Approach to WP1 in Niger

This is a long blog but we thought a synopsis of the Niger WP1 would facilitate some discussion. This also tackles some of the 5 items proposed by Eric. NB equitable access in the Niger was partially addressed by an analysis of ethnicity related poverty and we are hopeful the institutional research of the Niger Program will add to this. Overall the data was not longitudinal and hence temporal change was not analysed: accounting for change we see as a vital iteration. The fundamental research questions were to identify and spatially reference the three dimensions of poverty in the Niger Basin, identify where these were acute and with best available evidence and reliably estimate the relative importance of water related factors in explaining poverty. If water access or productivity was not a significant factor, we then sought to determine the significant non-water factors.

The four main questions we addressed were:

1) how to measure and spatially reference water related poverty?

2) Is scale important and if so what is an appropriate scale that captures the heterogeneity of poverty, identifies areas of critical need and the diversity of explanatory variables?

3) what is the appropriate, policy relevant scale that reveals opportunities for policy intervention and best targets the causal factors of water poverty?

4) how to address the subjective weighting of vulnerability or water poverty indices when expert opinion or participatory process are unreliable or not feasible?

The first question we tackled was how to measure poverty in the Niger Basin: ie metrics which account for a high proportion of subsistence livelihoods coupled with proportion of a non-market, hybrid economy. Hence many of the traditional monetised indicators may not be sufficiently precise to capture the magnitude of poverty and the geographic heterogeneity. Importantly, as an initial research assumption, we did not differentiate “water” poverty from any other type of poverty.

We selected the rate of child mortality (up to the age of 5), child morbidity (deviation of the ratio of height for age compared to a healthy median) and a composite wealth/asset index.

Child mortality is likely to be a function of a household’s ability to obtain essential services, nutrition and shelter. Secondly, it is, to some extent, more unidirectional than other measures: child mortality is caused by poverty, but does not in itself cause poverty to the same extent as do other socio-economic factors. Finally, child mortality provides a relatively direct method of quantifying water poverty, because poor water quality, caused by limited availability, limited access and poor infrastructure, is the direct cause of some of the most prevalent, fatal childhood diseases.

Child morbidity: There are many measures of child morbidity, however we utilise the ratio of height for age (stunting, measured as average height for age ratio (standard deviations below healthy reference median). This indicates long term, cumulative effects of inadequate nutrition and poor health, including that before birth. Height for age is thus relatively insensitive to short term seasonal variation in calorie intake, making it more appropriate when comparing data collected at different times.

Wealth Index: A wealth index, such as the DHS Wealth index was used and demonstrates the material standing of households. Wealth represents long term access to consumer and productive goods, indicating simultaneously the level of poverty and the capacity to earn a livelihood.

Metric selection was partly driven by literature based insights, a need to evaluate metrics that are a surrogate for the multi factorial nature of water related poverty and availability of reliable primary and ground-truthed data. The DHS data provided this. Hence all analysis quantified the effect of explanatory variables according to the 3 dimensions of poverty described below. We introduced and emphasised water access, water quality and water productivity variables, but did not introduce any a priori assumptions as to their relative importance in explaining poverty. We introduced many of the structural or non water variables that Jorge mentions to ensure a comprehensive vector of explanatory variables.

The 22 explanatory variables can be classified as:

Community vulnerability: the capabilities, assets and activities of the community. In the poverty model constructed here community vulnerability is represented by health variables, socio-economic variables, infrastructure and assets.

Situational vulnerability: how well these assets, capabilities and activities can withstand exposure to shock and stress. As this study focuses on water poverty in particular, situational vulnerability is represented here by the water variables.

Hazard threat: the events that challenge the community/situation. These are represented by natural disaster risk, climatic variables, population density and environmental damage.

Institutions, conflict and corruption can potentially effect all of the above vulnerabilities and threats.

