European Journal of Clinical Nutrition (1998) 52, 136±144
ß 1998 Stockton Press. All rights reserved 0954±3007/98 $12.00
Body mass index as indicator of standard of living
in developing countries
M NubeÂ1 , WK Asenso-Okyere2 and GJM van den Boom1
1
2
Centre for World Food Studies (SOW-VU), De Boelelaan 1105, Vrije Universiteit, 1081 HV Amsterdam, The Netherlands; and
Institute of Statistical, Social and Economic Research (ISSER), University of Ghana, Legon, Ghana
Objective: To assess the suitability of the body mass index (BMI) as an indicator of standard of living in
developing countries.
Design, setting and subjects: The analysis is based on data collected in the ®rst two rounds of the Ghana Living
Standards Survey, held in 1987=88 (GLSS-I) and 1988=89 (GLSS-II). The dataset provides information on a
wide range of socio-economic variables, at the individual, the household and the community level, including the
height and weight data of approximately 9000 adults in the 20±65 y age bracket.
Method: Bivariate analysis was used to assess, at the individual level and at the level of population groups, the
relationships between adult Body Mass Index and selected household characteristics such as income and
expenditure, years of schooling of head of household, access to services, quality of housing, and nutritional status
of children. Through multiple regression, indicative estimates have been derived of the effects of these variables
on adult BMI. For comparison, the same relationships were investigated for weight and height.
Results and conclusion: At the individual level, BMI shows a signi®cantly positive relation with the various
socio-economic indicators of living standard, though the correlation coef®cients indicate a poor ®t. However, at
the level of population groups, the relationship between BMI and other characteristics of socio-economic
development is strong, with a correlation coef®cient of 0.86 between mean BMI and mean per capita
expenditures of 12 population groups in Ghana, presumed to be at different levels of standard of living. The
relationships between weight and the various socioeconomic characteristics were comparable to those for BMI,
while height was poorly correlated with the selected household variables. Results suggest that in low-income
countries, information on adult BMI (mean and distribution) can be used for assessing differences in standards of
living between population groups or for monitoring changes over time.
Sponsorship: The project is co-®nanced by the Netherlands' Ministry of Development Cooperation, within the
context of the ReÂseau SADAOC-Programme (Food Security and Sustainable Development in West-Africa).
Descriptors: body mass index (BMI); standards of living; developing countries
Introduction
Various measures can be used for assessing the standard of
living of a population or population group. Commonly used
indicators are incomes or expenditures as derived from
national level aggregate GDP data or from household
budget surveys. Distributional information can be
expressed as, for example, the Gini-coef®cient, which
measures the extent to which the distribution of income
among households deviates from a perfectly equal distribution (World Bank, 1997), or as the percentage of people
whose level of income or expenditure is below a certain
poverty line (Blackwood & Lynch, 1994; Ravallion,
1996a).
While there is no doubt that such income variables are a
®rst and important indicator of overall development, they
also have their limitations (House, 1991; Ravallion, 1996b).
Comparing levels of incomes or expenditures between
countries is not unambiguous and may need complicated
adjustments for differences in purchasing power. Comparing incomes or expenditures of different population groups
within a country has its pitfalls when for example households in one group are largely subsistence oriented while
Correspondence: Dr M NubeÂ.
Received 14 March 1997; revised 29 September 1997; accepted 5 October
1997
households in the other group are market oriented. Similarly, when assessing trends or variability over time,
problems of ®nding a common denominator are at play.
Therefore, over past years various attempts have been made
to include variables other than incomes or expenditures,
when assessing the level of well-being of a population or
population group (Dasgupta & Weale, 1992). Among these
are characteristics such as literacy rates and information on
health related variables such as life expectancy and access
to drinking water. The Human Development Report
(UNDP, 1990), published yearly since 1990, in which a
country-level human development index (HDI) is de®ned
on the basis of life expectancy, literacy rate and income, is
an example of such an approach. Nutrition is also increasingly used as an indicator of standards of living. Information may be presented on per capita kcal availability, or on
average per capita kcal consumption and its distribution.
However, in developing countries, data on dietary energy
availability may be subject to wide margins of error, and
their use requires great caution. Food energy consumption
data derived from household surveys have the potential to
provide more reliable information, although here also the
quality of data varies, depending on the procedures used in
collecting and processing data (FAO, 1983; Bouis, 1994).
With respect to nutrition, information may also be available
on the prevalence of malnutrition in children on the basis of
anthropometry. While such information is relevant, it has
BMI indicator of standard of living in developing countries
M Nube et al
the limitation that it relates to a speci®c age-group, generally the under®ves.
In recent years, in developing countries information has
been increasingly collected on the adult body mass index
(BMI), de®ned as a person's weight in kilograms divided
by the square height in meters. In the ®rst place, BMI
provides direct nutritional information. The generally
accepted normal range for adult BMI is from 18.5±25
(WHO, 1995), and for af¯uent countries the mean adult
BMI is generally in the range 23±27 (Rotimi et al, 1995;
Bailey & Ferro-Luzzi, 1995; Simmons et al, 1996). In
developing countries, the mean BMI of adults is generally
lower than in high-income countries, and a considerable
proportion of the population may have a BMI below the
normal range (FAO, 1994; Alemu & Lindtjorn, 1995). Low
levels of BMI have been shown to be associated with
impaired physiological functions, such as poor pregnancy
outcome (FAO, 1994) and reduced work capacity (Deolalikar,1988; Haddad & Bouis,1991; Durnin,1994; van den
Boom et al, 1996), while susceptibility to illness may be
increased (Garcia & Kennedy,1994). At the other end of the
scale, overweight (BMI between 25 and 30) and obesity
(BMI > 30) are known to be associated with an increased
risk of chronic disorders such as diabetes mellitus, hypertension, coronary heart disease, and certain types of cancer
(WHO,1990,1995). In af¯uent countries, and increasingly
in high income strata in developing countries, high prevalence rates of overweight and obesity are a major public
health problem (INCLEN,1994).
