190 likes | 341 Views
Impact of Nonfarm Employments on Food Security in Rural Ghana: A Propensity-Score Matching Approach. by Victor Owusu Kwame Nkrumah University Science and Technology, Kumasi, Ghana Awudu Abdulai University of Kiel, Kiel, Germany Seini Yussif Abdul-Rahman
E N D
Impact of Nonfarm Employments on Food Security in Rural Ghana: A Propensity-Score Matching Approach by Victor Owusu Kwame Nkrumah University Science and Technology, Kumasi, Ghana Awudu Abdulai University of Kiel, Kiel, Germany Seini Yussif Abdul-Rahman Kwame Nkrumah University Science and Technology, Kumasi, Ghana Paper Presentation at Impact Evaluation Conference. Cairo, Egypt. 29th March-2nd April 2009
Outline of Presentation • Introduction and Problem Statement • Research Questions • Literature: Non-farm Employments and Food Security • Hypotheses • Relationship between Treatment and Outcome Variables • Matching Techniques • Data Description • Matching Results • Results on Sensitivity Analysis • Conclusions and Policy
Introduction and Problem Statement • Greatest challenge for sub-Saharan Africa is feeding ever-increasing population. • Agriculture has not been a sufficient vehicle in addressing the household-level malnutrition and food insecurity due to: • Hostile agro ecological factors • Low productivity • Reducing hunger and food insecurity has been part of developmental agenda since the World Food Summits in 1996 and 2001. • One possible pathways out of food insecurity in sub-Saharan Africa is the promotion and establishments of nonfarm employments (Barrett et al., 2001; Stamoulis and Zezza, 2003). • Most rural communities in Africa derive 42% of income from nonfarm employments (Haggbade, Hazell and Reardon (2002).
Research questions • Does nonfarm employments reduce household food insecurity? • To what extent does nonfarm employment contributes to household food security?
Nonfarm Employment and Food Security • Improved access to nonfarm sources of income could lead to household welfare and food security (Holden et al., 2004; Winters et al., 2006). • Most rural households in Ghana adopt various livelihood strategies to safeguard food poverty (Ashong and Smith, 2001). • Contribution of women to household food supplies and income in Ghana through nonfarm employments could complement agriculture (Canagarajah et al., 2001). • What is not clear in the empirical literature is the direct causal effect of nonfarm employment on household food security. • Causal effects allow us to make inferences about the outcome that would have been observed for participants had they not participated in nonfarm employments. • Main contribution of this paper is to examine the unobserved counterfactual outcome by employing matching techniques.
Hypotheses • Participation in non-farm employments by rural farmers increases household income. • Household food security is influenced by participation in non-farm employments.
Relationship Between Treatment and Outcome Variables Due to inadequate information on the counterfactual situation, we resort to randomization by collecting non-experimental data (Blundell and Dias, 2000). Self-selection bias arises because the decision to participate or not to participate may be dependent on benefits of participation. Higher income effect could lead to household food security which also influences non-farm employment participation. To account for the selection into treatment on observables, we use linear regression:
where is household income for husbands (i=1), wives (i=2) and joint (i=3) If only husband participates, j=1, only wife participates, j=2 If joint participation by husband and wife, j=3 is a treatment variable=1 if an individual participates and 0 otherwise is the corresponding treatment effect is a vector of confounding variables such as personal and household characteristics, and other location characteristics is a vector of unknown parameters . is the error term with
For participation in non-farm employments, we can specify an index function for an observed variable H as Selection bias occurs if the error term, ηof treatment equation and error term, εof outcome equation are correlated due to the influence of unobservable factors such that When , standard regression produces biased results. To avoid this bias, we resort to matching techniques through covariate adjustments (Dehejia and Wahba, 2002).
Matching Techniques • Propensity-score (p-score) • With experimental data, we employ the propensity score matching approach (Rosenbaum and Rubin, 1983). • Given the p-score, the Average Treatment Effect (ATT) as noted by Becker and Ichino(2002) is estimated as where and are two counterfactual outcomes of participation & non-participation 2.Implementation of the p-score • Estimation of the p-score • Choosing appropriate matching algorithm for the ATT • Satisfying the common support condition • Assessing the matching quality • Conditional Independence Assumption (CIA) • Re-estimation of p-score of matched and unmatched participants (Sianesi’s Approach,2004) • Sensitivity Analysis using the Bounding Approach (Rosenbaum,2002)
Data Description • The cross-sectional data was collected in 2007 among 150 farm households with 300 married individuals in 10 rural communities in Savelugu-Nanton District of Northern Ghana. • Treatment variables • Dummy variable=1 if husband participates, 0 otherwise. • Dummy variable=1 if wife participates, 0 otherwise. • Dummy variable=1 if there is joint participation by husband and wife, 0 otherwise. • Outcome variables • HHINC (continuous) denotes total household income (on-farm, non-farm & income from livestock sales and transfers). • Food security indicator: MSFC (binary)=1 if household does not mortgages standing field crops for current consumption. • Food insecurity indicator: DHC16(binary)=1 if household’s harvested crops last only for the first six months of the year.
