4. Analysis and interpretation
The statistical analysis of data should take place in two stages, as shown in figure 11.
Figure 11. Overview of Data Entry
descriptive statistical analysis
using the Nutrition Baseline software
of causes and determinants
using an advanced statistic software program
For preparing the questionnaire, entering and descriptive analysis of data the Nutrition Baseline Software is recommended. For further information please read the help in this program. The idea behind this program is to work only with one file for preparing the questionnaire, entering and analysing the data. It is also especially designed for performing a Nutrition Baseline survey. Compared to other procedures the computer work and possible errors are reduced.
For a more in-depth analysis of data a statistical program such as SPSS or SAS is recommended, since multivariate analyses such as multi-factorial analysis of variance or analysis of variance with a co-variant cannot be carried out with the Nutrition Baseline Software.
Note: One objective of nutrition surveys (see chapter 1.2) is that data should also be used for cross-sectional comparisons by country and measures. Consequently, data should be stored in a secure but accessible place.
An agreement has been reached with the Nutrition Unit of the WHO to professionally store the results of the anthropometric data from population-based nutritional surveys. It is highly recommended to send the results to the following address using a form attached in Annex 6.10.
20 Avenue Appia
CH-1211 Geneva 27
4.1.1 Data entry
Data entry should also be carried out using the Nutrition Baseline software. This program is suitable for this purpose because it is user-friendly and tests the entered data for its validity. The most important reason, however, is that it can calculate, as already mentioned, anthropometric indicators and is useful for descriptive analysis.
Furthermore it is easy to export the data files into programs such as D-Base, Excel or SPSS so that the possibility exists for switching to other computer packages later on.
For data entry first a questionnaire has to be created. These file contain the variable names, the question, the description of the variable name and the possible answers. It is strongly recommended to use the variable names provided in this handbook
Note: It is extremely important
that in each questionnaire a variable HOUSEHNO (a unique number
given to each surveyed household) is included. This variable is
the so-called "unique identifier". Furthermore, the data
from the child(ren) must contain a variable CHILDNO, which enables
one to distinguish between two or more children from the same household.
4.1.2 Plausibility check
A plausibility check must be carried out immediately after data input in order to correct erroneous data. Despite meticulous preparation and data input, erroneous data always appears in the database. Before any further processing of raw data, the data must be checked for plausibility. In part this can be done with the Nutrition Baseline software. For further information about the plausibility check in this program please read the help in this program and check the option sheet. The following procedures are recommended:
4.2 Anthropometric indices
The following are internationally used key indicators for determining the nutritional condition of children:
¯ Weight-for-age (wt/age),
¯ Height-for-age (ht/age),
¯ Weight-for-height (wt/ht).
A child lacking in energy or nutrients in relation to his height will not gain weight corresponding to his genetic potential. Conversely, an overfed child will be overweight. In addition, when a child is undernourished and/or suffers repeated infections over an extended period of time, the body growth is retarded and the genetic potential for height will not be reached.
In the population as a whole, there are always some individuals with unusual nutritional intakes who are heavier or lighter, or taller or shorter than others. If for example, the weight-for-height ratio for an adequately nourished community is determined, the ratio will have a biologically typical normal Gaussian distribution. Such a distribution has been derived for the weight-for-age, height-for-age and weight-for-height indicators for children in the USA aged from birth to 18 years of age. Separate curves have been developed for boys and for girls. These distributions are available from the National Center of Health Statistics (NCHS) for use as reference data.
If the distribution of the height-for-age ratios of the surveyed community group lies to the left of the corresponding distribution for the reference group, or if the distribution of the weight-for-height ratio lies to either side of that of the reference group, there is evidence of health or nutritional problems in the surveyed child population. Often this is estimated by determining the proportion of children at or below -2 Z-score units of the NCHS reference standard (or for weight-for-height ± 2 Z-score units). It must be kept in mind, however, that 2.3% of the children in the reference population fall naturally outside 2 Z-score units of the mean.