In addition to the direct nutritional interpretation of
BMI-data, in a number of studies in developing countries
a positive correlation has been reported between adult BMI
and common indicators of development such as incomes or
expenditures (Alderman & Garcia, 1993; Ghana Statistical
Service, 1994; Sichieri et al, 1994; Kennedy & Garcia,
1994; INCLEN, 1994; Cornu et al, 1995). Such results
suggest that the mean BMI, possibly in combination with
distributional information, can be used as a general indicator of development. However, it is important to note that
in af¯uent societies a positive correlation between BMI and
incomes or expenditures, or between BMI and other correlates of development is generally absent or the relationship
is even inverted (Dryson et al, 1992; Croft et al, 1992;
Burke et al, 1992; Paxton et al, 1994; Seidell, 1995).
Therefore, in low-income countries BMI could correlate
well with other indicators of development, but such a
correlation can be expected to disappear with rising
incomes and standards of living.
In the present paper we report on the relationships
between BMI in adults and various common correlates of
development in Ghana, on the basis of two large and
nationally representative samples of households, the 1988
and 1989 Ghana Living Standards Measurement Survey.
The study aims, in particular, to compare BMI with
commonly used indicators of standards of living, to identify
advantages and limitations with regards to BMI as indicator
of standards of living in developing countries, and to
provide guidelines for its use for such a purpose. For
comparison, also the two components from which BMI is
derived, weight and height, are included in the analysis.
Data and methodology
The main data source for the present study consists of the
®rst two rounds of the Ghana Living Standards Survey
(GLSS). The ®rst round (GLSS-I) was held from Septem-
ber 1987±1988, the second round (GLSS-II) from October
1988±August 1989. The surveys were conducted by the
Statistical Service of Ghana with assistance from the World
Bank. In both rounds the surveys cover over 3000 households, proportionally distributed over Ghana on the basis of
administrative and ecological zone, and on the basis of
rural=urban location. A panel of approximately 1200
households were surveyed in both rounds. Information
was collected on a wide range of socio-economic variables,
at the individual, the household and the community level.
Presented results are based on a combined sample of the
two datasets. Relationships are reported between individual
BMI, weight and height, and various household level
characteristics, presumed to re¯ect different aspects of
standards of living. While in total for 9215 adults (age
20±65 y) weight and height data were available, for the
main part of the study a smaller sample was constructed.
First, individuals from households with total expenditures
higher than 1.000.000 cedis per year, pregnant and lactating
women, and individuals with a BMI < 10 or a BMI > 40
were excluded from the sample. Then, from all households
that included at least one male and one female adult, the
youngest male and the youngest female adult were selected.
Thus, adults from single-adult households were excluded,
while also elder adults were excluded that belonged to
households from which already the youngest male and
female adult had been selected. This procedure resulted
in a subsample that consisted of 4228 adults (2114 males
and 2114 females) from 2114 households. For approximately 75% of the households the two adults selected from
each household were the head of the household and his
wife. The above procedure was followed in order to reduce
possible biases in the analysis resulting from uneven
representation of males and females and also in order to
achieve that all households were represented by the same
number of people. The exclusion of adults from households
with total expenditures above 1.000.000 cedis per year and
adults with a BMI below 10 or above 40 was done in order
to prevent large effects of outlyers on group means, in
particular when the size of population groups would be
rather small. Pregnant and lactating women were excluded
as their BMI may well be affected by their physiological
condition.
The BMI classi®cations used are those proposed by
FAO (FAO, 1994) and WHO (WHO, 1995). Mean household incomes and expenditures were calculated as reported
elsewhere (Seini et al, 1997). Incomes and expenditures
were de¯ated with a monthly consumer price index, and
refer to September 1987 prices (1 USD ˆ 170 cedis, September 1987). Household total expenditures were estimated
from the summation of cash expenditures and the imputed
value of consumed home produce.
Other variables considered in relation with BMI (and in
relation with weight and height), are the number of years of
schooling of the heads of households (fully completed
school years); the gender of heads of households; whether
households have access to electricity or treated water; the
quality of dwellings, expressed in a housing score on the
basis of the main building material of the house; the
number of acres of land a household is cultivating; a
seasonal dummy for the lean season (March±June); a
geographical dummy for the urban area; and for individuals
the age in years and gender (value 0 for male and 1 for
female). Further, for children under ®ve years of age,
height-for-age z-scores (number of standard deviations
below median of reference) were calculated, and for each
137
BMI indicator of standard of living in developing countries
M Nube et al
138
Table 1 Household socioeconomic variables
Variable
Units or scores
a
Per capita total expenditures
Per capita cash expenditures
Per capita value consumed home produce
Per capita total income
School years head of household
Gender head of household
Electricity
Water
Housing
Lean season
Urban
Height-for-age z-score of under®ves
Acres farmed
1.000s of cedis per year
1.000s of cedis per year
1.000s of cedis per year
1.000s of cedis per year
years
0 ˆ male
1 ˆ female
0 ˆ no electricity
1 ˆ electricity
0 ˆ no treated water (well, surface or rain water)
1 ˆ treated water (piped water, water from vendors)
1 ˆ mud, mudbrick or bamboo
2 ˆ galvanized iron or wooden plank
3 ˆ stone, bricks or cement
0 ˆ household not interviewed during lean season
1 ˆ household interviewed during lean season
0 ˆ rural
1 ˆ urban
Number of standard deviations below median of reference
Log (number of acres farmed ‡ 1)
a
September 1987, 1 USD ˆ 170 cedis.
household the mean height-for-age z-score of under®ves
was calculated. Correlation-coef®cients and associated Pvalues are given for all studied relationships. The household variables considered are listed in Table 1.
