Insurance, Maternal Care, and Early Childhood Development: Evidence from Kenya

Last registered on May 17, 2021


Trial Information

General Information

Insurance, Maternal Care, and Early Childhood Development: Evidence from Kenya
Initial registration date
May 15, 2021

Initial registration date is when the trial was registered.

It corresponds to when the registration was submitted to the Registry to be reviewed for publication.

First published
May 17, 2021, 10:30 AM EDT

First published corresponds to when the trial was first made public on the Registry after being reviewed.



Primary Investigator

University of Virginia

Other Primary Investigator(s)

Additional Trial Information

On going
Start date
End date
Secondary IDs
A new health center --- Brother Andre Maternal and Child Health Care Center --- recently opened in
Dandora, Kenya, a slum of approximately 150,000 inhabitants in Nairobi, the capital of Kenya.
Dandora is home to the city's largest garbage dump, and many struggle to find and keep consistent
The first goal of the project is to evaluate the impact of the health care center on the population. In
particular, we propose using randomization in the value and price of vouchers to experimentally vary
expectant mothers' propensity to use the health center, and then use this variation to measure the
impact of the hospital itself. Our hypothesis is that the hospital has significant effects on maternal
well-being after the birth, as measure in severity of hemorrhaging and child well-being as measured
by birth-weight.
The second goal of the project is to measure the effect of additional wealth on maternal and child
health outcomes. This is achieved through the randomization, since we will directly observe
households' characteristics and willingness-to-pay directly, but they will report their willingness-topay
before learning the price. This allows us to compare households who end up paying a lower
price to those who pay a higher one, and thereby learn the effect of additional wealth on average.
This provides important evidence for policies like maternal leave or even maternal grants that
financially support pregnancies. Our hypothesis is that the larger the difference is between a
household's stated willingness-to-pay and the the price the end up paying, the healthier the mother
and child will be.
The third goal of the project is to evaluate existing methodologies for understanding insurance
choice. Existing studies interpret data from markets in developed countries through a complicated
framework of statistical and economic assumptions to rationalize observed decisions and outcomes.
Our experimental study will, instead, measure many of these variables directly. This allows us to
apply existing methodologies to our experimental data and determine whether the answers are
correct or not. Our hypothesis is that a number of restrictive assumptions (multivariate logit errors,
ignoring adverse selection, neglecting the subjective beliefs of market participants) are important,
and lead to misleading results. We hope to either verify that these assumptions are problematic and
suggest remedies, or verify that existing approaches actually provide the right answers.

In particular, we want to measure the impact of the introduction of a new, safe, clean hospital to a
population who has previously reported experiencing obstetric violence at the hands of care-givers
and other serious challenges before, during, and after birth. Since we are exploiting experimental
variation to measure this effect, the results will tell us the benefits of introducing similar facilities into
comparable contexts. In many countries, traditional birth attendants and relatively inferior maternal
health clinics are still a challenge to maternal health, and the study will provide evidence about how
improved facilities can lead to better outcomes for mothers and children.
This experimental evidence will be achieved by randomizing the value and prices of vouchers for
care, and offering them to study participants (described in more detail below). This provides
economic data on how people in developing countries cope with the risk of incurring large
expenditures, particularly during a Cesarean section birth. We can then use these very rich data to
evaluate the methodology used by economists and public health specialists when trying to
understand insurance choice in developed countries, and provide a critique and possible remedies.
Since we are adopting a methodology
that allows us to observe household willingness-to-pay for insurance directly, we can provide
evidence on a well-defined and clear health risk (pregnancy), as opposed to a number of previous
studies that focus on crop insurance or general health insurance. The advantage of our study is that
it is health-related and a ubiquitous risk about which women face sincere risk. Unlike crop
insurance, it is a product many are familiar with (Kenya offers national health insurance, which is
unaffordable but desired by many in our targeted population). Unlike insurance experiments in the
US or Taiwan, our targeted population suffers a lack of institutions and outside options available to
those in developed countries. Finally, many households in developing countries hold their wealth in
relationships with other households, who share financial burdens and call in financial favors when
necessary. This makes the experimental evidence more useful for development policy compared to
studies from western countries, where financial assets or credit can be used to pay for care.
External Link(s)

Registration Citation

Johnson, Terence. 2021. "Insurance, Maternal Care, and Early Childhood Development: Evidence from Kenya." AEA RCT Registry. May 17.
Sponsors & Partners

There is information in this trial unavailable to the public. Use the button below to request access.

