In this lesson, we're going to be talking about how using household surveys and censuses estimate mortality, under five-mortality using direct and indirect methods. With the focus of talking about direct and indirect methods, what's up, strengths and weaknesses of each approach and how they could be used in your evaluation. So let's start with direct measurement of childhood mortality. That generally what happens is that most household surveys, demographic and health surveys use a direct measure of mortality and an indirect measure. But they collect both. It can be used in a census, but rarely. The idea is that the main difference is that to do a direct measure of childhood mortality, you have to have a full birth history of women of reproductive age. Usually that's 15-49. And the idea is that using this approach, you can estimate childhood mortality for periods up to 15 years preceding the surveys. The idea is it doesn't just give you a measure of mortality this year or current, but a kind of hold historic period for 15 years. This is important because when you run these kind of surveys, both with the direct and indirect measure, you can get you can run one survey and get a measure both of in line and baseline within an evaluation. So a single survey at the end could provide estimates of mortality over time. You'll see if you look at demographic and household, write ups of the surveys or if you go to their website that normally they use a five year reporting period to estimate mortality. So we'll discuss that later in the class about why they use five year period, but primarily it's to have a big enough sample size to get a reliable estimate. And that if you're trying to do this for an evaluation, you're going to have to use a larger survey that would allow you get mortality for one or two year period that more accurately reflects the end of or the beginning of your program. Let me explain to you what a full birth history happens. So when you do a household survey, you have questionnaires, a mother's questionnaire where you're going to ask her questions about all kinds of things the services for their children, where she gave birth. But one of the things in a full birth history is what you ask, a mother is to report all live births they've had in their lives. So you take all women between 15-49 and each one has to answer a series of about 10 or 12 questions. So let's just, you can look at this and say, talk about your first birth. What was the name given to your first child? Was this birth, was it twins or triplets? What was the name, what month was born? And then it goes on and ask about alive. And if the child still alive, if the child is dead, when the child died. The idea is you're doing this whole tracking of information and then down below for each, you're trying to get a list of all live births. And then basically of the live births, are they still alive? If they're still alive, fine, if they're not still alive, when did they die? So you're trying to track, get a birth occur, and was it a death or not within the five year period? And you have to do that for all children. So it's a fairly complex measure, and it's also really complex. You'll see this, I've just shown you the top part of the questionnaire, you have to do it for every person. So it's a very difficult and time consuming task. But this is kind of the gold standard, the full birth history to collect information on mortality. Now, let's just give you a little example when you think of a full birth history. So you think about if it's the year 2015 and you're running the survey, would you have a 30 year old woman. She can report information in the past, the previous 10 or 15 years. She go back and say, well if you look at the age 30, they have a birth, did the child die? That she can record almost 15 years or 12 years of information about births and deaths. An older woman whose age 45 gives you a longer history of births and deaths. And then a more recent birth, say a 17 year old woman really only has one birth to report and not a very long period. So the child may still be alive, but the child may be only two. So they're giving you incomplete information. But the idea is that if you collect this information, you can combine it and produce an overall estimate of births and deaths that occurred not only in 2015, but for at least a 15 year period in the past. So this is just giving you an idea again, of saying now, we're kind of combining information. You have women of different ages and different fertility rates and how they combine different information. So you can see nearly all the women are contributing information to what occurred in the last five years, the period around 2010 where when you go back another five years in the past, some of the women that was before they ever had birth. So they don't combine any information about births or deaths. And if you go even further in the past, there's fewer and fewer women. So in fact when you run a household survey, you're going to have women that are 45-49 reporting things up to 25 years in the past. Normally we don't use that to estimate mortality 25 years in the past. because you have bias sample. You only have a few women who were giving births, only the very youngest from those earlier periods. Well, what's been shown that this does a fairly good job, fairly representative and accurate job of estimating births and deaths in the last 15 years. I want to show you a little bit of here some data for under-five trends in Ethiopia using three Demographic and household surveys in the year 2000, 2005, 2011. All of thee are using direct measures and it's showing that the three different surveys have been plotted to show the estimates of mortality for different time periods. The bigger surveys, they actually broke it down into one year time periods. The smaller surveys are often in two or four year periods, sometimes even five year period. What you can see is that if we look back in the year 1995 the three, we actually have three different sources of information. All three surveys provide information related to the mortality rate in 1995. If we go the year 2000, again we still have three surveys providing information roughly around that time period. But if we move down to 2005, you say, now what we have is only the2005 survey gives one small estimates for that year. The 2000 survey provides no information. And then you have information in 2000. So the idea is as you run multiple surveys, you get a better estimate of mortality in the past because you have multiple data points. Please note a couple of things about here. Let's look at the year 1995. If we look at the bottom survey which is from 2005, you see there, they're estimating roughly about 130. Where the survey with the 2000, survey with the triangles, they're saying the estimate is closer to 200 and then the green survey is a little above 200. You actually have a fairly wide range of estimates from three different surveys that all use the same techniques both in sampling and in getting using a full birth history. So even though these are very reliable methods, you're still going to get a fairly widespread between mortality estimates from these kind of surveys. The second approach to estimating under-five mortality in children in surveys is using an indirect estimation method. This uses a summary birth history questions which developed by Brass in 75. And the real approach, this is that they were trying to come up with a much simpler approach, something that could be done more easily within a household survey. It's really based only on three questions you asked a mother, how many children