When plotting anthropometric measurements of groups of children, it is important to determine:
e.g. in the case of weight-for-age, height-for-age, and weight-for-height for children under five years of age:
- below -2 Z-score or above +2 Z-score units of the NCHS reference standard ( the later only in the case of weight-for-height)
in the case of mid-upper-arm-circumference (MUAC) of children:
- below 13.0 cm for children from 8 to 11 months
- below 13.5 cm for children from 12 to 23 months
- below 14.0 cm for children from 24 to 36 months
Children with a Z-score below -2 compared to the NCHS height-for-age reference are considered stunted (-2 Z-score is the value -2 standard deviations below the median of the reference population. By definition 2.3% of the reference population is below -2 Z-score).
The proportion of stunted children relative to all children examined is a valid reflection of the prevalence of the combination of disease prevalence and dietary inadequacy in light of physiological requirements in the whole community. The proportion of "stunted" children in most developing countries usually ranges between 20% and 33%. Higher figures are indicative of an "abnormal" nutritional situation, such as may arise because of long-term civil disorders, severe and chronic food shortage, or may reflect the combination of genetic selection and an abnormal geographic location, e.g., life in the Himalayas or the Andes.
Children with a weight-for-height less than -2 Z-score of the NCHS reference population are probably badly malnourished. They are only probably so, because in the "healthy" North American reference population from which the NCHS standards were derived, 2.3% were found to have this weight-for-height. If one finds significantly more than 2.3% of the children below this line, the children of the community are suffering from wasting. Since "wasting" has clearly been found to be associated with a depressed metabolism and decreased functional capacity, it is a serious sign. Wasted children experience higher mortality, probably because of reduced immune competence, from higher morbidity and from substantially reduced physical and mental abilities.
The third index used internationally as an indicator of undernutrition is the weight-for-age (wt/age) index. Most children below the threshold (-2 Z-score units below the mean for NCHS standards) are undernourished, those at or above the threshold are of normal nutritional status. This indicator combines both the ht/age and wt/ht indices. A child weighing too little for his age is too small and/or too lean (Gomez classification).
In the US reference population, the proportion of children falling below -2 Z-score units is 2.3%; however, in most developing countries in the absence of a severe food crisis the proportion is 20-40%. Higher proportions warn of a "worse-than-usual" situation.
It should be emphasized that these three indices are very useful for the classification of populations, but must be interpreted with care and require additional information for the diagnosis of the nutritional status of an individual.
4) Combination height-for-age and weight-for-height
Figure 12 presents an overview of the
thresholds of malnutrition using ht/age and wt/ht indices based on the
Figure 12. Thresholds of malnutrition
based on the Waterlow classification (Z-scores height-for-age and weight-for-height)
WATERLOW CLASSIFICATION SCHEME
||Interpretation of nutritional status|
stunted + wasted
stunted + obese
wasting: lean condition due to acute
nutritional and/or health problems
stuntin: reduced linear growth due to chronic nutritional and/or health problems
5) Mid-upper-arm-circumference (MUAC)
The Mid-Upper-Arm-Circumference (MUAC)
is easier to measure but suffers from a higher measurement error especially
if the investigator has little experience or takes insufficient care during
measurement. MUAC may be quite useful to rapidly screen for severely malnourished
children in emergency situations, in refugee camps, under severe drought
conditions, etc, or may serve in the context of a rapid assessment of the
health situation of an area. Its sensitivity, i.e. the ability of the test
to detect malnutrition, is quite high if the higher threshold value (13.5
cm) is used. At this level one may, however, miss borderline cases, i.e.
those moderately malnourished or at risk of becoming severely malnourished.
