Study site: Southwestern Bangladesh is a flood plain bounded by India to the west and the Bay of Bengal to the south. It is a low (<10 m asl), flat, and fertile deltaic plain dominated by alluvial soils (BBS 2004). The coastal plain is partly sandy and saline. The most dominant physiography characterizing the deltaic landscape is its extensive network of river systems. The region contains a large number of oxbow lakes. Approximately 23 million people inhabit the region with a density of approximately 650 persons km-2 and a mean family size of 6. The population growth rate is 2.1%. Sex ratio is 1:0.94 (male/female). Twenty percent of the population is urban, making southwestern Bangladesh’s population predominantly rural. Education is not compulsory, but is free for elementary level. The crude literacy rate of the region is 45%. Approximately 37% (two persons per household) of the total population of the region is economically active, 77% of which is male. The per capita income averages 350 US$ (1 US$ = 65 Taka in January 2006) with an approximate economic growth of rate of 8%. Approximately 15% of Bangladesh’s total land area (147,570 km2) is officially demarcated as public forests and no forests are accessible to households in the study area. Agriculture is the main occupation for most inhabitants of the region, followed by small to medium-scale business, daily labor, and civil service. Rice, wheat, jute, sugarcane, pulses, and potatoes are the principal agricultural crops in southwestern Bangladesh, but the region is also important for production of various types of vegetables, spices, fruits, and nuts. Intensive shrimp (tiger prawn) culture is a newly emerging economic activity along the coastal belt of the region. Approximately 5% of the total land area of the region is under homestead use, which includes the dwelling house, homegarden, animal pens, and a pond. The homestead area averages 500 m2 across the region. A tropical to sub-tropical monsoon climate characterizes the region. In the plain land areas of the region flooding occurs during the monsoon and severe drought during dry season (Johnson 1982; Rasid and Paul 1987; Rasid and Pramanik 1990), causing heavy damage to the lives and property in the region. The annual average rainfall is 1,800 ± 268 mm with an annual average relative humidity of 78% (BBS 2004). Three distinct seasons are summer (March to May), rainy (June to October), and winter (November to February). The mean annual temperature is 260C, in some places reaching a maximum of 400C or more during the summer and going down to 70C during the winter.
Sampling design: Six major regions/districts (administrative unit) locally called zilla were purposively selected from southwestern Bangladesh. Each district consists of a number of sub-districts locally called thana (administrative unit). Two sub-districts from each purposively selected district (n = 12) were selected randomly. In the absence of systematically arranged households and sub-district information on households, we used the following method to select study households in an unbiased manner. Every day a new local guide was hired to assist in selecting households for the inventory. The research team did not tell the local guide the specific reason for selecting households. Immediately after selecting a household, s/he was requested to stay away from the household in order to prevent bias with future household selection. A total of 402 households from six purposively selected districts were selected randomly for primary data collection. Data for this study were collected on two aspects: (1) household characteristics and (2) homegarden vegetation structure.
Data collection
Household characteristics: A face to face interview was conducted with the head of each household, in the presence of all family members if available. Data on 21 contextual attributes were collected on biophysical, demographic, and socioeconomic conditions from each sampled household, using a questionnaire administered during the interviews. Household biophysical data such as the type of road access (paved or mud) and distance (in km) from the household to the nearest market and urban center were recorded with the help of household members. Household demographic data such as household and household head’s age, head’s gender, education, headship tenure, and family size (number of total, adult, earning, and literate members) were recorded. Household socioeconomic data such as homestead size, agricultural landholdings, and total landholdings were recorded in local unit Katha (1 Katha = 67 m2). Data on the household head’s occupation, major source of family income, and total time invested (hours per week) for homegardening were also recorded from each sampled household. Data on household income such as income from agricultural activities, off-farm activities, and homegarden (both subsistence and sale) were collected mainly in consultation with all members of the household who had knowledge about it.
Homegarden vegetation structure: A botanical inventory was conducted in the homegarden and was conducted only once in each selected household; therefore the seasonal variation in stem density and species richness in the homegarden was not assessed. Every individual plant except small grasses (<25 cm height) was identified and recorded by botanical name whenever possible, or by local name if the botanical name was not immediately known and later confirmed at the Bangladesh National Herbarium. Every individual tree and shrub was counted, except those in hedgerows due to difficulty in differentiating the stems. Individuals of herbs and climbers (woody and non-woody) were not counted due to difficulty in differentiating the stems. However, the individuals of herbs and climbers of the IUCN Red Listed species were counted. Geographical location and altitude of each sample homegarden was recorded using a GARMIN GPS.
Data analysis: The central analytical method for this paper was multiple linear regression using household contextual characteristics as the independent variables, with number of species per homegarden, tree and shrub stem density per hectare, household income from homegarden, and the percent share of household income from homegarden as the dependent variables. A correlation matrix exhibited several correlations among the 21 selected independent variables. We therefore performed a principal component analysis (PCA) with Varimax Rotation and Kaiser Normalization to reduce the independent variables into factors or component variables (Hair Jr et al. 1998). PCA is powerful and widely used tool (Gonzalez and Woods 1992) to simplify complex multidimensional variables into factors, and gives uncorrelated variables by transforming data into factors (Cooley and Lohnes 1971). PCA factor loadings were then used to calculate the PCA scores of each factor for each household. Species richness per homegarden was calculated for each sampled household. The abundance (number of individuals per species) for each species of tree and shrub in the homegarden (except those planted in hedgerows) was computed for each household and later converted to a per hectare basis. Household annual income was calculated as the sum of income from agriculture, off-farm, and the gross margin of the homegarden. Gross margin of each homegarden was calculated by asking the respondent what percentage of homegarden products they sold and consumed, how much income was earned from the previous year’s sale of homegarden products, converting those earnings into value for the consumed products, and then summing the sales and consumption values. This allowed us to calculate the proportion of total income represented by homegardens. It should be noted that the homegardeners’ estimates of percent of product sale is an aggregate value across all products, and not an estimate for each product. This avoided error associated with owing to difficulties in recollection by the respondents, but may have reduced the precision of the gross margin through aggregation. Four multivariate generalized linear models (GLM) were performed to test the influence of the independent variables (PC scores) and regional influence on the species richness per homegarden, tree and shrub stem density, household income from homegarden, and the percent share of household income from homegarden. The multivariate GLM used region/district name as the fixed factor and the independent variable PC scores as covariates. A Levene’s Test for homogeneity of variances was performed for all dependent variables during all GLM analysis. Normality of residuals was confirmed post hoc by plotting the histogram of the standardized residuals and testing against a normal distribution with a one-sample Kolmogorov–Smirnov test. When variances or residuals did not meet test requirements, we performed a weighted least squares procedure, where the group weightage was the inverse of the residual variance (National Institute of Standards and Technology 2007). One backward stepwise binary logistic regression was conducted using PCA scores as independent variables and the presence–absence of IUCN Red Listed species for Bangladesh in homegarden as the binary dependent variable (absence = 0, presence = 1).