I cannot speak for the SMART Initiative and have little experience of their software but I will attempt to answer this question. I am not sure what you mean by "range mean -3 to +3". If this is the range outside of which a calculated WHZ is deemed to be unreliable then it is too restrictive since WHZ < -3 is a case-defining threshold for SAM and the software would only use cases of bilateral pitting oedema when estimating the prevalence of SAM. If this is a mean WHZ for the entire survey dataset then this is sensible (for standard survey work - see below) as a mean WHZ of -3 would correspond to a 50% prevalence of SAM. The EpiNut module of EpiInfo v6.xx uses these thresholds to identify individual cases that may have erroneous data. This makes sense since this thresholds are probably incompatible with life. I think you should check what SMART means by "range mean -3 to +3". If cases are just flagged then there are no consequences to this. If they are automatically excluded from reports then it will stop you using SMART software to perform such analyses as investigating the range of WHZ of children admitted to a therapeutic feeding program since the mean WHZ in this application will almost always be below -3. I think that you should be careful when interpreting automatic plausibility checks. They are useful but can mislead. As an example I will look at the age distribution. Plausibility checks on this data are a little limited because after about two years ages are often reported as whole years. The standard approach is to recode data into year-centred age groups and then check for a deviation from uniformity in the distribution. This is a reasonable approach but if there was a serious shock in the recent past that caused a great deal of infant mortality then the age distribution will show a gap where the dead children would have been had they not died. The data may be good but the plausibility check will suggest that they are not. A similar situation could arise if infant mortality differs between sexes or (e.g.) female children are preferentially aborted as has been reported from China and India. The checks may also provide unreliable results if there are older (stunted) children in the data causing the test to be based on small numbers in some strata or, if older children are treated as 5 year olds, reporting an excess of older children. There is nothing wrong with plausibility checks and I would encourage you to use them but you do have to treat them with caution and check whether there are alternative (i.e. to poor data quality) explanations for the "suspect" patterns in the data. The issue of digit preference has been discussed on a previous thread on this forum. As for "totally useless" ... data would have to be of very poor quality for it to be "totally useless". I think that digit preference is far less serious than a poor sample.
|