Income and Asset Poverty among Pastoralists in Northern Kenya

ABSTRACT

In this study we use household panel data collected in Marsabit district of Northern Kenya, to analyse the patterns of livelihood sources and poverty among pastoralists in that area. We estimate income poverty using imputed household income relative to the adjusted poverty line and asset poverty using a regression-based asset index and tropical livestock units (TLU) per capita. Our results indicate that keeping livestock is still the pastoralists’ main source of livelihood, although there is a notable trend of increasing livelihood diversification, especially among livestock-poor households. The majority of households (over 70%) are both income and livestock-poor with few having escaped poverty within the five-year study period. Disaggregating income and asset poverty also reveals an increasing trend of both structurally poor and stochastically nonpoor households. The findings show that the TLU-based asset poverty is a more appropriate measure of asset poverty in a pastoral setting.

View correction statement:

Corrigendum

Acknowledgments

We thank two anonymous reviewers for their valuable comments. We thank the German Academic Exchange Service (DAAD) for the PhD scholarship through the Food Security Center (FSC), University of Hohenheim, Germany. Data collection was made possible, in part, by support provided by the generous funding of the UK Department for International Development (DfID), the Australian Department of Foreign Affairs and Trade and the Agriculture and Rural Development Sector of the European Union through DfID, the United States Agency for International Development, the World Bank’s Trust Fund for Environmentally and Socially Sustainable Development, and the CGIAR Research Programs on Climate Change, Agriculture and Food Security and Dryland Systems. The data is freely available on request in the IBLI website at http://ibli.ilri.org/index/. The data synax code can be obtained from the corresponding author.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. The 16 sublocations are Dirib Gombo, Sagante, Dakabaricha, Kargi, Kurkum, Elgathe, Kalacha, Bubisa, Turbi, Ngurunit, Illaut, South Horr, Lontolio, Loyangalani, Logologo and Karare.

2. The TLUs help to quantify the different livestock types in a standardised manner. Under resource-driven grazing conditions, the average feed intake among species is quite similar, about 1.25 times the maintenance requirements (1 for maintenance, and 0.25 for production; that is, growth, reproduction, milk). Metabolic weight is thus considered the best unit for aggregating animals from different species, whether for the total amount of feed consumed, manure produced, or product produced. The standard used for one tropical livestock unit is one cow with a body weight of 250 kg (Heady, 1975), so that 1 TLU = 1 head of cattle, 0.7 of a camel, or 10 sheep or goats.

3. Typical climate conditions over the course of the year include a short (January–February) and long (June–September) dry season and two rainy seasons (March–May, also known as the long rainy season, and October–December, also known as the short rainy season).

4. The official poverty line is also used by Radeny et al. (2012), Suri, Tschirley, Irungu, Gitau, and Kariuki (2009) and Barrett et al. (2006).

5. We also estimate the asset index using a random effects and pooled OLS model. However, the Breusch-Pagan Lagrange multiplier test and the Hausman test both indicate that the random effects model is superior to the pooled OLS model and the fixed effects model superior to the random effects model, respectively.

6. The frequency and amount of the money given to the beneficiary households have changed over time.

7. The few households who engage in crop farming are located primarily in four sublocations: Dakabaricha, Dirib Gombo, Sagante, and South Horr.

8. To establish the factors that could influence income diversification, we compute the inverse of the Herfindahl index (H), which measures the degree of concentration of household income into various income sources. This inverse is defined as , where S is the share of income of source k. Households with more diversified income sources have the largest index, while households with one income source have an index equal to one. We also calculate the correlation coefficients between the index and several household variables: the household size variable is positive and significant, while education is negative and significant, indicating that educated households have fewer income generating activities.

9. Using the dependency ratio instead of the three different age categories still yields a similar negative and significant estimate.

10. To compare the predictive accuracy of the asset index and TLU per capita, we use Theil’s U-statistic, defined as , where is the actual value and is the predicted value from the model. This statistic measures how past asset index or TLU per capita (t-n) predicts the current asset index or TLU poverty (t) for household (i). Theil’s U-statistic uses a forecasting model to predict the accuracy of a given indicator measured on a range from 0 to 1, with lower values reflecting a more accurate prediction (Theil, 1966). We obtain U-values of 0.29 and 0.54 for the asset index and TLU per capita, respectively, which indicates that the asset index is a better predictor of future (asset index) poverty than TLU per capita. Nevertheless, one must take into account that Theil’s U statistic for the asset index is based on imputed values, whereas the corresponding value in the TLU case is based on actual values. As imputed values tend to have a lower variation, it comes as no surprise that Theil’s U statistic for such measures are larger than those based on actual observations.

Additional information

Funding

This work was supported by the DFID [202619-101];United States Agency for International Development [EDH-A-00-06-0003-00];World Bank Group [7156906].