Behavioral Response to Information?
Circumcision, Information, and HIV Prevention
February 2012
Susan Godlonton, University of Michigan
Alister Munthali, University of Malawi
Rebecca Thornton1, University of Michigan
Abstract
Understanding behavioral responses to changes in actual or perceived risk is important particularly in
the context of public health interventions where risk reduction goals can be undermined by risk
compensating behavior. We build on the existing literature by examining the behavioral responses by
men to an information campaign in which they learn new information about HIV risk. In a sample of
approximately 900 circumcised and 300 uncircumcised men living in rural Malawi, we randomly
disseminated information about HIV risk and male circumcision and measure behavioral responses one
year later. We find no evidence of risk compensation among circumcised men as measured by sexual
behavior and condom purchases while uncircumcised men significantly decreased risky sexual behavior.
There was no significant impact of the information on adult male circumcisions.
JEL classification: C93, D81, D84, I15, I18, J13
1
Corresponding author: Thornton: Department of Economics, University of Michigan, 213 Lorch Hall, 611 Tappan St. Ann
Arbor MI, 48109-1220, USA; [email protected]; Godlonton: Department of Economics, University of Michigan, 611
Tappan St. Ann Arbor MI, 48109-1220, USA; [email protected]; Munthali: P.O. Box 278, Zomba, Malawi,
[email protected]
Funding for this study was provided by Michigan Center for Demography of Aging (MiCDA), OVPR and Rackham at the
University of Michigan as well as the Institute for Research on Women and Gender. We acknowledge the extensive
contributions of the field team including James Amani, Sheena Kayira, Collins Kwizombe, Denise Matthijsse, Ernest
Mlenga, and Christopher Nyirenda. We also thank assistance from Kondwani Chidziwisano, Jessica Kraft, Erica Marks, Julie
Moran, Jason Stanley, and Kondwani Tomoko. We thank helpful comments from seminar participants at the Development
Day Conference at University of Michigan, NEUDC, Population Association of America, University of Illinois, Chicago,
Princeton University, Yale University, University of Georgia, Georgia State University, DePaul, University of California
Berkeley, and the CSWEP CeMENT Workshop.
1
1 Introduction
Understanding behavioral responses to changes in actual or perceived risk is critical to optimal policy
design given the possibility of risk compensation. Starting with the seminal work of Peltzman (1975),
this area has been of growing interest to economists. Building on Peltzman’s work, much of the existing
literature has focused on the net effect of mandatory seat belt laws that alter the exposure to risk.2 In
addition to examining responses to seat belt laws and other regulations that affect personal risk,
researchers have recently begun examining responses to health related technological innovations such as
protective sports gear, sunscreen, cholesterol treatment, and vaccines.3 There is also a growing literature
examining behavioral responses to information about risk (Smith et al. 1995). In the last decade, with
the large sums of money being spent on HIV prevention, there is a growing interest and resulting
literature focused on risk reduction and risk compensating behaviors in response to various HIV
interventions or HIV risk; this literature has found mixed results.4 This paper builds on this literature by
examining the behavioral response to a health information campaign in which men learn that male
circumcision is (partially) protective against HIV transmission.
As background, despite the substantial effort devoted to HIV prevention, HIV/AIDS has
continued to spread (USAID, 2007). Recently, attention has been placed on male circumcision as a
2
There is an extensive literature examining the safety belt regulations and car safety technological innovations (Peltzman,
1975; Evans and Graham 1991, Cohen and Einav, 2003). Blomquist (1988) provides an extensive review.
3
Other examples of responses to levels of risk include: the use of protective gear in riskier sports (Walker, 2007; Braun and
Fouts, 1998; Williams-Avery, 1996). There is evidence suggesting that the overall risk of melanoma may be higher with the
use of sunscreen due to compensatory behavior by individuals in increasing their exposure to the sun (Autier et al., 1998;
Dickie and Gerking, 1997). Individuals who begin treatment for high cholesterol experience changes in the perceived
riskiness of eating poorly, which is then associated with an increase in BMI (Kaplan, 2010). There is also a debate to whether
women will engage in unprotected sex at an earlier age if they receive the HPV vaccination. (Lo, 2006; Zimet et al., 2006;
Kapoor, 2008)
4
Overall risk reduction has been found in studies looking at the adoption of tenovir (HIV prevention trial drug in use in
medical randomized control trials) where no change in sexual behavior was observed after receiving tenovir (Guest et al.,
2008); and with the roll-out of anti-retroviral therapy (a meta analysis is provided by Crepaz et al (2004). Evidence of risk
compensation has been observed with the roll-out of anti-retroviral therapy (Stolte et al.,2001; de Walque et al., 2010); and in
some condom use promotion campaigns (Kajubi et al., 2005).
2
potentially important biomedical HIV prevention strategy. Randomized control trials in three countries,
South Africa, Kenya, and Uganda, provided evidence that male circumcision is up to 60 percent
effective in reducing transmission risk (Auvert, 2005; Bailey et al., 2007; Gray et al., 2007; NIH, 2006;
WHO, 2007).5 Organizations such as the WHO and UNAIDS have since suggested voluntary medical
male circumcision as an important HIV prevention strategy, and have placed focus on scaling up
services for adult men who are at highest risk
While the medical evidence points to male circumcision as a viable HIV prevention strategy,
there are several open questions that have prevented the rapid scale-up of male circumcision provision in
several African countries. In particular: how does a country scale-up the provision of voluntary medical
male circumcision services safely? What is the demand for male circumcision and how can
organizations target or alter this demand efficiently and ethically? What is the behavioral response to
learning that male circumcision is protective against HIV infection? Importantly, some policy makers
have noted the need to “proceed with caution” because of the ambiguous effect of how individuals may
adjust their sexual behavior upon learning about the relationship between HIV and circumcision
(Namangale, 2007) and researchers have noted that risk compensation may reduce the overall benefit of
male circumcision (Kalichman et al. 2007; Cassel et al., 2006).
5
Across the three trials there were 11,054 HIV negative men who were admitted into the studies. Eligibility was based on the
willingness to be circumcised. In each study, approximately half were randomly assigned to be offered circumcision surgery,
while the others remained uncircumcised. All participants were extensively counseled in HIV prevention. The South African
trial resulted in 61 percent reduction in risk in men who had received circumcision 12 months earlier, when adjusted for
behavior factors. The trial in Kenya resulted in a 53 percent reduction of HIV infection among circumcised men, while the
trial in Uganda resulted in a 48 percent reduction in HIV infection (see for example NIH 2006). The findings were so
dramatic that each study was halted earlier than scheduled due to ethics concerns of withholding lifesaving surgery to the
uncircumcised control group. Several mechanisms may account for why male circumcision reduces a man’s risk of HIV
infection. Because the foreskin’s surface has more immune cells vulnerable to HIV infection than the external surface, it may
promote entry of the virus. The foreskin may also allow the HIV virus to remain on the surface of the penis for a longer
period of time. The moist environment could allow the virus to survive longer, potentially increasing the risk of infection.
Small tears in the foreskin as a result of intercourse could also promote entry of the virus. Also, after circumcision, the penile
shaft and glands develop more epithelial keratinization, a process which may make the penis less susceptible to the virus.
Circumcision removes Langerhans cells from the underside of the foreskin which causes hardening of the surface and
promotes rapid drying (Szabo and Short, 2000).
3
More recently an expert panel of Nobel Prize economists ranked “scaling up male circumcision”
as priority 7 out of 18 other HIV prevention strategies, mainly ranked low due to a concern about disinhibition behaviors (RethinkHIV, 2012).6 The panel noted that “Circumcision is protective, but only to
a certain extent. It is possible that, as a consequence of large-scale male circumcision with an
accompanying information campaign about its protective effect, males and their partners opt for less
safe sexual practices and for example become less likely to use condoms or more likely to engage in
concurrent partnerships.”
Behavioral response to the circumcision surgery itself has been the subject of some research.
Several papers have found no evidence of increased risky sexual behavior after receiving a circumcision
(Agot, 2007; Mattson, 2008) while another paper examined differential responses after receiving a
medical circumcision by baseline subjective beliefs (Wilson et al., 2011). No papers, however, to our
knowledge look at the impact of information about male circumcision and HIV on behavior. In addition
to looking at the effects on sexual behavior, we also present some evidence of the demand for male
circumcisions among uncircumcised men in response to the health information campaign. This is also
the first study that provides a causal estimate of the impact of information about HIV and male
circumcision on actual circumcisions.
To measure the response to information, we interviewed a sample of 1,228 men living in 70
villages in rural Malawi and then randomly disseminated information, by village, about circumcision
and HIV. The information consisted of a sheet that was read to respondents and discussed. It contained a
description of what male circumcision is; a description of the studies that found that male circumcision
is associated with lower risk of HIV infection; and the mechanisms through which male circumcision is
6
The panel consisted of Paul Collier, Edward Prescott, Thomas Schelling, Vernon Smith, and Ernest Aryeetey (Vice
Chancellor of the University of Ghana).
4
protective. Respondents were also told that male circumcision is not completely protective.
Approximately one year after the baseline survey and information intervention, the project revisited
respondents, to measure sexual behavior and condom purchases, as well as to record circumcisions
received in the past year.
Among circumcised men who received the information, we find no evidence of risk
compensation, either with reported sexual behavior or purchases of condoms. On the other hand, our
findings suggest that uncircumcised men who received the information were practicing safer sex,
reducing the number of times of sex per month by 25%, and increasing condoms used by 58% percent.
To support our findings, we also observe significant changes in beliefs about male circumcision and
HIV, one year after learning about the correct link between circumcision and HIV infection.
We explore a variety of potential threats to validity to our results. We find no heterogeneous
treatment effects across a number of baseline characteristics. We find some evidence of spillovers which
could suggest that our estimates are lower bounds of the effects of the information. We also test whether
the information changed respondents’ subjective life-expectancy, another channel of behavioral change
that would predict opposite results compared to a risk compensation model. We find no evidence of this.
In terms of effects of the information campaign on circumcisions among uncircumcised men
during the year after the information campaign, the adoption of the surgery is correlated to the
information campaign with a total of 2 control men and 5 treatment men who had received a
circumcision at the time of the follow-up survey. We discuss some of the reasons they state for getting a
circumcision, which for the most part are orthogonal to reasons related to HIV prevention.
Our results have important policy implications about risk compensation concerns related to
future voluntary medical male circumcision roll outs. Our findings support the dissemination of
5
information about male circumcision and HIV. In addition, our results on the relatively low adoption of
male circumcision among uncircumcised men also point to the fact that additional factors may be
important for scaling up male circumcision other than individual demand (See also Chinkhumba,
Godlonton, and Thornton, 2012).
