Study design
We used a food systems framing to conceptually position our research to investigate how small-scale fisheries shape two key aspects of food environments – physical access to food via living in proximity to small-scale fisheries (fish as food pathway), and economic access to food via small-scale fisheries livelihoods (fish as income pathway).
We examined food system components of supply chains (small-scale fisheries livelihoods related to harvesting, processing and trade), food environments (proximity to small-scale fisheries and livelihoods), income poverty status, and household diets (fish consumption and annual food security) (Supplementary Fig 7)40,41. Small-scale fisheries are notably recognised for their safety net function during times of shocks and extreme events, increasing the ability of households to recover, exit poverty and afford food over the longer-term42.
Country selection and household survey data
We selected Malawi, Tanzania and Uganda, given these countries represent a region where small-scale fisheries provides the main supply of fish and are important for rural inland and coastal livelihoods24,43, and yet substantial data gaps remain in valuing small-scale fisheries in the regional food system. Small-scale fisheries, particularly inland fisheries, in this region are known to be highly productive with a linear increasing trend in catches over the last three decades25,35. On average 70% of the total catches consist of small pelagic species, which are largely driven by climate, and are highly productive, resilient, and under-exploited34. However, challenges do exist in fisheries governance and signs of over-exploitation of some few fish stocks44, as well as high post-harvest fish waste and loss across value-chains undermine the potential benefits from the sector23. We analysed the World Bank’s Living Standards Measurement Surveys and its Integrated Surveys on Agriculture (LSMS-ISA) from Malawi, Tanzania and Uganda. The LSMS-ISA surveys conducted in these countries collected georeferenced household-level data and had been designed and implemented with a dedicated fishery module39 which contained questions on household fish consumption (frequency, quantity, and form of fresh or dried fish) and small-scale fisheries livelihoods across value chains (harvesting, processing and trading). The fishery module was collected across different years in Malawi (2016–17), Tanzania (2014–15) and Uganda (2010–11), and accordingly these are the years analysed in this study. The LSMS-ISA surveys collects consumption data over a period of 12 months so that the indicator captures the intrinsic variability due to seasonality, such as low and high periods of food consumption.
Geospatial data and distance to fishing grounds
Georeferenced household data from LSMS-ISA surveys were matched with geospatial data on the location of inland water bodies and coastlines (Supplementary Table 11) to investigate geographic correlates (e.g., distance to fishing grounds – water bodies where fisheries occur) of poverty and food security. Data on inland water bodies were from the Global Lakes and Wetlands Database (GLWD)45, and the European Space Agency GlobCover databases for coastlines46. Inland water bodies from the GLWD database include permanent, open water bodies (e.g., lakes, reservoirs, rivers) with a surface area ≥0.1 km2 for each country, including cross-border water bodies. We selected water bodies to represent types of water bodies known to support fisheries, based on catch data24,43. We assume the entire coastline of Tanzania was accessible and used for marine small-scale fisheries. We use the term ‘water body’ to mean either freshwater or marine waters.
Distance between water bodies and households was calculated as the shortest, straight line, distance from the household location (identified through the GPS coordinates of the households) to any point of the nearest water body. The distance was expressed in km.
In our descriptive statistics, a cut-off threshold of 5 km from fishing grounds was used to compare the key indicators presented in this study (e.g., percent of poor and food insecure households, frequency and quantity of fish consumption, etc), for households proximate and distant (≤5 km was considered close and >5 km was considered far) from fished water bodies, as well as between fishing and non-fishing households. The choice of the cut-off threshold used for our descriptive statistics was guided by other studies16,17, in addition to reflecting the distribution of households by quintile of distance to water bodies. Concerning the latter, we found that the average distance from fishing ground of the first quintile was always lower than 5 km in all countries.
In the regression analyses, the distance to water bodies was included as a continuous variable (in km). This choice reflects the need to better understand dynamics for households that tend to live more distant from fishing grounds. These dynamics were captured by measuring the marginal increase in the probability of being poor or food insecure for a one-unit increase (1 km) from the mean distance to fishing grounds.
