You are on page 1of 15

Validity of food consumption indicators in the Lao

context: Moving toward cross-cultural standardization

Soo Mee Baumann, Patrick Webb, and Manfred Zeller

Abstract settings is required to generate more insight into the


extent to which universal thresholds can be applied to
Background. Cross-cultural validity of food security dietary diversity indicators with or without locally deter-
indicators is commonly presumed without questioning mined correction factors such as the exclusion of small
the suitability of generic indicators in different geographic amounts of food items.
settings. However, ethnic differences in the perception of,
and reporting on, food insecurity, as well as variations
in consumption patterns, may limit the comparability Key words: Cross-cultural validity, dietary diversity,
of results. Although research on correction factors for food consumption indicators, Food Consumption Score
standardization of food security indicators is in process,
so far no universal indicator has been identified.
Objective. The current paper considers the ability of Introduction
the Food Consumption Score (FCS) developed by the
World Food Programme in southern Africa in 1996 to Indicators for measuring food security are commonly
meet the requirement of local cultural validity in a Lao- used across regions following the implicit assumption
tian context. The analysis is based on research that seeks that a tool valid for one region may be applied in a dif-
to identify options for correcting possible biases linked to ferent cultural setting as well. However, food security
cultural disparities. indicators often fail to meet the requirement of cross-
Methods. Based on the results of a household survey cultural validity. A measurement tool valid across dif-
conducted in different agroecological zones of Laos ferent cultural environments should be appropriate in
in 2009, the FCS was validated against a benchmark multiple geographic settings with results comparable
of calorie consumption. Changing the thresholds and across multiple settings; i.e., anthropometric indica-
excluding small amounts of food items consumed were tors for children under five can be compared across
tested as options to correct for biases caused by cultural different regions of the world given the global reference
disparities. standards established by the World Health Organiza-
Results. The FCS in its original form underestimates tion (WHO) in 2006. However, ethnic differences in
the food insecurity level in the surveyed villages. How- the perception of, and reporting on, food insecurity,
ever, the closeness of fit of the FCS to the benchmark as well as variations in consumption patterns, may
classification improves when small amounts of food items limit the comparability of results [1]. This is potentially
are excluded from the assessment. problematic, since having standardized measures of
Conclusions. Further research in different cultural food insecurity indicators would constitute an impor-
tant tool for practitioners in development programs.
Although research on correction factors for stand-
Soo Mee Baumann and Manfred Zeller are affiliated with the ardization of food security indicators is in process, so
University of Hohenheim, Stuttgart, Germany; Patrick Webb is far no universal indicator has been identified [2, 3],
affiliated with Tufts University, Boston, Massachusetts, USA. in contrast to measurement instruments for poverty,
Please direct queries to the corresponding author: Man- which are corrected according to the effective purchas-
fred Zeller, Chair of Rural Development Theory and Policy,
Department of Agricultural Economics and Social Sciences ing power of local currencies. If correction factors for
in the Tropics and Subtropics, University of Hohenheim, different settings can be clearly identified, such factors
Wollgrasweg 43, 70599 Stuttgart, Germany; e-mail: manfred. may contribute to the development of standardized
zeller@uni-hohenheim.de. tools for food security.

Food and Nutrition Bulletin, vol. 34, no. 1 © 2013, The United Nations University. 105
106 S. Baumann et al.

The current paper considers whether the Food The Food Consumption Score
Consumption Score (FCS) developed by the World
Food Programme (WFP) in southern Africa in 1996 The FCS was developed by WFP in order to improve
meets the requirement of cultural validity in the the assessment of food security levels in a way that is
Laotian context. The analysis is based on research tailored to WFP information needs and as a stand-
that seeks to identify options for correcting possible ard tool to further harmonize WFP’s data analysis
biases linked to cultural disparities. In order to test the [4]. Among several tools to assess food security, the
suitability of the FCS to measure food consumption measurement of food consumption in kilocalories is
in Laos, the indicator was validated against a bench- considered one of the best indicators available. How-
mark of calorie consumption and the association of ever, determining food security levels by means of food
FCS, and the benchmark was tested through bivariate consumption in kilocalories is highly complicated and
descriptive analysis of food groups, individual food time consuming. In order to achieve a flexible data
items, and Calorie Consumption Groups (CCGs). By collection instrument that can be applied in different
doing so, the contribution of individual food items contexts and provide reliable and comparable results,
and groups to the ability of the FCS to differentiate by WFP uses proxies for actual caloric intake and diet
CCG was examined in detail. Changing the thresholds quality. For this reason, the FCS has been developed as
and excluding small amounts of food items have been a standard methodology. As a proxy indicator, the FCS
tested as options to correct for biases that might be “represents the dietary diversity, energy, and macro and
caused by cultural disparities. To test the suitability of micro (content) value of the food that people eat” [5].
the original FCS thresholds in the Laotian context, a The FCS is a composite score based on dietary diversity,
sensitivity–specificity analysis was conducted. food frequency, and relative nutritional importance of
The FCS is a proxy indicator for food consump- different food groups. Dietary diversity refers to the
tion: it measures food intake and diet quality. It is a number of food groups a household consumes over a
frequency-weighted diet diversity score multiplied by certain reference period (usually 7 days for the FCS),
the relative nutritional importance of different food without regard to the frequency of consumption; food
groups. The FCS has been developed by WFP as a frequency, in contrast, is defined as the number of days
harmonized measurement tool for rapid assessment on which a particular food group is consumed over a
of the current food security situation in a given area. reference period. The weights for the FCS were deter-
Correction factors for the FCS may contribute to the mined based on an interpretation by a team of analysts
overall process toward a standardization of food secu- of “nutrient density.” “Nutrient density is a term used
rity indicators. The FCS is based on dietary diversity to subjectively describe a food group’s quality in terms
and food frequency data that have proven to be strong of caloric density, macro and micro content, and actual
proxies for food security and are promising indicators quantities typically eaten” [4]. Pulses, for example,
with a reasonable data collection and time frame. The are given a weight of 3, reflecting their high protein
Food and Agriculture Organization (FAO) and WFP content, whereas sugar is given a weight of only 0.5,
both apply proxies based on dietary diversity or food reflecting its lack of micronutrients [4]. Tables 1 and 2
frequency information. Identifying correction factors show how the FCS is calculated with the food diversity
for the FCS will provide valuable information for the and frequency information and the respective weights
further development of a standardized tool based on for each food group.
diet diversity and food frequency information. The information required for calculating the FCS is