We proceed as follows: Section 2 presents background information about male circumcision and
Malawi. Section 3 outlines the data and experiment. Section 4 presents a conceptual framework about
the behavioral response to information about risk. Section 5 presents the economic strategy and results.
Section 6 discusses possible threats to validity, Section 7 presents results on subsequent adult male
circumcisions and Section 8 concludes.
2 Background on Circumcision, HIV, and Malawi
Circumcision is not only one of the oldest surgical procedures in the world it is also one of the most
commonly practiced for both religious and non-religious reasons (Marck, 1997; Doyle, 2005). Studies
have shown that overall, 62 percent of adult males in Africa are circumcised (Drain, 2004). Male
circumcision is commonly associated with the practice of Islam as well as with adolescent initiation
where the practice is seen as a rite of passage from childhood to manhood. Among certain groups, men
must become circumcised before they can marry or participate in making community decisions (Marck,
1997). Traditionally, circumcisions in Africa have been conducted in non-clinical settings and are often
combined with a ceremony lasting several weeks.
This paper is set in Malawi, a poor, landlocked country in Southern Africa. In Malawi, as in
other African countries, circumcision is highly correlated with religion and ethnicity. According to the
Malawi Demographic Health Survey in 2010, an average of 19 percent of men reported being
circumcised (National Statistics Office, Malawi, 2011). This is highly correlated with ethnic group, with
6
the majority (85%) of the Yao ethnic group being circumcised, as well as a significant percentage of
Lomwes (30%). Other ethnic groups have much lower rates of circumcision such as the Chewas (8%)
and Tumbukas (1%)7. Circumcision rates are also highly correlated with religion – approximately 94.2
percent of Muslims in Malawi are circumcised compared to 10.7 percent among non-Muslims (National
Statistics Office, Malawi, 2011).
Although male circumcision has been supported widely by international organizations such as
the WHO and UNAIDS and there has been a call for accelerating the scale-up of voluntary medical male
circumcision (United Nations, 2011), with the exception of Kenya, the expansion of services in most
high-priority countries, such as Malawi, has been slow (World Health Organization, 2011). In Malawi,
there have been some debates about the right way to approach male circumcision and HIV prevention.
In a country with one of the highest rates of HIV in the world – estimated at approximately 11.9%
(UNAIDS, 2007) – the potential benefits of rolling out male circumcision could be very high. However,
policy makers in Malawi have cautiously moved forward with integrating male circumcision into a
national HIV prevention policy. At a two-day conference of experts in February 2007 organized by the
National AIDS Commission the Health Minister, Marjorie Ngaunje, warned about the need for caution
in policy and program development because “misinformation over circumcision might erode the gains
made in the fight against HIV and AIDS” (Namangale, 2007). In September, 2010, the Principal
Secretary for HIV/AIDS in Malawi, Mary Shaba, reiterated the need for caution before the government
could universally support male circumcision for HIV prevention stating that “We have no scientific
evidence that circumcision is a sure way of slowing down the spread of Aids” (Tenthani, 2010).8
7
The Yao and the Lomwe typically practice initiation ceremonies for adolescent boys that include circumcision as well as
rituals involving receiving instruction for future life as a man (Stannus and Davey, 1913). Other groups in Malawi practice
initiation ceremonies such as the Gule wankulu or virombo among the Chewa, although this does not involve circumcision.
8
More recently in late 2011, Malawi adopted voluntary medical male circumcision as part of its HIV prevention strategy
7
3 Data and Experimental Design
This paper uses data from men, collected in two waves across a twelve month period. The baseline
survey was conducted in October/November 2008 in Traditional Authority Kuntumanji in the Zomba
district in Malawi. This is located in the Southern Region of Malawi and was selected because of its
diverse population consisting of both circumcised and uncircumcised male residents. To sample men
into the study, 70 villages were first randomly selected into the sample, stratified by the distance of each
village to the nearest mosque.9 Within each village a full household enumeration was conducted. The
second stage of sampling involved randomly selecting men ages 25 to 40 from the household
enumeration.10
To try to balance the sample across circumcision status and because of the high correlation
between religion and circumcision status in the aggregate data in the country (e.g. DHS Malawi), the
selection of men were stratified by their religious affiliation (Christian or Muslim). In each village, a
maximum of 20 Christian and 20 Muslim men were selected. Despite the attempts to stratify the sample
by religious affiliation in order to have a balanced sample of circumcised and uncircumcised
respondents, a considerable fraction of the men who were sampled were already circumcised (73.7
percent).11 Most of the respondents who were Muslim were circumcised (94.2 percent) while a
surprising large percentage of Christians was also circumcised (60.9 percent). The observed rates of
(Tunikekwathu News, 10 October 2011).
9
The process for selection and randomization included: taking a list from the 2008 census of all of the villages in Traditional
Authority Kuntumanji; dropping all duplicate village names in that list; merging in the locations of churches and mosques
provided by the National Statistics Office and listing the villages by distance to a mosque (from near to far). Villages were
then divided into groups of 10 villages and were randomly divided into half for treatment and half control.
10
While respondent refusals were low, men in the area are very mobile making it difficult to find and survey all the selected
men. In total 67 percent of those who were randomly selected from the household enumeration had completed baseline
surveys. There were no significant differences in the finding rate across basic demographics or treatment status. For example,
69 percent of Christians and 67 percent of Muslims who were initially sampled from the enumeration list had completed
baseline surveys.
11
Stratification was done by religious denomination rather than by circumcision status at the household listing. This was due
to the fact that the majority of respondents at the household listing consisted of women and may have been unable to
accurately report the circumcision states of men in the household.
8
circumcision are significantly higher for Christians as well as the Chewa and Nyanja tribes in the data
relative to national aggregate male circumcision statistics. However, the rates are similar to other
available data restricted to areas similar to our study site.
Appendix A presents the circumcision rates by religion and ethnicity for the data in Kuntamanji,
data from the Malawi Diffusion and Ideational Change Project (MDICP) in Balaka in 2008 (a nearby
and similar district) and the Demographic and Health Survey (DHS) in rural Zomba in 2004. In both of
these data sets the circumcision rate among non-Muslims is similar to the data in Kuntamanji. There are
a number of potential reasons that this area may have a higher percentage of circumcised non-Muslim
men when compared to the national averages. This area constitutes a sizeable population of men who are
both Christian, and belong to the Yao ethnic group (87% of the sample). The Yao tribe practices
circumcision as part of cultural initiation rites thus driving up the percentage of non-Muslim circumcised
men in the sample. Another potential factor for the higher percentage of non-Muslim circumcised men
in the sample could be due to spillovers within villages – that is, Christians, Chewas, and Nyanjas living
in villages where circumcisions are common, may be more likely to circumcise their sons. Alternatively,
if respondents practiced Islam as a child and were circumcised at that time, but later converted to
Christianity, we would also see a larger number of Christians who were circumcised in the data;
however our data provide limited support for this hypothesis.12
The total sample for the analysis in this paper consists of a total of 1,228 male respondents.
Table 1, Column 1 presents the summary statistics for these men. Overall, the men are on average 32
years old. Most are married (90 percent) and have had sex in the past year (96 percent).
The
12
Most of the respondents cite religion or culture as the reason why they were circumcised (89 percent). Related to this, the
majority (92 percent) of circumcisions took place in the bush as part of the initiation rites. In follow-up data, men were asked
about their religious conversion history. Among those men who claimed to “always be Christian”, 56 percent were
circumcised. Similarly, 58 percent of men whose mother was always Christian or father was always Christian were
circumcised (not shown). This suggests that religious conversions are not likely to explain all of the discrepancy between the
circumcision patterns in this area and that in Malawi overall.
9
respondents have an average of 1.9 children (conditional on having any children). The two majority
tribes consist of the Nyanja – similar to the Chewa ethnic group (40 percent) and the Yao (33 percent).
The men have only had on average 5.6 years of schooling (6.7 conditional on having ever attended
school). Over half (62.1 percent) of the respondents are farmers who report having on average 4.4 assets
out of a list of 13 different possible assets.13 Only 12.3 percent of the respondents report having salaried
employment. The average age at sexual debut was 17 years with an average of 4 lifetime sexual
partners. Less than half (40.1 percent) of the men have ever used a condom with any partner. While
most men report being sexually active in the last month (75.5 percent), only 11.3 percent of their sexual
encounters in the last month are considered as safe sexual acts defined as the fraction of encounters in
which they used a condom.
Half of the villages (35) were randomly assigned to the treatment group. All of the men residing
in treatment villages were assigned the information treatment. Immediately after the baseline survey,
respondents who were in the treatment group were informed that circumcision is partially protective
against HIV transmission. Interviewers discussed a standardized information sheet that explained the
three randomized control trials in Uganda, South Africa, and Kenya, as well as the results from these
trials. Information was also provided about some of the medical reasons why circumcision is partially
protective. Respondents were encouraged to ask questions during the discussion. To understand the
delivery of the information treatment we compare how much time it took to administer the survey
between treatment and control respondents. During the baseline survey, on average, treatment interviews
13
The full list of assets include: bed, mattress, sofa, table, chairs, paraffin lamp, television, radio, mosquito net, bicycle,
motorcycle, ox-cart and landline phone.
10
were approximately 4 minutes longer than control group interviews and this difference is statistically
significant (a difference of 4.33, standard error of 1.277; not shown).14
It is worth noting that villages were randomly allocated to treatment and control groups
independently of village characteristics and geography. Because of this, control villages could be located
in close proximity to treatment villages leading to a possibility of information spillovers about
circumcision from the treatment to the control respondents. If spillover effects were large enough, our
comparison of condom purchases between control and treatment groups would be a downwardly biased
estimate (See for example, Miguel and Kremer 2001). For this reason, to mitigate the possible effect of
information spillovers on the baseline survey, all control group interviews were conducted before the
treatment interviews. Note that in the presence of such spillovers, our results estimating the effects of
information would be a lower bound of effects without contamination. This is discussed further in
Section 6.
Approximately one year after the baseline survey, we attempted to re-interview all men
interviewed at baseline, asking about sexual behavior in the past year and circumcision status. In
contrast to the baseline survey, there was no specific order to treatment and control interviews during the
follow-up survey. Approximately 70 percent of the men who were interviewed at baseline were found at
the follow-up. Attrition was not significantly associated with the treatment status, or circumcision status
(Table 2). Attrition was also not significantly associated with other factors such as age; education;
marital status; nor perceptions about HIV and circumcision (results not shown).
14
Note that control interviews were conducted prior to treatment interviews. Over time, as interviewers became more familiar
with the questionnaire the time until completion was reduced. An alternative explanation driving the difference in the
treatment and control time of interview is that the interviewers skipped more questions over time or created more missing
values. We find no evidence that there was an increase in non-response or missing values correlated with either day of
interview or treatment status (not shown). Given the order of control and treatment interviews, the overall difference in time
to completion between treatment and control is likely to be a lower bound on the actual time allocated to the delivery of the
information script. In fact, controlling for day fixed effect increases the estimated difference in survey interview time.