Independent variables in Probit Model • Matching is based on variables which influence both treatment and outcome variables and are not affected by the treatment (Caliendo and Kopeinig, 2008). • Selection of explanatory variables is based on previous research and information and institutional settings. (Smith and Todd, 2005). • Explanatory variables include: • Personal and household characteristics • Household capital assets • Location characteristics Non-farm employment activities • Agro-processing: small-scale processing of sheanuts and groundnuts. • Cotton ginnery and soap manufacturing. • Trading in foodstuffs.
Results on ATT • Non-farm employment has a positive relationship with household income and food security indicator (MSFC). • The impact of non-farm employment participation on the food insecurity indicator (DHC16) was found to be negative. • Implication: Increased participation in nonfarm employments increases the household income as well as increasing the likelihood at which households do not mortgage their standing field crops for current consumption. • The causal effects of 10.84 for husbands and 8.85 for wives at 5% significant level indicate that husband’s participation yield an increase in the household income by ¢10,845,800 ($1147) while the wife’s participation yield an increase of ¢8,853,500($936). • The ATT= 0.8571 for food security indicator MSFC at 5% significant level, suggesting that husband’s participation increases the probability at which the households do not mortgage their standing field crops for current consumption by 86%. The reduction in mean absolute bias from 26.9% to 9.9% indicates a substantial reduction in bias of 63% as a result of employing the matching technique. • The effects are 0.70 and 0.86 for wives and joint participation for the outcome variable (MSFC) at 10% and 1% significant levels. A reduction in mean absolute bias of 21.2% to 2.8% (86% reduction in bias) for wives is an indication that the covariates are balanced by using the propensity score matching approach.
Results on ATT (cont’d) • Participation in non-farm employments by wives reduces household food insecurity (DHC16) by 0.44. Balance checks show mean absolute bias reduction from 19.2% to 4.7%. • Joint participation reduces the likelihood of food depletion within the first six months (DHC16) by 0.46 (ATT=0.4643) at 10% significant level. The corresponding mean absolute bias reduced from 27.9% to 3.5% representing 85% of removal of bias as result of the randomization procedure.
Results on Sensitivity Analysis • The critical levels of gamma at which causal inference are significant are investigated. • We employ the bounding approach (Rosenbaum, 2002) in the sensitivity analysis. • Upper bound scenarios indicating over-estimation of ATT are reported. • The lower bound scenarios underscoring the under-estimation of the treatment effect are less interesting. • For ATT which is significant, we increase the level of gamma until the treatment effect inference is changed (Hujer et al, 2004). • The critical value of 1.10 for the impact of husband’s nonfarm employment on the MSFC, implies that individuals with the same Z vector differ in their odds of participation by a factor 10%. • For wives, the positive significant impact of nonfarm employment on food security (MSFC) requires a hidden bias of 2.95. • The impact of nonfarm employments on food insecurity (DHC16) requires a critical value of 1.35 to render the significant negative effect spurious. • It is interesting to note that our findings compare favorably with results from other studies and are generally insensitive to hidden bias.
Conclusions • Non-farm employment has a significant positive relationship with household income and the food security indicator where households do not mortgage their standing field crops for current consumption. • Significant negative relationship was also found between non-farm employments and the food insecurity indicator where household’s duration of food crops last only for the first six months of the year. • Joint participation by couples rather provides higher direct causal effects. • The differences in impact on food security by husband’s and wife’s participation were not much but husband’s participation appears to be higher. • The results confirm Nkurunziza (2006)assertionthat women in Africa appear to be disadvantaged in financial capital and time which are key factors to job entry. Policy • Ensuring food security at the household-level should involve strategies that create opportunities and expansion of non-farm micro-enterprises in the rural economy. • Policy efforts should be geared towards: • Encouraging easy entry into the non-farm sector by both males and females through improvement of human capital endowments. • Access to credit through assistance from NGO’S and donors involved in provision of microfinance schemes.