Raising the threshold to 14.5 cm will increase sensitivity but reduce specificity,
i.e. the ability of the test to correctly identify those not at risk. Arm
circumference has the advantage that it varies relatively little between
one and three years of age, i.e. it is relatively age-independent, hence
does not require knowledge of the precise age, as do weight-for-age, or
height-for-age. Above three years and below one year of age, especially
below 6 months, however, it is of limited value and should be used and
interpreted only with caution.
The forming of clusters or groups based
on a variable, say age, can help to identify statistical variations or
influences on nutritional indicators.
If clusters are to be formed, care must be taken to ensure that there is enough variation and that the number of individual cases in each group is roughly the same. If marital status is to be used as the basis for clustering and of all adults 42% are married, 30% single, 16% separated, 8% widowed and 4% divorced, the three smallest sample sizes are too small to build meaningful clusters since they are too small for statistical interpretation. Either these three groups should be excluded from the statistical analysis or they should be consolidated into one group. In this example that would be the group of the separated + widowed + divorced = 28%. In clustering it is first necessary to carry out a descriptive statistical analysis. This is easy to do with the Nutrition Baseline Software
With some variables, the basis for
clusters can easily be selected. This allows the clusters to be established
from the beginning of the project.
0-5, 6-11, 12-17, 18-23, 24-35, 36-47, 48-59
Once the clusters have been formed,
they can then be compared statistically with respect to the nutritional
indicators. An example is shown below:
wasting + stunting
In order to draw conclusions on the adequacy of nutrition, it is necessary to determine the individual requirements for a healthy nutrient intake. Such a relationship is, however, in practice very difficult to determine during a routine field survey because of
With the Nutrition Baseline software it is possible to calculate the nutrient intake for the mother and the child but the food weighing method or the 24-hour-recall method are not recommended as part of the surveys discussed in these guidelines. Based on previous experience, the usability and accuracy of the findings do not justify the expense involved. Therefore, only the typical food customs in a country are documented via a food frequency recall. Consequently, only qualitative conclusions concerning nutrient supply can be drawn, but it is possible to derive qualitative statements on the possible nutrition problems within a family.
Before analyzing a survey on feeding practices, inquiries should be made to the health authorities concerning the nutrient value tables used in their country.
FAO/WHO recommendations for nutrient
requirements are generally used for developing countries. Chapter 6.4 in
the Appendix contains an overview of nutrient recommendations for children
up to six years. However, authorities of many countries have published
their own recommendations.
Breast-feeding, supplementary feeding, and weaning practices
A major cause of undernutrition and malnutrition in developing countries lies in improper breast-feeding, supplementary feeding, and weaning practices. The following principles are recommended:
Table 8. Definition of breast-feeding
categories according to WHO
|Category||A child must be given:||A child may be given:||A child may not be given:|
|Exclusive breast-feeding||Human breast milk||Medicines, vitamins, trace elements||Anything else|
|Predominant breast-feeding||Mostly human breast milk||Fluids (water, or water based fluids)||Anything else (including substitutes for human breast milk)|
|Complementary feeding||Human breast milk and semi-solid or solid foods||Usual foods and beverages|
|Bottle feeding||Usual fluids or semifluid foods from bottles with nipples||Also human breast milk from a bottle|
The following indicators are generally used for breast-feeding, supplementary feeding, and weaning practices:
Exclusive colostrum feeding rate
The proportion of infants reported to have received exclusively colostrum during the first 12 hours after birth.
(Infants <12 months (<365 days)
of age reported to have been exclusively breast-fed with colostrum in the
first 12 hours after birth) /
(All surveyed infants <12 months (<365 days)of age)
Exclusive breast-feeding rate
The proportion of infants under 4 months of age exclusively breast-fed.
(Infants <4 months (<120 days)
of age who were exclusively breast-fed in the last 24 hours) /
(All surveyed infants <4 months (<120 days) of age)
Predominant breast-feeding rate
The proportion of infants under 4 months of age predominantly breast-fed.
