TRAITEMENT OF NUTRITIONAL AND MORTALITY SURVEYS
DATA ANALYSIS WITH NUTRISURVEY
ACKNOWLEDGEMENTS
We would like to thank Yvonne Grellety and Michal Golden for revising the present work.
TABLE OF CONTENTS
- Description & Preparation………………………………………….4
- Opening…….……………………………………………………….4
- Nutrisurvey………………………………………………….4
- An existing file……………………………………………..4
- Useful Icons……………………………………………………..…4
- Planning………………………………….…………………………5
- Naming the Survey…..……………………………...…..5
- Sampling…………………………………………………….6
- Sample size calculation………………………………….6
- Sample size for Mortality Rate Survey………………7
- Sample size for both anthropometry
and mortality analysis…………………………………..8
3.6.Calculating Clusters……………………………………...8
3.7.Selecting Clusters ...….……………………………….…9
- Training………………………………………………………….…11
- Options……………………………………………………………..13
- Data Entry…………………………………………………..13
- Plausibility Check……………………….………….…….14
- Report…………………………………………………..…..14
- Data Entry Anthropometry……………………………………16
6.1. Variable View………………………………………………16
- Data Entry Mortality……………………………………………18
B. Introducing Data
- Anthropometry
- Data Entry ………………….………..…….…………...19
- Pasting data from Excel……………………………….20
- Mortality…………………..……………………………………...21
C. Data Quality Check………..…………………………………….....23
- Results & Data Analysis
- Anthropometry…….....……..………………………………...24
- Mortality………………………………………………………….24
- Analysis……………………………………………………………25
- Excel………………………………………………………..25
- EPI Info…………………………………………………….25
- Annexes…..………………………………………………..…..……..26
NUTRISURVEY
This program can be downloaded free from or . (If it is already installed in a computer, it is important to check that is the latest update).
- DESCRIPTION AND PREPARATION
- OPENING
1.1.Nutrisurvey ena
Once installed, the first things that will appear on the screen are the names of the persons who designed this software, as well as the e-mail and website from where ena has been downloaded. Press click on <OK>.
1.2.An existing file
When closing down the software, after saving your file, the way to open it (or any other existing Nutrisurvey file) is by opening first Nutrisurvey ena,secondly by clicking onand thirdly by searching the file.asthat you want to open.
Or using the icon shown in chapter 2 (below), after opening Nutrisurvey ena.
- USEFUL ICONS
New Opensa new filein .asformat
Open Opens an existing.as file
Save Saves the actualfile.as
Save as Saves file.as with a chosen name
Import Importsfiles .rec EPI-Info 5/6 or DBase .dbf
Exit Exitsfrom actual module and returns to principal menu
The first page to appear will be Data Entry, but you have to go to:
- PLANNING
STEPS TO FOLLOW
3.1.Naming the survey
You need to give a unique name to your survey. However, it is important to be consistent in the naming of files and directories and to give all files names that can be recognized later by any team member.
The name of the file should start with a three letter code for the country (e.g., SUD for Sudan, ZAM for Zambia, ANG for Angola, etc.), then the file-name should have the date of the survey in YYMM format (year, year, month, month). In certain circumstances the region, type of subject (refugee, IDP, resident) or the agency involved can be usefully included in the name of all the files. Then there is a code for the type of file: REP for report, DAT for the data file, etc.
Thus, a file named <LIB_0409_rep.doc> would be the report of a survey taken in August 2004 inLiberia. There may be several simultaneous surveys conducted in Liberia around that time, <LIB_0409_IDP_Buchanan_AAH_dat.xls> would be the data file for a survey with IDPs in Buchanan, Liberia in September 2004 conducted by Action Against Hunger.
3.2.Sampling
The next step is to choose the type of sample you will use: 1) random survey or 2) clustersurvey.
3.3.Sample size calculation
Calculate the sample size for the anthropometric survey. Introduce the target population size. The population will be children under 5-years-old. If that number is unknown, an estimate of 20% or less (in case of having a high mortality rate in the area) of total population will be used as our population size.