The sample was divided into socio-economic groups
on the basis of four criteria to make comparisons
between the different segments of the population. Firstly,
on the basis of geographical location, households were
classi®ed as belonging to the coastal, the forest, or the
savannah agro-ecological zone. In the second place, households living in communities with less than 5000 inhabitants
(on the basis of the 1984 population census) were classi®ed
as rural, while households living in communities with over
5000 inhabitants were classi®ed as urban. Finally, rural
households were divided, on the basis of their main
economic activity, into farming and non-farming households, while urban households were divided, on the basis of
record of training and experience of the head of the
household, into skilled and unskilled households. These
classi®cations resulted in a total of 12 different socioeconomic groups (abbreviated as rursavfarm, rursavnonf,
rurforfarm, rurfornonf, rurcoafarm, rurcoanonf, urbsavunsk, urbsavskil, urbforunsk, urbforskil, urbcoaunsk and
urbcoaskil). At the level of these 12 socio-economic
groups, mean values were calculated for BMI, weight,
height and the various household characteristics, and correlation coef®cients between group means for BMI, weight
and height, and the other variables are presented. The
correlation coef®cients were obtained from weighted least
square estimation, with weights equal to the reciprocal of
the group mean variances of respectively BMI, weight and
height. Finally, an analysis was made, through multiple
regression, of the effects of the various socioeconomic
characteristics on BMI, weight and height. The
ANTHRO-software (Centre for Diseases Control, Division
of Nutrition, Atlanta, Nutrition Unit, and WHO, Geneva)
was used for calculating height-for-age z-scores; the software is based on the NCHS (National Center for Health
Statistics) growth reference (Dibley et al, 1987). SASsoftware was used for all other data processing and for
correlation and regression analysis.
Results and discussion
BMI of adults in Ghana
In Table 2 data are presented on means and distribution of
adult BMI, separately for men and women, for the rural and
urban population, and for the whole sample of 9215 adults.
Mean BMIs are higher for female than for male adults, and
higher for adults living in urban areas in comparison with
those living in rural areas. Overweight and obesity
(BMI > 25) is mainly prevalent in women living in urban
areas, while the prevalence of chronic energy de®ciency
(BMI < 18.5) is much higher in the rural areas. The prevalence of a BMI lower than 18.5 is rather similar in men
(16.2%) and women (16.6%) despite the considerably
higher mean BMI of women. The results are in line with
previously presented data on only GLSS-I (Alderman,
1990). It may be noted that the prevalence rate of 16.5%
of individuals with a BMI < 18.5 in combination with a
mean BMI of 21.5 is in con¯ict with a suggestion made by
a WHO Study Group (WHO, 1990) that for a population
group with a mean BMI in the 20±22 range, rates of chronic
energy de®ciency would be very low.
Table 2 BMI of adults (age 20±65 y) in Ghana, for men and women, for
rural and urban population, and for whole sample.
Nb
mean BMI
BMI-distribution (%)c
<16
16±16.9
17±18.4
18.5±24.9
25±29.9
30±39.9
40
a
Males
Females
Rural
Urban
All a
4961
20.8
4253
22.1
5788
20.7
3427
22.5
9215
21.5
1.2
2.7
12.3
78.5
4.7
0.5
0.1
1.6
3.2
11.8
65.3
12.0
5.7
0.4
1.7
3.4
14.0
74.7
4.8
1.3
0.1
0.8
2.0
8.7
68.6
13.6
5.6
0.6
1.4
3.0
12.1
71.2
8.4
3.1
0.3
Mean and distribution adjusted to a male±female ratio of one (average of
columns 1 and 2);
b
Size of (sub) sample:
c
BMI-ranges ``<16'', ``16±16.9'' and ``17±18.5'' also referred to as Grade
III, II and I Chronic Energy De®ciency (Ferro-Luzzi et al, 1992); BMI 25±
30: overweight, BMI > 30: obesity (WHO, 1995, p 452).
Source: GLSS-I and GLSS-II.
BMI indicator of standard of living in developing countries
M Nube et al
Correlation between adult BMI and household
income=expenditure variables
In Table 3 data are presented on the correlation between
adult BMI and four household per capita (annual) income=expenditure variables, viz household per capita total
expenditures, household per capita cash expenditures,
imputed value of households' per capita consumed home
produce, and household per capita total income.
The data show for per capita total and per capita cash
expenditures highly signi®cant positive relationships
(P < 0.0001) with adult BMI. It should be noted that cash
expenditures are stronger correlated with adult BMI than
total expenditures. An explanation for these ®ndings might
be that in the GLSS-dataset cash expenditures data are
more accurate and a better proxy for households' wellbeing than total expenditures. Total expenditures include
the imputed value of consumed home produce, which is
dif®cult to estimate through questionnaire methods. Results
further indicate that the per capita value of consumed home
produce is negatively correlated with adult BMI. This may
be explained by the fact that low-income households are
more likely to derive part of their food supply from home
food production than high-income households.
The income variable is weaker correlated with BMI than
the two expenditure variables. Redistribution of income
between low and high income households may be partially
responsible for these results (Minhas, 1991; van den Boom
et al, 1996). In addition, it is a common experience that
income data derived from household surveys are less
reliable than expenditure data. In both GLSS-rounds,
household incomes are, on average, approximately 70%
of expenditures, probably partially as a result of serious
underreporting of incomes (see also Table 6 for sample
means of per capita incomes and expenditures). Table 2
further includes the correlation coef®cients of weight and
height with the same four income=expenditure variables.
