Stage 2 of a SQUEAC assessment is (as you say) about hypothesis testing. A very common hypothesis to test will be that coverage is high in some places and low in other places. You might test this hypothesis with a small scale survey in a place or places where you believe coverage to be high and a place or places where you believe coverage to be low. It is important to note that you might choose an indicator other that coverage (e.g. awareness of CMAM program, recognition of MUAC strap, recognition of the signs of malnutrition). This could be anything that arises in Stage 1 of the SQUEAC assessment. There are examples in the FANTA SQUEAC/SLEAC technical reference looking at screening activities, default, DNA, &c. Let us look at the mechanics of data analysis using an example of distance from facility and awareness of an iCMAM program from Kenya (data courtesy of ACF-Kenya and the local and national MoH in Kenya). In this study, the use of "awareness" rather than "coverage" simplified data collection in a low prevalence setting as the sample was of mothers of children aged between 6 and 36 months which are very much easier to find than SAM cases. Here are the data: Distance from Mothers Team Village iCMAM facility Interviewed Aware Not Aware ---- ------------ -------------- ----------- ----- --------- 1 Lakole Near = 1 km 5 5 0 Mlandanoor Far = 6 km 5 1 4 2 Bilikomarara Near = 1 km 5 5 0 Martaba Far = 13 km 5 0 5 ---- ------------ -------------- ----------- ----- --------- Note : Each team started at a different iCMAM facility The first hypothesis (i.e. good awareness if near to an iCMAM facility) would be confirmed if more than: d = floor(10 * (50 / 100)) = 5 respondents were aware of the program in the near villages. The study found ten respondents who were aware of the program. The first hypothesis was, therefore, confirmed. The second hypothesis (i.e. poor awareness if far from an iCMAM facility) would be confirmed if: d = floor(10 * (50 / 100)) = 5 or fewer respondents were aware of the program in the far villages. The study found one respondent who was aware of the program. The second hypothesis was, therefore, confirmed. Given these results, the SQUEAC assessment team concluded that distance was a factor affecting program awareness and was likely to be a factor affecting coverage. The team concluded that coverage was likely to be patchy. What was done in this example is a "rough and ready" hypothesis test and it confirmed the initial hypothesis / hypotheses. A more complicated test procedure is not required. You can do a formal test if you like. If you do this than you should be aware that a chi-square test is often not reasonable with SQUEAC data because of small numbers in table cells. A Fisher Exact test (as is used below) could be used. You should also be aware that SQUEAC hypothesese are usually one-tailed and you will need to adjust the p-value accordingly. There are simple and free statistical calculators that work with tabular data (e.g. EpiCalc, EpiTable & Statcalc in EpiInfo, and [url=http://www.openepi.com/]OpenEpi[/url]. Usually all that is required is to arrange the data in a two-by-two table. With the data above, for example: Aware of the program | Yes | No | ----------------+----------+----------+ Distance : Near | 10 | 0 | ----------------+----------+----------+ Far | 1 | 9 | ----------------+----------+----------+ Epidemiological calculators might label the rows as "exposure" and the columns as "disease". In the example small study presented above, the association between proximity and awareness is very marked. A formal test of the null hypothesis that program awareness was independent of proximity to the program returns a p-value of p < 0.0001 (one-tailed Fisher Exact Test). This is very strong evidence against the null hypothesis. An estimation approach would return a risk ratio of 10.00 (95% CI = 1.56; 64.20) with proximity as the “risk exposure”. Continuing the example from Kenya, we have evidence that coverage is likely to be patchy. We could go ahead with a Stage 3 survey. We would have to be sure to report that coverage was likely to be patchy and the overall coverage estimate was an average that might not apply anywhere. It is quite common that you will be required to provide an overall coverage estimate even when it might be of very little use. Usually we have enough information from the first two stages of a SQUEAC investigation to institute reforms that will improve program coverage. In the example from Kenya used here we might (e.g.) increase outreach / mobilisation / sensitisation and increase the number and geographical spread of sites offering iCMAM services. We have sufficient information to make that case. A low overall coverage estimate might, however, also help you make that case. I hope this is of some help.

Thank you Mark for your detail responses. My question is to know the thresholds for patchy/heterogeneous. Or when do say the difference between the high and low is within acceptable range or above? Let’s take your example from Kenya. The awareness coverage in one area (nearby villages) was 100% whereas in far villages it was 10%. The difference between the two areas is 90. Then you conclude that it is patchy. In this case, doing stage3 is has little value. So, my question is now, what if the difference between the high and low coverage is let’s say 80 or 70, or 60 or 50 and so on? Can we say all these are heterogeneous?

From Mark Myatt: Your question is difficult to answer categorically. I think a difference of 50% is wide particularly if the lower limit is well below a reasonable standard. That is "patchy" means a real qualitative difference such as some areas above a standard and some areas well below a standard. If coverage is markedly patchy then an overall estimate will not have much technical merit. Sometimes you just have to do the stage III survey because donors, planners, &c. expect a survey and see all other data as less useful. The stage III survey may not have much technical merit but it may have "political" merit. Best wishes, Mark