11
During the follow-up survey, men were asked questions about their sexual behavior over the past
year, as well as beliefs about male circumcision and HIV. In addition, uncircumcised men were asked
about whether or not they had undergone a circumcision surgery.
Collecting accurate outcome measures on sexual behavior can be challenging given the sensitive
nature of discussing sexual behavior.15 However, in addition to asking respondents about their reported
sexual behavior, we also collect data on actual condom purchases as a proxy for the demand for safe
sex.16 Immediately after the survey, each respondent was given 30 Kwacha (approximately 20 cents or
just less than one days wage) for their participation in the study. Each respondent was then offered the
opportunity to purchase subsidized condoms: 5 Kwacha (3 cents) for a package of three condoms or 2
Kwacha (1 cent) for one condom.17 The number of condoms that were purchased was then recorded by
each interviewer. On average, 41 percent of the respondents purchased at least one condom. Among
those purchasing condoms, the average number of condoms purchased was 6.3 condoms. It is important
to note that villages in the study are relatively remote and condoms are not easily accessible. On
average, respondents would need approximately 32 minutes to travel (walk) to shops to purchase
condoms. Also, because the respondents needed to give up a small amount of money in order to
purchase condoms from the interviewer, the number of condoms purchased could be interpreted as one
15
In some studies that ask respondents to report their sexual behavior, such as number of sexual partnerships, frequency of
sex or condom use, it has been suggested that respondents tend to under-report sexual encounters and over-report preventive
behaviors such as condom use or abstinence (Fenton, Johnson et al., 2001). Respondents may feel social pressure to report
behavior they believe is acceptable, especially given that interviewers may themselves emphasize the importance of safe
sexual practices. On the other hand, other research suggests that for non-casual partners, accounts of sexual behavior within
reported relationships can be generally reliable (Clark et al., 2011).
16
This method has been used in previous work by providing respondents with a small financial endowment to purchase
condoms at a reduced rate (Thornton, 2008). Condoms were also sold immediately after the baseline questionnaire for the
control group and immediately after the provision of information for the treatment group during the baseline survey. We do
not present these results as any effects on condom purchases immediately after the treatment would be more likely to be due
to framing effects rather than longer-term impacts of learning new information (Zaller and Feldman, 1992).
17
The condoms sold were Chishango brand, the most widely available and popular condom in Malawi, and were offered at
approximately half the retail price at the time of the study. To reduce the risk of purchasing condoms for resale, respondents
were limited to purchase 18 condoms from the money received by interviewers. Only 44 individuals purchased the maximum
number at follow-up
12
proxy indicator of the demand for safe sex. Using condoms alone as a proxy measure for the demand for
safe sex is somewhat problematic if men decide to abstain from sex entirely. In that case they may have
no demand for condoms, and yet are practicing safe sex. In this case however, the majority of our
sample is sexually active (75 percent had sex in the previous month at baseline). Alternatively, it could
be that men increase the number of individuals with whom they have sex, or the number of times they
have sex offsetting any increase in the uptake of condoms. To address this issue we use both self-reports
and condom purchases as outcome measures.
4 Conceptual Framework: Behavioral Responses to Information
In the risk response literature there are competing theories or approaches that affect the predicted net
behavioral response with the introduction of a new technological innovation or new information that
affects an individuals’ risk. The first approach is mainly used as a simplification when modeling the
impact of technology or information that assumes the technology/information has no effect on behavior.
That is, effects are purely technological rather than also including behavioral responses. The second
approach, following Peltzman (1975), predicts that individuals facing higher risks should optimize by
engaging in safer behavior. This has been examined empirically by a number of researchers who find a
general pattern of support of this theory (Chesney et al., 1997; Johnson, 1999; Francis, 2007; Oster,
2007; Nikolov, 2011 Auld, 2003; Boozer and Philipson, 2000). The extreme case of this argues that
individuals have a target level of acceptable risk and will adjust their behavior in response to new
innovations or information to maintain this target level (Wilde, 1982). Some have argued that risk
compensation can go in either direction – either towards more risky behavior, or towards more
protective behavior. Evans (1985) proposed that any “behavioral feedback” may be possible including a
protective response to innovation or information that puts the individual at lower risk. Lastly, other
models based in psychology or behavioral economics suggest that increasing individuals’ beliefs of
13
infection may be counter-productive for motivating safe sexual behavior (Levine and Ross, 2002).
Instead of encouraging people to practice safer sexual behavior, having high beliefs of risk may
encourage denial and fatalism (Daoreung, 1997; Stein, 1999; Crain, 2005).18
Each of these competing approaches predicts differential behavioral responses to learning new
information. The existing empirical literature also presents mixed results. Households in Bangladesh that
were told that their wells were unsafe due to high levels of arsenic switched water sources despite a
fifteen fold increase in the amount of time needed to collect water (Madajewicz et al., 2007). On the
other hand, providing information about the specific levels of arsenic yielded differential results
depending on the levels of arsenic in the well with less switching among those who faced the most risk.
(Tarozzi, 2011). Within the context of safe-sex interventions there are also mixed results. In Kenya,
school children were randomly told specific information about HIV risk – that older men (as opposed to
younger men) were more likely to be infected with HIV (Dupas, 2011). This information had a large
effect on reducing sexual activity of adolescent school girls with older men. In Uganda, a randomized
condom promotion trial finds that participants in the treatment group were more likely to use condoms
but also increased their number of sexual partners; the increased condom use was insufficient in
reducing overall exposure to risk (Kajubi et al., 2005). Several studies have examined risk compensation
in the presence of HAART (anti-retroviral therapy) and some also presents mixed results. Stolte (2010)
and de Walque et al. (2010) find evidence in support of risk compensation, while Crepaz et al, (2004) in
their meta-analysis do not find support for risk compensation.
One way in which researchers have sought to shed light on these mixed results within similar
contexts is to measure how individuals respond to information based on their subjective beliefs (Khwaja
18
For example, Kaler argues that sexually active men in Malawi who believe they are already infected with HIV use this
excuse to justify risky sexual activity (Kaler, 2004). Another report of teenagers in several countries across Africa found that
respondents reported “little point in [changing sexual behavior] since ‘we are all dead already’” (Bennell, 2003).
14
et al., 2007; Stoler, 2004; Salm, 2010).19 Economists have suggested that in response to an information
intervention, only individuals who learn new information should change their behavior; those with
correct baseline beliefs should have already optimized their behavior and would exhibit no additional
behavioral change. On the other hand, in other disciplines, such as psychology and political science,
theories of confirmation bias predict that individuals assess information with a bias towards reinforcing
existing beliefs. (Kunda, 1990 and Molden and Higgins, 2005). Nyhan and Reifler (2010) provide
evidence of an extreme case where they observe a “backfire effect” in an experimental setting.
Individuals with incorrect priors were even more likely to hold inaccurate beliefs after they were
provided with correct information. These two theories predict opposite responses to information based
on subjective beliefs.20
Not only are there different theories predicting asymmetric responses to information, measuring
subjective beliefs can be difficult (Bertrand and Mullanaithan, 2001), especially in a setting where levels
of education are low (Delavande et al. 2011). In addition, how one treats the data in analysis can have
large effects on the results; for example, in how researchers code “Don’t Know’s” (Thornton, 2012) or
use multiple questions that elicit beliefs in different ways. Because of these challenges, in this paper,
rather than examining the differential response to information based on baseline subjective beliefs about
male circumcision or personal risk, we study the differential responses to information across men by
their circumcision status. Our assumption is that circumcision status is the best predictor of how
individuals will interpret the information about male circumcision and HIV.
19
There is a growing literature in economics on measuring subjective beliefs and relating these beliefs to behavior
(Delavande et al. 2011; Engelberg et al., 2006; Gan et al., 2003; Gan et al., 2004; Hamermesh, 1985; Lusardi, 1999; Manski
and Dominitz, 1994; Schunk, 2005; Smith et al., 2001).
20
A good example in which there are mixed results on the provision of information by baseline beliefs is in the context of the
behavioral response to receiving an HIV test. Several papers have found evidence supporting the economic theory that only
those who are surprised change their behavior (Boozer and Philipson, 2000; Gong, 2011; de Paula, Shapira and Todd, 2010),
others have found no significant effects (Thornton, 2008, Goldstein et al., 2008). Publication bias also likely plays a role in
the body of evidence available.
15
Despite the challenges with measuring and using subjective beliefs data, because this paper
evaluates an information intervention, it is important to address how exposed men living in rural Malawi
were likely to have been to the information about HIV transmission and circumcision before the
intervention. At the time of our baseline survey in October/November of 2008, the information about
HIV transmission and circumcision was not widely known in Malawi. However, at that time, several
national meetings of experts, sponsored by the National AIDS Commission in Malawi had been held to
discuss a national plan in relation to circumcision. After each of these meetings, several newspaper
articles were printed and there was some radio coverage, discussing the findings from the randomized
controlled trials as well as the outcome of the national meetings. Respondents could have been exposed
to some of this information.21 Although information about male circumcision and HIV infection was
available to individuals in Malawi via radio and newspapers in a limited manner at the time of the
baseline survey, we have reason to believe that the effects of receiving the information directly and
individually from interviewers during this study are stronger than the effects of other media sources.
First, the information may not have reached the sample respondents through newspapers or the
radio. Our baseline data indicates that only 75 percent of our respondents could read in Chichewa, and
of those who were literate, only 17 percent read the newspaper more than once per month.
Second, even if respondents had access to information from other sources about the relationship
between circumcision and HIV infection, research suggests that even among individuals with prior
information exposure, comprehension of the information increases after receiving it directly from an
interviewer (Guadagno and Cialdina, 2002; Valley, Thompson et al., 2002; Guadagno and Cialdina,
2007; Valley et al., 1998).
21
Despite some newspaper coverage, there was no accepted national policy regarding male circumcision for HIV prevention.
In fact, even as late as September 2010, one of the leading HIV/AIDS officials in Malawi claimed that there was not enough
evidence supporting the medical benefits of male circumcision (Tenthani, 2010).
16
Lastly, our baseline data indicate that the majority of individuals had the wrong baseline prior
knowledge about the relationship between HIV and circumcision. We measure knowledge of the
relationship between male circumcision and HIV infection in two ways. Respondents were asked
whether they believed that circumcision increased, decreased, or had no effect on the risk of HIV
infection. When asked this, 13 percent of men stated that circumcision increased the risk of infection, 37
percent believed that circumcision decreased the risk, and 48 percent said that circumcised and
uncircumcised men were at equal risk (the remaining 2% reported not knowing).22 Respondents were
also asked the question: “If 100 circumcised men slept with an HIV positive women last night, how
many of them would acquire HIV?” A similar question was asked in reference to uncircumcised men.
