Skip to main content
Search form
  • 00:05

    [Researching the Value of Health Insurance in IndiaUsing a Large Scale Field Experiment]

  • 00:10

    PROFESSOR KOSUKE IMAI, PHD: I'm Kosuke Imai.[Kosuke Imai, PhD, Professor, Government & Statistics,Harvard University] I'm a professor of governmentand statistics at Harvard University.I specialize in developing statistical methodsfor social sciences.My current project is actually randomized evaluationof health insurance program in India.So this is a big project that we areworking with the government of Indiato evaluate how effective to introduce the health insurance

  • 00:34

    PROFESSOR KOSUKE IMAI, PHD [continued]: program to the poor people in India,and then how that's going to improve their healthand financial situation.[How did you become interested in the research?]I got interested in it because, from the methodological pointof views, so I'm interested in developing statistical methods

  • 00:55

    PROFESSOR KOSUKE IMAI, PHD [continued]: for evaluating the effectiveness of policies and programs.And this is one of the--health insurance program-- is the oneof the program that's very important.It affects people's life and not just the young, but oldand every generation.And so it's a high impact policy.And so it was a great opportunity for me

  • 01:15

    PROFESSOR KOSUKE IMAI, PHD [continued]: to use my skills, statistical skills, to do this evaluation.[What was the purpose of the research?]And another nice thing about thisis that research connects to the real impact.So the government is doing this evaluationin order to figure out what's the best

  • 01:37

    PROFESSOR KOSUKE IMAI, PHD [continued]: way to deliver health insurance program to the people in India.So the research output--our research output will have a huge impacton how they actually is going to implement this program.[What research methods were used,and how was data collected?]If you want to evaluate the health insuranceprogram in a scientific way, you have

  • 01:60

    PROFESSOR KOSUKE IMAI, PHD [continued]: to use randomized experiment.So ideally, you want to spread the household into two groups,randomly flipping a coin.And some household will receive the health insurance program,and the other household won't.But doing that is unethical.If you withhold the health insurance programfrom some households, that is unethical thing to do.

  • 02:22

    PROFESSOR KOSUKE IMAI, PHD [continued]: So what we instead did was to randomly encouragesome people to sign up for health insurance,but not deny the right to sign up to other households.So we randomly encouraged some householdsto sign up for this health insurance program.And then, while others, we just didn't say anything.

  • 02:44

    PROFESSOR KOSUKE IMAI, PHD [continued]: So it's a long-term program.So you first assign which householdreceives encouragement to sign up for the health insuranceprogram.And then you follow up later to take some measurementson their health status, their utilization of the hospitalsand medication, and also, the household financial situation.

  • 03:08

    PROFESSOR KOSUKE IMAI, PHD [continued]: Because a lot of poor families, if youdon't have health insurance, you basicallydon't go to the hospitals for minor injuriesor minor disease.But then when you go there, you have a serious healthconditions.So you end up costing a lot more, because there'sno preventive care.So we want to assess how the health insuranceprogram actually protect them from a catastrophic health

  • 03:30

    PROFESSOR KOSUKE IMAI, PHD [continued]: expenditure.So we did a survey to measure some of the output,like a year and two years later.So in this, it's a typical study.You might spread the group into 50% achieving group and 50%control group by randomizing.In this study, we did it a little bit differently.So the reason why we did it differentlyis that we had a sense from the prior study that introducing

  • 03:54

    PROFESSOR KOSUKE IMAI, PHD [continued]: the health insurance program to a small communitywould have a spillover effect.So even if some people didn't receive the health insuranceprogram, they may be affected by the factthat your neighbors or your friendsare receiving this additional assistance.So what we did is, instead of just--in a typical way, where you just assign

  • 04:15

    PROFESSOR KOSUKE IMAI, PHD [continued]: 50% of households to the achieving group and 50%to the control, we vary that proportion.So for some villages, we only assigned 20% of householdto the health insurance program.In other villages, we assign 80% of the householdsto the health insurance program.So in that way, we can compare how

