Volume 5, Number 1
July 1999

Economic Roles and Statistical Procedures

Michael Warner (M.Warner@wye.ac.uk), research associate in the department of economics at Wye College in England, initiated a discussion based on an abstract of his paper titled "Who Does What in Rural Africa? Economic Roles and Statistical Procedures." The paper examines a range of statistical procedures available for analyzing categorical data related to the basic economic roles of rural peoples in Sub-Saharan Africa. By setting out to answer the question "who does what in rural Africa?," Warner attempts to sidestep recent debates about the relative importance of gender and other social constructs in determining the economic roles rural people play. Using data from northern Ghana, the paper begins by exploring the usefulness of basic statistical procedures, such as frequencies, cross-tabulation, and measures of association, in answering the question of who does what. The paper then examines other, more computer-intensive procedures, such as loglinear modeling and logistic regression analysis.

The analysis shows that the existing statistical toolkit is more than capable of describing the economic roles of rural peoples in Africa. Loglinear models, in particular, are found to be highly useful in describing the complex interplay between involvement in production activities, asset ownership, and contributions to the household by rural women and men. Warner suggests that these models may have real value in assisting in the accurate targeting of production and consumption initiatives.

Given advances in affordable computing power, he concludes that a statistically-based approach is now relatively cheap, rapid, and easy to undertake. Finally, Warner suggests that the implications of adopting such an approach, in terms of data collection costs, are modest, that a survey-based statistical approach is entirely complementary with other research methodologies, and, therefore, that such an approach holds great potential for making useful information available to policymakers.

Agnes Quisumbing responded that she thoroughly supports the use of survey-based quantitative approaches to examine how economic (and other) roles are assigned to individuals. When the design of surveys is informed by the anthropological and ethnographic literature, it is possible to come up with quantifiable measures of individuals' propensity to undertake particular activities, and to analyze, as Warner does, the contributions of a variety of factors, such as gender and marital status. Although the paper is directed towards a nonstatistical audience, Quisumbing felt that it might be useful to make a number of clarifications that would guide the choice of technique as well as provide a better understanding of the broader literature on time allocation. (References are included below).

1. There is actually quite a lot out there on the determinants of time spent in various activities, or the probability that an individual will undertake a particular activity. It is true that there are relatively few studies on Sub-Saharan Africa, but Warner’s study is certainly not the first. Examples are Appleton et al. (1991), who examine, among other things, determinants of attending school or fetching water; Ram and Singh (1988) and Singh and Morey (1987), who look at the value of wives' time in Burkina Faso; and Jacoby (1993) on polygyny in Côte d'Ivoire. Although the latter two use a lot more econometrics (and impute an implicit value to the time wives spend on the farm or in home production), they ask essentially the same question: who does what?
2. Warner also compares frequencies between groups to see who is more likely to do what. Such comparisons could be made more robust with a t-test of differences in means or differences in proportions.
3. Stepwise regressions can pose problems because it is easy to put a whole lot of variables without any underlying theory of what is going on. It might be useful, as Kevane and others have done, to use the anthropological evidence to guide one in the inclusion of right-hand-side variables.
4. Multivariate analysis might also be a more general approach than loglinear modeling. If you have data only on dichotomous (yes or no) observations you could use a probit or a logit. If you have data on actual time spent, there are a variety of available multivariate techniques.
5. Finally, such statistical programs are not the monopoly of SPSS. SAS and Stata are easy-to-use packages, and Stata is particularly easy to use because it has a good online help facility. Bola Akanji commented that she found Warner's abstract extremely interesting as she has been recently researching the patterns of gender responsibility in agriculture in different farming systems in Nigeria. Her study does not focus, however, on methodologies for analyzing roles, but assesses the effects of economic reforms on changing gender roles, especially in tradable crop production. Akanji said she would be keen to see the paper to find out how her findings fit with what Warner’s models imply for similar systems in rural Nigeria.

References:

Appleton, S., D. L. Bevan, K. Burger, P. Collier, J. Willem Gunning, L. Haddad, and J. Hoddinott. 1991. Public Services and Household Allocation in Africa: Does Gender Matter? Oxford: Oxford University, Center for African Studies.

Jacoby, H. 1993. “The Economics of Polygyny in Sub-Saharan Africa: Female Productivity and the Demand for Wives in Côte d'Ivoire.” Department of Economics, University of Rochester, Rochester, N.Y., U.S.A. (This has been published in the Journal of Political Economy.)

Ram, R., and R. Singh. 1988. “Farm Households in Rural Burkina Faso: Some Evidence on Allocation and Direct Returns to Schooling, and Male-female Labor Productivity Differentials.” World Development 16 (3): 419-424.