TABLE 1. Food groups and weights


No. Food items Food group Weight
1 Rice, maize, cassava, other roots and tubers Main staples 2
2 Fish, other aquatic animals, poultry, pork, red meat (goat, Animal protein 4
sheep, beef, buffalo), wild meat (large- and small-bodied
wildlife), eggs
3 Pulses, lentils, tofu, bean curd Pulses 3
4 Vegetables, leafy vegetables, bamboo shoots, mushrooms Vegetables 1
5 Oils, fats, and butter Oil 0.5
6 Fruits Fruit 1
7 Sugar Sugar 0.5
8 Milk, milk products Milk 4
Source: Adapted from World Food Programme [4].
Cross-cultural validity of food consumption indicators 107

collected by asking households to recall on how many thresholds used are: poor food consumption (0–21),
days certain food items have been consumed within the borderline food consumption (21.5–35), and acceptable
past 7 days. The FCS measures the frequency of con- food consumption (> 35). If the consumption of oil and
sumption in days instead of calculating single meals. sugar is frequent in a population, while the consump-
No matter whether an item is eaten once or several tion of other food groups is rare, WFP recommends
times during one day, each day is counted as a fre- raising the thresholds from 21 and 35 to 28 and 42
quency of one (one day). In the next step, the frequency by adding 7 to each threshold. A daily consumption
of consumption of each food item is summed up into of oil and sugar adds 7 points to the FCS: poor food
eight standard food groups and truncated at 7 days. If consumption (0–28), borderline food consumption
the sum of frequencies of all food groups amounts to (28.5–42), and acceptable food consumption (> 42).
more than 7, this value is recoded as 7, constituting the Households in the poor group and the borderline group
maximum number of days a food group can have been are usually considered food insecure, while households
consumed within the past 7 days. The consumption in the acceptable group are categorized as food secure.
frequency of each food group is then multiplied by an The major advantage of the FCS is its convenience.
assigned weight that is based on its nutrient content. Data collection is easy and fast, and the respondent
Finally, the weighted frequency of consumption of each burden is comparably low. Respondents are generally
food group is summed up into the FCS. The maximum able to answer the questions without difficulty, and
value for a household score is 112; this score can only enumerators can be trained within a short period of
be reached if a household consumes food from each time. Moreover, data analysis follows a recurring pat-
food group on every day within the reference period. tern and thus does not require more than basic data
Consequently, the FCS can be described with the analysis skills. The FCS not only shows the current
following function: status of food insecurity, but may be used for food
security monitoring. If surveyed regularly, the FCS
FCS = astaple xstaple+ apulse xpulse+ aveg xveg+ afruit xfruit
may display seasonal variations in food consump-
+ aanimal xanimal+ asugar xsugar+ adairy xdairy+ aoil xoil
tion and changes over several years. The FCS can be
useful for depicting changes, as it not only estimates
where
which households are food insecure or secure, but also
FCS = Food Consumption Score; distinguishes between the degrees of food insecurity.
xi = frequency of food consumption = number Thus, the FCS also provides information on whether a
of days on which a food group was con- food-insecure household is close to being food secure,
sumed during the past 7 days; and and likewise if a food-secure household is on the brink
ai = weight of each food group. of food insecurity.
The major limitation of the FCS is that it does not
The Food Consumption Group (FCG) is determined record quantities. Thus, the food security level of
by comparing each calculated household score with households consuming very small amounts of various
pre-established thresholds to determine which FCG food items is easily overestimated. Furthermore, the
the household should be assigned to and whether FCS does not provide information on intrahousehold
the household is food secure or insecure. The typical food consumption. Within a household displaying

TABLE 2. Completed Food Consumption Score (FCS) template


Days eaten in
past 7 days,
truncated at Weight Score
Food items Food group 7 (A) (B) (A × B)
Glutinous rice, white rice, maize, cassava, other roots and Staples 7 2 14
tubers
Fish, other aquatic animals, poultry, pork, red meat, wild Animal protein 2 4 8
meat, eggs
Pulses, nuts, bean curd Pulses 0 3 0
Green leafy vegetables, bamboo, other vegetables Vegetables 5 1 5
Oil and fats Oil and fats 2 0.5 1
Fresh fruits Fruit 2 1 2
Sugar Sugar 0 0.5 0
Milk and milk products Dairy products 0 4 0
Composite score 30
Source: Adapted from World Food Programme [4].
108 S. Baumann et al.

food-secure consumption according to the FCS, single food items commonly consumed in Lao PDR. All tra-
members might nevertheless be food insecure. Another ditional measurement units were converted into metric
difficulty associated with the FCS is the necessity for units either through direct on-the-spot weighing or by
respondents to remember all food items consumed measuring amounts reported on the market.** In rural
during the reference period. Food-poor households areas of Laos, food items are often sold at fixed prices
usually display a much more repetitive consumption in varying amounts.*** If respondents could not recall
pattern and can easily recall which foods were bought the quantity of a food item they bought at the market,
at the market. Nevertheless, at times respondents may they usually remembered the amount of money they
forget to report single food items. Moreover, the FCS spent on it. With that information, the interviewing
displays limitations by design: due to truncation, the team went to the respective markets and weighed the
FCS is less precise for households with a higher score, quantity of the food item sold at this price. As men-
and weights given are based on the appraisal of a nutri- tioned before, quantities vary, especially over seasons
tion expert group only. In the overall validation study and years. This increases the chance of the assessments
of the FCS, conducted by the International Food Policy to be imprecise. Yet, it is unlikely that quantities sold
Research Institute (IFPRI) [6], the correlation of the at a certain price changed drastically over a few days.
FCS with the benchmark increased when truncation Moreover, rural households in Laos usually have few
and weighting were omitted.* options to store fresh food items. Thus, foodstuffs are
The number of food items asked about during the usually purchased shortly before consumption.
FCS data collection is not predetermined and may After translating all food items consumed by all
vary between countries. Yet, the number of food items households interviewed into metric measurement
asked about influences the resulting score of a house- units, the quantity of each food item was converted
hold. Increasing the number of food items increases into kilocalories with the help of the ASEAN Food
the probability that households will report all foods Composition Table from 2000 [7] and the Food Com-
eaten but may inflate the FCS because different food position Table for Use in East Asia from 1972 [8].
items from the same food group eaten on the same Before converting into kilocalories, the so-called refuse
day are double counted. For example, if a household was subtracted from the food quantities. Refuse refers
consumes both cabbage and mushrooms in one day, to the nonedible parts of a food item such as peels,
because mushrooms are recorded separately from other bones, or kernels and is usually given as a percentage.
vegetables, together their frequency will be recorded as Unfortunately, there is no food composition table for
2, even though the vegetable food group was consumed Laos in particular. However, a compendium from the
on only one day. Lao Ministry of Health with certain Lao food items that
Lastly, the FCS only captures a snapshot of the cur- are commonly consumed in Laos and their respective
rent food consumption and cannot distinguish whether nutrients was consulted [9].
the food consumption reported for the recall period In order to take varying individual food needs into
reflects average food consumption or might have account, adult equivalents were calculated by using
resulted from exceptional consumption due to special the method applied in the IFPRI validation study [6].
events or a certain period of the year. More information The requirements for a household with newborns and
is needed if consumption practices and trends are to be young children will be less than those for a household
fully understood. FCS data collection should be com- of the same size with adults only. Applying adult
plemented by questions on usual food consumption equivalents allows for comparison between house-
and context questions if a more in-depth understand- holds of different age and sex structures. Households
ing of the reported consumption is required. with calorie consumption/adult equivalent/day above
5,000 kcal or below 500 kcal were considered outliers
and excluded from the analysis in line with the iden-
Methods tification of outliers in the IFPRI validation study [6].
The newly created variable of food consumption in
Kilocalorie benchmark kilocalories per adult equivalent (Calorie Consumption

