Estimates and Interpretation Acknowledgements

5. Conclusions

This paper is motivated by the increasing attention that has been given by demographic analysts and family planning program supporters to the possible roles of social interactions in the diffusion of knowledge, attitudes and behaviors related to family planning. The few previous empirical studies of this topic suggest that if social interactions are important, their omission from empirical models is likely to result in distorted estimates of the direct effects of family planning programs. Our analysis adds to this literature by showing that there are some important implications of using nonlinear models for measuring program effects and for evaluating the roles of social interaction that have not been explicitly considered in the previous empirical literature. We demonstrate some of these properties formally, and investigate them empirically using data that includes measures of social interactions. We find that for Nyanza, Kenya, the nonlinear versus the linear specifications indeed lead to different substantive results with different implications for demographic analysts and program supporters.

First, we distinguish between the direct effects of a family planning program on an individual's probability of using family planning and the indirect effects due to social interaction. Our empirical estimates show that the nonlinear model of the relations among program effects, social interaction and of modern family planning leads to some fairly large differences in the estimates of program effects from those obtained with the linear model - e.g., with estimated direct program effects on the ever use of family planning from 20% lower to 27% higher for the linear than the nonlinear model. We then show empirically that in our data as much as 43% of total program effects are due to social interaction [Note 9]. This social multiplier effect, as it has been named by Montgomery and Casterline [1993], is due to a feedback loop that occurs because social interaction renders the family decisions of community members interdependent. Because of this social multiplier effect, attributing all of the total change in contraceptive behavior to a direct impact of changes in program effort is a substantial overestimate of the direct program effect.

In addition, the linear specification assumes that the total program effect and the social multiplier are identical across subpopulations with different levels of contraceptive use. This need not be the case in the nonlinear model, and our results show important differences in these effects between the women with low and high education: social interaction leads to substantially larger multiplier effects in the high education subpopulation with a higher overall propensity of using family planning.

Second, we show formally that if the model is nonlinear, there may be both a low-level Malthusian equilibrium in which contraceptive use remains relatively low despite ongoing program efforts as well as an equilibrium where contraceptive use is high [Note 10]. If a population is at a low-contraceptive-use and high-fertility equilibrium - a situation that may characterize much of sub-Saharan Africa, including places with family planning programs - small program changes have relatively small effects. However, large increases in program efforts - even if transitory - may cause a shift to a high-contraceptive-use and low-fertility equilibrium. In a linear model, in contrast, large program efforts can lead to high contraceptive use, but the program efforts must be maintained at high levels to sustain high contraceptive use. Our empirical analysis does not indicate the presence of multiple equilibria in our data. Thus, these estimates suggest that there is little likelihood that a sharp transitory increase in program activities in Nyanza would lead to a rapid shift to much higher sustained levels of contraceptive use. But such possibilities may exist in other contexts.

Third, we show formally that intensified social interactions may either increase or decrease the total effect and social multiplier effect resulting from family planning program efforts, and `more' social interaction can thus reinforce or retard the diffusion of an innovation. When a nonlinear (logistic) model is used, increasing the impact of social interactions is status quo reinforcing close to a stable equilibria (whether at low or high contraceptive use) in a multiple-equilibria situation. Therefore, if a new program effort were to intensify social interactions near the stable equilibria, the total-or long-term-change in contraceptive use resulting from the program effort is reduced and these more intensive social interactions would retard the diffusion of family planning after the program interventions. Our nonlinear empirical estimates for S. Nyanza District imply that when social interactions are intensified, they reduce the total effect associated with program interventions, but slightly increase the social multiplier effect. These findings are in contrast to the linear estimates that imply that more intense social interaction leads to a larger social multiplier effect and an increased total effect after the program interventions.

Thus, we show formally that there are some important implications of nonlinear models of social interactions that have not been emphasized in the previous literature and how they contrast with the implications of linear models, and we show empirically that in the Nyanza case there are some substantial differences in estimating program effects required for a sustainable fertility transition between the nonlinear and the linear specifications The value of having the right model may be considerable and the implications of nonlinear models in this context need to be understood to interpret fully their results.

 

Estimates and Interpretation Acknowledgements

Empirical Assessments of Social Networks, Fertility and Family Planning Programs: Nonlinearities and their Implications
Hans-Peter Kohler, Jere R. Behrman, Susan Cotts Watkins
© 2000 Max-Planck-Gesellschaft ISSN 1435-9871
http://www.demographic-research.org/Volumes/Vol3/7