Next we addressed the issue of appropriate scale of measurement. The common measures of water stress (per capita) measured at the national scale indicate that most of the Niger basin countries are not suffering water stress now or projected to 2025. Burkina Faso was the exception. Secondly the water stress or water poverty indices did not correlate with traditional national measures of poverty (HDI, social vulnerability index or genuine savings indicator). We hypothesised this was a matter of scale: i.e. national scale statistics are not sufficiently precise by compressing valuable poverty information. Hence we analysed poverty (by the 3 metrics) at basin scale, national scale and at the smallest administrative unit for each country that remained policy relevant where we had reliable data. There were 631 data points for the Niger countries. Data was interpolated and validated for those units with insufficient or low sampling points. Data was collected from a number of sources (available if required) and filtered to ensure GIS compatibility. We constructed GIS layers for analysis and to facilitate interpretation. The Null hypothesis was:

H₀ :Water Poverty Basin = Water poverty national = water poverty hot spot

Water poverty represents a vector of the coefficients of the 22 explanatory variables of child mortality, child morbidity or the wealth index.

Next we deconstructed the construct of vulnerability or the equivalent of the water poverty index. Our primary motivation was to estimate evidence based coefficients of the factors of vulnerability rather than the subjective or expert determined weights commonly applied. That is we decomposed vulnerability or composite water index and, post analysis, reconstructed them with evidence based values.

Analysis utilised two spatial statistical techniques: geographic weighted regression (GWR) and a spatial lag model. We can further discuss if required. Essentially failure to account for spatial auto correlation biases traditional regression OLS models and reduces the reliability of results. In all cases spatial correlation was significant. In the spatial lag model, specific and validated spatial weightings were calculated for each poverty metric applied to each spatial scale. There was no one weighting that was valid across all spatial scales, metrics and regions. We selected from 1^st to 4^th nearest neighbour as the geographic weighting matrix. Only non-correlated Independent variables (variance inflation factor < style="">i, using data from neighbouring points. The extent to which proximate administrative units contributed to a particular local analysis ws dependent on the distance from i.

We were able to explain water related poverty for the basin with GWR at R² of approx 0.89, with coefficients specific for each admin district. We reconstructed “water vulnerability” at the admin unit scale, revealing poverty heterogeneity across the Basin and the explanatory variables specific to each unit. These varied significantly. All results have been generated in map form to aid interpretation. We view this as an evidence based visual policy tool for poverty reduction in Basin sub regions. For example the question may be: with a given aid budget, how would the funds be prioritised and best distributed amongst a bundle of investment areas (education, water access, livestock densities, water quality, irrigation intensity, human footprint). With the analysis we can reliably estimate that for example a prioritised investment in education that resulted in a 1 year increase in schooling results in a 0.32 % reduction in child mortality for a specific sub region. In other areas prioritising water access has the greatest effect on reducing child morbidity or mortality.

We also used a spatial lag model that accounts for significant spatial auto correlation at the basin, national and hot spot scale. Hot spots were established as those admin units with high poverty and high spatial correlation, eg central Mali, Northern Nigeria and Eastern Burkina Faso. We rejected the null hypothesis and found that a different vector of explanatory variables were estimated for basin, national and hot spot analyses. This allows for a more and tailored response through policy intervention or priotised investment that targets high poverty areas.

There is some contradiction in the results when comparing the 3 poverty dimensions. Pragmatically we concentrated on those areas where the results were consistent for all 3 metrics or at least 2 of the 3. For GWR analysis we consider mortality and morbidity only as the wealth index cannot be compared internationally with confidence.

Below are some graphics and example applications we anticipate can be used as a participatory tool for future poverty analysis in the Niger Basin. These predictions are based on the GWR results.

A reduction in the proportion of people using surface water or unprotected well water represents a likely improvement in the quality of water used. Our model suggests that this approach could substantially alleviate child mortality in north west and central Nigeria, although other areas are affected only very slightly.

Education appears to be the most universally effective means of reducing poverty, with improvements in health predicted for much of Mali, north west Nigeria, east Burkina Faso and central Nigeria. In particular, the poverty hotspot around the Inner Niger Delta appears substantially alleviated by an additional 4 years in average education.

Also considered in this manner was both the predicted impacts from an increase in irrigation intensity and a decrease in the average distance to a dam. Neither approach led to a prediction of substantial reductions in child mortality and child morbidity (stunting). We haven’t shown these graphics because 1) they show little noticeable reduction in poverty and 2) to keep this file size to a minimum.

We would be happy to chat with anyone interested in our approach or anyone who has suggestions to make. Cheers, John Ward.

>