Similar to the categorical measure of beliefs, only 32 percent of the men’s individual corresponded to
the belief that circumcised men are at less risk than uncircumcised men.
Given the low levels of knowledge about male circumcision at baseline, we predict responses to
the information intervention. Moreover, we predict that circumcised and uncircumcised men will
respond differentially to the information. In our study, uncircumcised men who learn that circumcision
reduces the transmission rate of HIV learn that they are more at risk than they had believed; if they reoptimize their behavior, these individuals may practice safer sex. Circumcised men who learn they are at
less risk, believing they are safer may practice riskier sex. On the other hand, a different channel of
behavioral change might be that circumcised men learning they are less at risk may also infer they will
live longer thus providing them with the motivation to practice more protective behavior. We explore
this further below in Section 6. Regardless of the channel, each of these theories alone predict
asymmetric responses to the information by circumcision status.
22
Specifically, respondents were asked: “What is the likelihood of getting HIV if you are circumcised (rather than
uncircumcised)? Increased, Decreased, Same, or Don’t know”.
17
5 Results: Responses to Information Campaign
Econometric Strategy
To study the impact of information on sexual behavior and condom purchases, we utilize the fact that
the information about circumcision and HIV was randomized across villages. Our main empirical
specification estimates the following regression:
(1)
,
∗
′
,
We examine the effects of individual i living in village j, as a function of being assigned to a treatment
village. We examine several main outcome variables, Y. First, several different measures of self-reported
sexual behavior recorded at the time of the follow-up survey such as condom use, number of partners,
and any sexual activity in the previous month and year. To address the issue of multiple inferences we
also create an index of risky sexual behavior which is the mean of the normalized value for nine selfreported measures of sexual behavior (Kling, Liebman and Katz, 2007).23 Our second main set of
outcome variables measure purchases of condoms.
We include a vector of individual controls which include age, marital status, logged income,
years of education, number of assets (asset index), whether the respondent had sex in the last week at
baseline, tribe dummies, and an indicator of religious denomination. In addition, when we estimate the
effects of the treatment on condom purchases we include interviewer fixed effects in some
specifications.24 In the analysis we cluster standard errors by village.
23
The nine measures used in constructing this index are: Whether wife is pregnant; Had sex in last month; Number of sexual
interactions in last month; Total number of condoms used in last month; Fraction of sexual encounters that used a condom;
Number of sexual partners in last month; Number of partners in the last year; Number of condoms purchased in the last
month and the number of condoms that respondent received free in the past month.
24
Some interviewers were more effective in selling condoms to respondents. Accounting for interviewers have small effects
on the results, slightly reducing the impact of the treatment on condom purchases, but not affecting the main results.
18
Because of the randomization of the treatment, the error term is uncorrelated to treatment status
(due to unobservables or selection of individuals into having the information). Baseline statistics
compared across treatment status indicate that men in the treatment and control villages are not
systematically different which is reassuring for our identification strategy and randomization (Table 1).
Along some aspects, there are significant differences between treatment and control men – for example,
respondents assigned to the control group are slightly wealthier as indicated by logged reported
expenditures and number of assets. There are no differences across ethnicity or religious composition as
can be expected as we stratified the randomization across religion. Baseline characteristics are also
balanced across treatment status, separately across circumcised and circumcised men (Appendix B).
Recall that the sample was initially stratified by religious denomination – the proportion of
Muslims vs. Christians within a village – when assigning the treatment. Our results are consistent when
running specifications separately by religious denomination rather than by circumcision status (not
shown). The results are also robust to probit models in the case of indicator outcome variables (not
shown).
The coefficient on the treatment indicator,
, represents the difference in outcomes between the
treatment and control group among uncircumcised men. If uncircumcised men practice safer sex because
they learn that they are more at risk than circumcised men, then
sex. The coefficient on the interaction term,
0 when Y is a measure of risky
, represents the differential effect of the treatment across
circumcised and uncircumcised men. Given that circumcised and uncircumcised men are learning that
they face asymmetric risk of infection from the information, we would predict that
measure of risky sex). Lastly,
0 (when Y is a
indicates the difference in sexual behavior between circumcised and
uncircumcised men in the control group. Recall that male circumcision is strongly linked to ethnicity
19
and religion and thus we might expect underlying differences in sexual behavior between circumcised
and uncircumcised men. The extent that either circumcised or uncircumcised men are engaging in riskier
or safer sex would predict the sign of
. It is important to note that our empirical strategy does not
involve comparing circumcised men to uncircumcised men, rather the differential responses to the
randomized information campaign between these two sets of men.
To determine whether there is risk compensation among circumcised men who learn the
information, we compare circumcised men who learn the information with circumcised men who do not,
testing the null hypothesis that there is no risk compensation. We present the p-values testing
0. If we can reject the null hypothesis (with a p-value less than 0.10), then there is evidence for risk
compensation. If we cannot reject the null, we may either be under-powered, or else there is no risk
compensation.
Results: Sexual behavior and condom purchases
We report the effects of the media campaign on sexual behavior in Table 3. Examining first the
coefficients on the treatment indicator, the results indicate that uncircumcised men exposed to the
information intervention report practicing safer sex. Along several dimensions, we find large and
significant differences in being less likely to have sex in the past month and having it fewer times.
Uncircumcised men exposed to the information were more likely to report using and purchasing
condoms. The results are also strong in terms of the composite index of risky sexual behavior (Column
11).
We find that for many measures of sexual behavior there are significant differences in the
response to the treatment between circumcised and uncircumcised men. The sign of the interaction term
20
is in the direction predicted, consistent with the fact that the information about risk is asymmetric in
circumcision status.
To test for risk compensation among circumcised men, p-values are presented testing the
hypothesis that
0. We cannot reject the null that there is no impact of the treatment among
circumcised men for any of the measures of sexual behavior. While the coefficients are small, and fairly
tight around zero, our sample size only allows for detecting significant effects larger than 0.25 standard
deviations (with 90 percent power). While we cannot say with full confidence that there was no risk
compensation, we can reject large changes in sexual behavior among circumcised men.
Column 12 presents the effects of the information on condom purchases which include additional
controls for interviewer fixed effects (Results are robust to excluding these controls). While statistically
insignificant, the coefficients are consistent with the sexual behavior results. Uncircumcised men are 7.5
percentage points more likely to purchase a condom after learning the information, which is not
statistically significant at the traditional levels, but is with 80 percent power.
These results are robust to either excluding or including baseline controls or measuring treatment
effects on differences (results not shown).
In all, the evidence suggests no large responses in terms of behavioral dis-inhibition among the
circumcised men learning about the relationship between circumcision and HIV risk. Among
uncircumcised men in this study, however, we find significant effects on sexual behavior one year after
the information campaign. Uncircumcised men learning the information reduce the number of times
having sex per month by almost 2 acts or 25%, and increase condoms used by 1 condom, or 58%
percent. It is useful examining how reasonable the sizes of these effects are and how they compare to
other successful HIV prevention studies. Robinson and Yeh (2011) study how female sex workers in
21
Kenya respond to individual and household shocks. Women who were sick were 8.5 percentage points
less likely to have sex. Dupas (2011) find that girls who learn about HIV transmission risk with older
men are 12 percentage points more likely to use a condom at last intercourse, and Kohler and Thornton
(2011) find that women receiving a moderate cash grant were 7.4 percentage points less likely to have
sex. Are these effects small or large? While difficult to quantify, it has been shown in simulations that
even small increases in condom use or safe sex can make a difference in HIV prevention (Bracher et al.,
2004).
Results: Ex-Post Knowledge about Male Circumcision and HIV
Our results suggest that the information provided to the men caused reductions in risky sexual behavior
among uncircumcised men and no effects on sexual behavior among circumcised men. Is there evidence
that the information was understood or recalled by those men in the treatment group? Table 4 suggests
that even 12 months after learning the information, those in the treatment group had different levels of
knowledge about male circumcision and HIV transmission rates.
At the follow-up survey men in both the treatment and control groups were asked “If there were
to be a scientific study about HIV, would it show that circumcised men (relative to uncircumcised men)
are at an increased risk of contracting HIV, a decreased risk of contracting HIV, or an equal risk?” The
majority in both treatment and control groups report that circumcision would lead to a decreased risk of
HIV; however, the treatment men are significantly more likely to think that circumcision leads to a
decreased risk and less likely to report that circumcision is either an increased risk or that there is no
impact. Treatment men were 7.1 percentage points more likely to believe that the study would say that
circumcision is related to a lower risk, while 8.7 and 3.2 percentage points less likely to think that the
study would reveal circumcision is related to equal or higher risk, respectively (Table 4, Columns 1-3).
22
Respondents were also better able to accurately identify the countries in which the circumcision studies
took place. In particular, treatment respondents were significantly less likely to state that the trials were
conducted in Malawi and Nigeria (not shown).
Respondents were asked if they believed that circumcision was related to an increased HIV risk,
decreased HIV risk, or had no impact on HIV infection, and were also asked to give the number of
circumcised or uncircumcised men (out of 100) who would be infected by sleeping with an HIV positive
woman.25 Overall, men in the treatment group are 7.1 percentage points less likely to believe that there
is no relationship between circumcision and HIV risk, with a shift into believing that male circumcision
is related to reduced risk (Table 4, Columns 4-6). In addition, men in the treatment group believe that
circumcised men face a 7.8 percentage point reduced risk than uncircumcised men (Table 4, Column 7).
The impact of the treatment is also robust to measuring effects on relative risk (as measured by
differences, ratios, changes from differences at baseline, and changes from ratios at baseline). In sum,
there were persistent significant differences in knowledge and beliefs between the treatment and control
one year after the information campaign.
6 Potential Threats to Validity
Heterogeneous Treatment Effects
To test the robustness of our findings we examine some potential heterogeneous treatment effects that
may affect the results. Among both the circumcised men and uncircumcised men we find no significant
differential effects by education, exposure to media, wealth, ethnicity, religion, prior beliefs about HIV
and male circumcision, or prior condom use (not shown).
Spillovers
25
The wording of the question was “If 100 circumcised men slept with an HIV positive woman last night, how many of them
would acquire HIV?” A similar question was asked in reference to uncircumcised men.
23
Another potential threat to validity is spillovers. Information, unlike other interventions, is more prone
to spillover effects given the ease with which information can be shared. Many steps were taken to
reduce the chance of spillovers – the randomization was conducted at the village level and control
interviews at baseline were conducted before all of the treatment interviews. However, if there were
spillovers, this would mean that comparing treatment and control outcomes would underestimate the
true impact of the information.
To determine whether spillovers are a concern, we first compare changes in knowledge about
male circumcision across baseline and follow-up among those men who were in the control villages.
