Sunteți pe pagina 1din 11

Food Policy 36 (2011) 108–118

Contents lists available at ScienceDirect

Food Policy
journal homepage: www.elsevier.com/locate/foodpol

Non-farm work and food security among farm households in Northern Ghana q
Victor Owusu a, Awudu Abdulai b,⇑, Seini Abdul-Rahman a
a
Kwame Nkrumah University of Science and Technology, Kumasi, Ghana
b
University of Kiel, Kiel, Germany

a r t i c l e i n f o a b s t r a c t

Article history: This paper examines the impact of non-farm work on household income and food security among farm
Received 7 October 2009 households in the Northern Region of Ghana. We analyze the impact by employing propensity score
Received in revised form 9 June 2010 matching method that accounts for self-selection bias. The matching results show that participation in
Accepted 13 September 2010
non-farm work exerts a positive and statistically significant effect on household income and food security
Available online 16 October 2010
status, supporting the widely held view that income from non-farm work is crucial to food security and
poverty alleviation in rural areas of developing countries.
Keywords:
Ó 2010 Elsevier Ltd. All rights reserved.
Africa
Non-farm employment
Household income
Food security
Impact assessment

Introduction their income from rural non-farm activities – a high share consid-
ering that only about 10% of the rural labor force is employed in the
One of the greatest challenges still facing most countries in sub- rural non-farm sector. Non-farm work offers the poor a potential
Saharan Africa is how to feed their ever-increasing populations. escape route from poverty, since they usually require little capital
Agriculture which is the main source of income for about 90% of or training to set up and are labor intensive. The extra income from
the rural population in the region, has not been a sufficient vehicle agricultural growth can create demand for these goods and ser-
for solving household-level malnutrition and food insecurity, be- vices, thus starting a virtuous cycle in which agricultural and rural
cause of low productivity and hostile agro-ecological factors. off-farm income grow and sustain each other’s growth (Stamoulis
Reducing hunger and food insecurity has therefore remained an and Zezza, 2003).
essential part of the international development agenda since the Several studies have examined the nature and determinants of
World Food Summits in 1996 and 2001. Food security exists when non-farm work in rural areas of developing countries (Abdulai
all people, at all times, have physical and economic access to suffi- and Delgado, 1999; Abdulai and CroleRees, 2001; Barrett et al.,
cient, safe and nutritious food to meet their dietary needs and food 2001; Canagarajah et al., 2001). Studies examining the impacts of
preferences for an active and healthy life. As noted by Rukuni non-farm income on household incomes have focused on the in-
(2002), food security at the household level is a balance between come-equity effects of non-farm income (e.g., Deininger and Olin-
availability of and access to sufficient food. to, 2001). However, work on the direct impact of non-farm work on
Barrett et al. (2001) have noted that one of the possible path- household income and food security is quite scarce (Lanjouw,
ways out of the vicious circle of food insecurity in sub-Saharan 2001; Holden et al., 2004; Chang and Mishra, 2008). Moreover,
Africa is the promotion and establishments of non-farm work in most of the studies that have examined the relationship between
the rural sector. Evidence provided by Haggblade et al. (2002) sug- non-farm work and household income have normally analyzed
gests that most rural communities in Africa derive about 42% of correlations rather than causal effects (e.g., Lanjouw, 2001). Using
bio-economic modelling, Holden et al. (2004) found that better ac-
q
The authors thank two anonymous reviewers and the journal editor for very
cess to low-wage non-farm income has a substantial positive effect
useful comments and suggestions on an earlier draft. The authors thank the German on household income.
Research Foundation for financial support for the data collection. The first author is This study contributes to the literature by empirically examin-
grateful to the Kiel Institute for World Economy for financial assistance. ing the impacts of non-farm work on household income and food
⇑ Corresponding author. Address: Department of Food Economics and Consump-
security status, using a sample of farm households in Northern
tion Studies, University of Kiel, Olshausenstrasse 40, 24118 Kiel, Germany. Tel.: +49
431 880 4426; fax: +49 (0) 431 880 7308. Ghana. We employ propensity score matching method to control
E-mail address: aabdula@food-econ.uni-kiel.de (A. Abdulai). for self-selection that normally arises when participation in

