From Shea Rutstein: I believe that could be three potential biases: 1. Young women who are in a relationship for a short time may believe that it will last quite a while and therefore give the date of when they started living with him. 2. On the contrary, the young women are not sure that the relationship will last but later on have a baby with the man and therefore later on, say the date when they first met or had sex with the man. 3. For women under age 20, they may not be in a union but when the cohort ages, they choose the heaped age of 18 even though they "married" at a higher age, especially if they are now older and do not remember the age or date well. The strength of each bias may depend on how well age is an important factor in the society, social norms about out of wedlock children and sexual relationships with multiple partners, etc. My suggestions: 1. As was said, track age at first birth for cohorts over the surveys. (Note that age at first union and first birth should include all respondents, not just those ever in union or had a child, otherwise there is a bias). 2. For both age at first union and first birth, check current status report by single year of age to see if there is significant heaping. 3. If there are enough cases, analyze dividing the sample by level of education (none and less than primary complete versus primary complete or more). I would expect more stability in reporting with more education.

From Kerry MacQuarrie: Marriages among the youngest cohorts are under-reported because they are not necessarily construed as marriages in real-time, but those surviving as marriages are retrospectively reclassified as such, as you explain. This would cause a detectable difference in the % for the same birth year estimated from 2 different surveys (1) among the age groups where marriage is concentrated (i.e. younger ages) and/or (2) among age groups for whom relationships are most likely to experience instability early on (i.e. younger ages). This would suggest that the estimate taken later on is more accurate and perhaps should be weighted more. Or perhaps taking a larger age span rather than single year brackets might smooth over some of the inter-survey variation? I'd also point out that there is a variable corresponding to question 801, which disaggregates not in union, living with a man (as if married), and "actually" (formally) married.

From Tom Pullum: I don't think this is a real inconsistency, that is, a data quality issue. I believe these tend to be unions that did not begin with a formal ceremony. The young women and a partner have living arrangements that can either dissolve or become stable over a period of time. In a sense there is a transition during which the woman's "boy friend" becomes her "spouse". I believe that is what is going on--just an evolving definition of whether a relationship is tentative or confirmed. When it is "confirmed", then the start and status are back-dated. You could compare with a setting in which unions do tend to begin with a formal ceremony. I think the pattern would not be observed there. Also, in Benin, you could look at the reporting of age at first birth. If you see the same pattern for age at first birth in successive surveys, then there could indeed be a data quality issue. Thoughts from other forum users would be welcome.