Income and its variability in a drought-prone region: seasonality, location and household characteristics

Abstract

This paper investigates three questions in a poor and drought-prone region of western Orissa, India. Is aggregate income stable? How do a household's characteristics and its local environment influence both the level of its income and its variability? Does a short-lived, common shock cause significant income variability? The study is based on an original, three-season panel data set of 240 households, in which one monsoon season was marked by a severe drought. Aggregate income varied little, but there were considerable fluctuations in individual household incomes across seasons, the main sources of which were idiosyncratic shocks. This suggests that although current programmes are effective in stabilizing aggregate income against drought, massive additional intervention along the same lines is not an efficient way to combat a drought's effects. Rather, it is desirable to promote suitable insurance arrangements to deal with idiosyncratic shocks, measures which would complement those designed to raise incomes permanently.

Cet article s'intéresse à trois questions dans la région pauvre et sujette aux sécheresses d'Orissa occidentale, en Inde. Le revenu global y est-il stable ? Comment les caractéristiques d'un ménage et son environnement local influencent-ils le niveau de son revenu et de sa variabilité ? Est-ce qu'un choc commun de courte durée peut être la cause d'une variabilité significative des revenus ? L'étude est basée sur un ensemble de données de panel de trois saisons, composé de 240 ménages, dans lequel une saison de mousson a été marquée par une sécheresse sévère. Le revenu global varie peu mais il y eut des variations considérables dans les revenus des ménages d'une saison à l'autre, dont les raisons furent des chocs idiosyncratiques. On peut en déduire que, malgré l'efficacité des programmes actuels sur le revenu global en cas de sécheresse, une intervention généralisée allant dans le même sens ne serait pas un moyen efficace de combattre les effets de la sécheresse. Il serait plutôt préférable de promouvoir des dispositifs d'assurance adéquats pour faire face aux chocs idiosyncratiques, mesures qui compléteraient celles destinées à accroître les revenus de façon permanente.

Keywords:

• poverty
• natural hazards
• income variability
• India

Acknowledgements

The author thanks Clive Bell and Annegret Steinmetz for extensive discussions and written comments. Valuable comments were also made by participants in seminars at the World Bank, the University of East Anglia and Heidelberg University, and by an anonymous referee of this journal. Financial support by the German Research Foundation (DFG) is gratefully acknowledged. All responsibility for the opinions expressed and the errors that remain is the author's alone.

Notes

 1. The DPAP was the first area development programme launched by the Government of India (in 1973–74) to tackle the specific problems of the country's extensive drought-prone regions. A whole collection of such schemes and programmes is now in operation, such as the Desert Development Programme (DDP) or the Integrated Wasteland Development Programme (IWDP). For more details see http://wrmin.nic.in/development/drought.htm (accessed January 2004).

 2. The Indian crop year runs from July, when the southwest monsoon sets in, to the end of the following June, and is further divided into the kharif and rabi seasons. In western Orissa, about 80% of annual rainfall is received in the months of July, August, September and October. The dry rabi season, which begins in January, is relieved by occasional showers in winter and the weeks leading up to the outbreak of the monsoon.

 3. According to the NSSO 55th round, no fewer than 87% of the rural population in the southern region of the state are below the poverty line (Glinskaya 2003). Glinskaya argues that this estimate is almost certainly too high: the NSSO does not consider differences in relative prices for cereals across regions, nor does its consumption module adapt well to the peculiarities of consumption patterns in the tribal-dominated southern region (see also section 4).

 4. Source: Taluk-Office, Titlagarh (Bolangir) and Kesinga (Kalahandi).

 5. Where the individual characteristics are concerned, the population of villages in the area can be thought of as an archipelago of islands, from which we have drawn a random sample. As members of the archipelago, the islands share some common features, but they also vary among themselves, and it is the influence of these variations that we investigate. Viewed in this way, it is clear that the role of spatial factors in this study has nothing to do with what is called the phenomenon of spatial autocorrelation in the econometrics literature (for an extensive treatment, see Anselin and Bera 1998). This interpretation is strongly supported by the fact that the plotted residuals of the spatial variables are randomly distributed.

 6. Source: Taluk-Office, Titlagarh (Bolangir) and Kesinga (Kalahandi).

 7. With only three periods and a modest sample size, it is reasonable to keep the model fairly simple, and we assume additive separability for the error term.

 8. The distributions are clearly non-normal, but since the fraction of landless households is virtually the same for all three groups, the t-test should be robust enough.

 9. The cumulative distributions, which are available from the author on request, are tightly bunched.

10. They are substantially lower than the HCR of 87% calculated by Glinskaya (2003) on that basis for the southern region of Orissa in 1999–2000. Using Deaton's (2003) price-adjusted poverty line (Rs.300 per month per head for rural Orissa), the number of poor households falls further, by almost 6 percentage points in each season.

11. Not a single regressor involving the seasonal shift dummies is significant at conventional levels, and the hypothesis that all such coefficients are jointly zero cannot be rejected: the value of F (38, 663) is 0.95, with an associated p-value of 0.554. The same holds for the hypothesis that the constant term does not vary over time: the value of F (2, 663) is 0.75, with an associated p-value of 0.473. Over the three seasons in question, therefore, this specification yields the finding that where household income is concerned, the term representing the time-varying constant did not in fact vary, and neither did the values of the coefficients of the included regressors, which are listed in Table 5.

12. The ascriptive caste dummies and the variables that characterize the village (irrigated and forested area, population density, transport links and marketing facilities) do not change over the span of three seasons. Also dropped is the value of farm equipment, which happens not to vary over the period in question.

13. For a detailed study of this connection in southern Tamil Nadu, see van Dillen (2004).

14. Scheduled caste households earned, on average, Rs.985 more than scheduled tribe households in kharif 2000, but only Rs.82 more in kharif 2001.

15. For the calculation of livestock income, see Appendix 2.

16. It is possible that this finding is attributable to the rather strong correlation between landholding and draught-power – at 0.50, the said correlation coefficient is the largest in the correlation matrix.

17. In addition to those employed in estimating eq. (1), one could also introduce village dummies to represent all the rest, measurable or otherwise. This was tried, but none of the said dummies was at all significant, which suggests that the characteristics in Table 5 capture the local environment rather well.

18. Note that this variable does not include forest land outside the immediate area of each village, as defined in the land registry maps. The inclusion of forest areas bordering on the study villages, which are often jointly used by the inhabitants of the surrounding villages, might actually ‘improve’ the regressor.

19. The severity of pest attacks is also connected with climatic conditions. Precipitation was abundant in kharif 2001, and while paddy output was high, cotton failed almost entirely throughout the region. Interviews revealed that cotton is particularly susceptible to pest under wet conditions, that is, often in years with high paddy yield. Farmers therefore choose a portfolio of paddy and cotton, if the size and quality of their landholding supports this strategy.