While the correlation coef®cients for weight are practically
similar to those for BMI, the correlation coef®cients for
height are much lower, though still signi®cant (except for
the correlation coef®cient between height and the per capita
value of consumed home produce, P ˆ 0.0530).
Comparing adult BMI with household per capita total
expenditures and with household cash expenditures as
indicator of household well-being
For adult BMI to be a useful indicator of standards of
living, it should not only be signi®cantly and positively
correlated with common household economic variables
(income, expenditures), but also with other aspects of
household well-being. Therefore, from available information in GLSS ®ve other household characteristics have been
selected which are presumed to re¯ect various aspects of
households' standards of living. The ®ve household characteristics are the number of years of schooling of the head
of the household, whether the household has access to
electricity, whether the household has access to treated
water, the quality of the household's dwelling on the basis
of the type of main building material of the house, and the
nutritional status of children (under®ves) living in the
household, expressed by the z-scores for height-for-age.
Table 4 provides the correlation coef®cients between adult
BMI and these ®ve household characteristics, and also, for
comparison, between two selected income=expenditure
variables and the ®ve household characteristics. The two
income=expenditure variables are selected from the four
variables in Table 2. Per capita total expenditures is
included in Table 3 as it is the most common indicator of
a household's economic status, per capita cash expenditures
as it gave the highest correlation-coef®cients with adult
BMI in the previous section.
Table 4 shows that per capita cash expenditures gives
the highest correlation coef®cients with four out of the ®ve
household characteristics. Nevertheless, within the context
of the present study, it is important to note that adult BMI is
positively and signi®cantly correlated with all ®ve selected
household characteristics. It is further noteworthy that the
correlation coef®cients between BMI and the ®ve selected
household characteristics are practically the same as the
corresponding correlation coef®cients between per capita
total expenditures and the ®ve household characteristics.
Finally, Table 4 shows that also with respect to these
household variables, there is very little difference between
the correlation coef®cients of BMI and weight with the
selected household characteristics, while the correlation
coef®cients of height with the same variables are again
much lower.
Mean BMI and means of other correlates of standards of
living at the level of population groups
In the previous sections relationships between BMI and
various household characteristics have been analyzed with
individual and household level data. On the basis of the
reported positive correlations between BMI and the various
household characteristics (Tables 3 and 4), it can be
expected that at the level of population groups differences
in standards of living will also be re¯ected in differences in
mean BMI between population groups.
Therefore, the sample has been divided into different
socio-economic groups, on the basis of agroecological and
urban=rural location of the household. Rural households
were divided into farming and nonfarming households, and
urban households into skilled and unskilled households (see
Table 3 Correlation between BMI, weight, and height, and annual income=expenditure variables
N ˆ 4228a
BMI
Weight
Height
a
Per capita
total expenditures
0.17b
0.0001c
0.18
0.0001
0.05
0.0001
Size of (sub)sample.
Correlation coef®cient.
c
P-value.
Source: GLSS-I and GLSS-II.
b
Per capita
cash expenditures
0.23
0.0001
0.23
0.0001
0.07
0.0001
Per capita value of consumed
home produce
70.13
0.0001
70.13
0.0001
70.03
0.0530
Per capita
total income
0.10
0.0001
0.11
0.0001
0.03
0.0281
139
BMI indicator of standard of living in developing countries
M Nube et al
140
Table 4 Adult BMI, weight, height, per capita total expenditures and per capita cash expenditures correlated with other household
characteristics
Na
Schoolyears
head of household
Electricityd
Treated watere
Building
material of housef
Ng
Height-for-age
z-score under®vesh
BMI
Weight
Height
4228
0.22b
0.0001c
0.25
0.0001
0.27
0.0001
0.27
0.0001
2361
0.14
0.0001
4228
0.22
0.0001
0.26
0.0001
0.25
0.0001
0.27
0.0001
2361
0.18
0.0001
4228
0.05
0.0001
0.06
0.001
0.07
0.0001
0.08
0.0001
2361
0.11
0.0001
Per capita total
expenditures
Per capita cash
expenditures
4228
0.25
0.0001
0.27
0.0001
0.27
0.0001
0.24
0.0001
2361
0.10
0.0001
4228
0.34
0.0001
0.40
0.0001
0.40
0.0001
0.37
0.0001
2361
0.13
0.0001
a
Size of (sub)sample.
Correlation coef®cient.
c
P-value;
d
Electricity dummy, 0 ˆ no electricity, 1 ˆ electricity.
e
Water dummy, 0 ˆ no treated water, 1 ˆ treated water.
f
Building material of dwelling, quality score from 1±3.
g
Smaller sample sizes for correlation coef®cients with height-for-age z-scores as not all households have children in the 0±5 y age
group.
h Height-for-age z-score de®ned as number of standard deviations below mean of reference population.
Source: GLSS-I and II.
b
Data and methodology). This resulted in twelve socioeconomic groups, expected to have different levels of
standards of living.
