From Blessing Mureverwi: I have used both SMART and WHO flags before.It is ok to use WHO flags as long as the quality of data is good,as in this case.SMART flags may underreport malnutrition in such a case.

Dear Jose, SMART has the option to change the setting to WHO standards and WHO flags. It seems that what you applied in your analysis is WHO standards but SMART flags (+/-3 SD of observed mean). For comparability purposes we would really encourage SMART users to apply the WHO standards with recommended flag limits which are: WHZ -5 +5 HAZ -6 +6 WAZ -6 +5 BAZ -5 +5 (BMI-for-age z-score) For details on WHO standards see [url]www.who.int/childgrowth/standards[/url] For details on standard analysis for the WHO Global Database on Child Growth and Malnutrition, please see Table at the bottom of [url]http://www.who.int/nutgrowthdb/software/en/[/url]

Why don't you use the "No Exclusion" option in case if you think data was properly taken and the check the weight and height of cases which are below -3SD, if they are well taken as well as those cases are really depicted, I think No exclusion would be good choice.

This note from the plausibility check notes answers your question: "In SMART there are two methods of identifying results from children that are unlikely to be correct measurements. The first is the same as that recommended by WHO and used in Epi-info. This function is included to make it compatible with Epi-info and also because the values that are absolute abnormalities can be identified for individuals as the data are being entered. The red highlights show these values. The absolute values have been chosen because they are so extremely abnormal that they are very unlikely to be correct – indeed some are hardly compatible with life. The default values can be set on the Options page and excluded automatically in all the calculations by selecting the “Exclusion of Z scores from Zero (reference mean)” button. These values should be examined and, if there has been a recording error, corrected. Any uncorrected values should definitely NOT be used for the analysis – and the number of exclusions, that can be found using the Filter option on the Data Entry Antropometry Screen shoud be recorded in the report (they should not be simply eliminated from the data-base). The second method is conceptually different. Here we want to exclude from the analysis data that are “more likely to be errors than real values”. So, if the measurements from the members of the population form a normal distribution (bell shaped curve). Then we can estimate what proportion of the observations should be outside a window around the mean of the population. Thus, the table below shows the number of children that would be excluded incorrectly out of a sample of 1,000 subjects, with various settings of the Plausibility check flags. Flags SD away from the mean Number excluded from 1000 children -2.0 23.0 -2.5 6.0 -3.0 (default value) 1.3 -3.1 1.0 -3.3 0.5 -3.5 0.2 Normally, when we examine survey data we find many more children who have extreme values than is given in this table. It would clearly be incorrect to set the limits at ± 2 SD from the mean because a large number of correct measurements would be excluded. If we set it at ±3.0 we will exclude just over one out of a thousand children incorrectly, and the other children we will have excluded correctly because they were bad measurements. If the boundaries are set at -3.3 or even -3.5 then almost no child will be incorrectly excluded. Excluding one of a thousand measurements when we should include that measurement makes almost no difference to the final result, but including a lot of children outside these boundaries can have a major effect upon the result. Thus, for the final analysis it is recommended that the exclusions based upon the “Exclusion of Z scores from Observed Mean” set on the Options page should be used and the numbers excluded reported. : (Taken from: SMART: Ensuring data quality – is the survey result usable? By Micheal Golden)

In this discussion, it should be acknowledged that there is an intrinsic weakness in the second mode of quality check in the SMART methodology. It assumes that all populations have a bell shape curve distribution with an SD close to 1. This is based on observations of past studies showing this is usually the case, but this does not mean in any way that it should always be so. It is clear for instance that if you have a socially heterogeneous population, the SD will be above 1, even if measures are perfectly correct. So, in practice, there is no way to distinguish a survey done in heterogeneous population from a survey with measures including a large random error component by examining the shape of the curve. There is a real risk with this second approach to reject real malnutrition cases and to underestimate malnutrition. Also, you can reject as incorrect a survey which just shows unequal distribution of malnutrition in the surveyed population, with an increased prevalence in some subgroups. So when you are sure of the quality of your measures, you should use the “no exclusion option” as suggested by Tariq or just reject measures which are clearly not plausible. Else, you will underestimate the prevalence of malnutrition and miss the existence of some subgroups with high malnutrition prevalence.

If the population we are surveying are completely heterogeneous, and if it is possible to spatially distinguish this different groups SMART recommends to do a separate survey. In practice, no one applies this and the overall GAM rate is considered to represent the whole population, which it did not. I remember doing a survey in such a place and there are two distinct population in terms of livelihood, coping mechanism and what not. The challenge is interventions usually follow an administration boundary but livelihood zone may not. In that survey the result shows a relatively stable situation. But the fact is part of the rural population were highly affected (to the point you do not need a survey to know that) while the highland population are resilient enough not to be affected by the shock. The good thing is in SMART there are new functions that help us understand the hetrogenity problem and and understand the results accordingly (the index of dispersion).

I do not see the button "Exclusion of Z-scores" under Options. Under "check of z-scores for plausibility report from the mean" I used the default of -3 to 3. However, the results output includes all values outside this range as well. Am I missing something? How do I use the WHO flags in the SMART software?

Dear Beulah, The plausibility report was designed using SMART flags (weight for height, +/- 3 SD from the observed mean of the survey pop). It is not possible to generate the plausibility report using WHO flags (weight for height, +/- 5 SD from the reference population). WHO flags are used in ENA software under the following tabs: Data entry anthropometry: if any z-score is entered outside of WHO ranges it will immediately be flagged (purple). Results anthropometry: under the “exclusion of z-scores with” option you can choose, SMART flags, WHO flags, or No exclusion. If you choose ‘WHO flags’ you can copy and paste the results into another Word document.