0306-9192/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.foodpol.2010.09.002
V. Owusu et al. / Food Policy 36 (2011) 108–118 109

non-farm work is not randomly assigned and self-selection into food poverty. This livelihood strategy has been widely documented
participation occurs. By explicitly considering the causal relation- for other parts of the world by several authors (Barrett et al., 2001;
ship between participation in non-farm work and household wel- Lanjouw and Feder, 2001; Reardon et al., 1998; Reardon et al.,
fare, the paper seeks to address counterfactual questions that 2001). Available evidence suggests that rural non-farm income
may be significant in predicting the impacts of policy changes. represents on average 42% of rural income in Africa, 32% in Asia,
We estimate separately for male and female-headed households 40% in Latin America and 44% in Eastern Europe (FAO, 1998; Davis,
to account for gender differences in the impact of participation in 2004). Canagarajah et al. (2001) found in Ghana that participation
non-farm work. in non-farm activities increased more rapidly for married women
The remainder of the paper is organized as follows: Section 2 and female-headed households than for men and asserted that wo-
discusses non-farm work and food security in Ghana. Section 3 men are able to combine agriculture and non-farm employment as
outlines the conceptual framework and empirical procedure, and a way out of poverty and household food insecurity.
the propensity score and treatment effects. Section 4 presents Gladwin et al. (2001) have noted that addressing food insecurity
the data used in the empirical analysis. Section 5 discusses the in Africa through increased food production may be inadequate so
empirical results, while the final section provides some concluding efforts must be geared toward enhancing farmers’ access to addi-
remarks. tional income through non-farm employment. Rural non-farm
enterprises promotion is one of the Government’s poverty reduc-
tion strategies set out in the Ghana Poverty Reduction Strategy
Non-farm work and food security in Ghana Paper II (MoFaEP, 2003). The activities of Non-Governmental
Organisations are complementary to these efforts of the govern-
Ghana has been characterized over the years by regional ment in the promotion of rural non-farm enterprise as a livelihood
inequality in terms of wealth and resource endowments that is strategy for enhancing food security in rural areas of Northern
basically geographical and political (Shepherd et al., 2004). The Ghana.3 The non-farm activities include agro-processing, commerce,
northern part of Ghana which is predominately of savannah veg- transport services, charcoal production, firewood gathering, repair
etation is associated with extremely poor quality soils, short uni- services, wage work, and seasonal migration, among others. Agro-
modal rainfall season and periodic drought (Dickson and Benneh, processing, which is mostly carried out by women is generally pur-
1988). This situation contrasts with the natural soil fertility and sued through traditional methods and on very small-scale bases.
bimodal rainfall seasons of the southern part of the country. Cou- These activities generally include processing of sheanuts, ground-
pled with these adverse environmental conditions of Northern nuts, rice, cotton ginnery, and soap manufacturing. Other activities
Ghana is the poor access to markets and well functioning finan- include trading in foodstuff such as maize, beans, rice and other
cial institutions. These conditions have contributed to a higher grains, sand winning, the bulk of which is used for construction work
level of food insecurity in the region (Whitehead, 2006). The re- in Tamale Metropolis and Fishing along the Black Volta (District Pro-
gion has actually remained the poorest in terms of living stan- file, 2006).
dards, literacy levels, health, and nutrition status for several
decades (Whitehead, 2006).
Conceptual framework and empirical procedure
Although poverty levels have declined in the country over the
last decade, progress has been much slower among food crop farm-
The model employed is essentially the one suggested by Huff-
ers than for other livelihood groups (Devereux, 2008). In particular,
man (1991), where farm households allocate their time to individ-
rural households in the north suffer seasonal strains in well-being
ual activities including non-farm employment. A farm household is
and seasonal pressures are found to be worst where the house-
assumed to maximize a utility function defined over consumption
holds face declining food stocks. Despite the efforts by policy mak-
of goods Q and leisure, H, i.e., U = U(Q, H). Utility is maximized sub-
ers to combat poverty by improving farmers’ access to improved
ject to time, budget, production, and non-negativity constraints.
technology, poverty remains prevalent in the region. Devereux
The time constraint is T = L1 + L2 + H, where T is total time endow-
(2008) points out that the ‘poorest’ groups in Northern Ghana
ment, L1 and L2 are respectively time allocated to farm work and
who are normally vulnerable to shocks such as drought, bush fire,
non-farm work, and H is leisure as defined above. The budget con-
and loss of animals may no longer engage in agriculture at all and
straint on household cash income can be expressed as
for that matter struggle to obtain enough food, especially during
the ‘hungry season’ months of the year.1 The ‘less poor’ who are PQ ¼ p1 y1  w1 L1 þ w2 L2 þ R ð1Þ
more dependent on agriculture than the ‘poorest’ or the ‘vulnerable’ where P is the price for the consumption good purchased in the
may also face land and labor input constraints that limit their ability market, w1 and w2 denote returns to labor from farm work and
to accumulate enough capital. As a recently conducted survey by non-farm work, respectively, y1 and p1 are annual quantity of farm
Ghana’s Ministry of Food and Agriculture (MoFA) revealed, the ‘less output produced and sold and price for farm output, respectively
poor’ tend to pursue a ‘survival strategy’ rather than a ‘development and R represent non-labor income.
strategy’.2 The first order condition for optimal time allocation for farm
As noted by Ashong and Smith (2001), most rural households in work, non-farm work and leisure is given as oU/oLi = wioU/
Ghana adopt various livelihood strategies in order to overcome oQ  oU/oL = 0. This first order condition can be rearranged to ob-
tain the returns to labor from farm work and non-farm work:
1
The ‘‘hungrier season” months in Northern Ghana as indicated by Devereux wi = (oU/oL)/(oU/(oQ). When farm households allocate their time
(2008) is between March and July but from January, less food-endowed households
to the three activities, the labor supply functions for farm work
have started experiencing food shortages.
2
A ‘development strategy’ as noted by Devereux (2008) is based on savings
and non-farm work can be derived as
through livestock (with resources acquired from crop sales or livestock husbandry), L1 ¼ L1 ðw1 ; w2 ; p1 ; p2 ; ZÞ ð2Þ
leading to investment in both agricultural and non-farm livelihood activities. Most
households in this group have adequate labor capacity, so are responsive to
commercial farming opportunities. When the ‘poorest’ households are hit by a shock,
3
they are forced to adopt ‘survival strategies’ that include the sale of assets (livestock); Active NGOs in Northern Ghana who promote rural non-farm enterprises as a
reduction in food intake; engaging in petty trade; migration; withdrawal of children means of ensuring food security include the Catholic Relief Services, Opportunities
from school; self-medication; and reliance on families, community-based organiza- Industrialization Centre (O.I.C), ACTION Aid-Ghana, Danish International Develop-
tions or NGOs for assistance. ment Agency, Ghana Danish Community Project, TECHNOserve and NEWENERGY.
110 V. Owusu et al. / Food Policy 36 (2011) 108–118

