Menu ENN Search
Language: English Français

developing a methodology for a highly security compromised area. need guidence

This question was posted the Assessment forum area and has 1 replies. You can also reply via email – be sure to leave the subject unchanged.

» Post a reply

Dr Mohammad Najeeb

Nutrition Cluster coordinator

Normal user

26 Feb 2014, 07:24

Hi Mark, Salute
We are in process of designing a nutrition survey in area which is highly security compromised, infact active war zone in many places. The survey is to establish nutritional status of children under five among IDPs and Host communities. A large number of IDPs have left their original home area and moved to comparatively safer areas. Most of them are residing on temporary basis with friends, relatives, hired or so on. We do not have any definitive number or more information on IDP movement which is still very much dynamic.. But we collected as much as possible information, listed out all communities with IDPs, and calculated our sample size based on regions with the following formula
n = { t2 x p x q }
d2
With desired accuracy of 1.2% the number of children is 2400 among 2857 families, which was increased to 3000 children to accommodate contingency clusters and data quality issues.
The methodology of selection of families is two fold
1. In areas of symmetrical IDP/Host community residences, CSAS sampling from lists of communities sorted by a spatial factor has been used.
2. In areas with non-symmetrical residences of IDPs among general population, following procedure is decided
B. If the IDPs and/or Host families are living randomly, than the team will follow the following procedure
Upon reaching an area of IDP concentration, the team will randomly pick a house on the right and enquire if any IDP family is residing in that house with at least one child under five year of age.
a. If they find any IDP, they will interview the IDP and the host family separately and will enquire from that household of any other IDP family residing close by and proceed to that house.
b. If they do not find any IDP in that house, they will enquire from that house of any IDP family residing close and proceed to that house
c. If the family do not know of any IDP, the team will pick next randomly picked house and enquire of presence or knowledge of any IDP family residing close by and proceed to that house to follow steps as (a) or (b)
I will be happy to have your comments and guidance on our methodology.

Mark Myatt

Consultant Epideomiologist

Frequent user

26 Feb 2014, 16:21

Good to hear from you Najeeb (I believe we met last year in Pakistan).

I will try to comment on all you sent.

Sample size : A precision of 1.2% is very precise. At typical precision is 3% or even 5% (I think 1.2% is too precise for this sort of application). I will use a 95% CI of +/-3% on a 10% prevalence, a design effect of 2.0 (we do not know better ... you might), and a finite population correction in the example calculations below. The base formula is:

    n = DEFF * [(p * (1 - p)) / (precision / 1.96)^2]
    n = 2.0 * ((0.1 * (1 - 0.1)) / (0.03 / 1.96)^2)
    n = 768

We can apply a finite population correction (FPC) to this:
    new.n = (old.n * population) / (old.n + (population - 1))
    new.n = (768 * 2400) / (768 + (2400 - 1))
    new.n = 582

When you analyse the data you can apply another FPC as described in this post. If you plan to take many small samples then you might use DEFF = 1.5 in your sample size calculation.

In these sorts of missions I prefer to take the smallest sample size (in terms of both the number of PSUs and the within-PSU sample size) that can meet my needs as it exposes me and my staff to less risk.

WRT your (1) ... The setting makes PPS sample very difficult as you have no good prior information on populations. This is typical of emergency settings with displacement. I think your idea of a CSAS or other spatial stratification is a good one. Since this is not a PPS sample you will need to weight after data-collection. This means that you need to have some way of estimating the eligible population in each primary sampling unit (PSU). Posterior weighting is described in this post.

[b]WRT your (2) ...[/url] This looks like it might work. I would segment the community and take a part of the overall PSU sample from each segment. I think it is important to pilot the sampling method. A rule-of-thumb is that the sample should come from all over the PSU. If you do this then you will help to keep the DEFF low.

In insecure areas it is a good idea to select a number of contingency clusters / locations that can be "swapped in" should a particular community be inaccessible. With a spatial sample you can chose the nearest (in terms of distance) sampling opportunity.

Please pay close attention to your personal safety and that of your survey staff.

Feel free to ask follow-up questions. I think you will need to think hard about how you will get the weights. We can review that here. I am in the field myself and may take a day to answer.

I hope this helps.

Back to top

» Post a reply