In Table 5 results are presented for the means of BMI for
the twelve socio-economic groups. The table reveals considerable differences in mean BMI between the various
socio-economic groups, with the lowest value (20.0) for
farm households in the savannah region in northern Ghana
and the highest for skilled urban households in the coastal
region of southern Ghana (23.2). The difference in mean
BMI of 3.2 points between these two socio-economic
groups corresponds to a difference in mean body weight
of almost 10 kg. Table 5 further includes the means of
various household characteristics for the twelve socio-
Table 5 Means for adult BMI and other household characteristics, by socio-economic group (socio-economic groups ranked by increasing mean BMI)
Groups means by socioeconomic group
Socio-economic
group
c
Rursavfarm
Rurforfarm
Rurfornonf
Rurcoafarm
Urbsavunsk
Rursavnonf
Rurcoanonf
Urbforunsk
Urbsavskil
Urbcoaunsk
Urbforskil
Urbcoaskil
N
a
772
878
276
420
142
124
190
352
52
558
90
374
mean
BMI b
20.0
20.7
21.1
21.2
21.4
21.5
21.8
22.0
22.3
22.5
23.1
23.2
mean mean
weight height
53.5
54.3
55.8
56.3
57.8
58.9
58.2
58.6
59.0
60.8
62.0
63.0
163.2
162.1
162.6
162.7
164.2
165.1
163.0
163.4
162.8
164.5
163.9
164.9
Corr-coeff. between group mean of
column variable and group mean for BMId
Corr-coeff. between group mean of
column variable and group mean for weight
Corr-coeff. between group mean of
column variable and group mean for height
a
mean per
capita total
expenditures
mean per
capita cash
expenditures
45.3
54.2
56.0
55.4
48.7
52.6
62.5
61.1
68.5
78.7
78.5
98.2
20.7
35.7
43.1
39.3
34.2
36.3
53.1
52.2
61.9
74.0
69.5
94.8
0.86
0.0001
0.84
0.0001
0.49
0.0115
0.94
0.0001
0.90
0.0001
0.48
0.0119
mean school
years head
of household
1.6
4.9
9.0
4.7
3.6
6.4
7.2
5.3
11.5
6.9
11.8
12.7
0.71
0.0006
0.59
0.0035
0.22
0.1214
mean
electricity
< 0.01
0.06
0.25
0.03
0.27
0.06
0.29
0.51
0.54
0.78
0.71
0.88
0.85
0.0001
0.89
0.0001
0.59
0.0034
mean building mean height formaterial of
age z-score
dwelling
under®ves
1.1
1.2
1.6
1.5
1.4
1.3
1.9
1.7
2.2
2.6
2.4
2.8
0.88
0.0001
0.92
0.0001
0.58
0.0042
71.46
71.65
71.15
71.17
71.69
71.08
70.89
71.19
70.26
70.79
71.15
70.71
0.59
0.0034
0.68
0.0010
0.50
0.0100
Size of (sub)sample;
Units: BMI in kg=m2; weight in kg; height in cm; per capita total expenditures and per capita cash expenditures in `000's cedis per year, schoolyears head
of household in years; electricity dummy, 0 ˆ no electricity, 1 ˆ electricity; building material of dwelling, score from 1±3; height-for-age z-score in
standard deviations below mean of reference population.
c
Socioeconomic groups de®ned as rural savannah farm households (rursavfarm), rural forest farm households (rurforfarm), rural coastal farm households
(rurcoafarm), rural savannah non-farm households (rursavnonf), rural forest non-farm households (rurfornonf), rural coastal non-farm households
(rurcoanonf), urban savannah unskilled households (urbsavunsk), urban forest unskilled households (urbforunsk), urban coastal unskilled households
(urbcoaunsk), urban savannah skilled households (urbsavskil), urban forest skilled households (urbforskil), and urban coastal skilled households
(urbcoaskil).
d
Correlation coef®cients obtained from weighted least squares estimation, with weights equal to the reciprocal of the group mean variances of BMI, weight
and height of each socioeconomic group.
Source: GLSS-I and II.
b
BMI indicator of standard of living in developing countries
M Nube et al
groups. For example, for the whole sample the mean BMIs
of the agegroups 31±35 y (mean age 33.1 y) and 36±40 y
(mean age 38.3 y), were respectively 22.1 and 22.0. With
respect to the male to female ratio, all groups have a male
to female ratio of one as a result of the procedure followed
in the sample construction (see Data and methodology).
It may ®nally be noted that prevalence rates of chronic
energy de®ciency (BMI < 18.5) and overweight and obesity (BMI > 25) differ widely between the various socioeconomic groups. For the socioeconomic group with the
lowest mean BMI (farmers in the rural savannah region,
mean BMI 20.0), the prevalence rates of chronic energy
de®ciency and overweight plus obesity were respectively
25% and 9%, while these rates were respectively 3% and
26% for the socioeconomic group with the highest mean
BMI (skilled people in the coastal urban area, mean BMI
23.2). These ®gures clearly reveal that a `price' is to be
paid for increasing standards of living.
Figure 1 Relationship between group mean BMI (kg=m2) and group
mean per capita cash expenditures (`000' cedis per year) for 12 different
socioeconomic groups (data from Table 5). Source: GLSS-II and GLSS-II.
economic groups. Correlation coef®cients (obtained from
weighted least square estimation, see Data and methodology) between the means of BMI and the means of these
household characteristics are high and vary between 0.71
and 0.94, except for the under®ves' height-for-age z-score
which is rather weakly correlated with adult BMI (correlation-coef®cient 0.59). Figure 1 gives a graphical representation of the relationship between the mean values in the
twelve subgroups for BMI and household per capita cash
expenditures. It might be interesting to mention that the
slope of the relationship between mean BMI and mean per
capita cash expenditure is estimated to be 0.047 (intercept
19,14). This implies that an increase in group mean per
capita cash expenditures from, for example, 40.000 to
60.000 cedis per year would be associated with an increase
in group mean BMI from 21.0±21.9. Similarly, it could be
calculated that an increase in group mean per capita total
expenditures from, for example, 50.000 to 75.000 cedis per
year would be associated with an increase in group mean
BMI from 20.6±22.4 (slope of relationship between group
mean BMI and group mean per capita total expenditure
0.067, intercept 17.23).
Again, for comparison, Table 5 gives also the correlation coef®cients between mean the group means of weight
and height and the selected household characteristics. As in
the previous tables, BMI and weight behave rather similar
with respect to their correlation coef®cients with the
selected household variables, while for height the correlation coef®cients are much lower.