We acknowledge two limitations behind the calculation of the straight-line distance to water bodies. First, using the straight-line distance to water bodies may introduce biases in the statistical analyses presented, especially for households located in any particular landscapes within the country. The walking or travel time distance over a road network would provide a better alternative, however there is lack of data on road networks. Despite the straight-line distance to water bodies encompasses some limitations, we still believe that this method of calculation provides a good proxy to categorize household in relation to their distance to water bodies, and the results from the analyses should not deviate substantially from other method of calculation. For example, a study51 found that the straight-line distance tends to be highly correlated (R > 0.91) with both walking and travel time distance.
Second, an additional bias in the presented analyses may be introduced due to the modification strategy of the households GPS coordinates. This strategy was implemented before dissemination of household level data to avoid the risk of disclosure of sampled households. In its essence, the modification strategy relies on random offset of cluster center-point within a specified range. For urban areas a range of 0–2 km is used. In rural areas, where risk of disclosure may be higher due to more dispersed communities, a range of 0–5 km offset was used. While we had no control over this modification strategy, we believe that the modification of the GPS coordinates does not affect the way households are classified in relation to their distance to fishing grounds: considering that the modification strategy was applied to both distant and proximate households, we expect that the distribution between households close and distant to water bodies has remained unchanged and, hence, the presented statistics are still valid for the analysis.
Variable construction
We used a range of socio-economic indicators across food system components (Supplementary Table 11). As a measure of physical and economic access to food we used two indicators of small-scale fisheries: proximity to fishing grounds and fishing households. Household livelihoods were assigned according to whether households primarily, but not exclusively, engaged in small-scale fisheries (fishing, harvesting, processing and/or trading which varied by survey), agriculture (e.g., crop or livestock), or neither fisheries or agriculture. For each country survey, households were categorised according to their engagement in fishing and/or agriculture activities in the prior 12 months. Households in which one or more member engaged in fish-related activities were defined as ‘fishing households’. Fish-related livelihood activities were defined as fish harvesting, processing, and trading in Malawi and Tanzania, whilst in Uganda they were defined only as fishing. Households with one or more member engaged in agriculture, but not in fish-related activities, were defined as ‘agriculture households’. Through data exploration of livelihood categories, we found that 96% of all fishing households in our study combine fish-related and agricultural activities, with only 4% engaging exclusively in small-scale fisheries. Examination of diverse livelihood typologies within fishing household category (e.g., fisher-farmer, which is common in the region or exclusive fisher) was deemed out of the scope of this study and not feasible due to the small number of observations of exclusive fishers.
Household poverty was measured using the per-capita monthly expenditure (equivalized using the adult equivalent scale). Poor households were defined as those households with a per-capita monthly expenditure below the national poverty line. The national poverty line –which was defined by national authorities in the three countries analysed–is a country-specific monetary threshold below which a household (and its members) cannot meet their basic needs. The poverty metric, as defined above, was used across physical, natural and human capital: asset wealth, distance to markets, access to land and education level of head of household. Since the asset wealth captures the typologies and number of assets owned by the household (durable goods – radio, bicycle, TV; utilities and infrastructure – access to protected water source and electricity), we developed an index for assets using the principal component analysis. This technique reduced the multi-dimensionality of the asset’s variables, and it allowed the data to identify the linear combinations of the assets components that explain the greatest share of the variation in wealth. As the final wealth index was standardised across households, this index allowed providing a ranking of households which reflected their ownership of assets.
Food security was measured using two indicators; household-level food consumption profile – using the Food Consumption Score (FCS) index20, and subjective food insecurity defined as the number of months during a year that a household reported not having enough food to feed the household. Together, these indicators provide a more comprehensive understanding of household food security over a longer period than other surveys (e.g. Demographic and Health)47,48,49. The LSMS-ISA surveys collects food consumption data over a 7-day recall period. To capture seasonal variation in the food consumption indicators, sampled households were interviewed over a 12-month period: for each month of the year, a different portion of sampled households was interviewed so that the derived indicators reflect the intrinsic variability in food consumption, which may be due to seasonality. We used the FCS index as a food security indicator as it is akin to the data collected via the LSMS-ISA surveys, and that there was a need for comparison across select countries. The FCS index measures the frequency (number of days) and diversity of food groups consumed over a 7-day recall period, with weights given to groups based on nutritional value. The FCS index is validated as a proxy for energy sufficiency (quantity of food) and food access, and is associated with other household-level diet diversity measures (e.g. household dietary diversity score (HDDS))20,48. The difference between FCS and other indicators such as HDDS is the recall period (7-days versus 24 h), diversity of groups, weights assigned based on nutrition, and use of frequency together with diversity of groups consumed. The FCS with a longer recall period can show more habitual consumption but can also have limitations with people’s recall reliability. Although it has not been validated yet as an indicator for micronutrient intake, it does provide weights to nutrient-rich food groups and accounts for frequency of consumption, which other indicators do not. Fish consumption was described in terms of the (i) quantity (kg of wet weight equivalent per household per week), (ii) form (fresh, dried, smoked, other) and (iii) source (purchased, own consumption, gift) of fish consumed. The share of households reporting consumption of other animal source foods was also calculated to examine the relative role of fish in overall diets.
