Study area The fieldwork was conducted in two selected villages in Bangladesh: Falshatia and Jorbaria. Falshatia, in the Manikganj thana (subdistrict) is a flooded, double-cropped area, west of Dhaka and Jorbaria. Mymensingh is a non-flooded, triple-cropped area, north-east of Dhaka. In the Manikganj village, the agricultural production system can be characterized as traditional, whereas in the Mymensingh village the farming system practiced can be characterized as modern, with the use of high-yielding rice varieties, irrigation, fertilizers and pesticides. In Bangladesh, the three rice harvest seasons comprise the Aman (November – December), the Aus (March – April), and the Boro (May – June). The Aman harvest was traditionally the most important, but in areas where high-yielding rice varieties and modern technology have been introduced, the Boro harvest has become more important during the last decade. The present study was undertaken to cover two seasons with respect to rice production: the lean season which includes October – November and the peak season which includes December – March. The study was conducted in 1981 – 1982 and 1995 – 1996. Study population and sampling design In 1981 – 1982, the study population comprised 50 households in each village in both seasons (October – November 1981 and December 1981 – January 1982) and in 1995 – 1996, 152 households in each village in both seasons (October – November 1995 and January – March 96). The data collected in 1981 – 1982 was a part of the national nutrition survey of rural Bangladesh in 1981 – 1982 (Ahmad & Hassan, 1983). In 1995 – 1996, households were sampled and data collected using the same technique as in 1981 – 1982. The households were selected from four socioeconomic subgroups following the probability proportional to size method. The four subgroups were based on the Bangladesh rural population tax system: group A is the poorest, most certainly landless and not required to pay any local tax; group B is the less poor and pays a token tax. Group C may be regarded as a rural upper-middle-class and pay an appreciable amount of tax, while group D is the richest group and pays the highest proportion of tax. In Manikganj, the proportions of sampled households in the two wealthiest groups were 14% in 1981 – 1982 and 36% in 1995 – 1996. In Mymensingh, the proportions of the wealthiest groups were 24% in 1981 – 1982 and 10% in 1995 – 1996, perhaps due to the concentration of landholding on fewer households. Of the 100 households studied in 1981 – 1982, data were missing from one household in December – January. Of the 304 households studied in October – November 1995, data were missing from 13 households in January – March. The study covered a total of 510 individuals in October – November 1981, 512 individuals in December – January 1981 – 1982, 1373 individuals in October – November 1995 and 1281 individuals in January – March 1996. The characteristics of the surveyed households are shown in Table 1. Although the average household size was larger in 1981 – 1982 than in 1995 – 1996, the difference was not significant. The average number of consumption units for each season and year ranged from 4.4 to 5.3 and decreased from 5.2 (95% CI 4.9 – 5.6) in 1981 – 1982 to 4.5 (95% CI 4.3 – 4.7) in 1995 – 1996. Dietary intake Dietary intakes of all members of the selected households were obtained by 24 h food weighing. Training of investigators comprised practicing the techniques of food weighing and recording at the Institute of Nutrition and Food Science, University of Dhaka, Bangladesh for 2 weeks, followed by a pilot survey in a village close to Dhaka city. Exactly the same method was used in 1981 – 1982 and 1995 – 1996. All food that went into the household cooking pots was weighed as raw weight. Intake per person was calculated at household level by dividing the total intake of each single food in every household by the total number of consumption units in the household. An individual present at every meal within the one-day period was counted as a full-time consumer, with the value of one consumption unit. All family members, including children and breast-fed infants, were valued similarly in terms of consumption unit. Arithmetical adjustments of the household’s number of consumption units were made for household members who were partially absent. If a household member was absent at one out of the usual three meals consumed in the household, two-thirds of a consumption unit, instead of one, was allocated to this member (FAO, 1967). To avoid changes in energy and nutrient intakes caused by differences in food composition values, the values of all foods were computed using the same food composition table pertinent for the Bangladeshi diet for both 1981 – 1982 and 1995 – 1996 (Darnton-Hill et al, 1988). The foods consumed were divided into selected food groups on the basis of the above food composition table with the following modifications: pulses and legumes, green leafy vegetables and vegetables were pooled in the food group of vegetables; fats, oils and oilseeds were pooled in the food group of fats and oils and fish, meat, eggs, milk and milk products were pooled in the group of animal products. Fish were divided into small and large fish, of which all the small fish are indigenous and most of the large fish are cultured. Data analysis All analyses were performed with the Statistical Analysis Systems statistical software package version 8.1 (SAS, 2000). Prevalence of stunting, wasting and underweight among children and CED among adults were analyzed by logistic regressions using the SAS procedure GENMOD. Year, village, season, gender and socioeconomic group were included as independent fixed variables. Individuals nested within both year, village and household and households nested within year and village were included as random effects in the statistical model. MUAC for children and adults and BMI for adults were analyzed by ANOVA as described below. All main effects and two-factor interactions between fixed effects were tested. Only P-values for a year and significant fixed effects are presented. Univariate analyses of variance (ANOVA) were performed in the procedure MIXED (Littell et al, 1996). In the statistical model, consumption of food by food group was evaluated as the dependent variable. Year, village, season and the socioeconomic group were included as independent fixed variables. Households nested within both year and village were included as a random effect in the statistical model to take into account that repeated measures were done in households within years but the households between years were different. All main effects were evaluated as well as two- and three-factor interactions between all fixed effects. To obtain sufficient numbers of individuals in each socioeconomic group, the two wealthiest groups were pooled (group C and D). Preliminary statistics showed that the two wealthiest groups were not significantly different. Homogeneity of variance and normal distribution among random effects were investigated by plots and histograms of residuals. Shapiro –Wilk’s test for normal distribution was performed. The investigations showed it necessary to log transform intakes of protein, fat iron, calcium, vitamin A and the food groups fats and oils and spices. The least-square means (l.s. means) and standard errors (s.e.) were estimated and transformed back where necessary before the presentation. The significance level was set to 0.05 for main effects, 0.01 for two-factor interactions and 0.001 for three-factor interaction to reduce the probability of type I error. Only P-values for main fixed effects, significant interactions and the three-factor interaction between year, village and season are presented.