Methods

Description of the research site

The data used to operationalize household food security come from the principal author's dissertation research, conducted from November 1986 through August 1987 in Pahargaon (a pseudonym) Village Development Committee (formerly called a panchayat) in the western hills of Nepal. A total of 115 households were randomly sampled, representing 767 individuals in six villages. The villages included in the study area lie along the slopes of hills at altitudes ranging from 3,500 to 4,800 feet. Agricultural fields range from approximately 10,00 feet (down in the river valley) up to 5,000 feet. The lower river valley fields (irrigated cropland, or khet) are considered more valuable because they are more productive. All study households owned some land, but for many the amount was inadequate for subsistence. A system of land rental (adhiyaa) is well established in Pahargaon, in which landowners permit villagers to cultivate plots of land and receive half of the harvested produce as payment.

FIG. 2. Detailed framework of food security within the household

FIG. 3. Food flow through the Nepali household

Villages in Pahargaon are largely isolated from the larger market areas and centres of power for the region. The area around Pahargaon is heavily deforested, and villagers must walk three to five hours to obtain firewood. Water is available mainly from ground springs, which vary in distance from a few minutes to a half-hour round-trip walk from Pahargaon households. Ethnically, Pahargaon is composed of members of all four main caste groups in the Hindu Varna system: Brahmin, Chhetri, Vaisya, and Shudra. There are notable differences between higher and lower castes in terms of education, occupation, wealth, and, consequently, political power. In rural Nepal these differences extend also to food proscriptions related to caste status and thereby to diet.

Data-collection techniques

Before the initiation of data collection using structured instruments, exploratory qualitative research was conducted using key informant interviewing, focus groups, and unstructured observation techniques. This period of preliminary ethnographic data collection assisted in developing culturally appropriate and valid quantitative instruments for later phases of the research and contributed to the final interpretation of the quantitative data results.

Structured data collection was focused on four key areas. All four instruments were administered from January to April 1996. This is generally the period of greatest food availability in the panchayat. The household food-frequency instrument was administered a second time in almost all study households approximately three to five months later from June to August 1996. From June to August is the pre-monsoon and early monsoon season, which is generally regarded as the period of greatest food scarcity in the panchayat.

Household and individual demographic data, including information on caste status, age, and sex, were collected using a survey administered to the male head of the household.

Economic status indicators were collected at the household level from the male head of the household, including ownership of land, animals, and material possessions and quality of the house.

Household food stores and usage patterns were obtained through a structured interview. The male head of the household was asked to estimate the amounts of 20 key foods (identified in the ethnographic survey) acquired by the household over the preceding 12 months and how the food had been used by household members (an indicator of past food security). He was also asked to describe the amounts of each food currently in storage (an indicator of current food security), limited to "storable" foods, such as grains and tubers. In addition, the respondent was asked to describe the amount of land currently planted and the numbers of each kind of animal currently owned (indicators of future food security).

The accuracy of recall by the informant over an extended period of time was of concern. Accuracy was enhanced by several methods:

» Different means of food inflow and outflow were identified by ethnographic methods and distinguished from one another. For instance, respondents were asked not only the total amount of rice that came into the household, but also how much rice they produced on their land, received as payment, received as gifts, received in trade, or purchased.

» Information was cross-checked during the interview, both within and between foods, and with other questions. The total food coming into the household should be roughly equal to the amount reportedly flowing out of the household, plus the amounts reported as eaten and stored. A household owning a lot of rice-producing land but reporting very low rice production would be asked to explain the inconsistency.

» Respondents were encouraged to report quantities using a variety of local measures, which were later translated into grams.

» Other household members, especially those involved in agricultural production, were encouraged to participate during the interview and often served to refresh the memory of the principal respondent.

Household food consumption patterns were estimated using a weekly food-frequency instrument. This instrument was administered twice in each study household. The female head of the household was asked to report the number of times any household members had consumed 70 different foods during the previous week and to give an estimate of the amount of food consumed by household members each time (familiar household measures were used to estimate quantity). The 70 foods were identified as the most commonly consumed through preliminary ethnographic interviews with key informants; however, additional spaces were provided for other foods.

Scale and score construction

This section describes how we operationalized food security at the micro level of the household. A key method was the construction of scales and scores that captured the complexity of the factors that make up household food security [30]. Separate exploratory factor analyses were conducted to identify key components of three different scales representing past, current, and future household food security. Factor analysis is an appropriate analytic method when the investigator wants to identify key constructs underlying a set of data [31]. The method has been used in dietary studies to identify patterns of food consumption for specific populations [32-34]. Although we initially experimented with developing our own scoring system, we soon discovered that the complexity of the data (multiple sources of food, multiple ways that food could leave the household, multiple styles of managing food resources) necessitated an analysis strategy that would permit underlying patterns to emerge, effectively summarize data, and provide optimal weights for component variables.

The principal-factor method was used to identify components of each scale [31]. A combination of scree test (a plot of the eigenvalues of the factors) and assessment of the proportion of the variance accounted for by the factors was used to determine the number of factors to be retained for rotation (conducted using the varimax method). In interpreting the rotated factor pattern, a selected item was considered to load on a given factor if the loading was 0.40 or greater for that factor and was less than 0.40 for all other factors. No item was permitted to load on more than one factor. Factor scores for each item in the three scales were computed by multiplying its value by its factor weighting. Reliability for all scales was assessed by calculating coefficient alpha [35].