The quantity information required for calculating the ** All food quantities refer to uncooked and raw food items,
benchmark variable calorie consumption was collected with the exception of processed foods such as the commonly
consumed fermented fish paste. Usually refuse was included
using a comprehensive 7-day recall module covering all when measuring food items. If only edible parts were report-
ed, this was noted and considered later during the calculation
* IFPRI has conducted an overall validation study of the of the calorie consumption indicator.
FCS based on data from Burundi, Haiti, and Sri Lanka in *** For example, people buying mustard leaves on the mar-
which alternative dietary diversity and food frequency indica- ket would not ask to buy a certain amount but would state
tors were constructed and compared with the performance how much money they were willing to spend. The number
of the FCS. The study recommends further validation of the of mustard leaves they would receive would then depend on
FCS at the country or (sub)region level. the price.
Cross-cultural validity of food consumption indicators 109

TABLE 3. Cutoff points for creating Calorie Consumption


purpose of gaining information on whether the FCS is
Groups (CCGs)
suitable as an outcome indicator for WFP operations.
Calorie consumption Therefore, villages receiving WFP food assistance and
(kcal/adult equivalent/ day) Shortfall (%) Profile villages that had never participated in WFP operations
< 1,470 > 30 Poor were chosen. Furthermore, villages that were included
1,470 to < 2,100 30 to > 0 Borderline in the CFSVA based on a countrywide two-stage cluster
≥ 2,100 0 Acceptable sample were chosen in order to be able to compare FCS
Source: Adapted from World Food Programme [10]
results from 2006 with those from 2009.** From each
village, 12 households were randomly selected using a
list of all households provided by the village heads.*** In
per Adult Equivalent) was then transformed from total, 96 households were interviewed.
a continuous to a discrete variable by using existing
cutoff points for poor, borderline, and acceptable food Questionnaire
consumption. The food consumption module of the
Emergency Food Security Assessment Handbook sug- The questionnaire for the validation study included a
gests two cutoff points derived from a basic minimum section on the FCS taken from the Lao CFSVA in 2006
dietary energy requirement of approximately 2,100 and extended by a column that allows separate collec-
kcal/capita/day, corresponding with shortfalls of 0% tion of data on the number of days in which a food
and 30% relative to requirements (see table 3). item was consumed in small amounts only. Those days
were later subtracted from the total number of days a
Sampling and questionnaires food item was consumed to test the effect of excluding
small amounts on the predictive power of the FCS.
In order to capture the prevailing diversity within the Small amounts were defined as 1 level tablespoon per
country, households in different agroecological zones person per day or less. The tablespoon was shown to
in the north and south of Laos were selected. According the respondents during the interview. For the section
to the Comprehensive Food Security and Vulnerabil- on calorie consumption, the general questionnaire
ity Analysis (CFSVA) by WFP [11], several variables, design was taken from the IFPRI validation study;
such as wealth, access to safe water, livelihoods, access however, food items were adapted to include only
to forest resources, and prevalence of nutritional those commonly consumed in Lao PDR. During the
status, vary tremendously between agroecological Lao Expenditure and Consumption Survey III (LECS
zones. The classification used in the CFSVA is based III) [13], all households were asked to fill out a trans-
on a World Bank study [12]. Districts are assigned to action diary, which, among other things, included all
zones according to indicators related to agricultural the food items acquired by the household during an
potential and ecological similarity or difference, which entire month. The LECS III was conducted over a
provide a more useful frame than a distinction along whole year, which reduced the chance that food items
administrative units. The classification distinguishes would be excluded due to seasonality. Taking from all
between six major agroecological zones: Vientiane the transaction diaries of the LECS III resulted in a list
Plain, Central and Southern Highlands, Bolaven Pla- of more than 150 food items reported to be consumed
teau, Mekong Corridor, Northern Lowlands, and by households all over Laos. Furthermore, the Depart-
Northern Highlands. ment of Statistics set up a list of all measurement units
Data from three different agroecological zones were used by the households to report food quantities. Both
collected. Muang Sing belongs to the Northern High- lists were made available to be incorporated into the
lands, Laongam is part of the Central and Southern questionnaire for the FCS validation study in Laos.****
Highlands, and Toumlan is in the Mekong corridor.
The Vientiane Plain, the Bolaven Plateau, and the
Northern Lowlands were not covered. These three
agroecological zones are relatively wealthy. Working ** The sampling frame for the CFSVA included all rural
in the better-off zones might have increased the pos- households but excluded Vientiane Capital and Vientiane
sibility of no single household displaying food-poor Province.
consumption, which would have been less favorable *** From each village 12 households were chosen, and
another 5 households were selected to replace the chosen
for the validation. households if necessary.
Out of the three districts, eight villages were select- **** During the interview, all nonmetric measurement units
ed.* The village selection was oriented toward the reported were asked to be shown and if possible directly
measured, and both the nonmetric and the metric equivalents
were noted. For measurement units that were not available
* The villages selected were Ban Nammai, Ban Xo, Ban in the village, corresponding amounts were measured in
Tapao-Donpoi, Ban Sandong, Ban Mouamterb, Ban Donm- the market and cross-checked with prices reported by the
ouang, and Ban Hongloaong. households.
110 S. Baumann et al.

TABLE 4. Accuracy table


“True” food insecure “True” food secure
(benchmark) (benchmark)
P N Total
Predicted food insecure (FCS) pʹ True positives (TP) False positives (FP) Pʹ
Predicted food secure (FCS) nʹ False negatives (FN) True negatives (TN) Nʹ
Total P N
Source: Adapted from International Food Policy Research Institute [6].