Request Information
Experimental Details


Intervention Start Date
Intervention End Date

Primary Outcomes

Primary Outcomes (end points)
The main end points of interest are:
1. Maternal health: self-reported basic measures include, were additional returns to the hospital required; did mothers have to be transferred to a more advanced hospital because of complications; how long did it take for mothers to return to a normal level of activity. We ask additional questions about symptoms --- fever, swelling in legs, vaginal hemorrhage or discharge, etc. --- to get additional information about what conditions the women might have been suffering from, and whether those conditions are related to quality of care or sanitary conditions at the hospital they chose.
2. Quality of care received: self-reported measures of cleanliness and sanitation; responsiveness to mothers' requests for pain relief or discomfort; obstetric violence; forced use of beds by multiple mothers during labor or after the birth; treatment of newborns (e.g. use of incubators)
3. Child health: whether, how frequently, and how long breastfeeding took place; weight, height at birth and subsequent ages; mortality (though we are underpowered for any kind of treatment effects on this variable); and various additional diagnostic conditions shortly after birth such as distressed breathing, discolored skin, diarrhea or vomiting.
4. Care decisions: do women select higher quality places to receive care, and how does the expected price of care shift this decision. In particular, do women select to give birth at home and opt out of the formal health care system.

As this is is not a standard design with a treatment arm and a control arm but does allow selection into treatment via the BDM mechanism, we plan to use an instrumental variables approach to analyze these outcomes as well as more structural approaches.

Define the redemptive value of the voucher as the face value minus the amount paid by the household: the net benefit of getting the voucher through the BDM mechanism. To measure the effect of getting care at this specific hospital, we will run a first stage including the voucher's price and redemptive value and measures of individual/household wealth on the decision to use the maternal center, and then regress mother and child health outcomes on fitted values and the same set of covariates in the second stage. This controls for household wealth but the random price and redemptive value experimentally shifts, conditional on x, the propensity of a household to use the maternal center. Since the price and redemptive value are random (randomized on enumerator tablet or through the BDM experiment), they satisfy the exclusion restriction. Whether they are sufficiently strong instruments will be determined by the outcome of the experiment.

Similarly, to analyze the impact of quality on outcomes, we gather data on community perceptions ("If the price of delivery were zero at each hospital, how would you rank them") versus revealed preferences ("The last time you gave birth, where did you go? How much did you pay?"). This allows us to rank the available options, and see how the ranking correlated with questions about the sanitary conditions, quality of doctors and staff, and other questions. We can then substitute the decision of whether the participant goes to the particular maternal health center with a quality index or ranking in the first stage, and then regress health and economic outcomes in the second stage on the predicted quality of care from the first stage.
Primary Outcomes (explanation)
We will look for heterogeneity in treatment effects by interacting with baseline variables, particularly whether the subject is the head of household or self-reported measures of influence with the head of household; their own financial status (income, ownership of a business). We do not construct variables like "empowerment", but instead use observable household roles like HoH or measures of financial independence (owning and operating her own business, private wealth and income) to understand what gives women more choice and influence over their care decisions.

Secondary Outcomes

Secondary Outcomes (end points)
In terms of mechanisms, we are interested in a number of secondary outcomes or channels through which treatment effects might be operating:

Household stress and bargaining: did vouchers reduce divorce rates and self-reported conflict within the household?
Investments in child health and education: did the money saved on delivery allow for more investment in early childhood development, particularly nutrition and education?
Secondary Outcomes (explanation)

Experimental Design

Experimental Design
We randomize the redemptive value and prices of vouchers for treatment at a specialized maternal and child care facility, and elicit demand for the vouchers from participants using the Becker-DeGroot-Marschak mechanism. This provides economic data on how people in developing countries cope with the risk of incurring large expenditures, particularly when a Cesarean section is a possibility. This also provides data on how women trade off the price with the quality of care, determining what facility they give birth at or whether they elect a home birth.