Past food stability scale

To obtain some indication of past food supply stability (PASTFDSC), respondents were asked to recall the flow of 20 key foods into and out of the household during the 12 months leading up to the interview date. The foods were rice, shuto (dried ginger), wheat, corn, mustard, potatoes, barley, lentils, millet, soya beans, peanuts, vegetables, fruit, milk, eggs, goat, chicken, buffalo, and pig. The respondents were asked to estimate the amount of each food coming into household stores through five specific pathways: production, purchase, gift, payment, and trade. The respondents were then asked to estimate how much of these foods left household stores through six pathways: consumed by household members, sold, traded, given to others, paid in rent, and fed to animals. Payment includes food received as rent for land use. Trade indicates food traded for other types of food. Gifts can mean food received either as a gift or, as many low-caste families do, as compensation for services rendered (e.g., leatherwork, blacksmithing, or tailoring). All 20 food categories were combined on the basis of source (how they came into the household) and use (how they left the household). Each of these scores was then adjusted according to household caloric requirements (to account for age and sex composition differences between households). These adjusted variables were then converted into common "units" by recoding each score into quartiles.

Factor analysis was then used to identify the main patterning in the scores. Most loaded on factor 1 (amount of food stored, sold, given in rent, produced, fed to animals, or given as gifts). Traded (either received or given) food consistently loaded on its own factor (factor 2). Food purchased (bought) and food received as pay consistently loaded on their own factor (factor 3). A second round of correlation analysis was conducted to verify the factor analysis findings. The final Cronbach's alpha of the six-item PASTFDSC scale was 0.747, indicating a reliable unidimensional scale. Finally, confirmatory factor analysis was used to generate standardized scoring coefficients for these items to use as weights when combining the items into a single-scale score. All items loaded onto one factor. The final PASTFDSC variable had a mean of 1.61, a standard deviation of 1.06, a median of 1.63, and a range of 0 to 3.42 and was approximately normally distributed. A high score on the PASTFDSC therefore indicates that in comparison with households with lower scores, the household produced a lot of its food, had a lot of food in stores, gave out a lot of food in rent (and therefore had people working on its land), gave out a lot of food as gifts, and used a lot of food to feed its animals.

Current food supply/stores scale

The current food security scale (CURRFDSC) reflects household food stores at the time of the household interview. The 20 foods recorded in the household food stores and usage instrument were combined into 8 food groups. For example, rice, corn, wheat, and millet stores were combined into the grains group. Factor analysis and correlation analysis were used to select food-store variables to constitute a unidimensional scale. The final scale included grains, vegetables, nuts and beans, and milk (based on current productivity estimates of milk-producing animals) and had a Cronbach's alpha of 0.711. Factor analysis was then used to generate weights that were used to combine the four food groups into one scale. Univariate statistics on the scale CURRFDSC indicated a fairly normal distribution, with a mean of 1.5, a standard deviation of 0.96, and a range of 0 to 3.26.

Future food productivity scale

This scale reflects the amount of land currently planted in a variety of crops and the numbers of work animals and meat- or milk-producing animals currently owned as a means of indicating the potential of the household to produce food in the near future (FUTUFDSC). For each of 11 planted food crops, the amount planted in seed (e.g., the amount of rice seed) in the current year was weighted by the proportion of total land that was owned or rented by the household. Plantings on rented land were weighted by 0.5, since the household would only receive half of what they planted. These foods included those crops that are most commonly planted in large quantities and not in kitchen gardens (except tirmilo [an indigenous black oilseed] and mustard).

For fruits and vegetables, households were only asked whether or not they grew a particular variety on their own land. Correlation analysis was done to construct additive fruit variety (13 items, alpha=0.825) and vegetable variety (18 items, alpha=0.881) subscales. In terms of animals, correlational analysis resulted in an additive subscale that included numbers of cows, bulls, goats, and buffaloes (alpha=0.530).

Each of these scores - 13 planted foods (amounts planted), fruit variety subscale, vegetable variety subscale, and animal ownership subscale - was then converted into quartiles. Correlation analysis was done on the converted variables to construct a scale for future household food security. Thirteen items remained in the final scale: fruits subscale, vegetables subscale, animals subscale, and the following planted crops: tirmilo/baari, peanuts/baari, millet/baari, lentils/baari, potatoes/baari, mustard/baari, corn/baari, wheat/baari, wheat/khet, and rice/khet. (Baari is unirrigated cropland and khet is irrigated cropland.) The final scale (FUTUFDSC) has an acceptable Cronbach's alpha of 0.784. The scale values have a normal distribution, with a mean of 14.7, a standard deviation of 7.09, and a range of 0 to 30.

Data analysis

The effect of the three measures of household food security on household food consumption patterns was examined using multiple regression. Separate models were run to examine the effects of past, current, and future food security on the frequency of consumption of different food groups and on the variety of foods consumed by the household (both between and within food groups). Scale scores for each of the three measures were converted into quartiles, with the second, third, and fourth quartiles entered into the models as dummy variables. The primary outcome variables for the analyses were based on the food-frequency results. These data were summarized by calculating additive scores by food group (grains, beans, green leafy vegetables, tubers, other vegetables, fruits, meats, and dairy products). Dietary variety, a proxy for dietary quality, was calculated in two ways: as total food group variety (whether or not one or more foods were consumed within each food group; maximum score, 8) and as total food group intensity (summing all foods in all food groups; maximum score, 30).

Other variables included in the models were dummy variables for caste (Brahmin, Chhetri, and Vaisya were included; Sudra, the lowest-caste group, was not included) and socio-economic status (the second and third terciles were included; the lowest tercile was not included), based on the total value of all possessions. In addition, an independency ratio (number of adult male and female household members aged 15 to 60/number of children and elderly in the household) was calculated and incorporated into the models. Standardized beta coefficients were generated for each of the models. Statistical analysis was performed using the SAS statistical package (SAS/STST version 6.11, SAS Institute, Cary, NC, USA).