Data analysis methods analysis is a more sophisticated way of assessing the


goodness of the FCS classification, and is at the same
In order to assess how well the FCS predicts the food time applicable for the optimization of the thresholds
security level for the Lao context, the FCS is vali- because it systematically examines errors of exclusion
dated against the benchmark of calorie consumption. and inclusion with reference to the benchmark.
Validation is understood as the process of determining In order to assess the general suitability of a proxy
whether a method provides a useful analytical meas- indicator, the Area Under the Curve (AUC) from
urement for a given purpose and context [14]. Follow- Receiver Operating Curve (ROC) analysis is an impor-
ing the validation study conducted by IFPRI [6], the tant test statistic. The ROC plots sensitivity against 1
predictive power of the FCS was tested with a range of – specificity for all thresholds of the test variable (e.g.,
analytical techniques, in particular, descriptive analysis, the number of food groups) based on the results of
scatter plots, correlation analysis, bivariate analysis, cross-classifications with the dichotomous benchmark
cross tabulation, and sensitivity–specificity analysis. variable. Sensitivity refers to the proportion of food-
The bivariate descriptive analysis was undertaken insecure households also classified by the proxy indica-
for a more in-depth understanding of the association tor as food insecure; specificity, in contrast, describes
between the FCS and the benchmark variable. With the the proportion of food-secure households also classi-
help of this analysis, the contribution of individual food fied by the proxy indicator as food secure. Sensitivity
items and groups to the ability of the FCS to differenti- increases with higher thresholds (errors of exclusion
ate by CCG may be examined. The bivariate analysis decline), while specificity falls (errors of inclusion rise).
is carried out for the original FCS (without quantity Thus, there is always a tradeoff between sensitivity and
restrictions) as well as after exclusion of small amounts. specificity. If the FCS shows good properties for the
In order to compare classifications based on the FCS classification, it will produce an ROC well above the
with those of the benchmark, the two indicators were diagonal of the graph, indicating combinations of high
cross tabulated. Because of its design, cross tabulation sensitivity and specificity for a range of thresholds. If
may, however, produce a relatively high proportion the ROC coincides with the diagonal, the association
of “close matches,” even though the observations are between the variables is purely random. The AUC will
equally distributed across cells and no association of be 0.5, and no acceptable combinations of sensitivity
the two variables exists. and specificity will be found. A perfect classification
For testing the ability and extent to which the FCS by the proxy indicator would result in an AUC of 1.00.
may predict the calorie consumption benchmark, As a general rule of thumb, an AUC below 0.6 is con-
different options to improve the predictive power sidered not acceptable, above 0.7 is good, higher than
of the FCS were examined. First of all, the impact of 0.8 is very good, and greater than 0.9 is excellent [6].
excluding small amounts from the analysis was tested. The components relevant for testing the suitability
In order to show the goodness of the FCS when small of FCS thresholds in order to optimize the FCS pre-
amounts were excluded, the analysis, which included diction of the benchmark classification are shown in
descriptive analysis, scatter plots, correlation analysis, table 4 [15]. A suitable cutoff point for classifying the
and cross tabulation, was repeated with an FCS for food security status can be found by seeking to balance
which small amounts had been excluded. Moreover, we sensitivity and specificity and to minimize the propor-
assessed whether thresholds can be found for which the tion of households misclassified [16].** As a rule of
FCS mimics the food security level of the benchmark thumb, sensitivity and specificity should amount to at
better, and if so, which thresholds are most suitable.
Therefore, the FCS was calibrated against the bench- are also referred to as leakage (households receive food
mark. Considering, though, that calibration does not assistance even though they are not in need) and errors of
provide information on errors of inclusion or exclusion, exclusion as undercoverage (households in need of food as-
sistance are not targeted).
a sensitivity–specificity analysis was conducted.* This ** Minimizing the proportion of households misclassified
will at the same time maximize total accuracy. The proportion
* When using an indicator for targeting, errors of inclusion of misclassified equals 1– total accuracy.
Cross-cultural validity of food consumption indicators 111

50

40

Cumulative consumption frequency


30

20

10

0
21
23
24
25
26
27
28
29
30
31
32
33
34
35
37
38
40
41
42
43
44
45
47
48
49
50
52
53
54
55
56
58
59
60
61
62
63
64
66
72
76
Borderline Acceptable

Food consumption score


Staples Vegetables Pulses Fruit Animal protein Dairy Sugar Oil and fat

FIG. 1. Stacked food frequency of food groups. FCS, Food Consumption Score

least 60%, and the percentage misclassified should not adult equivalent/day would be considered outliers and
exceed 30%. However, sensitivity–specificity analysis would be excluded from the analysis.
does not guarantee that chosen cutoff points will yield Figure 1 presents the stacked food frequency of the
realistic estimates of the prevalence of food security food groups as it evolves with an increasing FCS.** The
[6]. Sensitivity–specificity analysis requires dichoto- figure shows that during the reference period, almost
mous variables for the benchmark as well as the proxy all households interviewed reported consumption of
indicator, dividing the households into either food staples on a daily basis and a high consumption of
insecure or food secure. In order to test for thresholds vegetables. The food groups that were barely consumed
between all three FCGs, the analysis was done for the by food-poor households and were more frequently
lower threshold (between poor and borderline) and consumed with an increasing FCS are animal protein
upper threshold (between borderline and acceptable) and oil. Fruit and sugar were very rarely consumed by
separately. households with poor or borderline food consump-
tion, and only those in the high end of the acceptable
FCG seem to consume fruit and sugar on several days
Results per week. Pulses were barely consumed, and no single
household reported having consumed dairy products.***
The data collected were entirely usable, with one Figure 2 shows that the FCS calculated as recom-
exception.* A single household was excluded from mended by WFP gave much lower estimates of food
the analysis as an outlier, with its members reporting insecurity than did the calorie consumption indicator.
a consumption of only 231 kcal/adult equivalent/day. According to the FCS, only 1% of households were food
Even though this low calorie consumption accurately poor, in contrast to the benchmark that indicated 21%
reflects a small food consumption obtained by begging, to have poor food consumption.**** In the borderline
it was determined beforehand that all households with consumption group, the FCS came up with 27%, which
calorie consumption above 5,000 or below 500 kcal/ is a higher share of households than the 21% estimated
by the benchmark. WFP considers all households in the
* It is important to stress that the results shown are not rep- poor and borderline groups to be food insecure. Thus,
resentative of the entire country, nor is the study necessarily
reflective of average household food consumption throughout
the year. Data were only collected in 96 households, which ** For each FCS value, a running average of the surrounding
does not allow for generalization of the findings across all values for that food group and the value in question was used
regions and ethnic groups of Laos. Moreover, the data were to smooth the graph.
collected at the end of the rainy season, which is the lean *** Condensed milk, which is common in Laos, was not
season in Laos and the beginning of the rice harvest. Informa- considered in this survey. Whether it should be included
tion on consumption patterns is therefore only valid for the and weighted equally to fresh dairy products requires a more
end of the rainy season and may not reflect average annual detailed nutritional examination.
household consumption. Data for the validation study were **** Households in the food-poor CCG consume fewer than
collected between August and October 2009. 1,470 kcal/adult equivalent/day.
112 S. Baumann et al.

consumption increased from poor to acceptable calorie


Poor Borderline Acceptable
80
consumption, regardless of whether small amounts
were excluded.*** The consumption of sticky rice or
70 white rice was not included in the analysis, because
60 the frequency of consumption depends mainly on
50
preference of the different ethnic groups. When food
items were summed up as food groups, five out of seven
Percent