  • 04:38

    PROFESSOR KOSUKE IMAI, PHD [continued]: the number of households around you who receive the treatmentaffects how you react.So you can compare two households--one household where who live in the villagewhere there are very few people got the treatment.And then you can compare with another householdin another village where a lot more people

  • 05:01

    PROFESSOR KOSUKE IMAI, PHD [continued]: receive the treatment.And then see whether there is any difference between the two.So we had this two-stage randomization design,where some villages were assigned to a larger fractionof--some villages, larger fraction of householdswere assigned to the achieving group.And in other villages, smaller fraction of households

  • 05:22

    PROFESSOR KOSUKE IMAI, PHD [continued]: assigned to the achieving group.And that, which village you're going to be in,were also randomized.[What were some unexpected challenges you hadto overcome?]A lot of the difficulties that we faced in the projectis that people talk to each other.So these are literally small villages in India.

  • 05:44

    PROFESSOR KOSUKE IMAI, PHD [continued]: So the community is relatively small.They know each other.So even though we randomly spreadsome households to receive the encouragementabout this new program, and in other householdwere left without any notification,people talk to each other.So it turns out that some people said, "Oh, Igot this encouragement.

  • 06:05

    PROFESSOR KOSUKE IMAI, PHD [continued]: Somebody told me about this new program.It looks great, so I'm going to sign up.Would you want to sign up too?"And so it turns out that some of the control group whoideally you don't want them to sign up for this programended up signing up.The other problem we faced is that whenyou introduce a new program like this into a small community,

  • 06:26

    PROFESSOR KOSUKE IMAI, PHD [continued]: it affects everybody.So it turns out some people having the health insuranceaffects the people who didn't receive the health insurance.So one potential reason could be that, nowthat a lot of people in the village has health insurance,they used hospitals quite a bit.And as a result, it might get overcrowded,

  • 06:47

    PROFESSOR KOSUKE IMAI, PHD [continued]: which reduces the amount of usage for the peoplewho didn't get the health insurance.So the difficulty is that when you do experiment, ideally,people don't talk to each other.They don't interfere with each other.But in the real society, everybody talks to each other,

  • 07:08

    PROFESSOR KOSUKE IMAI, PHD [continued]: and everybody's behavior affects each other.So how to deal with that in a statistical ways is difficult.I think there's so many challenges thatare specific to the local conditions.And this is why this project is conductedby a group of a large number of people, scholars who--

  • 07:31

    PROFESSOR KOSUKE IMAI, PHD [continued]: some of them don't specialize in the study of statistics.You know I'm specializing in the statistical aspectof the study.But others actually study India and the health carepolicy in India, so they really knowwhat's going on on the ground.For example, even in order to make surethat whether the household is actually sign up for the health

  • 07:53

    PROFESSOR KOSUKE IMAI, PHD [continued]: insurance program, you have to actually go visit and checkto see if they actually have the health insurance card.And so that type of process is collectingdata is very difficult. It's not the data comes in spreadsheet.And people don't report.There are certain things that they don'tfeel comfortable reporting.

  • 08:14

    PROFESSOR KOSUKE IMAI, PHD [continued]: And they might be also misreporting,like if you ask about their income or sensitive questionsabout diseases, people may not be forthcoming telling that.So you have to think about creativelyhow to elicit truthful responses while preservingsome privacy so that the respondents feel

  • 08:36

    PROFESSOR KOSUKE IMAI, PHD [continued]: comfortable telling you the truth.[What are some implications of missing data?]Missing data is a very difficult problem.I mean most important thing is try to reduce the missing datawhen you are asking questions.So depending on how you ask questions,that could read the different degree of missing data.