(Infants <4 months (<120 days)
of age who were predominantly breast-fed in the last 24 hours) /
(All surveyed infants <4 months (<120 days) of age)
Timely, complementary feeding rate
Proportion of infants 6-9 months of age receiving breast milk and complementary foods
(Infants 6-9 months (180-299 days)
of age receiving complementary foods in addition to breast milk
in the last 24 hours) /
(All surveyed infants 6-9 months (180-299 days) of age)
Continued breast-feeding rate (1
Proportion of children 12-15 months of age who are breast-feeding
(Children 12-15 months of age breast-fed
in the last 24 hours) /
(All surveyed infants 12-15 months of age)
Continued breast-feeding rate (2
Proportion of children 20-23 months of age who are breast-feeding
(Children 20-23 months of age breast-fed
in the last 24 hours) /
(All surveyed infants 20-23 months of age)
Proportion of infants less than 12 months of age who are receiving any food or drink from a bottle
(Infants <12 months (<365 days)
bottle-fed during the last 24 hours) /
(All surveyed infants <12 months (<365 days) of age)
Reliability of a survey
There are three basic types of variables, and the method used to check the reliability of a survey depends on the type of variables. Reliability is a measure of the repeatability or reproducibility of the results.
Variables with continuous values of measurement
For variables with continuous values, such as height, weight, or hemoglobin levels, the mean and the standard deviation of the difference between the recorded measurements of the enumerators and the supervisor should be calculated. A small mean and small SD are desirable. In annex 6.7 an example of the calculation of an intra- and inter observation is described.
Variables with yes/no answers
For variables permitting
only yes/no answers, such as the presence of goiter, anemia, or undernutrition
(Z-score < 2), the inter and intra observer errors of
the surveying of these variables should be calculated according to the
a: number of identified observations
detected jointly by the enumerator and the supervisor
b: number of observations not detected by the enumerator but detected by the supervisor
c: number of observations detected by the enumerator but not detected by the supervisor
d: number of observations neither detected by the enumerator nor and the supervisor
Sensitivity of finding (proportion of true positive): a / (a + c) x 100
The sensitivity indicates the proportion of surveyed individuals actually possessing an observed attribute (undernourishment, belonging to a certain income group, etc.) who were correctly identified.
Specificity of finding (proportion of true negative): d / (b + d) x 100
The specificity indicates the proportion of surveyed individuals not possessing an observed attribute who were correctly identified.
Ideally, both sensitivity and specificity should be 100. The lower the value, the less precise was the surveying of the data.
Variables allowing more than two answers
In determining the reliability of answers to questions with more than two answers, the percentage deviation between the enumerators and the supervisor is calculated. The smaller the deviation, the better the survey.
% deviation = (Number of answers with
differences between the responses obtained by enumerators and those obtained
by the supervisor) /
(Number of all answers with responses) x 100
Analysis of causes and predictors
With statistical methods it is possible to identify the relationship between the variables investigated in the survey. If there is a statistically significant association between two variables, it is called a predictor. A predictor suggests a causal relationship between two variables, but does not prove that one exists.
Three goals can be achieved in the analysis of predictors.
Significant associations between a nutritional indicator and a socioeconomic indicator can help to identify social risk groups, such as the landless and illiterates.
It is also possible to arrive at predictors based on observed socioeconomic data in which several individual predictors are included. Thus it was possible in the slums of a Brazilian city to find a higher risk of malnutrition in young children from families who obtain their electrical power supply from neighbors. Undoubtedly the cause of the malnutrition is not in the type of electrical power supply, and therefore an improvement in type of electricity supply would not result in an improved nutritional condition for the children. Rather, there are other causal factors involved in these predictors that are responsible for the poor nutritional conditions, such as low income and little education. Therefore, although the means of obtaining electrical power is not the cause of the malnutrition, defining these predictors can help the consultants to identify risk groups more quickly and at less cost. For example, in this way families who have no electrical supply for their households can be selected for individual discussion on health and nutrition.