Overall, 32 percent of the respondents in the control at baseline reported that male circumcision
decreased the risk of HIV transmission. At follow-up, this rate had increased to 61 percent in the control.
Similarly, there were large increases in correct beliefs about male circumcision and HIV among the
control group when using the continuous measure of beliefs; Figure 1 presents this CDF.26 This figure
graphs baseline answers pooled among treatment and control, as well as answers at the follow-up
separately among treatment and control men. The treatment and control CDFs at follow-up illustrate the
effect of the treatment on moving the distribution of beliefs of circumcision and HIV transmission.
However, note that the control group distributions at follow-up are also different from the distribution of
beliefs at baseline.
These results could indicate an increase in media exposure about circumcision and HIV among
both treatment and control groups between the baseline and follow-up survey. Alternatively, this also
could be an indication of spillovers from the treatment to the control group between the time of the
26
Recall that men were asked how many circumcised men would become infected (out of 100) after sleeping with an HIVpositive woman. On average circumcised respondents thought that the probability of infection was 81 percent for circumcised
men and 91 percent for uncircumcised men. Uncircumcised men perceive the rate of infection for circumcised men to be 87
percent and 92 percent for uncircumcised men. The actual per-act rate of transmission from female to male is orders of
magnitude lower. Across a meta-study of 43 publications and 25 different populations the rate was 0.04% (95% confidence
interval 0.01-0.14) (Baily et al. 2009).
24
baseline and follow-up survey in which case our estimates would be lower bounds of effects without
spillovers. There is some evidence that both of these might be occurring in our data.
Overall media exposure increased overall among all respondents. For example, respondents were
asked how many messages they had heard (either in the newspaper or on the radio) about HIV/AIDS in
the past 30 days at both baseline and follow-up. At baseline, respondents reported hearing on average
11.9 messages. At the follow-up this increased by approximately one additional message (not shown).
On the other hand, spillovers might have been important as well. We can measure some
spillovers by regressing beliefs about male circumcision and HIV on the fraction of neighbors living
within 0.5 kilometers who were treated with the information (controlling for total number of neighbors)
among only the control group (See Godlonton and Thornton, forthcoming for further discussion on this
methodology). We find that for every additional 5 treated respondents residing within a 500m
geographic band around a control respondent, the control respondent was 4 percentage points more
likely to report that circumcision reduces HIV risk, and almost 4 percentage points less likely to report
that circumcision has no impact on HIV risk. This spillover effect is also captured in the continuous
measure of beliefs (results not shown). To the extent that there were some informational spillovers, our
results provide lower bounds of the true effects.
Life Expectancy
One possible channel through which information about HIV risk could affect behavior is that individuals
could change their perceptions about their life expectancy. If this were the case, circumcised men
learning the information might then learn they had a longer life expectancy and have the motivation to
protect themselves more. We asked individuals about life expectancy at the follow-up (by asking yes or
no questions to elicit beliefs of surviving up to various ages) and find no significant differences in life
25
expectancy between treatment and control men either overall or separately by circumcision status (not
shown). We next turn to presenting the impact of the information campaign on adult male circumcisions
among those who were uncircumcised at baseline.
7 Adult Male Circumcisions
Demand for/Acceptability of Circumcision
Several studies have examined the demand for or acceptability of circumcision. One study in South
Africa found that 70 percent of the uncircumcised men reported that they would want to be circumcised
if it were proven protective against sexually transmitted diseases (Lagarde, Dirk et al. 2003). Other
similar studies in South Africa, Kenya, Malawi and Zimbabwe found positive attitudes towards
circumcision (Nnko, Washija et al. 2001; Bailey, Muga et al. 2002; Halperin, 2002; Rain-taljaard,
Lagarde et al. 2003; Ngalande, Levy et al. 2006). Informational campaigns have also been found to have
aggregate effects on increasing reported desire for circumcisions. For example, in Botswana respondents
were surveyed before and after an informational session that described the risks and benefits of male
circumcision and the proportion reporting they would definitely or probably circumcise a child if free of
charge in a hospital setting rose from 68 percent before the informational session to 89 percent after the
session. This increase was similar to the change among the uncircumcised men when asked about going
for their own circumcision (Kebbabetswe, Lockman et al., 2003). While these studies suggests that
information may have a large effect on willingness to become circumcised, most previous studies have
only collected reported willingness to be circumcised, rather than actual circumcisions.27
Data of actual circumcisions indicate that the rate of circumcision is increasing. For example, the
number of circumcisions performed at a hospital in Zambia increased from 20 cases per month up to 90
27
An exception is Godlonton, Thornton, and Chinkhumba (2012) who measure actual circumcisions at a clinic in Malawi.
26
cases per month from 2004 to 2006. Of those coming to be circumcised, 90 percent reported that they
were doing so to protect themselves from HIV (Bowa and Lukobo, 2006). It is reported that this
increase, as well as other increases in demand (e.g., in Swaziland) have been in response to learning that
circumcision is protective against HIV (WHO 2006; Schoofs, Lueck et al. 2005; McNeil, 2006; BBC,
2007). However, the causal relationship between information and an increase in demand for male
circumcision has not yet been established. Each of the above-mentioned studies suffers either from the
limitations of using only cross-sectional data in the analysis or of relying on hypothetical answers rather
than actual behaviors. In contrast, we measure actual circumcision behavior and the effects of exogenous
dissemination of information about male circumcision.
One year after the information campaign, in addition to asking men about their sexual behavior,
we asked men who reported being uncircumcised at the baseline their circumcision status again. In all,
we identified 7 men who were circumcised between the time of the baseline and follow-up survey. Five
men were in the treatment group and two were in the control group. We present some summary results
on circumcision in Table 5. Panel A present regressions predicting whether uncircumcised men were
likely to get circumcised based on their treatment status. There is a positive effect of being in the
treatment group on getting a male circumcision, although this is not statistically significant at traditional
levels. We also present several descriptive statistics about the type of men who got circumcised (Table
5, Panel B). These men were slightly younger than average, slightly poorer (although more likely to be
self-employed). Notably, all of the respondents were married. This is relevant to some reasons men
stated for why they received a circumcision; health reasons, pleasing their wife, or to setting an example
to others. Only one out of the seven men was circumcised in the clinic, the remaining six were either
circumcised at home or traditionally.
27
8 Conclusions
This paper presents results from a study that randomized information about the relationship between
HIV and circumcision among already circumcised and uncircumcised adult men in rural Malawi. From a
policy perspective, one concern is that already circumcised men will practice riskier sex upon learning
that circumcision reduces the transmission rate of HIV.
First, we find that rural Malawian men are not well informed about the relationship between
circumcision and HIV transmission risk. In other words, as part of an HIV prevention strategy there is
scope to provide this information.
We then show that there is no significant effect on risk compensation among circumcised men
who learn the information that circumcision is partially protective against HIV transmission, and
uncircumcised men practice safer sex. The effects on uncircumcised men learning the information are
fairly large, reducing the number of times having sex per month by almost 2 acts or 25%, and increasing
condoms used by 1 condom, or 58% percent.
We also find modest, although not statistically significant effects on adults in the treatment group
receiving a circumcision. In general however, the total of 7 adult male circumcisions that were received
after the study are small, indicating that providing information alone is not enough to increase demand
for voluntary medical male circumcision. This is important in considering which strategies to invest in
for scaling up medical male circumcision.
Our results from Malawi indicate that information can have lasting effects on knowledge and
behavior. However there are reasons to believe that the responses to information may depend on the
setting. In particular, in countries or areas in which there is widespread government support coupled
with devoted male circumcision resources, we may expect the response to information to create a higher
28
demand for circumcisions. It is also important to note that our results were from data among men who
were circumcised at young ages, due to cultural or ethnic reasons. The results may not generalize to risk
compensation among men who become circumcised as adults as a strategy for HIV prevention although
there is evidence that this is also not a significant concern (Mattson, 2008).
This paper also discusses some of the literature on risk compensation that has been studied
within economics. The literature provides mixed empirical evidence, with no clear theoretical
predictions of when, or in which setting individuals respond to risk information. While subjective beliefs
are important within theoretical models of decision-making, measuring these beliefs and empirically
testing how they affect behavior is challenging. Publication bias is likely to also bias what we learn
about these behavioral responses. We believe that there is a need for an extensive review of the existing
literature on risk compensation and further research to understand under what circumstances one
observes risk compensation and when one might not. However this is beyond the scope of this paper.
29
References
Agot, Kawango, James N Kiarie, Huong Q Nguyen, Jacob Odhiambo, Tom M Onyango, Noel S Weiss.
(2007). “Male Circumcision in Siaya and Bondo Districts, Kenya: Prospective Cohort Study to
Assess Behavioral Disinhibition Following Circumcision”. JAIDS Journal of Acquired Immune
Deficiency Syndromes: 1 January 2007 - Volume 44 - Issue 1 - pp 66-70.
Anglewicz, P. and H.-P. Kohler (2005). "The construction of self-assessed likelihood HIV/AIDS
infection: an examination from rural Malawi." Meeting of the International Union for the Study
of Population.
Arrow, Kenneth (1963). "Uncertainty and the Welfare Economics of Medical Care". American
Economic Review (American Economic Association) 53 (5): 941–973.
Auld MC (2003). "Choices, beliefs, and infectious disease dynamics". Journal of Health Economics 22
(3): 361–377.
Autier, P. et al. (1998). “Sunscreen Use, Wearing Clothes, and Number of Nevi in 6- to 7-Year-Old
European Children.” Journal of the National Cancer Institute, Vol. 90. No. 24.
Auvert B, T. D., Lagarde E, Sobngwi-Tambekou J, Sitta R, et al. (2005). "Randomized, controlled
intervention trial of male circumcision for reduction of HIV infection risk: The ANRS 1265
trial." PLoS Med 2: e298.
Bailey, R. C., Muga, R., Poulussen, R., and Abicht, H. (2002). “The acceptability of male circumcision
to reduce HIV infections in Nyanza Province, Kenya. AIDS Care, 14(1), 27–40.
Bailey, Robert C.; Stephen Moses, Corette B Parker, Kawango Agot, Ian Maclean, John N Krieger,
Carolyn F M Williams, Richard T Campbell, Jeckoniah O Ndinya-Achola (2007). "Male
circumcision for HIV prevention in young men in Kisumu, Kenya: a randomised controlled
trial". The Lancet (London: Elsevier) 369 (9562): 643–656.
BBC (2007). “Ugandan men getting circumcised”. London: British Broadcasting Company (BBC),
2007, http://news.bbc.co.uk/2/hi/africa/6227533.stm.
Bennell, P. (2003). "The Impact of the AIDS Epidemic on Schooling in sub-Saharan Africa."
Background paper for the Biennial Meeting of the Association for the Development of Education
in Africa.
