M&E specialiste

Normal user

7 Jun 2014, 08:02

Comment déterminer la taille de l'échantillon dans une enquête CAP d'évaluation d'intervention de nutrition avec une méthodologie LQAS?
How do you determine the sample size for a KAP survey to evaluate a nutrition intervention using the LQAS methodology?

Epidemiologist at Brixton Health

Frequent user

9 Jun 2014, 09:49

You need to choose:
(1) The levels of the indicator(s) that are considered to be bad and good. You might (e.g.) have < 50% as bad and > 80% as good.
(2) Acceptable levels of error (type I / type II). I usually set these both at 5%.
(3) The population size. This is important for small populations such as when you assess coverage of SAM treatment. IF this is large then you can specify a population size of 5000 or higher.
I have an on-line calculator ... click

this link. Here is an example sample size calculation:
I assume a large population so I set 'Population size' to 5000
I define good as > 80% so I set 'Upper threshold' to 0.8
I define bad as < 50% so I set 'Upper threshold' to 0.5
I want both errors to be <= 5% so I set both 'Maximum tolerable a error' and 'Maximum tolerable ß error' to 0.05
I click 'Find sample size and decision rule'. The calculator returns the LQAS sampling plan n = 28, d = 18. I need a sample size of 28. If I find >= 18 'good' cases then I classify the situation as good. If I find < 18 'good cases' then I classify the situation as bad.
I hope this helps.
BTW .. A three class LQAS (as in SLEAC) is possible. A sample size of n = 40 is usually large enough for most applications (see the SQUEAC / SLEAC technical reference for more information).

Public Health Nutritionist

Normal user

9 Jun 2014, 10:03

Dear Mark,
Is this sample size of 28 for one indicator? if the number of indicators to be collected are lets say five. does it mean we have to have 28 sample size for each indicator.

Epidemiologist at Brixton Health

Frequent user

9 Jun 2014, 10:18

As with any survey ... you can have many indicators.
It is typical to calculate the sample size for all indicators and then pick the largest required sample size and collect that for all indicators. The one complication is that you need to recalculate the decision rules for the larger sample sizes for all of the indicators.
You can find a "reverse" calculator

here. This needs you to enter the same information

**and** the sample size.
A very easy method is to pick a reasonably large sample size (e.g. n = 40) and then use the simplified methods found in the SQUAC / SLEAC technical reference.

Epidemiologist at Brixton Health

Frequent user

9 Jun 2014, 10:19

To be clear ... you can do many indicators with one sample.

Public Health Nutritionist

Normal user

9 Jun 2014, 11:34

What if the target groups for the indicators are not overlapping?

Epidemiologist at Brixton Health

Frequent user

9 Jun 2014, 11:41

Then you would need a separate sample from each target group.