were ever born? Among those who are born, how many died? And then you get the age of the mother or age of their first marriage or union. And you kind of ask yourself, how do you then translate that into mortality? When did it occur? And at what time period? The idea is that they try to link these things according to mothers, age of mother. If you remember back we did had the graph one, how older women contribute deaths over a longer period of time. Young women only contribute deaths in the last four or five years. That they try to use age of mothers to convert these number of children born, number of children died, into probabilities of death during different time periods. The idea is that you use mother's age as a way to translate the information from these three questions into approximate estimates of mortality over different time periods. One of the things about this is that because the 15-19 year old, you can't really use that age group well. Even sometimes there's 19 to 24 because it depends on agent for sex in a country. So that if you're in a country where the average sexual debut is 19 or 20, then the 15-19 year olds are going to be different. They're not going to be representative. So one of the problems with the indirect measure is it's not very good at measuring mortality now, that they don't like to say, well, I can use the youngest children to represent mortality in the last year. Generally, what happens within direct estimates, they assume that it's centered five years in the past. So here give you a little idea of how much simpler it is when you do a summary birth question these from the DHS. That simply instead of the long set of questions and asking about specifics, there is now just a few questions that have to be asked and it allows people to do an estimate of under-five mortality again, giving time trends. But the key is there's two weaknesses of this method. That one, it doesn't give you an estimate of deaths, it gives under-five estimates that doesn't give you information about neonatal deaths or infant deaths simply because you're not collecting any information about what age they were when they died. And the other problem is even with a large sample, you can't use summary birth histories to get estimates of mortality in the past year. Because it's working on the demographics and it's not representative for the one-year period. Give you some examples is how well does it work? Here is indirect estimation of under-five mortality in Malawi. This is the data from childmortality.org. And it shows you different surveys. Now we have censuses. You'll see a census from 87 DHS. DHS by the way, normally does full birth histories but also calculates indirect methods as well. And it's plotted all of the data again showing that there's fairly good consistency but then there's some years around the year 2000 there's quite a bit of spread of data. The blue line is signees estimate of their best estimate of mortality of combining all of the information and using Abbasian approach to kind of smoothing out the noise within the system. So let's just kind of quickly review the advantages and disadvantages of indirect methods. And over on the right you'll see in the slide is kind of giving different information, show how well it responds to different kind of external things happening in the country. Indirect and direct methods, they're generally consistent when they're adjusted for fertility and birth transference. We're not really going to talk about that but the idea is higher fertility rates give you better information, more reliable information. Birth transference relates to mothers. Are they accurately with direct measures? You have to say when it occurred, when the birth occurred, when the death occurred. And there's a tendency that to spread those out to more kind of rough same time period. And you don't have that problem with indirect measures. It's easy, very easy to train surveyors to do correctly with these three questions. It's simple, very fast, very few arrows made by the data collectors. But if you are interested in neonatal or infant mortality, you have to use model life tables. So basically you're estimating under-five mortality and then you're looking at a standard relationship between that and neonatal. So it's really not providing you any direct information of that. The disadvantages is that one of the problems you can have if there's political or economic uncertainty in a country, or if there's a hurricane or some other things that happens. In fact that may not be accurately captured within these surveys, that you'll get different trends. People may move their work a few years ago in Zimbabwe that there's quite a bit of political dislocation and suddenly you found very different estimates of mortality in part because there was a huge 400,000 people in slums around Harare and it had all been torn down. People have been moved out. So being able to find those same people over time and if mothers die or move, they can't report births and deaths in the early time period. Also longer recall period. There can be questions about one we talked earlier with longer recall periods. You can get a less representative sample of mothers who are giving birth to that period. And there could be recall issues for mother accurately reporting date of birth and death of a child 15 years in the past. So I want to talk a little bit about the most common sources of both direct and indirect estimates of mortality. These come from household surveys. And there's kind of two primary groups that run household surveys in low and middle income countries. Or at least funded by external sources, some countries run their own systems, but for most countries the big one and kind of the oldest is demographic and health surveys. DHS it's been sort supported by USAID. They started running these surveys in 1985. They actually were doing originally fertility surveys and they switched over to do more than fertility and they started calling demographic and health surveys. They primarily use the full birth history approach within their survey. It also they have the questions to estimate both the direct and indirect estimates of mortality, but they always use the full birth history. The same organization that does DHS also does AIDS Indicator Surveys and Malaria Indicator Surveys. So again, household surveys, but here they're focused on much kind of smaller domains. Only things about AIDS or only thing about malaria, in many of these surveys now they use the summary birth history to get an indirect estimates of under-five mortality. The second kind of major player in terms of household surveys is UNICEF who has sponsored the MICS survey. The Multiple Cluster Indicator Survey, MICS was basically started to supplement DHS because DHS runs in about 35 or 40 countries. UNICEF wanted the same kind of information from other low and middle income countries. It started in 95, but the first two rounds were very small and weak. But in the last 15 years they've run it every first they were on a five year cycle, now it's on a three year cycle. They've worked very hard to make these surveys kind of equivalent to, or at least consistent with demographic and health surveys. However, they have less money to provide technical support. They primarily do general technical support to the National Statistics Office in the country, but less support than you would get from a DHS. Therefore, one of their approaches has been primarily to use the indirect methods of estimating under-five mortality because it's easier for both in terms of the training and in the actual implementation of the survey.