Experimental Design Details
We will offer access to insurance to a random sample of women of child-bearing age in and around
Dandora, and use randomization in prices to determine the impact of the insurance and the health
center. Insurance will be offered during a baseline interview. At the endline, height-for-age and
weight-for-age measures of child health along with other basic indicators will be collected in addition
to a survey.
I will now describe the voucher distribution and demand elicitation procedure. The vouchers are
insurance contracts: respondents who receive a voucher will pay a 20 - 40 percent deposit within a
week of the survey, and the rest of the cost will be deducted from the face value of the voucher. The
deposit is a premium, and the price of a birth (normal or cesarean) less the remaining face value of
the voucher is a co-pay. We want to see how these respondents value the insurance contracts
presented to them. However, if we simply presented them with a fixed, take-it-or-leave-it price and
asked them to tell us whether they wanted to buy or not, we would only learn whether or not the
contract is acceptable at that price, not what the respondent's actual value is for the contract; we
would then have to make a variety of unrealistic statistical assumptions, such as, for example, that
the unobserved component of the respondent's value is normally distributed. So we want the
respondent to make us an "offer" for a contract with a given face value and deposit. Suppose we
said, "We're going to ask you to make us an offer for this voucher. The voucher has a face value of
KES X and requires a deposit of Y percent of your offer. Whatever you answer will be the price, and
KES X and requires a deposit of Y percent of your offer. Whatever you answer will be the price, and
you have to pay the deposit within a week by mPesa (a very popular mobile money program offered
by Safaricom)." The respondent has a best strategy of answering zero and paying nothing: this is an
even worse demand elicitation procedure, since it gives respondents a clear incentive to understate
their values. In situations like this, economists instead use the Becker-deGroot-Marschak (BDM)
demand elicitation method. It recognizes that when the price paid is out of the respondent's control -
-- as with the take-it-or-leave-it prices --- they behave in a fashion consistent with their true values,
but we want them to report a value, as in the pay-as-reported demand elicitation method that
performs so poorly. In short, BDM works like this: "I have this voucher for KES X that requires a Y
percent deposit. I am going to ask you for an offer, and then I am going to ask you draw a price out
of this bag. If the price is above your offer, you do not get the voucher. If the price is below your
offer, you can buy the voucher at that price. You have to pay a deposit of Y percent of the price
today, and the rest will be deducted from the face value of the voucher. So you will never pay more
than what you offer, and you can end up paying significantly less." The intuition is that if I
deliberately understate my value, I can only lose in the scenario when the price drawn is between
my stated offer and my value, so I lose when it would have been profitable to win; similarly, if I
deliberately overstate my value, I might win in the scenario when the price drawn is between my
stated offer and my value, so I win when I would have preferred to lose. This gives participants a
weakly dominant strategy to set their offer equal to their true value. The face value will be randomly
selected from $30, $50, $100, $150, $200, and $300. For reference, the price of a normal birth at
the clinic is $50, and the price of a cesarian is $300, so that we are offering a range of different risk
levels. The deposit will be randomly selected from 20, 30, or 40 percent. For statistical reasons,
variation in the deposit tells us more about the respondent's risk preferences than varying the value
value, but for economic reasons including full insurance and some under-insurance even for a
normal birth provides information about how various insurance schemes would be received by the
The BDM procedure has a few consequences to mention. First, some people will not get a voucher
despite reporting a high offer, while others will get a voucher despite making a low offer. This is the
experimental variation that we want. The strategy is that winning a voucher will push people who
are "on the fence" about going to the new hospital into going to the hospital, and they can be
compared with people of similar observables and offers, but who did not get the voucher. This is
similar to a standard randomized controlled trial, where some subjects get a treatment and others do
not. The exact statistical methodology is called an instrumental variables regression, or two-stageleast-
squares: we use controlled variation in the likelihood of using the hospital to predict hospital
utilization, and then infer hospital impact from utilization. Second, we cannot deviate from the BDM
procedure ex post and provide vouchers even if the reported offer was below the randomly drawn
price. This undermines the demand elicitation procedure because people might reveal to neighbors
or relations that we sell the voucher afterwards regardless of the outcome; everyone then also has a
weakly dominant strategy to report a price of zero, observe the outcome of the price randomization,
and then buy if the price is favorable.

