Poverty Dynamics in Rural Orissa: Transitions in Assets and Occupations over Generations

Abstract

We investigate whether historic land distribution determines stagnation or development of Indian villages. The empirical analysis is motivated by the Banerjee and Newman (1993) model of occupational choice and economic development. Family histories are collected for a random sample of 800 households. Households are classified into economic categories according to the assets–occupations mix at present and at grandfather’s time. Transitions are described, and for a remote district explained, by the historic village land distribution. We also investigate the role of social identity, and find that scheduled tribes are more likely trapped in poverty than scheduled castes.

Acknowledgement

Data was collected by RAC, Bhubaneswar, Orissa. The research is part of a larger collaboration with Basanta Pradhan, IEG-Delhi, that is sponsored by the Norwegian Embassy in Delhi. The paper was completed during a visit to the Dyson School of Applied Economics and Management at Cornell University.

Notes

1. For a review of village studies, see Jayaraman and Lanjouw (1999).

2. Note that this static part of the model has parallels to, for example, Eswaran and Kotwal (1986).

3. Note that we do not attempt to collect information on all eight grandparents of the couple, only the father and father’s father of the husband. Furthermore, it is natural to focus on the male line, as land is in most cases inherited by the sons and they normally stay in the same village. For a detailed analysis of the lack of female rights to land in India, see Agarwal (1994).

4. We use principal component analysis as described in Filmer and Pritchett (2001). The self-employed are split at the zero value of the first component asset index.

5. Here we do not use the survey commands, as we consider this just a ranking of the observations in the data and not a description of the population.

6. If our transition table had been a standard transition matrix with equal-sized wealth categories and a random sample of households at grandfather’s time, then the 0.7 proportion would be the normalised (as in Dercon and Krishnan [2000]) Shorrocks index.

7. These regressions with a dichotomous dependent variable were included in earlier versions of the article, and are available on request.

8. In reality the main differences turn up for the so-called upper castes, so we only add a dummy for this ‘general’ caste.

9. We have omitted one outlier from the analysis that had more than 400 hectares of initial land. The second largest was less than 100 hectares. Including the single outlier drastically reduces the importance of initial land. The observation is included in calculation of the mean and median, though, as the household may have had an effect on the development of the village. The latter explains the higher mean of the means in Kalahandi.

10. The mean of the means for this variable is also slightly different from the mean, as the proportions are calculated for all households while three households are dropped; one outlier and two because we do not have the necessary information to calculate the initial landholding.