I am familiar with DHS surveys and the way ASFR and TFR are calculated. My question is, would these indicators be inaccurate in the cases when the exact date of birth of women cannot be used for whatever reasons but instead their reported age (in years) is only available?

Thank you!

]]>

Let's say you have the cmc of interview, equivalent to v008. You need v011 (cmc of birth) and all you have is current age, in completed years, which is equivalent to v012. The simplest thing would be to calculate v011_hat=v008-(6+12*v012). That is, you estimate the woman's months of age and subtract from the cmc of interview to estimate the cmc of birth. I have stated this in terms of DHS variables but you may be working with other data.

]]>

I have a further question that follows up from the one above, mainly estimating the TFR as accurately as possible while only having age in years of respondents.

What would be your advice to deal with age heaping that likely occurs on 0s and 5s in age reporting?

Thank you!

]]>

I do not recommend any adjustment for heaping. The woman's current age may be heaped at multiples of 5, and especially multiples of 10, but the rates are based age at the time of the birth and exposure to age within a time interval. Her current age is backdated, and by different amounts for different time intervals. The effect of an adjustment would be very small.

Also we don't really understand the mechanism behind heaping. Usually, it is probably just rounding. But if, say, a tendency to move up (for example from 29 to 30) is more common than a tendency to move down (for example from 31 to 30), then there's a bias. If there is a bias, it could be different at younger ages versus older ages.

Without a good model for the pattern of misreporting, we don't have a basis for adjustments.

]]>