Deviating from the BDM procedure and selling the voucher ex
post at the randomly drawn price is also not in standard experimental practice in economics:
misleading participants or misrepresenting experiments disqualifies research at most journals.
The reason is that it encourages participants to "meta-reason" about the
interaction and undermines their trust in experimenters in general. Thus, instead of measuring how
health uncertainty determines willingness-to-pay for insurance, we are measuring the subject's
uncertainty about our intentions and what the outcome of the experiment is. Economics as a
discipline tries to avoid introducing this kind of meta-uncertainty into subject pools since it has
negative consequences for all economics experiments. During a trip to Kenya, we piloted this kind
of procedure for simpler goods like bags of sugar and tea, and participants understood relatively
quickly how it worked, especially after doing one or two practice rounds, which are part of the

In addition to the BDM procedure, we have a simple procedure for eliciting beliefs about the risk of a
cesarian and the likelihood of becoming pregnant. The procedure involves a sheet of paper with two
boxes and ten beans or other counters. One box is labelled "Cesarian Section Birth" and the other
is labelled "Normal Birth". The enumerator begins by saying, "If you think you will have a normal
birth for sure, place all the beans in the Normal Birth box", and does so. Then she says, "If you think
you will have a Cesarian section birth for sure, place all the beans in the Cesarian Section Birth
box," and does so. Then she says, "If you think a normal birth and a Cesarian section birth are
equally likely, put the same number of beans in each box," and does so. Then she says, "So now,
move the beans back and forth between the two boxes until you think it reflects the chance you need
a Cesarian section." This is also done where normal birth and Cesarian section birth are replaced
with the likelihood of becoming pregnant. This helps us measure the subjective beliefs of the
subjects about the likelihood they become pregnant or require much more expensive care.
In the absence of the vouchers, care at the Health Center requires payment. There are free
government facilities, but these are commonly known to be low quality and stressful places to give
birth. There are several facilities nearby that offer better services for prices comparable to what the
Health Center is charging. The vouchers will therefore deduct from the price of care that households
must pay, but will mostly not fully reimburse the household for care received.
I will now describe follow-up procedures. Women who receive a voucher and deliver at the Health
Center will be indicated in bills received by the hospital, but we will have no indication whether a
woman who does not receive a voucher or who receives the voucher but does not deliver at the
Health Center gave birth. We will return to survey areas at the endline and contact parents to
schedule a survey (we will record GPS coordinates of the place of the initial interview and whether
this is the woman's current residence or place of business). The endline survey will be identical to
the baseline survey, where the current questions about all previous births are restricted to births
since the baseline, and other questions whose answers do not change over time are dropped.
Randomization Method
Randomization on Tablets and a Public lottery: the face value of the voucher and the deposit percentage of the price that must be paid up front are randomized on the tablet at the start of the baseline survey. Study participants are shown the possible prices as laminated pieces of paper, the pieces of paper are dropped into an opaque cloth bag and shaken, and then they select a price themselves. The participant pays the deposit percentage up front, and the rest is deducted from the face value of the voucher. The redemptive value of the voucher is the face value minus the balance after the deposit is paid.
Randomization Unit
The randomization unit is the individual woman in the study: the price each woman faces is randomized.
We record whether women are members of the same household, and will cluster results at the household level.
Was the treatment clustered?

Experiment Characteristics

Sample size: planned number of clusters
7000 women
Each woman is presented with a BDM demand elicitation game, where the prices and generosity of the care being offered are randomized on the enumerator's tablet at the start of the baseline interview.
Sample size: planned number of observations
7000 women Each woman is presented with a BDM demand elicitation game, where the prices and generosity of the care being offered are randomized on the enumerator's tablet at the start of the baseline interview.
Sample size (or number of clusters) by treatment arms
7000 women
Each woman is presented with a BDM demand elicitation game, where the prices and generosity of the care being offered are randomized on the enumerator's tablet at the start of the baseline interview.
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
Supporting Documents and Materials

There is information in this trial unavailable to the public. Use the button below to request access.

Request Information

Institutional Review Boards (IRBs)

IRB Name
University of Notre Dame
IRB Approval Date
IRB Approval Number


Post Trial Information

Study Withdrawal

There is information in this trial unavailable to the public. Use the button below to request access.

Request Information


Is the intervention completed?
Data Collection Complete
Data Publication

Data Publication

Is public data available?

Program Files

Program Files
Reports, Papers & Other Materials

Relevant Paper(s)

Reports & Other Materials