40 groups for which consumption was reported**** showed


30 an increasing frequency along CCGs. After exclusion
of small amounts, the frequency of consumption for all
20
reported food groups increased from poor to accept-
10 able calorie consumption.
0 In order to assess which food items or food groups
CCGs FCGs contributed to the ability of the FCS to differentiate
FIG. 2. Calorie Consumption Groups (CCGs) compared with by CCG, it was tested which food items or groups
Food Consumption Groups (FCGs) with consumption frequency increasing along CCGs
displayed a significant difference across CCGs. The
food items were maize, poultry, pork, eggs, vegetables
according to the benchmark, 42% of all households (excluding green leafy vegetables, bamboo shoots, and
were classified as food insecure, whereas the FCS only mushrooms), oil and butter, and sugar; after exclu-
ranked 28% of households as food insecure.* The share sion of small amounts, the food items were poultry,
of households considered food secure is correspond- pork, and eggs. The food groups were oil and fats,
ingly 72% according to the FCS and 58% according to and sugar; after exclusion of small amounts, the food
the calorie consumption indicator. groups were animal protein and fruit. WFP recom-
The scatter plot exploring the relationship between mends not changing the weights in order to maintain
the FCS and the calorie consumption per adult bench- comparability. However, raising the weights of the food
mark (fig. 3) showed a positive relationship between groups that were able to differentiate well between
the two indicators, with most cases lying fairly close CCGs relative to other food groups may improve the
to the line of best fit and an R-squared of 0.258, mean- predictive power of the FCS. Animal protein is already
ing that the FCS explained 25.8% of the variance of weighted very high (weight = 4), but an increase in the
the calorie consumption variable. One household weights for oil and fats, sugar (weight = 0.5), and fruit
displayed remarkably high calorie consumption, close (weight = 1), for which the weights as of now are very
to 5,000 kcal per adult equivalent/day. As it was still low, would possibly improve the ability of the FCS to
within the acceptable 500 to 5,000 kcal range, this case predict calorie consumption. Excluding small amounts
was not excluded. But the quantities consumed in that did not improve the ability of food items or groups to
household were reassessed. The high number of calo- differentiate across CCGs.
ries resulted from a very high rice consumption, which Cross-tabulating the FCGs based on the FCS with
could be reproduced through reassessment. Pearson’s CCGs shows large errors of exclusion. A total of 8
correlation coefficient (0.508) and Spearman’s nonpara- households that the FCS ranked as borderline and 11
metric correlation coefficient (0.504), both significant households considered to have acceptable food con-
at the .01 level (2-tailed), confirmed the positive rela- sumption were categorized as food poor according to
tionship between the FCS and the benchmark variable. the benchmark. There are also some errors of inclusion;
In the course of the bivariate descriptive analysis, for instance, eight households that were considered to
the food frequency of each food item, the frequency be in the borderline consumption group were ranked
of all food groups, and the FCS were examined across as acceptable by the benchmark. The only household
the three CCGs. This analysis was done for the origi- that the FCS ranked as being food poor also seems to
nal FCS and for the FCS excluding small amounts be calorie deficient, but the households considered to
(1 tablespoon/person/day or less). For more than half be borderline (8 households) and acceptable (11 house-
of all food items for which consumption was reported holds) instead of food poor were likewise severely calo-
during the reference period,** the average frequency of rie deficient (fewer than 1,470 kcal/adult equivalent/
day). Table 5 shows in more detail how only 1% of the
* Households considered as food insecure (poor and extremely calorie-deficient households were captured
borderline CCG) consume fewer than 2,100 kcal/adult
equivalent/day. *** Note that no data on small amounts were collected for
** None of the households reported consumption of goat or staples. Therefore, a comparison with and without small
sheep meat, large-bodied wildlife, or milk products during amounts is not available for staples.
the reference period. **** The only food group not reported at all was dairy.
Cross-cultural validity of food consumption indicators 113

5,000

4,000

Kilocalories per adult equivalent


3,000

2,000

1,000

0
20 30 40 50 60 70 80
FCS
FIG. 3. Calorie consumption per adult equivalent plotted against Food Con-
sumption Score (FCS)

TABLE 5. Cross tabulation of Food Consumption Groups (FCGs) against Calorie Consumption Groups
(CCGs)
CCG
Poor, Borderline, and Acceptable Poor Borderline Acceptable Total
FCG Poor ( ≤ 21) Count 1 0 0 1
% of total 1.1% 0.0% 0.0% 1.1%
Borderline ( > 21–35) Count 8 10 8 26
% of total 8.4% 10.5% 8.4% 27.4%
Acceptable ( > 35) Count 11 10 47 68
% of total 11.6% 10.5% 49.5% 71.6%
Total Count Total 20 20 55 95
% of total 21.1% 21.1% 57.9% 100.0%

Good match 61.1%


Close match 27.3%
Poor match 11.6%

by the FCS classification. This means that the FCS did were considered acceptable while they should have
not recognize 19% of households that were severely been ranked as borderline, whereas at the same time
calorie deficient and thus, as expected, underestimated other households were considered borderline instead
the food insecurity level. Furthermore, 10 out of the of acceptable.*
20 households with borderline consumption were
estimated as being food secure by the FCS. The 8% * However, because of limitations of the method in an ex-
ranked as borderline by the FCS despite acceptable treme case, if all households might be considered food secure
consumption were less alarming, but indicate a prob- by the FCS, the cross-tabulation would nonetheless result in
lem of the FCS at the threshold between borderline and 57.9% good matches and 21.1% close and poor matches each.
acceptable consumption. It did not simply rank house- Therefore, aiming at a high percentage of good matches is
not sufficient to improve the predictive power of the proxy
holds as having a better food security status than they indicator and was thus complemented by other data analysis
actually have, but it was less precise. Some households methods.
114 S. Baumann et al.