We also examined the prices of foods consumed to investigate the accessibility of fish as food in terms of affordability compared to animal source foods. The LSMS-ISA survey collects data on the value and volume of food that were purchased and consumed. Those two variables were further used to construct the average price for each food item. To control for price level differences between countries, food prices data calculated from the survey were converted from local currency unit to international USD, using the Purchase power parity conversion factor corresponding to the year of the survey (Source: World Development Indicators database, World Bank). Moreover, since the surveys were conducted in different years, nominal prices corresponding to the years of the surveys were converted into real, inflation-adjusted prices using the Consumer Price Index (CPI, base year: 2010). This allowed to control for potential inflation patterns within countries and provide a better comparison of food prices per Kg. across the three countries analyzed (Source: World Development Indicators database, World Bank).
Finally, we drew upon nutritional databases (food composition tables, FishBase and Illuminating Hidden Harvest Initiative) to understand the relative nutritional value of fish; by species, size (small or large) and form (e.g., fresh or dried), compared to other animal source foods (Supplementary Table 12). This enables us to contextualise the nutritional importance of consumption patterns.
Descriptive statistics
We created a harmonized multi-country dataset for Malawi, Tanzania and Uganda with 18,715 nationally representative household-level observations. The sample included in this study represents more than 19 million households corresponding to a population of 93.8 million people across the three countries (Supplementary Information).
Descriptive statistics were calculated to compare poverty and food security indicators among households proximate and distant from fished water bodies, and between fishing and non-fishing households (see full details in Supplementary Information). For this analysis, households distant and proximate from fished water bodies were clustered into two groups based on a cut-off threshold of 5 km (distant > 5 km; proximate ≤5 km). The Welch’s t-test was then used throughout to assess the statistical significance of mean statistics between these two groups.
Econometric model
The estimated probabilities of being poor (household living below the national poverty line) and food insecure (household with a poor food consumption profile) were modelled through two separate probit regression models, where the outcome variable was equal to 1 for poor and food insecure households and 0 otherwise. The independent variables in both models included the household’s distance to water bodies and the distance to food market. Both variables are expressed as continuous variables (in km), reflecting the need to measure the marginal increase in the probability of being poor or food insecure (i.e., the estimated β coefficients) given a one-unit change (1 km) in the distance to fishing ground (or food markets) from its mean. Both models also included an interaction variable which measured the household’s distance to water bodies but restricted to only those households who were unable to reach the food market. We tested this interaction as we expected that living in proximity to water bodies could mitigate the negative effects on poverty and food insecurity when households are unable to access food markets. In order to measure the conditional difference in the average probability to be poor and food insecure between households who engaged in fisheries and households who engaged in other non-fishing activities, we constructed a categorical variable that classified households according to their main livelihood activity, namely (1) neither fishing, nor agriculture households (i.e., the reference baseline household category), (2) fishing households and (3) agriculture households. This categorical variable was further restricted to only households living in proximity to water bodies to better measure for which typology of household the proximity to fishing grounds is most beneficial. Both models were controlled for the age, sex and the highest level of education attained by the head of the household, as well as the total number of household members employed (over total household members) and the wealth index of the household.
For each model (poverty and food insecurity), we examined associations at the cross-country, national and rural levels (Tables 1 and 2, also available as Supplementary Data 1 and 2). Stata 15 was used for all statistical analyses. Both descriptive statistics and the regression coefficients were estimated using the household probability weight, the latter instrumental to make the derived indicators from the surveys representative of the population of interest thus allowing general inference for the three countries.
Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this article.
Source: Ecology - nature.com