Bertrand, M. and Mullanaithan, S. (2001). “Do People Mean What They Say? Implications for
Subjective Survey Data.” Economics and Social Behavior. Vol. 91(2) pp. 67 – 72.
Blomquist, G.C. (1982) “The Regulation of Motor Vehicle and Traffic Safety.” Boston, MA: Kluwer
Academic.
Boily, Marie-Claude, Rebecca F Baggaley, Lei Wang, Benoit Masse, Richard G White, Richard J
Hayes, Michel Alary. (2009). “Heterosexual risk of HIV-1 infection per sexual act: systematic
review and meta-analysis of observational studies”. The Lancet Infectious Diseases, Volume 9,
Issue 2, Pages 118 – 129.
Boozer, M.A. and Philipson, T.J. (2000) “The Impact of Public Testing for Human Immunodefixiency
Virus.” Journal of Human Resources, Vol. 31(3) pp. 419 – 446.
Bowa, K. and M. Lukobo (2006). "Male Circumcision and HIV in Zambia." East and Central African
Journal of Surgery 11(2): 66-71.
30
Bracher, Michael and Gigi Santow, Susan Cotts Watkins. 2004. “Assessing the Potential of Condom
Use to Prevent the Spread of HIV: A Microsimulation” Studies in Family Planning, Vol. 35, No.
1 (Mar., 2004), pp. 48-64
Braun, C., Fouts, J., (1998). Behavioral response to the presence of personal protective equipment. Hum.
Factors Ergon. Soc. 2, 1058–1063.
Cassell, M. M., D. T. Halperin, et al. (2006). "Risk compensation: the Achilles' heel of innovations in
HIV prevention?" BMJ 332(7541): 605-607.
Chesney, M.A. et al. (1997). “Risk Behavior for HIV Infection in Participants in Preventavie HIV
Vaccine Trials: A Cautionary Note.” Journal of Acquired Immunodeficiency Syndrome. Vol.
16(4) pp. 266 – 271.
Chinkumba, J., Godlonton, S, and Thornton, R. (2012) “The Demand for Male Circumcision” Working
Paper: University of Michigan.
Clark, Shelley, Caroline Kabiru, and Eliya Zulu. “Do Men and Women Report Their Sexual
Partnerships Differently?: Evidence From Kisumu, Kenya.” Forthcoming, International
Perspectives on Sexual and Reproductive Health.Cohen, Alma and Einav, Liran. (2003). “The
Effects of Mandatory Seat Belt Laws on Driving Behavior and Traffic Fatalities.” Review of
Economics and Statistics, Vol. 85, pp. 828-843.
Crain, C. (2005). "Chicken Little scare tactics won't stop HIV." Washington Blade.
Crepaz, N. et al (2004) . “Highly active antiretroviral therapy and sexual risk behavior: a meta-analytic
review” JAMA Vol. 292 pp. 224 – 236.
Crowley, I. P. and K. M. Mesner (1990). "Ritual circumcision (Umkhwetha) among the Xhosa of the
Ciskei." British J of Urology 66: 318-21.
Daoreung, P. (1997). "THAILAND-AIDS: No More Scare Tactics." InterPress News Service.
Delavande, A. et al. (2011). “Measuring Subjetive Expectations in Developing Countries: A Critical
Review and New Evidence.” Journal of Development Economics. Vol. 94(2) pp. 151 – 163.
de Paula, Aureo, Shapira, Gil and Petra Todd. (2010). “How Beliefs About HIV Status Affect Risky
Behaviors: Evidence from Malawi, Second Version”. PIER Working Paper No. 10-004.
de Walque, Damien, Harounan Kazianga, and Mead Over. (2010). “The Impact of HAART Awareness
on Risk Behavior: Evidence from Mozambique. Working Paper.
Dickie, Mark and Shelby Gerking (1997). “Genetic Risk Factors and Offsetting Behavior: The Case of
Skin Cancer” Journal of Risk and Uncertainty, 1997, vol. 15, issue 1, pages 81-97.
Doyle, D. (2005). "Ritual Male Circumcision: A Brief History." J R Coll Physicians Edinb 35: 279-285.
Drain PK, S. J., Hughes JP, Halperin DT, Holmes KK (2004). "Correlates of national HIV
seroprevalence: An ecologic analysis of 122 developing countries." J Acquir Immune Defic
Syndr 35: 407-420.
Dupas, P. (2011). “Do Teenagers Respond to HIV Risk Information? Evidence from a Field Experiment
in Kenya”. American Economic Journal: Applied Economics 3 (1), pp.1-36.
Engelberg, J., C. Manski and J. Williams (2006). “Comparing the Point Predictions and Subjective
Probability Distributions of Professional Forecasters.” NBER Working Papers 11978.
Evans, L. (1985) “Human Behavior Feedback and Traffic Safety”, Human Factors 27, pp. 555-556.
31
Evans, L. and Graham, J.D. (1991) “Risk Reduction or Risk Compensation? The Case of Mandatory
Safety-Belt Use Laws”. Journal of Risk and Uncertainty, Vol 4. pp. 61 - 73
Fenton, K. A., A. M. Johnson, et al. (2001). "Measuring sexual behaviour: methodological challenges in
survey research." Sexually Transmitted Infections 77: 84-92. Francis, A. M. (2007). "The
Economics of Sexuality: The Effect of HIV/AIDS on Homosexual Behavior, Desire, and Identity
in the United States."
Gan, L., G. Gong, M. Hurd and D. McFadden (2004). “Subjective Mortality Risk and Bequests.” NBER
Working Paper 10789.
Godlonton, Susan and Rebecca Thornton. Forthcoming. “Peer Effects in Learning HIV Results”. Journal
of Development Economics.
Goldstein, Markus, Zivin, Joshua, Habyarimana, James, Pop-Eleches, Christian, and Thirumurthy,
Harsha. (2008). “ Health Worker Absence, HIV Testing and Behavioral Change: Evidence from
Western Kenya.” Working Paper.
Gong , E. (2011) “HIV Testing and Risky Sexual Behavior.” Working Paper: Middlebury College.
Gray, R.H.; et al. (February 2007). "Male circumcision for HIV prevention in men in Rakai, Uganda: a
randomised trial". Lancet 369 (9562): 657–666.
Guadagno, R. E. and R. B. Cialdina (2002). "Online persuasion: an examination of gender differences in
computer mediated interpersonal influence." Group Dynamics: Theory Research and Practice
6(38-51).
Guadagno, R. E. and R. B. Cialdina (2007). "Persuade him by email, but see her in person: Online
persuasion revisited." Computers in Human Behavior 23: 999-1015.
Guest, G. et al (2008) “Changes in sexual risk behavior among participants in a PrEP HIV prevention
trial. Sexually Transmitted Disease. Vol. 35. Pp. 1002 – 1008.
Halperin, D. E. A. (2002). "Response to Ronald Gray, Male circumcision and HIV acquisition and
transmission: cohort studies in Rakai, Uganda." AIDS 16(5): 810-812.
Hamermesh, D. (1985). “Expectations, Life Expectancy, and Economic Behavior.” Quarterly Journal of
Economics 50(2): 389 - 408.
Johnson, M. (1999) “Belief of vaccine receipt in HIV vaccine trials: Further Cautions”, Journal of
Acquired Immune Deficiency Syndromes and Human Retrovirology, 21 pp. 413.
Kajubi, P. et al (2005). “Increasing Condom Use Without Reducing HIV Risk: Results of a Controlled
Community Trial in Uganda.” Journal of Acquired Immune Deficiency Syndrome, Vol. 40, No.
1.
Kaler, A. (2004). "AIDS-talk in everyday life: the presence of HIV/AIDS in men's informal
conversation in Southern Malawi." Social Science & Medicine 59: 285-297.
Kalichman S, Eaton L, Pinkerton S, 2007 Circumcision for HIV Prevention: Failure to Fully Account
for Behavioral Risk Compensation. PLoS Med 4(3): e138. doi:10.1371/journal.pmed.0040138
Kaplan, C. “Risk compensation and treatment for high cholesterol,” 2010. [Online]. Available:
http://www.econ.ucsb.edu/~kaplan/Research_files/Kaplan_StatinPaper.pdf.
Kapoor, S. (2008) “The HPV vaccine and behavioral disinhibition.” Journal of Adolescent health. Vol
42: pp. 105.
Kebbabetswe, P., S. Lockman, et al. (2003). "Male circumcision: an acceptable strategy for HIV
prevention in Botswana." Sex Transm Infect 79: 214-219.
32
Kohler, H.P. and Thornton, R.L. (2011). “Conditional Cash Transfers and HIV/Aids Prevention:
Unconditionally Promising?” World Bank Economic Review, forthcoming.
Khwaja, A., F. Sloan, and S. Chung. (2007) ‘The relationship between individual expectations and
behaviours: mortality expectations and smoking decisions,’ Journal of Risk and Uncertainty 35,
179–201
Kunda, Z. (1990) “The Case for Motivated Reasoning.” Psychological Bulletin Vol. 108(2) pp. 408 –
498.
Jeffrey R Kling & Jeffrey B Liebman & Lawrence F Katz, 2007. "Experimental Analysis of
Neighborhood Effects," Econometrica, Econometric Society, vol. 75(1), pages 83-119, 01
Lagarde, E., T. Dirk, et al. (2003). "Acceptability of male circumcision as a tool for preventing HIV
infection in a highly infected community in South Africa." AIDS 17: 89-95.
Levine, S. and F. Ross (2002). "Perceptions of and attitudes to HIV/AIDS among young adults at the
University of Cape Town." CSSR Working paper 14.
Lo, B. (2006) “HPV vaccine and adolescents’ sexual activity”. BMJ Vol. 332, pp. 1106 – 1107.
Madajewicz, Malgosia, Alexander Pfaff, Alexander van Geen, Joseph Graziano, Iftikhar Hussein,
Hasina Momotaj, Roksana Sylvi and Habibul Ahsan. (2007). “Can information alone change
behavior? Response to arsenic contamination of groundwater in Bangladesh”. Journal of
Development Economics. Volume 84, Issue 2, November 2007, Pages 731-754.
Manski, C. and J. Dominitz (1994). “Using Expectations Data to Study Subjective Income
Expectations.” Econometrics.
Mattson CL, Campbell RT, Bailey RC, Agot K, Ndinya-Achola JO, et al. (2008) Risk Compensation Is
Not Associated with Male Circumcision in Kisumu, Kenya: A Multi-Faceted Assessment of Men
Enrolled in a Randomized Controlled Trial. PLoS ONE 3(6): e2443.
Marck, J. (1997). "Aspects of male circumcision in sub-equatorial African culture history." Health
Transition Review Supplement to Volume 7: 337-359.
McNeil, Donald G. (2006). “Circumcision Halves H.I.V. Risk, U.S. Agency Finds”. New York Times.