where
Effect of excluding small amounts on association of
FCS and benchmark
FCSsmall amounts = Food Consumption Score after
Even though the FCS is fairly well correlated with the exclusion of small amounts;
benchmark, the classification into FCGs differs tremen- xi = frequency of food consumption
dously from the CCGs of the benchmark. Therefore, = number of days on which each
different options to improve the predictive power of food group was consumed during
the FCS and to identify correction factors were tested. the past 7 days;
The effect of excluding small amounts from the analy- ai = weight of each food group; and
sis on the relationship of the two continuous variables si = number of days a food group was
FCS and Calorie Consumption per Adult Equivalent, consumed in small amounts only
in addition to the ability of the FCS to mimic with its during the past 7 days.
FCGs the food security level estimated by the food con-
sumption benchmark, was tested. After consultation The scatter plot in figure 4 shows a clear positive
with different nutrition experts, small amounts were relationship between the FCS and the household
only excluded for some of the food groups. For animal calorie consumption indicator after exclusion of small
protein, pulses, vegetables, fruit, and dairy products, amounts. Inserting a line of best fit shows that the
small amounts were excluded. Small amounts were R-squared at 0.285 is slightly higher than without
not excluded for sugar and oil and fats, because from exclusion of small amounts. This slightly increased
a nutritional point of view, 1 tablespoon/person/day magnitude of the relationship between proxy indicator
or less of these food groups is not considered a small and benchmark is confirmed by Pearson’s correlation
enough amount to be excluded from the analysis. Fur- coefficient at 0.534 compared with 0.508 without exclu-
thermore, small amounts were not excluded for staples, sion of small amounts and Spearman’s rho, generating a
considering that they are usually not consumed in correlation of 0.542 compared with 0.504 before.
amounts of 1 tablespoon/person/day or less. A more significant change can be found when look-
The following equation shows how the FCS, after ing at the FCGs after exclusion of small amounts from
exclusion of small amounts for animal protein, pulses, the analysis. Table 6 shows the tremendous increase in
vegetables, fruit, and dairy, was calculated: the number of households now ranked as food poor.
Almost 14% more households are considered food poor
FCSsmall amounts = astaple (xstaple) + aanimal (xanimal – sani-
than when small amounts were not excluded. A similar
mal) + apulse (xpulse – spulse) + aveg (xveg share of households (11.6%) moved from the accept-
– sveg) + afruit (xfruit – sfruit) + adairy
able group to the borderline consumption group. Thus,
(xdairy – sdairy) + asugar xsugar + aoil xoil
we understand that excluding small amounts had an

5,000

4,000
Kilocalories per adult equivalent

3,000

2,000

1,000

0
10 20 30 40 50 60 70 80

FCS — excluding small amounts


FIG. 4. Calorie consumption per adult equivalent plotted against Food Con-
sumption Score (FCS) after exclusion of small amounts
Cross-cultural validity of food consumption indicators 115

TABLE 6. Effect of excluding small amounts on predictive


Excluding small amounts, moreover, shifts the mean
power of FCS
of the FCS. Without exclusion of small amounts, the
FCGs— mean values in the poor and borderline consumption
exclud- classes are barely distinguishable. After exclusion of
Predictive ing small small amounts, however, the FCS shows mean scores
power CCGs FCGs amounts
that reflect the increasing calorie consumption from
Poor 21.1% 1.1% 14.7% poor to acceptable much better (table 7).
Borderline 21.1% 27.4% 25.3%
Acceptable 57.9% 71.6% 60.0% Changing the thresholds
CCG, Calorie Consumption Group; FCG, Food Consumption Group;
FCS, Food Consumption Score Apart from exploring the impact of excluding small
amounts, we also looked at whether changing the
impact on all consumption groups. Households from thresholds improves the predictive power of the FCS.
all FCGs, not only the very food-poor households, ate A plausible method to match the FCGs to the CCGs
foodstuffs in small amounts. is simply to calibrate the FCS against the benchmark.
Small amounts were mainly consumed of certain To do this, the cumulative percentage of households
animal proteins and vegetables. Almost 19% of house- at each possible threshold is looked at, and the thresh-
holds reported eating wild fish in small amounts old for which the share of households below matches
(mostly on 1 or 2 days, but also up to 7 days out of the the percentage predicted by the calorie consumption
last 7 days), and more than 10% of households speci- benchmark, the best is chosen. Given that excluding
fied that on some days they ate vegetables (other than small amounts seems reasonable for Laos, new pos-
green leafy vegetables and bamboo or mushrooms) in sible thresholds for the original FCS as well as for an
small amounts only. Furthermore, households reported FCS after exclusion of small amounts were examined.
eating small amounts for most other food items in the Due to the fact that excluding small amounts mark-
animal protein group and for all vegetables and fruits. edly improves the ability of the FCS to mimic the food
When comparing the prevalence of households in consumption classes estimated by the benchmark, the
each FCG after exclusion of small amounts with the magnitude of adaptation of the thresholds is much
ranking of CCGs, we see an improvement in the pre- smaller if small amounts are actually excluded.
dictive power of the FCS. Table 6 shows the changes in If no small amounts were excluded, the thresholds
the FCG classification for the original FCGs and after producing FCGs closest to the food consumption
exclusion of small amounts in comparison with the classes proposed by the benchmark would be as high
CCGs. The ranking reflects the classification according as 32 and 43 (compared with the original thresholds
to the benchmark indicator much better when small of 21 and 35). If small amounts were excluded first,
amounts are excluded. It is remarkable that the size of the thresholds resulting from calibration would be
all three FCGs comes closer to the benchmark predic- 24.5 and 36.5 for FCGs most similar to the size of the
tion after exclusion of small amounts from the analysis. benchmark classes.
When cross-tabulating the FCS after the exclusion The major advantage of the calibration is that it helps
of small amounts with the calorie consumption bench- to select thresholds that produce as close as possible
mark, no major difference from the cross-tabulation an estimate of food insecurity that matches the “true”
with the FCS in its original form can be found. The prevalence of food insecurity. The main problem of
share of good matches is slightly lower and the poor conducting a simple calibration is that we do not know
matches are consequently marginally higher; however, whether the households in each group are the same as
the difference is negligible considering the size of the those assigned to the respective CCGs. In our case,
sample and the impact of each single household, as well for example, all households ranked as food poor by
as the discussed shortcomings of the method. The dif- the benchmark could possibly be considered to have
ference derives from one fewer household in the group borderline food consumption by the FCS. Similarly,
of good matches and one more in the poor matches. all households ranked as borderline by the benchmark

TABLE 7. Effect of excluding small amounts on mean values of CCGs


CCG Significance of
group difference
Total Poor Borderline Acceptable F-statistic
FCS 44.43 37.85 38.43 49.00 10.50***
FCS—excluding small amounts 40.65 30.75 36.43 45.78 9.81***
CCG, Calorie Consumption Group; FCS, Food Consumption Score
***Significant at the 1% level.
116 S. Baumann et al.