The various socioeconomic groups differ in their characteristics as a result of the criteria used for the classi®cation of households. However, the 12 socioeconomic groups
might also differ in other respects, which have the potential
to cause a bias in the results, and some checks on possible
differences have been done. For example, the population
groups differ in mean age ranging from a mean of 33.2 y of
adults in urban coastal skilled households to a mean of
37.9 y of adults in rural savannah farm households. However, the age differences between groups are likely to have
only a minor confounding effect on the mean BMIs of the
Multiple regression analysis
The main objective of the present study is to contribute to
the evaluation of BMI as an indicator of household wellbeing. Therefore, in the previous sections the emphasis has
been on the correlation coef®cients between BMI and other
indicators of standards of living. Yet, the various variables
studied are also, at least partially, causally related to BMI.
The GLSS database does not provide direct individual
information on the physiological determinants of adult
BMI, which are food energy intake, energy expenditure,
and individual metabolic characteristics. However, the
information available on various other variables may
serve as proxy's for these physiological determinants of
BMI. Through multiple regression, some indicative estimates have been derived of their effects on adult BMI.
Three types of variables are included in the regression
analysis: individual level, household level, and community
level characteristics. It is important to note that BMI is not
only determined by the balance between food energy intake
and energy expenditure, but that BMI also affects these
factors. For example, BMI may have a positive effect on
capacity for physical work, which on its turn may result in a
higher level of food consumption (Strauss, 1993; van den
Boom et al, 1996). Yet, in the present study we do not
consider such reverse causality, nor do we deviate otherwise from the common assumption that disturbances of the
regression equation have common variance and are independent among individuals. Relaxation of such assumptions would require a much more elaborate analysis and go
beyond the scope of the paper (see Thomas et al, 1996, for
an application with disturbances that have cluster-dependent variances). As the results of the multiple regression
analysis are conditional on assumptions that are admittedly
restrictive, some prudence should be give to their interpretation.
Results of the multiple regression analysis are given in
Table 6. As in the previous sections, for comparison results
are also given for weight and height. With respect to the
individual characteristics, all three anthropometric indicators are as expected signi®cantly related with gender. Also
the age relationships are in line with the generally observed
life cycle pattern of increasing BMI and weight over the
age period of approximately 20±50 y, and decreasing BMI
beyond that age (Dhurandhar & Kulkarni, 1992; Rotimi et
al, 1995; WHO, 1995).
With respect to the household level characteristics, both
per capita cash expenditures and per capita total income
141
BMI indicator of standard of living in developing countries
M Nube et al
142
Table 6 Multiple regression of BMI, weight, and height, on individual, household and community level characteristics
N ˆ 4228
Intercept
Individual characteristics
Gender
Age (y)
Age (y2)
Household characteristics
Per cap. cash expenditures
Per cap. value home produce
Per cap. total income
Household size
Electricity, dummy
Water, dummy
Building material dwelling
Schoolyears household head
Log(no. of acres farmed ‡ 1)
Community characteristics
Lean season
Urban dummy
r-square
BMI a
parameter estimate
P-value
Weight
parameter estimate
P-value
Height
parameter estimate
P-value
Sample
mean
15.59
0.0001
43.91
0.0001
166.17
0.0001
Ð
0.979
0.185
70.002
0.0001
0.0001
0.0001
75.132
0.533
70.007
0.0001
0.0001
0.0001
711.059
0.070
70.001
0.0001
0.2307
0.0699
0.50
36.04
Ð
0.007
70.008
0.005
0.080
0.027
0.443
0.359
0.050
70.236
0.0001
0.0154
0.0001
0.0001
0.8743
0.0074
0.0001
0.0001
0.0001
0.022
70.022
0.017
0.276
70.326
1.151
1.282
0.135
70.665
0.0001
0.0187
0.0001
0.0001
0.5255
0.0199
0.0001
0.0001
0.0001
0.006
70.001
0.004
0.100
70.562
0.028
0.473
70.001
70.062
0.0087
0.9148
0.0543
0.0020
0.1062
0.9322
0.0015
0.9738
0.5822
47.49
14.18
43.66
6.41
0.30
0.34
1.67
5.87
1.11
70.266
0.377
0.16
0.0127
0.0108
70.628
1.606
0.20
0.0489
0.0003
0.137
0.972
0.44
0.5254
0.0012
0.32
0.36
a
Units: BMI in kg=m2, Weight in kg; Height in cm; Gender dummy, 0 ˆ male, 1 ˆ female; Per capita cash expenditures, per capita value home produce, and
per capita total income in `000's cedis per year; Electricity dummy, 0 ˆ no electricity, 1 ˆ electricity; Water dummy, 0 ˆ no treated water, 1 ˆ treated
water; Building material of dwelling, score from 1±3; Schoolyears head of household in years; Seasonal dummy, March-June (lean season) ˆ 1, other
months ˆ 0; Urban dummy, 0 ˆ rural, 1 ˆ urban.
Source: GLSS-I and GLSS-II.
indicate a signi®cantly positive effect on BMI and weight.
On the other hand, the per capita value of consumed home
produce has a negative effect on BMI and weight. As
mentioned, this probably re¯ects the fact that the consumption of home produce is in particular high in low-income
rural households. Per capital total expenditures was not
included in the regression as it is a linear combination of
per capita cash expenditures and per capita value of
consumed home produce. The rather large signi®cantly
positive effect of household size on all three anthropometric variables is probably an artifact. Per capita incomes
or expenditures were calculated by dividing total incomes
or expenditures by the number of householdmembers, and
for larger households the per capita results might have
some downward bias.
Of the two variables which re¯ect household amenities
related to access to electricity and piped water, only the last
one yields signi®cantly positive parameters. While such
variables may be proxy's for overall household wealth,
access to piped water also reduces the food energy requirements of household members for the collection of water.
