|Nutrition Guidelines (MSF, 1995, 191 p.)|
It is essential during the training of enumerators to test that they measure children in a standard and accurate way. The test can be used for assessing the quality of any measurements: weight, height as well as mid - upper arm circumference.
Accuracy: ability to obtain a measurement which will duplicate as closely as possible the reference value.
Precision: ability to repeat a measurement on the same subject with a minimum variation.
These two abilities are complementary. An enumerator may be precise but not accurate: he finds a wrong value for the measure, but he <<precisely>> finds the same wrong value every time. In the same way, an enumerator may be accurate but not precise, meaning the mean measure on a number of measures is close to the reference measure, but wide variation between measures exists.
The standardization test consists of repeating a measure twice on 10 different children, with a time interval between measures on the same child. The amplitude of the variation between repeated measures is calculated to assess precision, and the mean measure is calculated to assess accuracy. Each enumerator is then given some sort of a score of competence in performing measures. Misunderstanding and measuring errors can then be corrected during the training process.
3. Practical organization of the test
The test is carried out during the training process. Ten children aged 6 to 59 months are gathered in a room where the test will be performed. For example, if the measure of height/length is assessed, each enumerator performs the measure and records it for each child on a form. A second series of measures is performed and the enumerators once again record their measures, but on a different form. The supervisor performs the measure as well in order to obtain the reference measures, assuming he is the most precise and accurate measurer available.
For each enumerator and for the supervisor, the following steps are followed:
Step 1: results of the two measures for each child are entered in column a and b. In order to facilitate the calculations, the height and mid - upper arm circumference are entered in millimeters and the weight in 100 grams.
Step 2: column d is the difference between the two measures: d = a - b.
Step 3: in the column labelled d^2 the value of d is squared: d^2 = (a - b)^2.
Step 4: in the column labelled <<sign>>, the number of occurrence of the most frequent sign in column <<d>> is entered. For example 7 is entered if there are 7 + signs and 3 - signs.
Step 5: column <<s>> is the sum of column a + column b: s = a + b.
These first 5 steps are common to the supervisor and the enumerators. The following 4 steps are only carried out for the enumerators since the test is meant to compare measures of the enumerators to those of the supervisor.
Step 6: the column S of the supervisor is added to each of the enumerator's forms.
Step 7: the difference between the s of the enumerators and the S of the supervisor is entered in column D.
Step 8: this difference is squared and entered in the column labelled D^2. D^2 = (s - S)^2.
Step 9: the greatest number of identical signs (+ or - but the largest) is entered in the column labelled sign.
A summary form is established. In this form, the sum of the column d^2 and D^2 is calculated: sum(d^2) and sum(D^2).
sum(d^2) reflects the precision. It represents the sum of the squares of difference between 2 measures on the same child. The acceptable value for the sum(d^2) of an enumerator is equal to twice the value of the sum(d^2) of the supervisor. The sum(d^2) of the supervisor is not zero. One component of the lack of precision is related to the equipment used to perform the measures, and that is why the reference value is calculated from the <<performance>> of the supervisor.
sum(D^2) measures the accuracy. It is the sum of the squares of difference between the sum of the 2 measures on the same child between the supervisor and the enumerator. The acceptable value for the sum(D^2) is fixed at three times the sum(d^2) of the supervisor.
Analysis of the signs allows us to assess whether the lack of precision or accuracy is always in the same direction. For example, an enumerator overestimating the height of a child will have most of the signs for precision as positive (i.e. case of the enumerator No 5 in our example). Such a systematic error can be quantified. Its value is sum(/20) (2 times 10 measures = 20 measures), being 8.9 millimeters in our example.