could possibly be ranked as food poor by the FCS. Nev- a different picture. The AUC of the ROC using the
ertheless, we would end up having the same, or almost original FCS amounted to 0.76, and after exclusion of
the same, number of households in each group accord- small amounts the ROC showed an AUC of 0.78, both
ing to benchmark and proxy indicator. In this case, we being higher than that for the lower threshold (fig. 6).
could not consider the FCS to be a good or valid proxy Without exclusion of small amounts, eight thresholds
indicator. Yet, knowing the correlations and scatter were identified between 42.5 and 47.25 with balanced
plots shown before, it is unlikely that the calibration in sensitivity and specificity above 60%, and a proportion
the case of the FCS for Laos produces such poor results. of misclassified between 30% and 32%. In total, only
Nonetheless, a sensitivity–specificity analysis was also three possible thresholds (between 36.75 and 38.75)
performed in order to compare the results with those displayed a proportion of misclassified below 30%. Yet,
from the calibration. A sensitivity–specificity analysis when small amounts were excluded before conducting
systematically examines errors of exclusion and inclu- the sensitivity–specificity analysis, around half of the
sion with reference to the benchmark. values showed a proportion of misclassified close to
When looking at different possible thresholds and 30%, but none below. This reflects the earlier finding
their sensitivity, specificity, and percentage of mis- that for the surveyed sample, the prediction of the FCS
classified for the lower threshold (between poor and for the distinction between borderline and acceptable
borderline), no single threshold could be found that is less precise than that between poor and borderline.
would meet all requirements (balanced sensitivity A balanced sensitivity and specificity with values at or
and specificity, both above 60%, and a proportion of above 60% can be found for 10 thresholds within a wide
misclassified below 30%). At the thresholds 43.35 and range between 35.25 and 46.25. At these thresholds, the
43.75, sensitivity and specificity are balanced and are proportion of misclassified varies between 30% and 33%.
both above 60%, but the percentage of misclassified Considering that thresholds selected based on sen-
amounts to 38% and 39%, respectively. A proportion of sitivity–specificity analysis results will not necessar-
misclassified households below 30% could be found for ily yield realistic estimates of the prevalence of food
thresholds between 20 and 34.75, but sensitivity is only security, it was tested whether the thresholds from
between 0% and 45%. If small amounts were excluded calibration also show acceptable sensitivity, specificity,
before conducting a sensitivity–specificity analysis, six and proportion of misclassified. For the lower thresh-
thresholds could be found between 28 and 32.5 that olds suggested by calibration (24.5), sensitivity is fairly
meet the balanced sensitivity and specificity, either at low. If we shift the lower threshold to 25.5, sensitivity
or above 60%, and a proportion of misclassified of 30% increases to 55% and is thus extremely close to an
or below. Figure 5 shows the results for the ROC of the acceptable level of sensitivity. The prevalence of food-
lower threshold. poor households is shifted to 23.2% instead of 22.1%
The sensitivity–specificity analysis for the upper (threshold of 24.5). Yet, for the purpose of assessing
threshold (between borderline and acceptable) yielded food insecurity, errors of inclusion are preferable to

Original FCS Excluding small amounts


1.0 1.0

0.8 0.8
Sensitivity

Sensitivity

0.6 0.6

0.4 0.4

0.2 0.2

0 0
0 0.2 0.4 0.6 0.8 1.0 0 0.2 0.4 0.6 0.8 1.0
1 – Specificity 1 – Specificity
AUC = 0.692 AUC = 0.746

FIG. 5. Receiver Operating Curve (ROC) for lower threshold (between poor and borderline). FCS, Food
Consumption Score; AUC, Area Under the Curve
Cross-cultural validity of food consumption indicators 117

errors of exclusion. At the threshold of 25.5, sensitivity borderline) is identical to the benchmark prediction,
and specificity are not yet balanced, as expected since total accuracy is still acceptable, sensitivity and speci-
the “true” prevalence of food poor is much lower than ficity are fairly balanced and both above 60%, and the
50%, but the proportion of misclassified remains very proportion of misclassified is slightly above the accept-
low. Achieving acceptable sensitivity and specificity able 30%. The suggested thresholds improve the ability
would require a threshold above 28, which in turn of the FCS to predict the calorie consumption bench-
would produce a prevalence of food-poor households mark. Yet, it should be remembered that the sample for
of 28.4%, which is much higher than the prevalence the FCS validation study in Laos was very small and not
predicted by the benchmark. The upper threshold of representative of the entire country. In order to decide
36.5 according to the calibration against the benchmark on definitive thresholds, a replication of the present
shows acceptable sensitivity and specificity, with fairly study with a representative sample would be required.
balanced errors of inclusion and exclusion. However,
the proportion of misclassified is slightly above the 30%
considered as acceptable. Yet, for the upper threshold Conclusions and discussion
there is no single option that would display a propor-
tion of misclassified below 30%. The validation of the FCS has disclosed the following
For assessing the general food security level in Lao main results:
PDR, it will be important to decide on thresholds that The FCS in its original form does underestimate the
guarantee that the FCS matches the “true” prevalence of food insecurity level for the surveyed villages. It dis-
calorie-deficient households and that it displays a high plays a comparably strong, positive, highly significant
total accuracy in predicting the benchmark. Sensitivity correlation with the calorie consumption benchmark,
and specificity should be balanced and display accept- but the classification of households into FCGs does not
able values. The proportion of misclassified should mimic the food security status predicted by the CCGs
remain below 30%. Considering, though, that the pro- of the benchmark.
portion of misclassified equals 1 minus total accuracy, Excluding small amounts has no major impact on
accounting for a high total accuracy will automatically the relationship of the two continuous variables, Calo-
produce a low proportion of misclassified. rie Consumption per Adult Equivalent and FCS, but
For the lower threshold, 25.5 matches most of the certainly improves the ability of the FCS to mimic the
mentioned criteria, displaying a prevalence of food number of food-insecure and/or -secure households
poor similar to the benchmark, high total accuracy, predicted by the benchmark. According to the find-
unbalanced but nevertheless fairly acceptable sensitivity ings of this study, excluding small amounts for the Lao
and specificity, and a very low proportion of misclassi- context seems advisable.
fied. The most suitable upper threshold seems to be the After exclusion of small amounts, the original
one resulting from calibration. For a threshold of 36.5, thresholds produce acceptable results for the sensi-
the prevalence of food-insecure households (poor + tivity–specificity analysis, but the prevalence of food

Original FCS Excluding small amounts


1.0 1.0

0.8 0.8
Sensitivity

Sensitivity

0.6 0.6

0.4 0.4

0.2 0.2

0 0
0 0.2 0.4 0.6 0.8 1.0 0 0.2 0.4 0.6 0.8 1.0
1 – Specificity 1 – Specificity
AUC = 0.763 AUC = 0.781

FIG. 6. Receiver Operating Curve (ROC) for upper threshold (between borderline and acceptable). FCS,
Food Consumption Score; AUC, Area Under the Curve
118 S. Baumann et al.

