of New Informa-
tion Technologies. Palgrave MacMillan, Washington, D.C.
Chong, A., Karlan, D., Shapiro, J., Zinman, J., 2013. (Ineffective) messages to encour-
age recycling: evidence from a randomized evaluation in Peru. World Bank
Economic (forthcoming).
Dupas, P., 2009. What matters (and what does not) in households’ decision to invest
in malaria prevention? American Economic Review Papers and Proceedings 99
(2), 224–230.
Dupas, P., 2011a. Do teenagers respond to HIV risk information? Evidence from a
field experiment in Kenya. American Economic Journal: Applied Economics 3
(1), 1–34.
Dupas, P., 2011b. Health behavior in developing countries. Annual Review of Eco-
nomics 3, 425–449.
Dupas, P., 2013. Short-Run Subsidies and Long-Run Adoption of New Health Prod-
ucts: Evidence from a Field Experiment, Mimeo.
Elder, J., Lloyd, L., 2006. Achieving Behaviour Change for Dengue Control: Methods,
Scaling-Up, and Sustainability. In: Working Paper for the Scientific Working
Group on Dengue Research, Convened by the Special Programme for Research
and Training in Tropical Diseases, 1–5 October 2006, Geneva.
Espinoza-Gomez, F., Hernandez-Suarez, M., Coll-Cardenas, R., 2002. Educational
campaign versus malathion spraying for the control of Aedes aegypti
in Colima, Mexico. Journal of Epidemiology Community Health 56, 148–
152.
Filmer, D., Pritchett, L., 2001. Estimating wealth effects without expenditure data –
or tears: an application of educational enrolment in states of India. Demography
38 (1), 115–132.
Fjelsoe, B., Marshall, A., Miller, Y., 2009. Behavior change interventions delivered by
mobile phone short message service. American Journal of Preventive Medicine
36 (2), 165–173.
Franklin, V., Waller, A., Pagliari, C., Greene, S.A., 2006. A randomized controlled trial
of Sweet Talk, a text-messaging system to support young people with diabetes.
Diabetic Medicine 23, 1332–1338.
Focks, D., 2004. A Review of Entomological Sampling Methods and Indicators for
Dengue Vectors. WHO.
INEI, 2012. Las Tecnologías de Información y Comunicación en los Hogares. Informe
Técnico Nro 2. INEI, Lima.
Jalan, 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.
Jensen, 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.
Jensen, R., 2010. Information, efficiency, and welfare in agricultural markets. Agri-
cultural Economics 41 (1), 203–216.
Joo, 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 of
Telemedicine and Telecare 13, 416–420.
Kahneman, D., Tversky, A., 1979. Prospect theory: an analysis of decision under risk.
Econometrica 47 (2), 263–291.
Karlan, D., McConnel, M., Mullainathan, S., Zinman, J., 2010. Getting to the Top of
Mind: How Reminders Increase Saving, NBER Working Papers 16205. National
Bureau of Economic Research, Inc.
Kling, J., Liebman, J., Katz, L., 2007. Experimental analysis of neighborhood effects.
Econometrica 75 (1), 83–119.
Koshy, E., Car, J., Majeed, A., 2008. Effectiveness of mobile-phone short message
service (SMS) reminders for ophthalmology outpatient appointments. BMC
Ophthalmology 8, 9.
Lester, 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. Effects
of a mobile phone short message service on antiretroviral treatment adher-
ence in Kenya (WelTel Kenya1): a randomised trial. The Lancet 376 (9755),
1838–1845.
Levin, 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 Human
Decision Processes 76 (2), 149–188.
Luby, 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-risk
communities in Pakistan: a randomized controlled trial. Journal of the American
Medical Association 291 (21), 2547–2554.
Madajewicz, M., Pfaff, A., van Geen, A., Graziano, J., Hussein, I., Momotaj, H., Sylvi, R.,
Ahsan, H., 2007. Can information alone change behavior? Response to arsenic
contamination of groundwater in Bangladesh. Journal of Development Eco-
nomics 84 (2), 731–754.
Meyerowitz, B., Chaiken, S., 1987. The effect of message framing on breast self-
examination attitudes, intentions, and behavior. Journal of Personality and
Social Psychology 52 (3), 500–510.
Miguel, E., Kremer, M., 2004. Worms: identifying impacts on education and
health in the presence of treatment externalities. Econometrica 72 (1), 159–
217.
Mutu, M., Yamano, T., 2009. The impact of mobile phone coverage expansion on
market participation: panel data evidence from Uganda. World Development
37 (12), 1887–1896.
PAHO, 2004. Planning Social Mobilization and Communication for Dengue Fever
Prevention and Control. A Step-by-Step Guide. Pan-American Health Organiza-
tion.
PAHO, 2007. Health in the Americas. Pan-American Health Organization.
Patrick, 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: randomized
controlled trial. Journal of International Medical Research 11, 1.
Pop-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 message
reminders. AIDS 25 (6), 825–834.
Ramírez, G., 2011. Situación Epidemiológica de Dengue en el Perú y Definiciones de
Caso de Dengue Hemorrágico de la OMS. Reporte preparado para el Ministerio
de Salud, Lima.
Reis, R., 2006. Inattentive consumers. Journal of Monetary Economics 53 (8), 1761–
1800.
Rhee, M., Sissoko, M., Perry, S., McFarland, W., Parsonnet, J., Doumbo, O., 2005. Use
of Insecticide-Treated Nets (ITNs) following a malaria education intervention
in Piron, Mali: a control trial with systematic allocation of households. Malaria
Journal 4, 35.
Rodgers, 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. Tobacco
Control 14, 255–261.
Rothman, A., Salovey, P., 1997. Shaping perceptions to motivate healthy behavior:
the role of message framing. Psychological Bulletin 121 (1), 3–19.
Rothman, A., Bartels, R., Wlaschin, J., Salovey, P., 2006. The strategic use
of gain- and loss-framed messages to promote healthy behavior:
how theory can inform practice. Journal of Communication 56, S202–
S220.
Salovey, P., Williams-Piehota, P., 2004. Field experiments in social psychology:
message framing and the promotion of health protective behaviors. American
Behavioral Scientist 47, 488–505.
Shah, S., Sani, J., Hassan, M., Safian, M., Aizuddin, A., Hod, R., 2012. Relationships
between Aedes indices and dengue outbreaks in Selangor, Malaysia. Dengue
Bulletin 36, 166–174.
Silver, J., 2008. Sampling the larval population. In: Mosquito Ecology Field Sampling
Methods.
Tarozzi, A., Mahajan, A., Blackburn, B., Kopf, D., Krishnan, L., Yoong, J., 2013. Micro-
loans, Insecticide-Treated Bednets and Malaria: Evidence from a Randomized
Controlled Trial in Orissa (India), Mimeo.
Thomas, D., 1990. Intra-household resource allocation: an inferential approach.
Journal of Human Resources 25 (4), 635–664.
 A.C. Dammert et al. / Journal of Health Economics 35 (2014) 147–161 161

Tontarawongsa, C., Mahajan, A., Tarozzi, A., 2011. (Limited) Diffusion of Health-
protecting Behaviors: Evidence from Non-beneficiaries of a Public Health
Program in Orissa, India, Mimeo.
WHO, 2009. Dengue Guidelines for Diagnosis, Treatment, Prevention, and Control.
World Health Organization.
Zurovac, 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. The
Lancet 378 (9793), 795–803.