The denominator in a mortality survey is person-time. In the sorts of mortality survey that we do this is usually person-days. It is common to set the recall period at (e.g.) 90 days and assume: (1) Each individual who is alive at the survey date contributes a full 90 person-days to the sample. (2) Each child born during the recall period contributed 45 days (that is half the recall period) to the sample. (3) Each individual that died contributed 45 days (that is half the recall period) to the sample. Parts (2) and (3) are “actuarial corrections” and avoid the need to use individual sums of person-days without introducing bias (under the assumption that people are born and people die pretty evenly throughout the recall period). If we do not correct for inward and outward migration and this is common and the inward and outward figures do not balance each other then we can introduce bias. The simplest way to correct for this is use the “actuarial correction” approach with each migrant adding (e.g.) 45 person-days to the sample. This may not be appropriate if (e.g.) a lot of outward migration occurred in the first 30 days of the recall period. In this case the actuarial correction causes each migrant to contribute 45 person-days to the sample when they probably contributed (on average) only 15 person-days to the sample. If we use too many person-days then we will introduce a downward bias. If we use too few person-days then we will introduce an upward bias. We usually do not bother too much about the exact at of arrival or exit and use the simple (i.e. half-recall period) actuarial correction. If there is little migration we may even ignore the problem. I hope this is of some help.

Just a few thoughts about these important questions: 1. Never use a recall period of exactly a certain number of days. You should follow the advice in the SMART manual: "The beginning of the recall period should always be a date that everyone in the population remembers, e.g., a major holiday or festival (Christmas, beginning of Ramadan, Duvali, etc.), an episode of catastrophic weather, an election, coup, political decree, or similar memorable event." In order to carry out a household census at some past point in time, you must give the respondent a well-remembered date as a memory anchor. The respondent can then think of this specific point in time when telling you who lived in the household at that time. 2. Probably the easiest way to account for in- and out-migration in a household, including births and deaths, is to do two household censuses - one at the beginning of the recall period and one at the time of the interview (the end of the recall period). The population denominator for that household is then the average between these two counts. Of course, as Mark has pointed out, this assumes that in- and out-migration occur evenly throughout the recall period which may not be true in certain sudden catastrophic situations; however, it is a reasonable assumption most of the time if the recall period does not cover too long a time period with too many changes within it. 3. Regarding the specific question of bias introduced by individuals leaving the household then returning within the recall period, this is probably a rare enough occurrence that it would not result in much bias. The most common problem with household break-up producing bias is when some individuals leave the household, and remaining household members do not know whether they are still alive or dead. This occurs in many situations where the household breaks up during the emergency, with some members remaining in dangerous situations. Examples include young men leaving the household to fight (Kosovo) or household members returning to their homes during the acute emergency (Darfur). This could leave many deaths unreported and uncounted.

Thanks Myatt for clarification, very helpful, and will read again and again to understand it better, about two points mentioned by Brad: - The first point: when yes we usually agreed with survey teams during training about the limits of recall period- when it begins and we usually ended it at the day before the survey? and we usually also relate it to local events known by community. - The second point: that HH remaining may not not know about those who left if they are a life or not- this confused me because we normally dont ask if they are alife or not but we ask about the number who left and and rated weather are <5 or adults - but may be we are mistaken? Please if you can advice again will be useful so that we improve the data we are collecting, because these surveys costing alot physically and financially

Yes, you are correct. You take a census of the household at the beginning and end of the recall period. But then for each person who no longer lives in the household, the interviewer must determine if they are currently alive or dead. Often, the interviewer will ask something like "Do you know where George is now?" rather than directly asking "Is George alive or dead?" in order to get this information.