  • 08:58

    PROFESSOR KOSUKE IMAI, PHD [continued]: So if people feel very uncomfortable,if the question is too direct, peoplemay not feel comfortable.In another study, where we study the public opinionin Afghanistan, and there, one of the most important thingwe had to ask was, do you support Taliban?

  • 09:22

    PROFESSOR KOSUKE IMAI, PHD [continued]: So we wanted to know how much support thereis among the public for the Taliban.But it turns out that answering that questionis very, very sensitive.And in Afghanistan, in particular,when you do a survey, you have to do it in a public sphere.So you go into a village and then you ask,can I ask a few questions, they don't let outsiders

  • 09:46

    PROFESSOR KOSUKE IMAI, PHD [continued]: ask questions in private.They want to monitor.So you end up interviewing respondentswhere everybody is surrounding you.So in those situations, if you ask, do you support Taliban?You might not want to say no or a yes.So more than 50% of people will refuse

  • 10:08

    PROFESSOR KOSUKE IMAI, PHD [continued]: to answer that kind of question.So in those situations, you have to think about whatother indirect way where you can still preserve the privacy,protect their anonymity, but try to getsome sense of whether there is any support.[What recommendations would you have for a student lookingto do similar research?]

  • 10:30

    PROFESSOR KOSUKE IMAI, PHD [continued]: So the best advice I would have, if youare interested in getting involved in this type of study,is to get involved.Actually, if you can.In this research group, we actuallyalso had undergraduate students doing some data workto help us do the analysis.And so data work could start from just checking the data

  • 10:52

    PROFESSOR KOSUKE IMAI, PHD [continued]: entry to sophisticated statistical analysis.So if you're a student, you can still find the role.It's a big project.It has a lot of things that has to be done.And so if you find a faculty memberwho is doing interesting project,you can volunteer and say, I wouldlike to be involved and be part of the project.

  • 11:14

    PROFESSOR KOSUKE IMAI, PHD [continued]: Getting involved in a project is a best wayto see how it's done.Participating in discussions and gettinginvolved from the beginning to the data collectionto the analysis.So I recommend that taking class is important,learning the statistical methods is obviously important.But getting involved with the actual project is, I think,

  • 11:39

    PROFESSOR KOSUKE IMAI, PHD [continued]: it's a valuable experience.I recommend that students try to do that.For me, the data science is using the datato answer scientific questions.It's not just analyzing the data,but the main goal is to answer scientific question.And social science has a huge opportunities

  • 11:59

    PROFESSOR KOSUKE IMAI, PHD [continued]: for data scientists.It's a great place where what you do actuallyhave an impact on a society.And to the extent that we humans care about how we live and howwe function within a society.Using the data to inform what's the best thing to dois going to be very important.So I want to encourage students for, maybe not

  • 12:23

    PROFESSOR KOSUKE IMAI, PHD [continued]: so much interested in the methods themselves,methods of data science themselves,but interested in applications, and especially applicationsthat are directly related to us like human behavior.Ideally encourage you to study data science,but think about how you can apply that to answersocial science questions.

  • 12:45

    PROFESSOR KOSUKE IMAI, PHD [continued]: There are too many students, I think,are just interested in method alonebut not necessarily the applications.But without application, I think,the statistical methods won't be helpful.


Kosuke Imai, PhD, Professor of Government & Statistics at Harvard University, discusses his research of the value of health insurance in India using a large-scale field experiment, including his interest in the research; the purpose of, methods used for, and data collection for this research; unexpected challenges; implications of missing data; recommendations for students considering similar research.

Looks like you do not have access to this content.

Researching the Value of Health Insurance in India using a Large-Scale Field Experiment

Kosuke Imai, PhD, Professor of Government & Statistics at Harvard University, discusses his research of the value of health insurance in India using a large-scale field experiment, including his interest in the research; the purpose of, methods used for, and data collection for this research; unexpected challenges; implications of missing data; recommendations for students considering similar research.

Copy and paste the following HTML into your website