3.3.1.Estimated prevalence
Enter estimated prevalence of GAM. With a fixed sample size, the higher the malnutrition prevalence, the lower the precision obtained. When making this assessment always decide upon a plausible range of values, rather than a single one. In many situations, a reasonable statement would be: “Given the situation, the malnutrition prevalence is unlikely to be above 20% or below 10%”.
In other words, if there is no certainty of this value, the higher (maximum) prevalence expected from a range of similar values must be introduced.
Nevertheless, if you are interested in a particular prevalence (e.g. the level that would trigger an emergency response), and you suspect the actual prevalence is below this threshold, enter the threshold number.
3.3.2.Desired precision
The first consideration is the minimum precision needed to meet the objectives of the survey. (For further explanation –descriptions and tables-, please go to the SMART METHODOLOGY manual.)
The desirable precision and expected malnutrition prevalence rate are interconnected. If there is a very high prevalence of acute malnutrition (e.g. 40%) the precision does not need to be high to enable agencies to make appropriate decisions. At a prevalence of over 35% or so, services will be overwhelmed and urgent and substantial intervention will be needed. A confidence interval of plus or minus 10% (25-45%) is perfectly acceptable under these circumstances. Normally, it can be set a 5% precision or more for high prevalence, falling to about 2.5% precision for lower prevalence.
In general, the lower the prevalence the greater the precision needed.
Ex. For 5% PREVALENCE, you will need 2.5 to 3% precision
3.3.3.Design effect
If it’s a random sampling survey, then the design effect will always be 1.
In cluster sampling, design effects can vary from 1 (if the population is homogeneous so that all the clusters are similar to one another) to 4 or higher where some clusters are not affected and others are severely affected.
In most nutritional emergencies, the design effect is about 1.5 increasing to 2 or more in more heterogeneous or large-scale surveys.
If the design effect is thought to be much greater than 2 then the population is sufficiently heterogeneous and therefore it is better to conduct two separate surveys, each focused upon more homogeneous sections of the population: e.g. Two cluster surveys, each with a design effect of 1.5, can be conducted with the same effort as one survey with a design effect of 3.
The adviceto anticipate a range of likely values for the prevalence and for the design effect, within which you anticipate the results will fall is important. In Nutrisurvey ena you should enter:
- The widest confidence interval that is acceptable = the minimum acceptable precision
- The highest prevalence that is anticipated
- The largest design effect that is likely to be encountered
(For further explanation –descriptions and tables-, please go to the SMART METHODOLOGY manual.)
Upon entering all these values in their respective boxes, the sample size will be automatically calculated and will appear in the turquoise box.
3.3.4.Increase sample size
By 5% or 10% to allow unforeseen contingencies.
EXAMPLE 1:
Let’s look at the following example:
Our sample size is 754. And if we increase this number 5%, according to the last step, then our sample size will be 792.
3.3.5.Divide the sample size by the average number of U5 children to have a household sample size.
e.g. if the average U5 children is 1.5, divide 792 by 1.5 = 528
3.4.Sample size for mortality rate survey
For the death rate component of the survey (you may need information from governmental and/or non-governmental organizations (NGOs) working in health):
- Enter an estimate of the total population that is targeted by the survey.
- Enter the expected mortality rate (x.xx/10,000 persons/day)
- Enter the required precision (x.xx/10,000 persons/day). For example, if your expected mortality rate is 2.0/10,000 persons/per day and you want a confidence interval of 1.4-2.6, enter a required precision of 0.6 (that is 2.0 -/+ 0.6 which gives 1.4 - 2.6). The precision chosen has a substantial effect upon the sample size needed.
- Enter the design effect. The default design effect for sample size calculations for mortality is 2.0. If violence-related-mortality is limited, a design effect of 1.5 for crude death rate may also be sufficient. You can also use the last survey raw data and check the design effect in ena: it calculates the design effect.