Predictors can be helpful in identifying indicators suitable for use in monitoring the impact of intervention measures.
Predictors can also provide important information regarding causal relationships and form the first steps for analysis of causes.
Three procedures can be employed for analysis of causes:
Statistically significant relationships between variables do not automatically provide information concerning the cause of nutritional problems. For this a separate analysis of causes is necessary.
1. Multiple statistical analysis of causes
In principle, in a statistical investigation of causes in order to obtain a clearer picture of a relationship, consideration should be given to the fact that a cause which cannot be eliminated has an integral part in the relationship. A check must therefore be made as to whether the available statistical information provides sufficient basis to be able to assume a relationship. Therefore data of all potential confounders must be available. Some important tests with the related assumptions are presented in the table of annex 6.8.2. Analyses of causes by comparison of literature
Special attention should be given to multiple regression analysis and multiple ANOVA. These most used statistical techniques allow among others the investigator to examine the effect of one factor while other factors in the regression equation are held constant mathematically.
A second important source for analysis of causes is literature. A cause-effect relationship can only be discussed with greater certainty, if a statistically established connection has been defined in another similar situation.3. Analysis of causes through specific investigations
The baseline survey must commence from the viewpoint that although nutritional problems have been solved in other locations more extensive research is needed to reveal causal relationships in the present location. The information gathered and relationships proposed in previous studies should be carefully considered during the planning for the baseline survey.If there is no statistically significant relationship between variables, either no association exists or one or more of the following factors may have influenced the results:
If there is a statistically significant relationship between two variables, the following questions must be considered:
1. Differences in the survey teams and their equipment
A survey must be implemented by a several survey groups. Although the teams along with their equipment should randomly select households for the survey, in practice the surveys by a given team will be limited to one geographical area to save time. Thus, differences in the results can result from different teams and equipment.2. Changes in demographic distribution
It is possible that during the project/program segments of the population in the designated region emigrate or immigrate. Those segments might not be representative of the general population. An example, they might be the poorest with the highest nutritional risk.3. Decreases or increases in the beneficiary group
During the project members of the beneficiary group enter or exit. This shortens the time available for the intervention, and the project/program measures cannot be sustained long enough to take effect.4. Excessive representation of socioeconomic data from families with more than one child under 5 years
Many households have more than one child under the age of 5 years. To secure the representation of all children under five years of age and to increase the coverage of sampling, a survey has to be taken of all children in a household. If the socioeconomic data of households is used in a statistical relationship to the socioeconomic data of the children, the data from the families with more than one child will be over-represented.5. Changes in external conditions
It is not always possible to arrange for a comparison group to evaluate the impact of intervention measures on the nutritional condition of a target group (see sub-chapter 126.96.36.199.). Without data from a comparison group, the interpretation of the impact of nutritional intervention based on longitudinal comparisons of baseline and follow-up surveys can be exceedingly difficult and must be undertaken with extreme caution, as other factors outside the control of the project/program can also have a bearing on nutritional status.6. Changes in the age distribution of the surveyed group
The nutritional status of babies from zero to five months is often better than that of babies from twelve to seventeen months. If the age distribution of the examined babies is different between the baseline and follow-up surveys, the impact of an intervention can be misrepresented.The last two points are applicable only to follow-up surveys.
Evaluation of indicators
Nutrition surveys yield important information about the nutritional condition of an investigated population. Many conclusions can be drawn on the type and distribution of evident nutritional problems using statistical analyses of the observed nutritional indicators (anthropometric indices, xerophthalmia, anemia, goiter, breast-feeding practices, etc.). However, these do not provide answers about how these findings are to be interpreted if questions arise concerning
There are no universally valid thresholds for nutritional indicators, which could then be used to decide which measures to apply. If one attempted to specify a certain set of economic and health conditions, a 10% stunting rate in one part of a country might be high, while in another part of the country, or in another country, it might seem low. It is therefore necessary to establish a list of criteria that can be used to evaluate the indicators.
