At this point we have sampled, we've stratified, we've sampled our clusters and we've sampled our households. Now I want to talk about how to identify eligible individuals in each sampled household. To identify eligible individuals, the first thing we need to do is actually to define who is eligible and who is not to respond to your household survey. This is going to vary by survey, depends on what you're trying to measure. Generally, eligibility criteria for a questionnaire are defined around by age and by gender. A household survey, as we'll talk about, should always include a household questionnaire. The respondent to the household questionnaire is usually any household member above a certain age, 15 or 18. You may also have a woman's questionnaire, a man's questionnaire may be a child questionnaire, a lesson questionnaire, and for each of your questionnaires, you're going to need to define which household members are eligible to respond to those questionnaires. So define your eligibility criteria. The other thing besides age and gender that you need to think about is whether you just want to include household members. Somebody who usually lives in the household, or whether you also want to include visitors who are generally defined as people who spent the previous night in the household. The general practice in the DHS and the mix is to include both household members and visitors who meet the eligibility criteria. Household members and visitors who meet the eligibility criteria are eligible to respond to the various questionnaires. One question that we get a lot is, should we sample individuals within households? We generally don't recommend doing that. We generally recommend including all eligible individuals in the sampled household. Why? The disadvantage or the reason for not doing this that people often give is, well then you get clustering. There's clustering of individuals within the household and so it's going to increase your variance. That's true, but there is also a substantial sample size advantage. If there are on average two women of reproductive age in each household then if you interview both of them, you have doubled your sample size. Usually, the sample size advantage outweighs the increased clustering that you get. The other downside to including all eligible individuals is that it creates a more complex dataset. It creates hierarchical data sets, where you have multiple women or multiple children nested within a household, and so it becomes a little bit more complicated in terms of data management and analysis, but we can deal with this through appropriate data processing and analysis. How would we actually go about identifying eligible individuals once we have established eligibility criteria when the interviewer enters the household? The first thing they do after greeting the household and administering the consent form is to administer a household member listing, sometimes called a household roster. The main objective of the household roster, one of the main objectives is to identify eligible individuals. Here you see an example of a household roster. This is from a household survey that we did. The structure of the roster here is based on the mix, so it's very similar to the mixed household questionnaire. You can see each row in the listing corresponds to a member of the household or a visitor. We collect some information about that person, their relationship to the head of household, their sex, date of birth, and age, and then did they stay here last night? Then if it's a child, we collect some information about whether their mother is in the household and who their mother is, so we can determine who to interview. You can see columns HL7A and HL7B identify or determine eligibility for women aged 15-49 years and children aged 0-4 years. This was a survey where those were our target populations, and essentially those columns are used to identify those eligible women and children and they all get interviewed or at least attempted to be interviewed. Sometimes, household surveys use screening questions to identify individuals who are eligible to be interviewed. What do I mean by screening questions? I mean, going into a household and asking, are there any children under the age of five years here? Can you tell me who they are? Are there any women aged 15-49 years? Are there any pregnant women? This is a problematic way of identifying eligible individuals for a couple of reasons. The main reason is that people tend to omit eligible members of the household. Now that sounds silly sometimes, how could you forget a member of your household? But this can happen for a number of reasons. Sometimes households are quite large. They may have lots of children and people really may just forget to mention, maybe the youngest child who was just born. Sometimes the person that you're talking to, their concept of who the household includes is different than your definition in the survey. They may not include, for example, household health in the definition where you would have included them. They may not include visitors, they may not include their cousins. Their adolescent cousin who's living as part of the household while they go to school. Doing the household roster allows you to take the respondent through the list of household members in a systematic way to ensure that you are including all household members, including those that are more vulnerable, those that are least likely to be listed or mentioned if you are using screening questions. We really recommend never using screening questions. Don't just ask, are there any kids under five, are there any pregnant women? You should do the household roster even though it is time-consuming. It plays a very important role in ensuring everybody is represented in the survey. Now we've gone through all the steps in sampling an individual in a household survey. We've stratified, we've sampled clusters, we've sampled households, we've identified eligible individuals in those households. Next, we are going to start talking about mapping and going into more detail about how to map and enumerate households within sampled clusters.