- Enter the chosen recall period in days. In most situations, 90 days (or from 30 to 120 days) will be used. However, the decision should be made individually for each emergency context according to the date of the last event that occurs in this area.
EXAMPLE 2
3.4.1. Divide the sample size by the average number of persons per household to have a household sample size.
e.g. 1646 divided by the average persons per HH e.g. 4 = 416
3.5.Sample size for both anthropometry and mortality analysis
3.5.1.For anthropometry analysis, e.g. 528
3.5.2.For mortality analysis, e.g. 416
The final sample size per household that will be retained will be 528
3.6.Calculating Clusters
3.6.1.For Anthropometric Survey
In cluster surveys, sample size should be divided by the number of households that can be visited each day. This will provide the number of clusters.
To continue with EXAMPLE1, let’s say that each of our teams can visit 14 households (HH) per day. (For further details on how to calculate this number, please go to SMART Methodology Manual). Dividing 528/14 = 37.7 Whenever the result has a decimal, it is advised to round up to the next whole number. Then we will have 38 clusters for our survey.
In pastoral/nomadic zones we can have difficulties finding children. If this is the case, it is advised to decrease the number of children/cluster and increase the number of clusters.
Statistically,the minimum amount of clusters that each of the surveys(anthropometric & mortality) can have for them to be valid is 26. However, in ACF we STRONGLY RECOMMEND having 30 clusters, minimum.
3.7.Selectingclusters
When designing a combined survey (with both components: nutrition & mortality), the sample size to estimate malnutrition prevalence, as well as the number of households needed to estimate mortality rate are calculated, and then:
The greater of these numbers is chosen for the survey.
3.7.1. Table for Cluster Sampling
- Enter number of clusters
2. Define what constituents a “village”: it should be the smallest unit with population figure. Then, under Geographical unit column enterthenames of all the towns, cities, districts or other areas that willpotentiallybe chosen to include in acluster. All potential areas have to be entered. It does not matter what order the areas are entered. But if any villageis omitted at this step, it then will not be part of the surveyed population.
3. Under Population size column enter the estimated population size for each “village”.
4. With theNumber of Clustersandtheir names entered, click on<Assign Cluster>
The computer will then select the areas where there will be clusters. This should be done only ONCE. It will potentially introduce a bias if they are reselected.
3.7.2.Random Number Table
If you want to do a random survey,then you need to generate a random table based in the box <Random Number Table>. For that you needthe following data: forRangefromandtothe total range of children of the total population, i.e. 1-1,350andenter forNumbersthe sample size to use. Then, clickon<Generate Table>.
The children to be interviewed will be shown in a Word sheet.
4. TRAINING
This page is used to standardize our teams in measuring weight and height (or length). The aim of the standardization test consists on improving the quality of the measurements. All the members of the teams should measure twice 10 or more children with a time interval between individual measures. (Check SMART METHODOLOGY to observe the appropriate procedure for standardization.)
The outcome of this exercise will be analyzedby Nutrisurvey Ena: you only need to enter the measures and then click on<Report>. Precision and accuracy are assessed by 1) calculating the variation between their repeated measurements (repeatability of measurements); 2) calculating the variation between each team member’s measurements and the ones of the supervisor. Each team member is then given a score of competence in performing measures (OKorPOOR). If the results areOK it means the enumerator is standardized, if the results are POOR the enumerator should repeat the exercise completely, perhaps with different people paired in teams.
Report for Evaluation of Enumerators
Weight:
Precision: Accuracy:
Sum of Square Sum of Square
[W2-W1] [Superv.(W1+W2)-
Enum.(W1+W2]
Supervisor 7,29
Enumerator 16,76 OK1,69 OK
Height:
Precision: Accuracy:
Sum of Square Sum of Square
[W2-W1] [Superv.(W1+W2)-
Enum.(W1+W2]
Supervisor 4,84
Enumerator 110,89 POOR0,25 OK
For evaluating the enumerators the precision and the accuracy of their measurements is calculated.