December 14, 2006.
Miguel, E. and M. Kremer (2004). "Worms: Identifying Impacts of Education and Health in the
Presence of Treatment Externalities." Econometrica 72(1): 159-217.
Molden, D.C. and Higgins, E.T. (2005). “Motivated Thinking.” In Holyoak, K.J. and Morrison, R.G.
eds., The Cambridge Handbook of Thinking and Reasoning. New York: Cambridge University
Press.
Namangale, Frank. (2007). “Approach Male Circumcision with Caution”. Malawi Daily Times. March
9, 2007.
National Statistical Office (NSO) [Malawi], and ORC Macro. 2005. Malawi Demographic and Health
Survey 2004. Calverton, Maryland: NSO and ORC Macro.
National Statistical Office (NSO) [Malawi], and ORC Macro. 2011. Malawi Demographic and Health
Survey 2010. Calverton, Maryland: NSO and ORC Macro.
Ngalande, R. C., J. Levy, et al. (2006). "Acceptability of Male Circumcision for Prevention of HIV
Infection in Malawi." AIDS and Behavior 10(4): 377-385.
Nikolov, P. (2011). “Does AIDS Treatment Stimulate Negative Behavioral Response? A Field
Experiment in South Africa”. Harvard University. Working Paper.
33
NIH (2006). "Adult Male Circumcision Significantly Reduces Risk of Acquiring HIV." National
Institutes of Health.
Nnko, S., R. Washija, et al. (2001). "Dynamics of Male Circumcision Practices in Northwest Tanzania."
Sexually Transmitted Diseases 28: 214-218.
Nyhan, B. and Reifler, J. (2010) “When Corrections Fail: The Persistence of Political Misperceptions”,
Political Behavior 32(2): 303-330.
Oster, E. (2007). "HIV and Sexual Behavior Change: Why not Africa?" University of Chicago Working
Paper.
Peltzman, S. (1975) “The Effects of Automobile Safety Regulation.” Journal of Political Economy. Vol.
83(4). Pp. 677 – 726.
Philipson, T. (2000). "Chapter 33: Economic epidemiology and infectious diseases." Handbook of
Health Economics. Vol 1, Part B. pp. 1761 – 1799.
Plamen Nikolov. 2011. “Does AIDS Treatment Stimulate Negative Behavioral Response? A Field
Experiment in South Africa.” Working Paper.
PlusNews. (2010). “Tracking the male circumcision rollout”. Available at:
http://www.irinnews.org/Report.aspx?ReportId=88286
ReThinkHIV (2012). Priorities: Georgetown University Expert Panel.
http://www.rethinkhiv.com/experts/113-rethinkhiv-expert-panel-outcome. Accessed January 30,
2012
Rain-taljaard, R. C., E. Lagarde, et al. (2003). "Potential for an intervention based on male circumcision
in a South African town with high levels of HIV infection." AIDS Care 15(3): 315-327.
Robinson, J. and Yeh, E. (2011) “Transactional Sex as a Response to Risk in Western Kenya” American
Economic Journal: Applied Economics. Vol. 3(1) pp. 35 – 64.
Salm, Martin. (2010). “Subjective mortality expectations and consumption and saving behaviours
among the elderly”. Canadian Journal of Economics/Revue canadienne d'économique Volume
43, Issue 3, pages 1040–1057.
Schoofs, M, Lueck, S, & Phillips, MM. (2005, July 05). “Study says circumcision reduces AIDS risk by
70%”. The Wall Street Journal, p. A1.
Schunk, D. (2005). “Subjective Life Expectancy, Savings Motives and Savings Behavior: Evidence
from the German SAVE Survey.” Working Paper, Mannheim Research Institute for the
Economics of Aging.
Smith, V.K. et al (1995)” Do risk information programs promote mitigating behavior?” Journal of Risk
and Uncertainty. Vol. 10(3) pp. 203 – 221. Stannus, H. S. and J. B. Davey (1913). "The Initiation Ceremony for Boys Among the Yao of
Nysaland." The Journal of the Royal Anthropological Institute of Great Britain and Ireland 43:
119-123.
Stein, J. (1999). "Scare-tactics don't work with HIV/AIDS." Sunday World News.
Stoler, A. (2004). “Mortality risk and economic behavior.” University of Chicago Mimeo.
Stolte, I.G. et al. (2001). “Increase in Sexually Transmitted Infections among Homosexual men in
Amsterdam in relation to HAART.” Sexually Transmitted Infections, Vol. 77. Pp 184 – 186.
34
Szabo, R. and R. V. Short (2000). "How does male circumcision protect against HIV infection?" BMJ
320: 1592-1594.
Tarozzi, A. et al. (2011). “Birght Lines , Risk Beliefs, and Risk Avoidance: Evidence from a
Randomized Intervention in Bangladesh.” Economic Research Initiatives at Duke Working
Paper no. 77.
Tethani, Raphael. (2010). “Malawi Rules Out Circumcision for AIDS Prevention”. Associated Press.
Thornton, R. L. (2008). "The Demand for and Impact of Learning HIV Status: Evidence from a Field
Experiment." American Economic Review. 98(5): 1829 – 63.
Thornton, R.L. (2012). “HIV Testing and Economic Behavior.” Working Paper: University of Michigan.
Tunikekwathu News 10 October 2011 (p.8)
UNAIDS (2007). “Country Situation Analysis: Malawi” available at www.unaids.org.
United Nations (2011). http://www.un.org/apps/news/story.asp?NewsID=40619
USAID (2007). "USAID/WHO AIDS Epidemic Update."
USAID (2009). “The Potential Cost and Impact of Expanding Male Circumcision in Malawi”. Health
Initiative Policy. Available at:
http://www.malecircumcision.org/programs/documents/Malawi103009.pdf .
Valley, K., J. Moag, et al. (1998). "'A matter of trust': Effects of communication on the efficiency and
distribution of outcomes." Journal of Economic Behavior and Organization 34: 221-238.
Valley, K., L. Thompson, et al. (2002). "How Communication Improves Efficiency in Bargaining
Games." Games and Economic Behavior 38: 127-155.
Walker, Ian. (2007). Drivers overtaking bicyclists: Objective data on the effects of riding position,
helmet use, vehicle type and apparent gender. Accident Analysis & Prevention Volume 39, Issue
2, March 2007, Pages 417-425.
WHO (2007). “WHO and UNAIDS announce recommendations from expert consultation on male
circumcision for HIV prevention”. Press release available at:
http://www.who.int/hiv/mediacentre/news68/en/index.html.
WHO (2011), Joint United Nations Programme on HIV/AIDS (2010) Progress in male circumcision
scale-up: country implementation and research update. Geneva: World Health Organization.
Available: http://www.malecircumcision.org/documents/MC_country_June2010.pdf. Accessed
25 August 2011.
Wilde, G.J.S. (1982). “The Theory of Risk Homeostasis: Implications for Safety and Health,” Risk
Analysis 2, pp. 209 – 225.
Williams-Avery R.M.; MacKinnon D.P. (1996). “Injuries and use of protective equipment among
college in-line skaters.” Accident Analysis and Prevention, Volume 28, Number 6, November
1996 , pp. 779-784(6).
Wilson, N., W. Xiong and Mattson, C.L. (2011). “Is Sex Like Driving? Risk Compensation Associated
with Male Circumcision in Kisumu, Kenya”. Williams College Working Paper.
Zaller, J. and Feldman, S. (1992). “A Simple Theory of the Survey Response: Answering Questions
versus Revealing Preferences”, American Journal of Political Science, Vol. 36, No. 3, pp. 579616
35
Zimet, G.D. et al. (2006) “Psychosocial aspects of vaccine acceptability”. Vaccine Vol 24(3) pp. S201 –
S209.
36
Figure 1: Baseline and Follow-up CDF of Perceived Circumcised Men Transmission Rate
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0
10
20
30
Baseline
40
50
60
Treatment
70
80
90
100
Control
Notes: This figure presents the distribution of the question at baseline and follow-up survey: “If 100
circumcised men each had unprotected sex with a woman who was HIV positive last night, how many of
them would become infected?”
37
Table 1: Sample Size and Baseline Characteristics
Panel A: Male Sample Description
Villages
Total
Respondents:
Panel B: Summary Statistics
Demographics: Age
Married
Years of Education
Circumcised
Literate in Chichewa
Literate in English
Tribe:
Chewa
Lomwe
Nyanja
Yao
Religion:
Christian
Muslim
Wealth:
Income (logged)
Assets
Farmer
Salaried
Self-Employed
Sexual Behavior Age at sexual debut
Had sex in the last month
Number of sexual partners across lifetime
Number of sexual partners in last 12 months
Ever used a condom
Fraction of safe sex encounters in past month
All
70
1228
Treatment
35
609
Control
35
619
All
31.779
0.896
5.878
0.737
0.736
0.328
0.050
0.169
0.409
0.355
0.539
0.389
9.133
4.436
0.621
0.132
0.357
17.116
0.755
4.294
1.115
0.401
0.113
Treatment
31.718
0.903
5.629
0.745
0.731
0.315
0.051
0.207
0.401
0.330
0.553
0.383
9.129
4.207
0.621
0.094
0.371
17.211
0.744
4.239
1.078
0.388
0.103
Control
31.838
0.889
6.123
0.728
0.742
0.341
0.050
0.131
0.417
0.380
0.525
0.396
9.136
4.661
0.621
0.170
0.342
17.022
0.765
4.348
1.151
0.414
0.122
Difference
-0.121
0.015
-0.494 *
0.017
-0.011
-0.026
0.001
0.076
**
-0.017
-0.050
0.028
-0.013
-0.007
-0.454 ***
0.000
-0.076 ***
0.029
0.189
-0.021
-0.109
-0.073
-0.026
-0.018
Notes:
The main sample consists of 1,228 men residing in Zomba district in Malawi who agreed to participate in the baseline survey
conducted in 2008.
Table 2: Follow-up Sample
Panel A: Follow-up Sample
Treatment
(1)
484
Respondents:
Finding rate
0.795
Panel B: Correlates of Finding Rate
Treatment
Treatment*Circumcised
Circumcised
Constant
Additional controls?
Observations
R-squared
Control
(2)
469
0.758
Difference
(3)
0.037
Interviewed in 2009
(1)
(2)
0.034
0.029
[0.032]
[0.050]
0.015
[0.049]
-0.043
[0.039]
0.828***
0.879***
[0.158]
[0.165]
Yes
Yes
1228
1216
0.010
0.010
Notes:
Robust standard errors clustered by village, additional controls used in Panel B include: age,
marital status, years of education, assets, logged income, whether they had sex in the past week (at
baseline), ethnicity dummies (Chewa, Nyanja, Lomwe and Yao), and an indicator for religious
denomination; exposure to HIV messages on the radio and TV as measured by number of
messages heard in past month relating to HIV/AIDS.
