CHAPTER SIX
CONCLUSIONS AND RECOMMENDATIONS
succeed in including and eliminating some factors that the other related models would not;
especially the independent variables with marginal significance values. With a carefully selected sample and meticulously supervised data collection, to eliminate or reduce missing cases to a bare minimum, it is guaranteed that the dataset will yield more accurate parameter estimates that can be used in building a model for predicting the risk of food insecurity.
6.1.2 At least eleven factors influenced food insecurity in Southern Sudan
The study in more than one set of analyses determined at least eleven factors as important predictors of food consumption; hence food insecurity in South Sudan (at least during 2006 or immediately after the 21-year civil war ceased). The factors to be included in the model are shown in Appendix 5. Statistical modelling techniques used differed in determination of predictors. The variables and the techniques used for selection are displayed in Table 6.1.
Table 6.1 Variable selection by statistical modelling techniques
Variable Cauchit Ordinal
Regression Model
Complementary Log-Log Model
Linear Regression Model a
1 State Yes Yes Yes
2 Number of household members No Yes Yes
3 Months of harvest food lasting Yes Yes Yes
4 Number of meals per day Yes Yes Yes
5 Farmland ownership Yesb Yes Yes
6 Farmland use Yes Yes Yes
7 Farmland planting Yes Yes No
8 Number of harvests in one year No Yes No
9 Livestock ownership Yes Yes Yes
10 Migration/movement of household Yes No Yes
11 Ownership of home garden Yes Yes Yes
12 Source of livelihoods Yes Yes Yes
13 Source of sorghum and millet Yes Yes No
14 Wealth Index Quintile No Yes Yes
a Used when the food consumption score is a continuous variable; b marginally non-significant
Although the model with Cauchit link function is recommended because of presence of extreme values of the food consumption scores, the Complementary Log-Log model determines more variables (13) as important predictors. The backward elimination model for fitting a continuous response variable does fairly well too in removing the non-significant (non-influential) variables.
6.1.3 At least eight factors could be used for food insecurity surveillance
By means of Table 6.1 above, it can be concluded that the three techniques are in agreement in determining eight independent variables as important in predicting the outcome of food consumption: state; number of months in which food harvest lasted; number of meals eaten per day; ownership of farmland; use of farmland; livestock ownership; home garden ownership and sources of livelihoods. With the post-conflict situation of Southern Sudan, some of these factors influenced food consumption negatively. Whereas ownership of certain production assets is supposed to improve the probability of being in better food consumption group, the relationship was quite the opposite. However, this finding provides basis for further investigation.
The eight variables could be included in food insecurity surveillance and routine monitoring exercises applying rapid data collection approaches.
6.1.4 Easily replicable methodology
Quite often agricultural economists tend to ask for quantitative data such as number of acres planted, quantity of food harvested, crop yields, quantity of food purchased, quantity of harvest food sold, quantity received from food aid, income earned from sale of crops and amount spent on buying food items. With rampant illiteracy, recall bias and the amount of enumerator training required, attempts to collect quantitative data synonymous to driving the research toward failure.
The study shows that even with the heavy and rampant missing data involving quantitative variables, it was possible to fit a model that determined the variables of importance. Indeed such quantitative variable as number of meals eaten per day, household size and number of months of food lasting, are not only easy to recall but they are also important parameters for predicting the outcome of food consumption and household food insecurity. It is, therefore, shown that the
methodology is replicable and adaptable for routine surveillance of food insecurity among highly illiterate populations.
6.1.5 Peculiar findings
The study uncovered some peculiar findings which could be typical of Southern Sudan. Of prominent interest is the finding that food aid was not an important determinant of food consumption! In other words, the difference between food aid recipients and non-recipients in their food consumption was not statistically significant. This is a revelation that although a sizeable 35 per cent of the households responded to have received food aid, their food consumption scores did not improve significantly over the non-recipients. This is evidence that reinforces the fact that food received from aid could provide relief from hunger but would not provide the real solution to food insecurity. Similar in peculiarity is the finding that households that reported to have experienced some sort of food shock (41%: 3783) did not differ significantly in their food consumption levels compared to those that did not experience any shock. This is a clear revelation of the endemic nature of food insecurity in Southern Sudan. In other words, poor food consumption and food shocks – at least for the period of the study – were characteristic of Southern Sudan.
Another finding of interest is the significant difference (p-value=0.018) between households receiving staple food items in form of “gift” from relatives and “another” source in relation to experiencing poor food consumption. With odds of 6 times getting staples from relatives than of getting from “other” sources being more probable to score low food consumption, it could be concluded that dependence on relatives, although affecting only 1 per cent (91 households), is manifestation of severe coping mechanism; hence the need for further investigation. A further cause of bewilderment is that the reported main sources of livelihood being livestock raring, farming and fishing are revealed to increase the probability of being in the lower food consumption categories. This outcome may also need to be further investigated.