security predicted with regard to the benchmark still the lean season, this might increase their diversity
has room for improvement. As a compromise between while calorie consumption could be reduced. How-
the thresholds suggested by calibration and the results ever, research has shown that at least dietary diversity
of sensitivity–specificity analysis, the lower thresh- indicators are able to predict calorie consumption
old of 25.5 matches most of the mentioned criteria, fairly accurately over different seasons [18]. Assess-
displaying a prevalence of food poor similar to the ing whether the relationship between the FCS and the
benchmark, high total accuracy, unbalanced but nev- calorie consumption benchmark varies over different
ertheless fairly acceptable sensitivity and specificity, seasons requires further research.
and a very low proportion of misclassified. The most The calculation of the benchmark was based on
suitable upper threshold seems to be the one resulting kilocalorie information from regional food composi-
from calibration. For a threshold of 36.5, the prevalence tion tables because no food composition table specifi-
of food-insecure households (poor + borderline) is cally for Laos is available. This might have impacted
identical to the benchmark prediction, total accuracy the precision of the benchmark. Yet, most food items
is still acceptable, sensitivity and specificity are fairly consumed on a regular basis are very similar to those
balanced, both are above 60%, and the proportion consumed in remote areas of the region.
of misclassified is slightly above the acceptable 30%. With regard to the question of cross-cultural validity,
Whether these proposed thresholds are suitable for all the FCS in its original form does not seem to be valid
parts of Laos can only be assessed through a nationally in Lao PDR without adaptation. The fact that excluding
representative sample. small amounts from the analysis improves the ability
For a final decision on whether small amounts of the FCS to predict the calorie consumption bench-
should be excluded from the analysis and which thresh- mark supports the assumption that due to differing
olds to choose for the FCS in Laos, the findings should consumption patterns (higher diversity in Laos but in
be reassessed with a nationwide representative sample. small amounts only), the FCS might overestimate the
In the present study, the FCS has not been validated food security level in Laos. Excluding small amounts
against diet quality due to the fact that individual die- for regions in which people are frequently consuming
tary assessments were beyond the scope of the research. wild foods and consumption is typically based on one
It should be considered that excluding small amounts staple product supplemented by various condiments in
might impact the evaluation of diet quality, despite the very small amounts, might constitute a suitable correc-
low contribution of small amounts to overall energy tion factor for the FCS.
intake. Apart from validating the FCS against diet After exclusion of small amounts in the Lao context,
quantity with a countrywide representative sample, a the thresholds proposed by WFP were very close to the
validation against a benchmark of diet quality is highly thresholds resulting from calibration and sensitivity–
recommended. A full measure of dietary quality should specificity analysis. A harmonized shift of FCS thresh-
go beyond measures of energy consumption and diver- olds based on dietary patterns (regular consumption
sity of foods consumed to a fuller understanding of of fat and sugar) is being tested in various countries.
amounts, content and nutrient density. Such metrics Nevertheless, the question remains whether universal
have still to be developed for field settings. thresholds can be applied to indicators based on dietary
Moreover, various studies have pointed out that wild diversity with or without locally determined correc-
foods, which are consumed frequently in Lao PDR, tion factors such as the exclusion of small amounts.
are considered inferior foods and their consumption This question might be better understood if excluding
increases as preferences are curtailed [17]. If this is the small amounts as a correction factor for the FCS was
case, an increased diversity of food from the consump- tested in different countries and the thresholds were
tion of wild foods cannot be considered as improved subsequently calibrated against a benchmark. Compar-
food security. In a countrywide representative sample, ing the results of these studies would then allow for a
it would be worthwhile to assess whether excluding more general statement on the suitability of excluding
small amounts may reduce the effect of ranking a small amounts as a correction factor and the possibility
household as more food insecure due to greater dietary of applying universal thresholds when small amounts
diversity resulting from increased consumption of wild are excluded.
foods. It is possible that households increase the con-
sumption of wild foods if other acquirable food items
are insufficient, but due to the limited availability, the Acknowledgements
consumption might not exceed what are considered as
small amounts in this study. The authors acknowledge the support of the WFP
Further shortcomings of the study are seasonal- Laos team and the WFP Regional Bureau for Asia; in
ity issues that might have impacted the association particular, the valuable inputs from Laura de Franchis,
between the proxy indicator and the benchmark. If, Ruangdech Poungprom, Michael Sheinkman, and Rita
for example, people consume more wild foods during Bhatia.
Cross-cultural validity of food consumption indicators 119

References
1. Frongillo EA. Commentary: assessing food insecurity in Health, 2007.
Trinidad and Tobago. Int J Epidemiol 2003;32:516–7. 10. World Food Programme. Emergency food security
2. Deitchler M, Ballard T, Swindale A, Coates J. Validation handbook, 1st ed. Rome: WFP, 2005.
of a measure of household hunger for cross-cultural use. 11. World Food Programme. Lao PDR: Comprehensive food
Washington, DC: Food and Nutrition Technical Assis- security and vulnerability analysis. Rome: WFP, 2007.
tance II Project (FANTA-2)/Academy for Educational 12. World Bank Lao PDR. Rural and agriculture sector issues
Development, 2010. paper. No. 37566. Washington, DC: World Bank, 2006.
3. Coates J. Experience and expression of food insecurity 13. National Statistics Centre, Lao PDR. Summary report:
across cultures: practical implications for valid measure- food insecurity assessment based on food consumption
ment. Washington, DC: Food and Nutrition Technical statistics derived from the 2002/03 Lao PDR Expendi-
Assistance Project/Academy for Educational Develop- ture and Consumption Survey. Vientiane: National
ment, 2004. Statistics Centre, 2007.
4. World Food Programme. Technical Guidance Sheet – 14. Frongillo EA. Validation of measures of food insecurity
Food consumption analysis: calculation and use of the and hunger. J Nutr 1999;129:506–9.
food consumption score in food security analysis. Rome: 15. IRIS Center. Developing poverty assessment tools project:
WFP, 2008. note on assessment and improvement of tool accuracy.
5. World Food Programme. Emergency food security Morrill Hall, Md, USA: IRIS Center, University of Mary-
handbook, 2nd ed. Rome: WFP, 2009. land, 2005.
6. Wiesmann D, Bassett L, Benson T, Hoddinott J. Valida- 16. Arimond M, Elin Torheim L, Wiesmann D, Joseph M,
tion of the World Food Programme’s Food Consump- Carriquiry A. Dietary diversity as a measure of women’s
tion Score and alternative indicators of household food diet quality in resource-poor areas: results from rural
security. Washington, DC: International Food Policy Bangladesh site. Washington, DC: Food and Nutrition
Research Institute, 2009. Technical Assistance Project/Academy for Educational
7. Pustawien P, Burlingame B, Raroengwichit M, Sungpuag Development, 2008.
P. ASEAN food composition tables. Bangkok: Institute 17. Coates J, Rogers BL, Webb P, Maxwell D, Houser R,
of Nutrition, Mahidol University, 2000. McDonald C. Diet diversity study. Rome: World Food
8. Leung WTW, Butrum RR, Chang FH, Rao MN, Polacchi Programme, 2007.
W. Food composition table for use in East Asia. Rome/ 18. Hoddinott J, Yohannes Y. Dietary diversity as a house-
Washington, DC: Food and Agriculture Organization/US hold food security indicator. Washington, DC: Food
Department of Health, Education, and Welfare, 1972. and Nutrition Technical Assistance Project/Academy
9. Ministry of Health, Lao PDR. Lao food book for dietary for Educational Development, 2002.
assessment. Vientiane: National Institute of Public

You might also like