The following criteria should be considered when evaluating nutritional indicators:
Reporting of survey results
5.1 Format of a technical report
A technical report should have the following structure:
Title page0. Abstract
Table of contents
1.1 Project/program description and overall framework2. Methodology
1.2 Nutritional situation based on previous reports
1.3 Objectives of the survey
2.1 Population and area surveyed3. Results
2.2 Study design
2.3 Sampling and data collection
2.4 Equipment utilized
2.5 Statistical methods
2.6 Sensitivity and specificity of the survey
2.7 Ethical considerations
3.1 Demographic and socioeconomic data4. Discussion
3.2 Food habits
3.3 Infectious diseases
3.3 Anthropometric data
3.5 Nutritional deficiencies
3.6 Existing intervention programs in the project/program area
3.7 Predictor analysis
4.1 Analysis of findings5. Recommendations
4.2 Problem tree
5.1 Possibilities for intervention6. List of references
5.2 Proposed in-depth studies
5.3 Proposed indicators for monitoring and evaluation
7.1 Questionnaires utilized
The report should include the following information:
l Title page
The purpose of the title page is to present a concise statement of the subject of the survey and to identify the responsible personnel . The title page is the "main gate" of the survey report that invites the reader to engage him/herself to study the document. The title is the summary of the summary.
The title page should contain the following information:
- Title of the reportl Acknowledgment
- Names of the principle enumerators and authors
- Date of submission of the report
- Name and address of the institution of the principle enumerators and authors
All persons, institutions and groups that assisted in carrying out the survey and without whose support the survey would not have been possible should be acknowledged.
The purpose of the abstract is to summarize in less than 400 words all important parts of the survey.
The abstract should
- describe the general objective of the survey (justification);l Introduction
- describe briefly the methodology used;
- report the main findings;
- discuss and analyze the main results;
- state the main conclusions and recommendations.
The purpose of the introduction is to put the survey in perspective.
The introduction includes several sections:
The description of the methodology should be so complete that it would be possible to repeat the survey using the same methodology.
When describing findings, data should be presented objectively without any comment or commentary.
The following tables, arranged according to communities or urban districts, should be included in the results section:
- height vs. age compared to the reference group
- weight vs. height compared to the reference group
- weight vs. age compared to the reference group
- frequency of distribution of Hb concentration in the blood
- Exclusive breast-feeding rate (proportion of infants under 4 months of age exclusively breast-fed) vs age of mother
Finally, the survey results should be analyzed in the light of other findings. The discussion of the results must be separated from their presentation. The clear separation of the presentation from the discussion of the results allows readers to draw their own unbiased conclusions from the survey findings.
Concrete information must be given concerning the survey goals described in the Introduction, as follows:
This section should contain a bibliography of the literature referred to in the report.
l Appendix section
The appendix section should include a copy of the actual questionnaire used in both the language of the investigators and the local language. The questionnaire is an important tool of a survey that has to be adapted to the local situation and must therefore be developed for each survey. The setting in which the data will be collected will influence the design and structure of the data recording form or questionnaire.
In addition, it is also possible to
use the appendix section to include tables useful for clarification but
too extensive for the main section of the report.
Considerations of style for writing the report
Tables and figures are important tools for information transfer in reports. However, there are some rules that should be considered when preparing a table or figure.
Boys n = 4
Girls n = 40
Total n = 44
Other style considerations include:
5.3 Information for the target groups
Once the survey is completed, the target group should quickly be informed of the findings. The presentation and contents should be tailored to the culture and educational level of the people.
A place and time for the meeting should be chosen so as to include as many people as possible. The questionnaire should include a question about when and where such a meeting should be held.
The following recommendations are given concerning the presentation and discussion of the results in a community: