Regression Analyses

The descriptive analyses presented in sections II and III of this paper point out significant differences between poor and non-poor households, which prompts research into the factors that determine the poverty status of households in 2003 given the endowments they had in 2001. A logit model was estimated to answer this question. The explanatory variables in 2003 were (Appendix III [ PDF 37.4KB | 3 page ]):

• Age and education of the household head
• Household size
• Place of residence
• Poverty status in 2001
• Farming area planted with rice
• Cash crop area
• Other crop area
• Credit availed
• Transportation asset ownership
• Production asset ownership
• Access to rural infrastructure
• Main crop planted

As outlined in the methodology section, a logit model is estimated where the probability of being non-poor is estimated by:

To test the fit of the model, the Hosmer and Lemeshow test was performed. The probability of 0.3972 suggests that the null hypothesis of the model being inadequate cannot be rejected. The 26 independent variables of the model also adequately explain the probability of households being poor or non-poor as evidenced by the likelihood ratio chi-squared.

Thirty-two percent of the actual poor were classified falsely as non-poor while 28% of nonpoor were considered as poor in the model. Overall, the model correctly classified 69.74% of the sample into their actual poverty status in 2003 (Table 14 [ PDF 34.1KB | 1 page ]). Details of the summary statistics and regression results are include in Appendix III [ PDF 37.4KB | 3 page ] (Tables A3.2 and A3.3.).

A. Interpretation of Results

Age and Education of Household Head. It appears that age is not a very important determinant of poverty status in agrarian communities, possibly due to household heads' passing on tasks to younger members of the household. Living conditions and personal characteristics associated with poverty risks are often shared among members of the family. The less educated the household head, the lower the probability of being non-poor in 2003.

Household Size. Consistent with most studies, the larger the family, the lower the chances of being non-poor. On average, predicted probabilities of households that were poor in 2001 were 0.21 percentage points lower than those of the non-poor households. The rate of change in probabilities for those above the poverty line in 2001 peaked at six household members, then tapered off as household size increased. For those below the poverty line, the biggest decrease occurred at one to three household members, but, as the family size grew, the probability of being non-poor in 2003 decreased (Figure 9 [ PDF 30.7KB | 1 page ]).

Place of Residence. Agricultural households in the ARMM have the lowest chance of being non-poor in 2003. Households from Visayas and Mindanao (excluding the ARMM) have very similar predicted probabilities while farmers in Luzon are significantly poorer. A possible explanation is that farmers in Luzon find it more difficult to earn high income due to the existence of natural calamities such as typhoons. Aside from being in mostly upland areas, the soil quality in Central Luzon is acidic due to the eruption of Mt. Pinatubo and receives little assistance from the government. In contrast, farmlands in Mindanao are rarely plagued by typhoons, and local governments receive ample assistance from foreign donors and the government.

Poverty Status in 2001. Fifty-four percent of the sample was classified as poor in 2001. All else held at the mean, the predicted probability of being non-poor in 2003 was 0.59 when the household was non-poor in 2001. If the household income was below the poverty line in 2001, the probability decreased to 0.33 (Figure 10 [ PDF 29.9KB | 1 page ]).

Farming Area Planted with Rice. The larger the area planted, the higher the probability of being non-poor. An additional hectare planted with rice increased the likelihood of being nonpoor by 0.23 points. All else held at the mean, a farmer who did not plant rice had 0.40 predicted probability while a farmer who planted rice in 13 hectares had 90% chance of being non-poor in 2003. As expected, the larger amount of land tilled, the higher the probability of being non-poor. The gap in probabilities between the two groups also decreased as the planted area increased (Figure 11 [ PDF 29.1KB | 1 page ]).

Cash Crop Area. Cash crops are composed of maize, sugar cane, and coconut. Of all crops, the cash crop group leads to the highest probability of being non-poor. For each additional hectare devoted to cash crops, the odds of being non-poor increased by a factor of 1.37, all else held constant (Figure 12 [ PDF 28.8KB | 1 page ]).

Other Crop Area. This pertains to land area planted with crops other than rice, maize, sugar cane, and coconut. These crops are mostly fruits (especially bananas) and vegetables. A one-hectare increase in areas planted with other crops will increase the odds of being nonpoor by a factor of 1.27, roughly similar to rice areas (Figure 13 [ PDF 28.7KB | 1 page ]).

Credit Availed. Although credit is not statistically significant, it is interesting to note that the higher the credit availed in 2001, the lower the probability of being non-poor in 2003. High credit amounts are particularly debilitating for households below the poverty line. A closer examination of the data shows poor households tend to obtain credit from informal sources (Figure 14 [ PDF 30.3KB | 1 page ]).

Transportation Asset Ownership. While ownership of trucks and jeepneys does not significantly affect poverty incidence, ownership of tricycles increases the odds of being nonpoor by a factor of 1.48. The difference in predicted probabilities for tricycle and non-tricycle owners, however, is only 0.096 points.

Production Asset Ownership. Ownership of generators, tractors, and threshers increases the likelihood of being non-poor by 35 points. Similar to transportation asset ownership, the difference between predicted probabilities among owners and non-owners is not very high. Non-owners are 43% likelier to be classified as non-poor while owners have a 50% likelihood.

Access to Rural Infrastructure. Access to FMRs and bridges in 2001 increased the chances of being non-poor by 0.34 points. It appeared, however, that access to irrigation did not have that much effect on poverty incidence.

Main Crop Planted. Coconut farmers are the most marginalized among all farmer groups. This is consistent with the findings of De Dios, 1993. Banana growers have the highest probability of being non-poor, followed by growers of other crops such as maize, rice, and sugar cane.

Download this Discussion Paper [ PDF 396.4KB| 35 pages ].

Comment(s)

There are [1] comment(s) for this entry. Post a comment.

1. Cris Sales
   (posted 10 February 2011 / 04:28:54 PM)
   
   In other words, CARP did in fact improve the lives of the prople in the Agrarian Reform Communities.