Disaggregating results by sub-areas

Home › Forums › Data Analysis and Reporting › Disaggregating results by sub-areas

This topic contains 2 replies, has 3 voices, and was last updated by  SMART 10 years, 9 months ago.

Viewing 3 posts - 1 through 3 (of 3 total)
  • Author
    Posts
  • July 8, 2015 at 9:39 am #1533

    kate.golden@concern.net
    Participant

    Dear SMART experts,

    We are frequently asked by our field teams if it is possible to disaggregate the results for e.g. GAM or SAM or vaccination for a sub-district if the SMART survey and sampling frame has been designed to represent the wider district (not stratified by sub-district).

    If one generates CI’s for each of the sub-districts and accepts that these might be quite wide for the estimates generated for each sub-district, could they still be valid?

    Furthermore for any comparison between subdistricts, if one is mindful of any overlap of CIs when interpreting, would this ever be acceptable?

    If not, what are the main reasons not to disaggregate results if the survey and sampling was originally designed to represent the district as a whole (and not sub-districts)?

    And would the same rules apply for disaggregating e.g. by reported main livelihood strategy (e.g.pastoralist, agriculturalist) of the HH?

    Thanks,
    Kate

    July 9, 2015 at 10:12 am #1538

    SeanFD
    Participant

    Hi Kate,

    You could do what you suggest, but as you note the CIs would probably be very wide – just how wide depending on the cluster total size and cluster sample size. If the sample size is small relative to the population, you may well not have a representative sample of the cluster even. You would need to calculate the CIs as for a simple or stratified random sample.

    The main reason for not doing this is the likely width of the CIs will preclude any meaningful conclusions.

    Why do you want to do this? If it is to identify areas that need specific attention, then a different approach, eg LQAS (Lot Quality Assurance Sampling) may be more appropriate. This will classify areas into those that reach or do not reach a pre-determined level. Those that surpass the level can be “ignored”; those that are only just above or below can be “watched” and those below can be targetted for support. Using this approach you can combine results from smaller areas to give an answer for a larger area, but NOT vice versa.

    Hope this is helpful.

    Sean

    July 22, 2015 at 6:33 pm #1543

    SMART
    Keymaster

    Just to reiterate what has already been stated; in your example if the data is disaggregated into sub-districts the confidence intervals will be wide because the sample size will be quite low. As a result, the findings will likely not be able to be used for programmatic decision making processes. During the planning stage of a survey(s) if the sampling universe contains two distinct livelihoods such as pastoralist and agriculturalist (and it is believed there will be a significance difference in prevalence) it would be best to conduct two separate SMART surveys.

  • Author
    Posts
Viewing 3 posts - 1 through 3 (of 3 total)

You must be logged in to reply to this topic.

Recent Replies