STATEMENT
OF WORK

methods

Climate Variability and Household Welfare in the Andes:
Farmer adaptation and
use of weather forecasts in decision making

Research Setting

We will work in two research sites one in the Province of Aroma in the Central Bolivian Altiplano and the other in Puno in Southern Peru. These sites were chosen because detailed production data exists on them for the early part of this decade. Our first proposed research site is the Community of San José Llanga in the Province of Aroma, Central Bolivian Altiplano. It is characterized by a high degree of abiotic perturbations-- periodic droughts, frequent frosts, occasional floods and seasonal wind erosion (Washington-Allen, 1993). Mean annual precipitation between 1943 and 1990 is about 400 millimeters, with a coefficient of variation exceeding 30 %.

Potential sites in Puno are two peasant communities, Santa María in Ilave, and the community of Apopata in Santa Rosa de Juli. The first is a crop livestock community near Titicaca Lake at an altitude between 3,827 and 3,931 masl. The second community, mostly of alpaca producers is located between 4,070 and 5,300 masl. These communities participated in the PISA project, a research program on agricultural systems funded by IDRC. CIRNMA our collaborating institution in Puno grew from this research program. Historical and in-depth information is available for sample selection (Arguelles and Estrada, 1991).

Household strategies in the Andean region

Framework for analysis

A household production systems and economic portfolio (Valdivia et al. 1996; Valdivia and Jette, 1995 and 1997). Data collection will be undertaken on a sample of households at selected sites in Bolivia and Peru. Criteria for selection include belonging to an area where climatic perturbations occur, where both crops and livestock are produced and are sensitive to climatic perturbations. Variability is required with respect to market linkages and transactions costs, and communal relations, and agroecological characteristics. We will build on the household approach for cluster analysis designed in previous research on economic strategies in Bolivia (Valdivia and Jetté, 1996).

Data collection

Two stages are part of the data collection process: In order to identify networks, climate and weather perceptions in production decisions and their timing related to each of the agricultural activities, and to measure transactions costs, farmer group meetings will take place in each identified community. Following these participatory rural appraisals, sections on these topics will be added to the original structured questionnaire about: crop variety information, spatial diversification and staggered planting by crop, traditional sources of information for production decisions, networks, costs of accessing information. Data will be collected from a random sample with replacement (Deaton, 1995), as was done for the original two samples in Bolivia. We will interview the same households in San José Llanga to evaluate the changing strategies and the impact of climatic events in changing the portfolio composition.

The existence of detailed production data from two previous years in Bolivia, and baseline data from Peru will permit us to gather more detailed information on the use of climatic data than would otherwise be possible. The data gathered in previous research, combined with the data gathered with the research proposed here provides a rich data set to study farmer decisions throughout this decade.

Strategies

Cluster analysis (Valdivia and Jetté, 1996; Aldenderfer and Blashfield, 1984) using the variables identified in previous analysis (Valdivia and Jetté, 1996), will be performed to identify current strategies and the consistency comparing the results to previous clusters (Do households persist in their initial group?). Previous analysis (Valdivia, forthcoming) showed the permanence of households in identified groups, by age and adult equivalent labor units first, and then by access to resources, by their quality, as well as total net income.

Comparisons between years will be performed with Discriminant Analysis (pair wise comparisons) and Canonical Analysis (more than two groups for three points in time) (Pedhazur, 1997; Aldrich and Nelson,1984). Income change between 1993 and 1995 was analyzed through regression analysis (Valdivia, forthcoming). Regression analysis to explain income change from one rainfall and frost year to another will be explained by factors believed to affect income: diversity index, assets, stage in the life cycle, and quality of resources. This analysis will be repeated to compare the effect of climate change with 1998 -1999. Other multivariate techniques will be used to analyze differences between clusters/groups (strategies) on variables hypothesized to affect or be affected by year (climate). Three production years with different rainfall patterns will be analyzed in Bolivia, and two in Peru.

Diversification and the portfolios

Diversification indices developed for production in 1993 and 1995 (Valdivia et al., 1996) will be revised to include climatic and weather risk (drought and frost) based on historical data and the simulation models. This index will include a measure of risk to each economic activity, in order for the index to reflect not evenness in spread of income, but also the amount of risk attached to each portfolio. This will allow us to test Rosenzweig and Binswanger hypothesis. We will measure the degree of covariance between economic activities in each portfolio to determine which ones are more risky (Rosenzweig and Binswanger, 1992; Reardon et al., 1992).

Impact of networks and social capital on risk management strategies

The household surveys will provide initial data on the structure and strength of networks. Questions will be adapted for use in Bolivia from the Acore discussion network items@ on the General Social Survey (Marsden, 1987; National Opinion Research Center, 1991). These items will be used to identify the weak and strong tie networks of farmers and correlated with response to climatic variation. Using cluster analysis, a typology of farmers will be developed and set of farmers will be identified for case studies in the year following the household survey. Measures of social capital and transactions costs will be constructed from the data to evaluate their effect on probability of using information.

Impacts of climatic events on household strategies

Evaluate if the risk reducing strategies without climatic information yield lower income than those with forecasts.

Models to predict outcome in crops and livestock with yields based on climatic forecasts

Experimentation in the real world, although absolutely necessary is in some circumstances an expensive and time consuming process with severe problems in controlling variables exogenous to the experiments (Dent, 1993). In computer modeling, by contrast, experimentation is easy, cheap and speedy, at least once a suitable model has been developed (Pandey and Hardaker, 1995). A word of caution is important at this level; modeling is only an analytical tool and not an end in itself. It is also worth mentioning that models are at best simple representations of reality thus they only should be seen as decision support tools (Quiroz et al., 1998).

A valid model is useful for evaluating scenarios that are difficult to be tested in the real world. This is especially true when assessing the long-term consequences of different management practices or the possible impact of climate change. In such cases, comprehensive models are extremely valuable for formulating and testing hypotheses that otherwise require many years of data.

In this project models of crop growth and livestock models developed and/or adapted to adequately simulate production in the Altiplano will be used to evaluate different production scenarios (see Arce et al., 1994; Quiroz et al., 1994;Tsuji et al., 1994).

Method to assess impact using a household portfolio approach

The set of strategies identified from the cluster analysis for each year will be the baseline used to compare with the net returns obtained with the models under same climatic and weather events. We will calculate the net economic returns of the household economic portfolio, including in-kind production for consumption, in each of the identified strategies, three in the case of Bolivia and two in the case of Peru. Sensitivity analysis will assist us in comparing the farmer strategy, with one that assumes a range of precision in forecast (Mjelde et al. 1996) as an approach to measure the potential impact of accessing climate and weather forecasts. The objective will be to determine which strategies (portfolio of economic activities, which included no-farming) perform better under climatic risk and why. The three years of data from Bolivia will show household strategies under three different rainfall and frost events. It will allow us to compare and identify the groups that are most vulnerable to climatic risk, and will allow us to evaluate the impact of current access to climatic and weather information. We will know from the study of networks, social capital and transactions costs which groups of farmers, identified by the strategies, are able to use information. This evaluation model will be developed through the end of year two, and will measure actual household decisions reflected in their portfolio of economic activities. A similar model will be constructed for the Peru site with the data gathered in year two. We expect the conditions that facilitate access to weather and climate information to differ because of distance to market, different crop and livestock mix, local conditions, government policies, and markets.

 Continue to Statement of Work - Plan of Work...