For precision, the sum of the square of the differences for the double measurements is calculated. This value should be less than two times the value of the supervisor.
For the accuracy, the sum of the square of the differences between the enumerator values (weight1+weight2) and the supervisor values (weight1+weight2) is calculated. This value should be less than three times the accuracy value of the supervisor.
5. OPTIONS
This is the last of the screens of Nutrisurvey, but is useful to fill out before entering the data. The software has an automatic way for entering data in DataEntry. We could agree and save these options or simply modify them based on our needs. If we save them, we should always remember that we are establishing these options for future surveys. However, they can be modified any time a new survey is planned.
5.1. Data Entry:
5.1.1. Automaticfilling out:
The program selects all these options; it is advised, nevertheless, thatthe number of household be introduced manually, because the number of children in one household could be greater than one.
We recommend not filling out Household No. automatically.
5.1.2. Entering of age mainly:
Either method can be selecteddepending on the document you find at household level: e.g. in some countries like Tanzania, every child has a birthday date…in other countries, which is mostly the case, you have no birthdate and you have then to estimate the age in months; in this case, you should use “with months”.
5.1.3. Entering of data:
You can select introducing data directly or“with Pull Down Editors”for some of the variables (sex, birthdate, edema). See the example below.
We recommend using ● directly as 1.1.99, 10199 or 010199 option.
5.1.4.
In this page we can also selectif we want the program to calculate anthropometric indicesafterpasting data from another file.
5.1.5. Weight of clothes
Here we can introduce average weightin gramsof clothingof survey subjects to be automatically subtracted from the weight introduced in Data Entrypage.
5.1.6. Correction for Edema: We can select automatic correction of weight for edema found:
“n/y” = average weight of body weight edema
“0,1,2,3 = mild, moderate or severe edema
We don’t recommend using it, unless strictly necessary.
5.2. Plausibility Check:
Enter the z-scores and the age range (in months) for the plausibility report in the box.
The plausibility check report, after entering all your data, will give you the questionable ID no.(s) out of range that you have selected, e.g. >3Z-score and <-3Z-score from the mean of the WH of the sample size.
In the Data Entry sheet, the software will highlight gross errors similar to flag in Epi Info in pink. but they are not the same as the ones in the plausibility check in this version.
5.3. Report:
If we want our report to begrouped by ages (in months).
We can change the age groups by only changing the first number in each range, ena is automatically changing the other one from the range above.
Once finished introducing desired parameters, click on which is located at the bottom right hand side of the page.
6. DATA ENTRY ANTHROPOMETRY
Fill out this page with the data gathered in the field questionnaire.
Take in account that the field questionnaire follows the program’s same order to facilitate data entry.
Nutrisurvey ena has a form that we can use and can be obtained clicking on:
Form for anthropometric survey
Other uses of this module are:
Form for mortality Formto gather mortality information
Paste Data Pastes data from clipboard
Copy Data Copiesdata from current survey to Excel format
Plausibility Check Verifies presenceof errors
Check of double entry Corroborates presence of errors when two persons are entering the same data
Report in Word Reproduces reportin Word format
6.1.Variable View:
This is one of the two sections of page Data Entry. In Variable Viewwe should save the file using the same instructions we used in Planning. (We can find these in the1st.step.) In <Variable View> section and before introducing datawe should establish ranks to be used in the survey.
Some variables in ena are automatically ranked; however, you can change some of them, like for example height ranks if the targeted population has growth retardation. These changes should be included in the report.
Ena will then check all those data which are out of range and highlight them in pink in <Data View> section.
In this latter section, data cleaning also can be made by using “Plausibility check” clicking onCheck Tableevery time you entersome data;this facilitates to underline errors.
The entered data have to be checked using the original written data collection sheets. Any error in data entry should be corrected immediately.
If a child is excluded from the survey, his information will disappear from the page and we should click again onCheck Table.
Column/Variable: