5    Results

Table 2 shows the descriptive statistics of the explanatory variables in the regression models. Table 3 reports results of the three Gibbs-sampling, data augmentation algorisms applied to the participation and sales data. The first column of Table 3 reports definitions of the covariates used in the various regression models and the second to fifth columns report, respectively, results from the single-equation probit model, results from the single-equation Tobit specification; and results from the two-equations probit and results from the two-equations Tobit specification. The dependent variable in the Tobit equations is livestock sales revenues in Philippine pesos. Livestock sales refer to sales of live animals only. The covariates are ordered in blocks corresponding, respectively, to the categories transactions costs, mobility, intellectual capital, financial capital, physical capital, credit and other excluded covariates. The focus on these groups is motivated as follows.

Table 2. Descriptive statistics of the explanatory variables used in the model.

Table 3. Estimation of market participation and sales of livestock in Don Montano, Umingan, Pangasinan, The Philippines.

First, we consider that transactions costs are likely to play a major role impeding entry by subsistence households into emerging markets. The problems of ensuring adequate demand, locating and negotiating a sale and transporting goods to market are anticipated to feature prominently in the household decision-making process. For these reasons and in the absence of precise information concerning the likely ranges of these costs, we use two proxies-return-time distance to market 'distance' and household labour availability or number of household members 'members'—as principal determinants of these costs. We assume that transactions costs increase with greater distance to market but may be reduced with increased labour abundance and the lower opportunity costs that this may imply. Consequently, for both participation and selling decisions we presume that the impacts of these two covariates are, respectively, negative and positive.

Second, because the transition to the new occupation in local markets requires freeing-up other resources, we desire some measure of the extent to which households in question may be more or less mobile than others. This degree is represented by the variable 'otheremp' which, in turn, measures the number of years that the household head devoted to non-farm employment activities in his/her current and previous occupations. In the absence of more precise information concerning previous employment prospects and the household's propensity to change occupations in response to these incentives, we use the covariate 'otheremp' as a proxy for mobility. The larger the value of 'otheremp', the more likely the household is to participate in markets. We do not have strong a priori expectations about the effect of 'otheremp' on sales.

Third, we assume that the level of intellectual capital stock in the household is positively related to the participation decision. However, this stock level may be related in a contradictory fashion when other employment activities are available, particularly when those employment opportunities require skill. In this way, a greater degree of intellectual capital-measured in terms of the number of years of formal schooling by both the household head and the spouse 'education', the number of years of farming experience 'farmex', farm experience by other household members 'otherex' and exposure to extension agents last year ('extension' = 1 if the household had contact with an extension agent, = 0 otherwise) may each exert a positive impact on the participation and selling decisions; although the precise impact of the non-farm specialist covariates ('education' in particular) are complicated by their opportunity costs in alternative enterprises. For this reason, unlike the farm-specific variables 'experience' and 'extension', we do not have strong prior beliefs about the likely sign of the coefficient of 'education'.

Fourth, we include measures of income⁵ derived from both farm and non-farm sources. The definitions of the variables are, respectively, income (in 100 thousand pesos (US$ 1 = Pesos 50) from sources other than farming 'otherinc', income from remittances 'remitinc', income from other household members 'memberinc' and income from crops 'cropinc'. Where the income relates to livestock enterprises we consider that this has a positive impact on participation and selling, but where the income relates to other farm activities and to other non-farm activities we consider that the impact will be negative. In the case of income earned by other household members, we assume that this diversification may lead to risk reduction in household decision-making and, with it, a likely increased propensity to undertake higher-risk activities, notably selling livestock. While this phenomenon may also explain a likelihood that returns from crop income may be positive, this sign is compounded by the fact that increased revenues from crop production may signal incentives to re-allocate away from livestock production and selling activities.

5. A Hausman test was performed to test for endogeneity across the four income variables, namely, otherinc, remitinc, memberinc and cropinc. The test results indicated that these are endogenous with respect to the variables age of the household head, age of the spouse, value of assets, farm size, other livestock numbers, and gender of the household head (Chi-squared = 4.1459, 4 d.f.). However, these covariates are not included in the set of covariates used in the estimation of the model. Hence, the issue of endogeneity need not be a concern in this case.

A more distinct, less diffuse set of prior beliefs are maintained with respect to the physical capital variables representing numbers of relevant livestock on the farm—cattle, buffalo, goat, chicken and pig. Each is expected to exert a positive impact on both the likelihood that participation will occur and the amount of selling that will be undertaken once the decision to participate has been made.

In the remaining category of debt, we expect the covariate 'credit' (representing the amount of indebtedness in 100 thousand pesos) to have ambiguous impacts on the participation and selling decisions. This is because debt can be interpreted in two ways. The first way pertains to the fact that increased debt in other activities may lead to lack of free collateral to secure loans for market selling activities. In this case, the sign of the coefficient of 'credit' is expected to be negative. In the second case, existing debt may be the result of previous borrowing that has occurred for the production and selling decisions and may, therefore signal greater propensity to sell. In this case, we expect the coefficient of 'credit' to be positive.

Finally, we have no reason to expect the impacts of other excluded factors 'constant' to be positive or negative.

Regarding the results reported in Table 3, three observations are apparent and important. First, with the exceptions of a few covariates, the participation and selling decisions are mostly affected by the same factors. Second, responses across the single-equation probit and Tobit specifications are slightly improved by moving to the multivariate specification in the sense that most of the marginal significance levels of each of the covariates is increased. Third, there is very strong evidence that the errors in the two equations are positively correlated. This observation is important because it suggests that it is most appropriate to consider the participation and selling decisions simultaneously. Hence, policy recommendations concerning market access, provision of infrastructure and estimates of minimum resource levels required to effect entry should be based on the two-equation formulation.

Regarding the impacts of the various covariates on participation, number of household members 'members', education levels of the head and spouse 'education', visits by extension agent 'extension', and the livestock assets—cattle, buffalo, pigs and chicken—are each highly significant. Each of the coefficient estimates of these factors has marginal significance levels in excess of 5% (that is, the 95% Bayesian highest posterior density regions corresponding to these coefficients do not contain zero). Propensity to participate declines with number of household members, education, and income from cropping, but increases with respect to increases in each of the other covariates. Distance to market 'distance', mobility 'otheremp', experience in farming by the household head, and other family members 'farmex' and 'otherex', respectively, indebtedness 'credit', and goat livestock numbers 'goat' are not significant determinants of participation. There is marginal significance of some of the income variables. Income from other sources 'otherinc', income from remittances 'remitinc' and income from other household members 'memberinc' have a positive influence on the participation decision, whereas increased crop revenues 'cropinc' lowers the likelihood of livestock-market participation. These results conform to prior expectations that income diversity lowers risk and, with it, the likelihood that (potentially risky) market development will occur, and that improved alternative production and marketing opportunities in other farm enterprises such as cropping may weaken participation incentives. However, perhaps the most interesting result is the strong negative impact that education exerts on the participation decision. When alternative employment opportunities exist, an increase in skills that resulted by increased education lowers the likelihood that market participation arises.

This last observation is confirmed somewhat by examining the impacts of education on sales in the single-equation Tobit regression (column three of Table 3). The effect of increased education on sales is insignificant, suggesting that the influence of education is more important in market entry than in market supply. Among the remaining covariates, all have the same signs of effects as those in the probit equation. However, the marginal significance levels are significantly lower than in the probit specification. This difference is most dramatic with respect to the extension covariate. Whereas in the probit equation this covariate was highly significant; in the Tobit equation it is not. Hence, extension activities appear to play an important role, but only in the set-up of market operations. Finally, most of the remaining covariates share significant levels that are similar to those in the probit model and, once again, indebtedness, farm experience and mobility do not appear to be significant factors explaining supply decisions.

The estimates pertaining to the two-equation model (columns four and five of Table 3) confirm two conjectures. The first conjecture is that the participation and sales decisions are strongly positively correlated. This fact is supported by the cross-equation error covariance reported in the lower part of Table 2. The implied t-statistic (4.54) corresponding to the covariance estimate (4249.24) suggests that the covariance is not close to zero. In other words, the probit and Tobit equations estimates are not independent. The second conjecture is that the inclusion of covariance information in the two-equation system significantly affects inferences about the participation and supply decisions; in this case, making considerably more precise derived statistical inferences.