Table 3: Impact of Information on Sexual Behavior
Treatment
Treatment * Circumcised
Circumcised
Constant
Observations
R-squared
Joint test of significance:
Treatment + Treatment *
Circumcised = 0
Average of dep var
Wife
Pregnant
(1)
-0.031
[0.051]
0.004
[0.056]
-0.02
[0.041]
0.384**
[0.175]
871
0.02
# condoms
Had sex
# Sex per # condoms Fraction
# condoms received
last
# Sex per month (all used past safe sex in # partners # partners purchased free last
month
month
partners)
month
past month last month last year last month
month
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
-0.071
-1.558*
-1.978**
1.101*
0.087*
-0.144*
0.037
0.902**
1.047
[0.052]
[0.892]
[0.916]
[0.558]
[0.045]
[0.078]
[0.091]
[0.441]
[1.070]
0.083
1.916*
2.268*
-0.824
-0.066
0.239**
0.017
-0.25
-1.461
[0.061]
[1.112]
[1.145]
[0.675]
[0.053]
[0.094]
[0.104]
[0.787]
[1.112]
-0.002
0.101
-0.235
0.820*
0.098**
-0.149**
-0.004
1.084*
1.673*
[0.050]
[0.790]
[0.812]
[0.459]
[0.040]
[0.074]
[0.069]
[0.634]
[0.842]
0.289
-0.429
-0.329
0.895
0.361**
0.385
1.308***
0.347
5.048
[0.180]
[2.632]
[2.794]
[1.331]
[0.179]
[0.319]
[0.479]
[1.983]
[3.284]
937
937
937
712
712
937
937
937
937
0.09
0.04
0.04
0.03
0.05
0.02
0.02
0.02
0.02
Any
RSM
Condoms
(11)
(1)
0.078
-0.176***
[0.055]
[0.059]
-0.027
0.184***
[0.069]
[0.058]
0.055
-0.124***
[0.052]
[0.044]
-0.437** 0.716***
[0.239]
[0.190]
939
937
0.09
0.07
0.340
0.707
0.571
0.633
0.500
0.513
0.270
0.495
0.299
0.559
0.830
0.574
0.150
0.762
7.375
7.677
1.972
0.219
0.869
1.117
1.477
3.386
0.000
0.411
Notes: Robust Standard Errors are clustered by village. Additional controls used include: age, marital status, years of education, assets, logged income, whether they had sex in the past week
(at baseline), ethnicity dummies (Chewa, Nyanja, Lomwe and Yao), and an indicator for religious denomination; exposure to HIV messages on the radio and TV as measured by number of
messages heard in past month relating to HIV/AIDS. RSM is a composite measure of 8 sexual behavior indicators: Wife is currently pregnant; Whether respondent had sex last month;
Number of times had sex last month (all partners); Number of partners in the last month; Number of partners in the last year; Number of condoms used in the past month; Fraction of safe
sexual encounters in the last month; Number of condoms purchased in the last month; Number of condoms received free in the last month. This is measured as the mean of the standardized
value for each of these measures of sexual behavior.
* significant at 10%; ** significant at 5%; *** significant at 1%
Table 4: Effectiveness of Delivery of Information
Study said that circumcised
men are at:
Believes circumcision is
related to:
Perceived HIV transmission rate of:
Circumcised men
Treatment
Constant
Observations
R-squared
Average of dep var
Higher
Risk
Lower
Risk
Equal
Risk
Higher
Risk
(1)
-0.032*
[0.019]
0.293**
[0.129]
947
0.03
0.102
(2)
(3)
(4)
0.071** -0.087*** 0.008
[0.033] [0.029] [0.028]
0.272* 0.288*
0.215
[0.161] [0.148] [0.139]
947
947
947
0.06
0.06
0.02
0.608
0.236
0.135
Lower
Risk
No
impact
on risk
Followup rate
Uncircumcised Men
Change in
Change in
rate
Follow-up
rate
(follow-up rate
(follow-up baseline)
baseline)
(5)
(6)
(7)
(8)
0.069* -0.071** -7.821***
-1.438
[0.039] [0.029] [2.320]
[1.277]
0.793*** -0.067 54.660*** 84.626***
[0.155] [0.129] [13.135]
[5.747]
947
947
938
939
0.05
0.06
0.07
0.02
0.640
0.208
67.030
-14.145
(9)
-6.476**
[2.796]
-26.764
[16.128]
936
0.03
88.575
(10)
-3.155*
[1.683]
-3.522
[8.650]
937
0.02
-2.644
Difference
Ratio
(7)-(9)
(7)/(9)
(11)
-6.360***
[1.986]
-29.959***
[11.175]
938
0.07
-21.529
(12)
0.075
[0.141]
0.124
[0.533]
938
0.02
0.808
Notes: Robust Standard Errors are clustered by village. Additional controls used include: age, marital status, years of education, assets, logged income, whether they had sex in the
past week (at baseline), ethnicity dummies (Chewa, Nyanja, Lomwe and Yao), and an indicator for religious denomination; exposure to HIV messages on the radio and TV as
measured by number of messages heard in past month relating to HIV/AIDS. The wording for the question in Columns (1)-(3) state: "If there were to be a scientific study about HIV,
would it show that circumcised men (relative to uncircumcised men) are at an increased risk of contracting HIV, a decreased risk of contracting HIV, or an equal risk?”. The wording
for the question in Column (7) asks: "The wording of the question was “If 100 circumcised men slept with an HIV positive women last night, how many of them would acquire
HIV?” A similar question was asked in reference to uncircumcised men for Column (9).
* significant at 10%; ** significant at 5%; *** significant at 1%
Table 5: Impact of Information on Circumcisions
Panel A: Circumcision by Treatment Group
Treatment
Constant
Includes Control Variables
Observations
R-squared
Average of dep var
Got circumcised
(1)
(2)
0.021
0.021
[0.017]
[0.019]
0.012
0.037
[0.008]
[0.052]
No
Yes
320
320
0.000
0.040
0.022
0.022
Panel B: Summary Statistics of those who got circumcised
Circumcised in
past year
N=7
(1)
Age
29.714
Married
1.000
Years of Education
6.143
Chewa
0.143
Lomwe
0.143
Nyanja
0.714
Yao
0.000
Income
70.012
Assets
5.286
Farmer
0.857
Self-Employed
0.857
Age at sexual debut
16.833
Had sex in the last month
0.667
Number of sexual partners across lifetime
3.333
Number of sexual partners in last 12 months
1.000
Ever used a condom
0.333
Panel C: Reasons for getting circumcised:
Health
Cultural reasons
Please wife
Reduce HIV/AIDS risk (and STI risk)
Set example for younger relative/child
Treatment
N=5
0.4
0.2
0.2
0.2
0.2
Not
circumcised
N = 312
(2)
32.540
0.863
6.655
0.061
0.284
0.530
0.099
107.375
4.406
0.641
0.312
17.638
0.745
3.804
1.118
0.382
Diff
(3)
-2.826
0.137
-0.512
0.082
-0.141
0.184
-0.099
-37.363
0.880
0.216
0.545
-0.805
-0.078
-0.471
-0.118
-0.049
*
***
***
**
***
*
Control
N=2
0.5
0
0
0
0.5
Notes: Robust Standard Errors are clustered by village. Additional controls used in Panel A column 2 include: age,
marital status, years of education, assets, logged income, whether they had sex in the past week (at baseline),
ethnicity dummies (chewa, nyanja, lomwe and yao), and an indicator for religious denomination; exposure to HIV
messages on the radio and TV as measured by number of messages heard in past month relating to HIV/AIDS.
* significant at 10%; ** significant at 5%; *** significant at 1%
Appendix A: Religion, Ethnicity and Circumcision
Fraction Circumcised
Religion: Christian
Muslim
Ethnicity: Chewa
Lomwe
Ngoni
Nyanja
Yao
Notes:
TA Kuntamanji
2008
N = 1214
(1)
0.606
0.942
MDICP 2008
(Balaka)
N = 391
(2)
0.587
0.994
MDHS 2004
(Zomba, Rural)
N = 167
(3)
0.319
0.906
0.667
0.561
0.400
0.657
0.928
0.720
0.459
0.368
0.737
0.992
0.160
0.370
0.167
0.242
0.830
The TA Kuntumanji sample constitutes all men in the baseline sample conducted in 2008 for which their
circumcision status and religion is known (1216 respondents in 70 different villages) or ethnicity and
circumcision status is know (1,214); there are 21 observations excluded as these respondents did not report
their circumcision status. The MDICP sample (2008) constitues all male respondents in the Balaka district
for which a VCT questionnaire and survey was administered totalling 391 observations. The Malawi DHS
sample constitutes only those men in rural Zomba district totalling 167 respondents.
Appendix B: Balancing Statistics by Circumcision Status
Demographics:
Tribe:
Religion:
Wealth:
Sexual Behavior:
Age
Married
Years of Education
Literate in Chichewa
Literate in English
Chewa
Lomwe
Nyanja
Yao
Christian
Muslim
Income
Assets
Farmer
Salaried
Self-Employed
Age at sexual debut
Had sex in the last month
Number of sexual partners across lifetime
Number of sexual partners in last 12 months
Ever used a condom
T
31.179
0.906
5.484
0.721
0.333
0.045
0.167
0.368
0.413
0.471
0.482
103.771
4.167
0.614
0.099
0.381
17.027
0.752
4.297
1.063
0.397
Circumcised
C
Diff
31.770
-0.592
0.906
0.000
5.728
-0.243
0.739
-0.018
0.344
-0.011
0.045
0.000
0.089
0.078
0.365
0.004
0.480
-0.067
0.420
0.051
0.509
-0.027
109.868
-6.097
4.696
-0.529
0.611
0.004
0.159
-0.060
0.358
0.023
16.809
0.219
0.770
-0.019
4.563
-0.266
1.146
-0.083
0.454
-0.057
**
***
***
*
Uncircumcised
T
C
Diff
33.059
31.946
1.113
0.889
0.844
0.045
6.072
7.168
-1.096 **
0.817
0.796
0.021
0.425
0.491
-0.066
0.072
0.054
0.018
0.320
0.246
0.075
0.503
0.563
-0.060
0.078
0.114
-0.035
0.810
0.808
0.002
0.072
0.096
-0.024
102.272 110.503
-8.230
4.307
4.533
-0.226
0.641
0.651
-0.010
0.078
0.205
-0.126 ***
0.355
0.295
0.060
17.636
17.610
0.026
0.728
0.758
-0.029
3.781
3.808
-0.027
1.125
1.108
0.017
0.399
0.365
0.034
Notes:
The main sample consists of 1,228 men residing in Zomba district in Malawi who agreed to participate in the baseline survey conducted in 2008.