Similarly, the positive effect of household heads' educational attainment on BMI and weight may be partially
explained by the fact that better education is generally
associated with physically less demanding jobs. Education
also contributes directly to income. The signi®cantly negative effect on BMI and weight of the number of acres
farmed may well be caused by the fact that work on larger
farms is more laborious.
Finally, the community level characteristics included in
the model are related to ®xed effects from the season at the
time of data collection, and from geographical location. A
signi®cantly negative effect on BMI and weight was
observed for the March±June period (1988 and 1989),
which is the lean season in Ghana. With respect to
geographical effects, being a resident in urban areas
appeared to have a signi®cantly positive effect on BMI,
weight, and height.
We thus reach the conclusion that, unless simultaneity
and inter-household and inter-community effects are pre-
dominant and have major effects on the sign and size of the
estimated effects, most factors related with standards of
living have signi®cantly positive effects on BMI and
weight, but not on height, thus con®rming the correlation
results in the previous sections. With respect to the explanatory power of the model, the R-squares for BMI, weight
and height are respectively 0.16, 0.20 and 0.44. It is
important to note that the high R-square for height is
largely the result of including gender as explanatory variable. When gender was omitted from the model, the Rsquares for BMI, weight and height decreased to respectively 0.12, 0.14 and 0.07.
Summary and conclusions
The present study aims to contribute to the assessment of
BMI as an indicator of standards of living. On the basis of a
large sample representing various socioeconomic groups in
Ghana, relationships between BMI and various household
characteristics, presumed to re¯ect different aspects of
standards of living, were studied. In particular, a crosssectional comparison was made between mean values of
adult BMI and various household characteristics of different socio-economic groups. For comparison, the same
relationships were investigated for weight and height.
The data revealed, at the individual level, signi®cantly
positive relationships between BMI and the various indicators of standard of the living, though the correlation
coef®cients indicate a poor ®t. At the population level,
the same relationships were much stronger, with most
correlation coef®cients in the 0.80±0.95 range. Weight
appeared to have similar characteristics as BMI with
respect to these relationships, while for height correlation
coef®cients were consistently lower. Also in the multiple
regression model most selected household variables
appeared to have a signi®cant effect on both BMI and
weight, but not on height. It can be concluded that, under
conditions as prevailed in Ghana at the time of the surveys,
differences in mean BMI or differences in mean weight
between population groups can be considered as indicative
BMI indicator of standard of living in developing countries
M Nube et al
of differences in standards of living, with the highest
standards of living in those population groups with the
highest BMI or highest weight. On the other hand, on the
basis of much weaker relationships, height is not considered suitable as indicator of standards of living.
When a comparison is to be made between BMI and
weight, weight is considered less appropriate as indicator
of well-being than BMI. The main reason is that differences in mean weight between population groups may
simply re¯ect differences in mean height. The heights of
individuals, apart from being strongly related to the height
of parents, are likely to re¯ect differences in health and
nutrition conditions in the more remote past, and are not
likely to be affected by present or recent conditions of
standards of living. BMI on the other hand re¯ects, by
de®nition, weight relative to height, and is at least partially
affected by present living conditions. It is further important to note that over past years some consensus has been
reached on a normal range for BMI, below or above which
health risks are increased. Such norms do not exist for
weight.
When comparing BMI with more common indicators of
standards of living such as incomes or expenditures, it may
be noted that data on weight and height are likely to be
more reliable than income or expenditure data. Clearly,
obtaining weight and height data has its own dif®culties
and pitfalls, but obtaining reliable income or expenditure
information is generally much more complicated, in particular in developing countries.
An important limitation with respect to BMI as indicator
of standards of living is that in more af¯uent societies the
positive associations between BMI and various aspects of
standards of living are generally no longer present. In other
words, beyond a certain level of standards of living, BMI
will no longer be indicative of household well-being.
However, when considering the present results, such a
point appears not yet to have been reached in Ghana. For
the twelve identi®ed socio-economic groups, the relationship between means of cash expenditures and BMI is of a
linear type, almost over the whole range of expenditures,
with the mean BMIs of the twelve socio-economic groups
varying from approximately 20±23. When the mean BMI of
a population or population group is within this range or
below, an increase in mean BMI can be considered as
strongly indicative for an increase in overall standards of
living.
References
Alderman H (1990): Nutritional status and its determinants, Social
Dimensions of Adjustment in Sub-Saharan Africa, Working Paper 3,
World Bank, Washington.
Alderman H & Garcia M (1993): Poverty, household food security, and
nutrition in rural Pakistan, Research Report 96, International Food
Policy and Research Institute, Washington DC.
Alemu T & Lindtjorn B (1995): Physical activity, illness and nutritional
status among adults in a rural Ethiopian community. Int. J. Epidemiol.
24, 977±983.
Bailey KV & Ferro-Luzzi A (1995): Use of body mass index of adults in
assessing individual and community nutritional status. Bull. World.
Hlth. Org. 73, 673±680.
Blackwood DL & Lynch RG (1994): The measurement of inequality and
poverty. World. Develop. 22, 567±578.
Boom GJM van den, Nube M & Asenso-Okyere WK (1996): Nutrition,
labour productivity and labour supply of men and women in Ghana.
J. Devt. Studies. 32, 801±829.
Bouis HE (1994): The effect of income on demand for food in poor
countries: Are our food consumption data bases giving us reliable
estimates? J. Devt. Econ. 44, 199±226.
Burke GL, Savage PJ, Manolio TA, Sprafka JM, Wagenknecht LE, Sidney
S, Perkins LL, Liu K & Jacobs DR (1992): Correlates of obesity in
young black and white women: The CARDIA Study. Am. J. Publ.
Health. 82, 1621±1625.
Cornu A, Masamba JP, Traissac P, Simondon F, Villeneuve P & Delpeuch
F (1995): Nutritional change and economic crisis in an urban Congolese
community. Int. J. Epidemiol. 24, 155±164.
