The use of "year-centred" age-groups pre-dates SMART by many years and is common practice reporting age-specific results in surveys undertaken in settings where ascertaining exact ages is difficult. The groups are centred around the months representing whole years. For example the mid-point of the 6-17 month is 12 months (i.e. one year). If ages are misreported there tends to be a bias towards whole years. For example, a 40 month old child might be reported as being 3 years old (a 10% measurement error). You can check whether this is happening by producing a histogram of your age variable. If you see big peaks at 12, 24, 26, 48, and 60 months (i.e. whole years), smaller peaks at 6, 18, 30, 42, and 54 months (i.e. half years), and very few cases at other ages then you have a reporting bias to whole and (to a lesser extent) half years. Recoding reported ages acts to smooth out peaks and admits that we are recording age with spurious accuracy. Smooothing out the peaks allows us to see the big picture with regard to our sample. The classic sample test table is age by sex. We expect to see roughly equal numbers of boys and girls in each age-group and we expect to see about 22% in each age-group apart from the oldest where we expect about 12% (it is half the width of the lower age-groups). If we see another pattern then we suspect that there may be something wrong with our sample. There is nothing to stop you using different coding schemes but you have to be aware that due to problems with ascertaining exact age they will not be as "crisp" as you might think. If you code to (e.g.) 6-23 months and 24-69 months you will (very probably) be creating groups that are, in reality, more like 6-18 months and 19-59 months. This will probably happen even if you decide to collect age in years. Additional observation ... misreporting problems makes estimation of the prevalence of stuntedness problematic (note that you cannot measure "stunting" or any process in a single cross-sectional survey). The H/A and W/A indicators are very sensitive to errors in age. If the tendency is (as in the UK) to round ages down (e.g. a 44 month old child is called 3 years) then the prevalence of stuntedness will be greatly underestimated because many children will be classed as being much younger than they are. I hope this helps.

Just to add, the new WHO infant and young child feeding indicators (WHO/UNICEF et al. 2008. 'Indicators for assessing infant and young child feeding practices. Part 1. Definitions') call for disaggregating age data in the following age groups: 0-5 months (i.e., birth up to 6 mos, a period of 6 completed months), 6-11 mos (6 up to 12 mos), 12-17 mos (12 up to 18 mos), 18-23 mos (18 up to 24 mos). For programming related to infant and young child feeding, some agencies now report using those breakdowns, sometimes aggregating results for 12-23 mos. The 'Part 1' WHO guidance on IYCF indicators is available at:http://www.who.int/nutrition/publications/infantfeeding/en/ Also at this link, the recently released Part II (measurement) and Part III (country profiles) .

If you want to use these age-groups then you will need to take care in collecting age data in your surveys (for the reasons outlined above). The good news is that errors tend to be small in younger chidlren when using (e.g.) local calendars to help decide age. This is an IMPORTANT ISSUE ... poor age data makes estimation of low H/A difficult and will also interfere with IYCF indicators. I'd be very interested to hear what people are doing to get good age data. The link above (clickable) is: [url]http://www.who.int/nutrition/publications/infantfeeding/en/[/url]

Very good point, Mark. Eliciting correct age is not only relevant for getting accurate data at survey level but also when individual assessing and counselling mothers around infant feeding. There is a good FAO guidance document that is dedicated just to estimating birth age in children, Guidelines for Estimating the Month and Year of Birth of Young Children. FAO, 2008. http://www.ennonline.net/pool/files/ife/fao-measuring-age.pdf

Marie, Thank you very much for that! This is pretty much what I have been doing (and seen others doing) for years. It is good to see it written down in such a clear form. I particularly like the algorithm on page 11. When I was working on the ASTRA survey method for trachoma prevalence, I worked with a Dutch paediatrician who had a method based on a simple set of observations (e.g. dentition, limb length (can the child touch the left earlobe with the right hand), &c.) to arrive at an estimate of age. I was sceptical at first ... but he (and nurses he taught it to) seemed to get it to within ± 3 months on two year old children (by comparison with GMP cards). Has anyone come across similar methods? A radical option is to abandon age and use height classes. These could be taken from a H/A reference or use sample-based quantiles. The link above (clickable) is: [url]http://www.ennonline.net/pool/files/ife/fao-measuring-age.pdf[/url]

Hi, I've heard about the touching the ear lobe and also the teeth method before. In fact I remember that the doc I had to go to before I entered primary school asked me to do the ear lobe thing (across the head) as a check whether I was ok to go to school or not. Howerver, I'm highly doubtful about these methods when we're dealing with populations/ kids that may be very underdeveloped. In that case a 3 year old may be as advanced as a 2.5 year old or 2 year old. This is where the snake bits itself in the tail - we're using an indicator that may be delayed due to undernutrition/ sub-optimal health and therefore is out of sync with the 'normal' age to actually estimate age. Don't think that's very wise. I would not want to discard the good experience that you've seen with this doctor and nurses but we have to be very careful with what we use. In my experience, a carefully developed, local events calendar, thorough training of the enumerators to apply it and sufficient time for the data collectors to apply the calendar with the carer in the way it should be applied is a critical mix for getting a good month-based age estimation -- but many times the survey teams don't have the time or expertise, or the supervisors/ survey leaders very quickly put together a local calendar last minute just before the training starts. I'm also a huge fan of promoting correct birth registration, b-dates on U5 cards etc. because this will pay off also with the next nutrition team coming in doing a survey...

I think that Anonymous477 is right to be sceptical. I am also sceptical that enumerators actually use the calendars we make and, instead, will record the maternal report or estimate age by eye which will bias ages down in populations in which stunting / stuntedness is common. We could use a combination of methods either as a means of checking data or for averaging. I am not very concerned about errors in age when using a measure of wasting such as MUAC or W/H. Indicators that include an age component (i.e. H/A, W/A, MUAC/A) are more sensitive to random errors in age than to random errors in anthropometry. See: Bairagi R, Effects of bias and random error in anthropometry and in age on estimation of malnutrition, Am J Epidemiol, 1986;123(1):185-91 Ascertaining age is also a problem when dates of birth are known. Studies with nurses and CHWs have reported survey and clinic staff having difficulty accurately performing the arithmetic required to calculate age from date of birth and date of examination. See: Hamer C, Kvatum K, Jeffries D, Allen S, Detection of severe protein-energy malnutrition by nurses in The Gambia, Arch Dis Child, 2004;89:181-184 Velzeboer MI, Selwyn BJ, Sargent F, Pollitt E, Delgado H, The use of arm circumference in simplified screening for acute malnutrition by minimally trained health workers, J Trop Pediatr, 1983;29(3):159-66 Probably best to record dates and have the computer do the calculations.