Further information on breastfeeding variables
Unit of measurement
i. Quantification of child's age Most parents will initially 'quantify' a child's age in weeks, but then after a while 're-calibrate' into months, and eventually into years. The usual transition for switching from weeks to months is around three months of age, so that from then on, each month on the ('anniversary') day of the child's birth, the parent will 'age' the child by a further month. That is, a baby is not generally described as four months old until its fourth 'month-birthday' ie. until it has completed four months of life.
ii. Units to be used For some event (eg. weaning) that has occurred at a specific age, the finer the units used in recording the 'age', the more accurate will be the analyses based on that age data. So, if the timing of the event is recorded not in weeks but in months, the fact that a month spans over four weeks will mean, inevitably, that the data value will, for a substantial proportion of cases, differ from the precise truth by up to two weeks. It might therefore seem that the age at which a particular event occurred should always be asked in weeks. However, the advantage of asking 'in weeks' depends on the age being accurately known in weeks, and as noted in (i), after the first three or four months of life it is not usual to continue to quantify a baby's age (recall 'age at' a past event) in weeks. Therefore if asking about an event that occurs about midway through the child's first year of life, then if requiring an answer in weeks (as was the case for starting solids in BC2), it is likely the respondent will have to convert into weeks the recalled age in months.
iii. Facility in converting units Conversion from months to weeks will require some mental arithmetic, which itself might not be an easy matter. A further concern is that even among the arithmetically adept, the relationship between months and weeks is very often misunderstood. This arises because most people assume four weeks per month, by which rule three months is converted to 3x4=12 weeks, whereas it is in fact 13 weeks, and six months to 6x4= 24 weeks, whereas it is 26 weeks. So if a respondent who recalls the age at some event of interest in months, is asked to report it in weeks, he/she will have to convert, and as shown above this is likely to give a figure that is biased down, by two weeks at around six months. This bias will be exacerbated by the tendency described in (i), to think of a child as, say, 'six months old' only when the baby is fully six months old. So, if solids are introduced a week before the six-month 'birthday', when the parent thinks of the child still as 'five months old', then if in a subsequent interview the age at introduction of solids is asked about, the respondent could well convert to 5x4=20 weeks, rather than the 25 weeks that was the baby's exact age.
iv. Flexibility in units that can be reported Duration of breastfeeding (at all, or exclusive), could be reported at interview in days, weeks or months, reflecting the fact that breastfeeding duration might be anything between a few days and over a year. Therefore the need for conversion by the respondent was avoided, and the recalled 'age' can be recorded exactly as stated.
Comparing BC1 and BC2 regarding age at starting solids
Age at starting solid food was recorded in BC1 in months, but in BC2 in weeks, so the only way to compare ages between cohorts is to convert one variable into the same units as the other (weeks into months, given the guidance hinges around six months). The question is what algorithm should be used for valid analysis? This is of considerable concern because in a data set of this size even a small difference, such as might arise largely by bias, could show as statistically significantly. Two conversion algorithms were used, as shown in the chapter. In the first case, for the coding to be x months, the age in weeks must be at least equivalent to x months (as per the usual quantification of a baby's age, considered in 1(i) above). So, for example, given the infant becomes 'four months old' at 17 weeks, the only values coded as 'four months old' are reported ages of 17 to 20 weeks. The second ('mid') algorithm takes a more 'centred' view of age, and considers the four or so weeks around the 'anniversary' date to define the age under consideration. So four months would be the conversion for weeks 15 to 18, those which are around the exact 'anniversary' - 17 weeks. The difference between the two algorithms is a shift of two weeks, and the 'mid' version would be more comparable if in BC2 respondents, their recall of age in months at starting solids is being converted mentally to weeks during the interview, via a simple multiplication by four (as discussed in 1(iii) above).
Manipulation of variables
i. Creating a binary '6 weeks or older' variable In 1(iv) above it has been noted that flexibility with respect to units is likely to have an advantage in terms of 'accuracy' of recorded breastfeeding durations, but this does create some uncertainty in terms of creating a variable to indicate the fact of having continued at least to the '6 week' threshold (as per one breastfeeding statistic in common use, as mentioned in the background (4.1)). There was no problem if the age was stated in weeks, obviously,
nor if stated in days (>= 42 days being unequivocally >= six weeks). However, if the age was stated in months, there was some uncertainty if the age was given as 'one month'. Arithmetically, one month is <six weeks, and so such responses have been recoded to not >= six weeks. However such a response is possible even if breastfeeding did continue for six weeks, if for example the baby received its last breastfeed at 43 days old, and this event has been remembered as having been at 'one month old' (as per 1(i) above, since the baby was not yet fully two months old). However, while there is a theoretical possibility that some babies who did complete six weeks breastfeeding have been coded as not having done so, the potential numbers are very small.
ii. Creating a binary '6 months or older' variable The guidance for parents states that babies need no other feeding than breastfeeding (or formula) until six months of age. If wishing to create a variable to indicate the fact of having continued at least to the 'six months' threshold (as per current guidance for age at starting solids) then there are concerns for BC2, since in this cohort the age was assessed in weeks, and hence the concerns of 1(i) and 1(iii) above apply. With regard to any comparison between BC1 and BC2, the concerns of 2 above also apply.
Email: Sharon Glen