Croft JB, Strogatz DS, James SA, Keenan NL, Ammerman AS, Malarcher
AM & Haines PS (1992): Socioeconomic and behavioral correlates of
Body Mass Index in black adults: The Pitt County Study. Am. J. Public
Health. 82, 821±826.
Dasgupta P & Weale M (1992): On measuring the quality of life. World
Develop. 20, 119±131.
Deolalikar A (1988): Do health and nutrition in¯uence labor productivity
in agriculture? Econometric estimates for rural South India. Rev. Econ.
Stat. 70, 406±413.
Dhurandhar NV & Kulkarni PR (1992): Prevalence of obesity in Bombay.
Int. J. Obes. and Rel. Metab. Dis. 16, 367±375.
Dibley MJ, Goldsby JB, Staehling NW & Trowbridge FL (1987):
Development of normalized curves for the international growth reference: historical and technical considerations. Am. J. Clin. Nutr. 46, 736±
748.
Dryson E, Metcalf P, Baker J & Scragg R (1992): The relationship
between body mass index and socio-economic status in New
Zealand: ethnic and occupational factors. New Zealand Med. J. 105,
233±235.
Durnin JVGA (1994): Low body mass index, physical work capacity and
physical activity levels. Eur J. Clin. Nutr. 48, Suppl 3, S39±S44.
FAO (1983): A comparative study of food consumption data from food
balance sheets and household surveys, FAO Economic and Social
Development Paper 34, FAO, Rome.
FAO (1994): Body Mass Index, A measure of chronic energy de®ciency in
adults, FAO Food and Nutrition Paper 56, FAO, Rome.
Ferro-Luzzi A, Sette S, Franklin M & James WP (1992): A simpli®ed
approach of assessing adult chronic energy de®ciency. Eur. J. Clin.
Nutr. 46, 173±186.
Garcia M & Kennedy E (1994): Assessing linkages between body mass
index and morbidity in adults: Evidence from four developing countries.
Eur. J. Clin. Nutr. 48, Suppl 3, S90±S97.
Ghana Statistical Service (1994): The Ghana Demographic and Health
Survey, 1993, Ghana Statistical Service, Accra, Ghana, and Macro
International Inc., Calverton, Maryland, USA.
Haddad L & Bouis H (1991): The impact of nutritional status on
agricultural productivity: wage evidence from the Philippines. Oxford
Bull. Econ. Stat. 53, 45±68.
House WJ (1991): The nature and determinants of socio-economic
inequality among peasant households in Southern Sudan. World
Develop. 19, 867±884.
INCLEN (1994): Socio-economic status and risk factors for cardiovascular
disease: A multicentre collaborative study in the international
clinical epidemiological network (INCLEN). J. Clin. Epidemiol. 47,
1401±1409.
Kennedy E & Garcia M (1994): Body index and economic productivity.
Eur. J. Clin. Nutr. 48, Suppl 3, S45±S55.
Minhas BS (1991): On estimating the inadequacy of energy intakes:
Revealed food consumption behavior versus nutritional norms. J.
Devt. Studies 28, 1±38.
Paxton SJ, Sculthorpe A & Gibbons K (1994): Weight-loss strategies and
beliefs in high and low socio-economic areas of Melbourne. Austr. J.
Public Health 18, 412±417.
Ravallion M (1996a): Poverty comparisons: A guide to concepts and
methods, Living Standards Measurement Study, Working Paper 122,
World Bank Washington.
Ravallion M (1996b): How well can method substitute for data? Five
experiments in poverty analysis. World Bank Observer 11, 199±
221.
Rotimi CN, Cooper RS, Ataman SL, Osotimehin B, Kadiri S, Muna W,
Kingue S, Fraser H & McGee D (1995): Distribution of anthropometric
variables and the prevalence of obesity in populations of west African
origin: International Collaborative Study on Hypertension in Blacks
(ICSHIB). Obes. Res. 3, Suppl 2, 95S±105S.
Seidell JC (1995): Obesity in Europe. Obes. Res. 3, Suppl 2, 89S±93S.
Seini AW, Nyanteng VK & van den Boom GJM (1997): Income and
expenditure pro®les and poverty in Ghana. In: Sustainable Food
Security in West Africa, WK Asenso-Okyere, G. Benneh & W Tims
(eds.). Kluwer Academic Publishers, Boston-USA, Dordrecht-Netherlands.
Sichieri R, Coitinho DC, Leao MM, Recine E & Everhart JE (1994): High
temporal, geographic and income variation in Body Mass Index among
adults in Brazil. Am. J. Public Health 84, 793±798.
143
BMI indicator of standard of living in developing countries
M Nube et al
144
Simmons G, Jackson R, Swinburn B & Yee RL (1996): The increasing
prevalence of obesity in New Zealand: is it related to recent trends in
smoking and physical activity? New Zealand Med. J. 109, 90±92.
Strauss J (1993): The impact of improved nutrition on labour productivity
and human resources development. In: The Political Economy of Food
and Nutrition Policies, P Pinstrup-Andersen (ed). Baltimore: John
Hopkins University Press.
Thomas D, Lavy V & Strauss J (1996): Public policy and anthropometric
outcomes in CoÃte d'Ivoire. J Publ. Econ. 61, 155±192.
UNDP (1990) (and subsequent years), Human Development Report,
Oxford University Press.
WHO (1990): Diet, nutrition, and the prevention of chronic diseases,
Techn. Rep. Series 797, WHO Geneva.
WHO (1995): Physical Status: The Use and Interpretation of Anthropometry, Techn. Rep. Series 854, WHO, Geneva.
World Bank (1997): World Development Report, Oxford University Press,
New York.