L2 ¼ L2 ðw1 ; w2 ; p1 ; p2 ; R; ZÞ ð3Þ However, Jalan and Ravallion (2003) argue that the assumption
of selection on observables is no more restrictive than assuming
where Z summarises household and location characteristics that
away problems of weak instruments, when Heckman two-step or
influence the individual’s reservation and non-farm wages. As noted
the IV approach is employed in cross-sectional data analysis.
by Huffman (1991), a positive number of non-farm hours will be
To address the selectivity bias problems associated with non-
observed for an individual i, if the potential market wage (wm i ) is
farm employment participation, we employ matching techniques
greater than the reservation wage (wri ).4 Thus, Li = 1 if wm i > wi
r

m r in assessing the impact of non-farm employment on household in-


and Li = 0 if wi 6 wi . However, these differential wages are not ob-
come and food security status (Dehejia and Wahba, 2002). The pro-
servable. What is observed is the decision to participate, or not to
pensity score matching approach addresses the problem of the
participate in non-farm work. This decision can be specified as an in-
limited distributional assumption of the errors, and more impor-
dex function, with unobserved variable,
tantly allows for a decomposition of the treatment effect on out-
Li ¼ bZ 0i þ li comes (Rosenbaum and Rubin, 1983; Heckman et al., 1999). Also
Li ¼ 1 if Li > 0 ð4Þ the counterfactual framework could detect ‘‘two important sources
Li ¼ 0 if Li 6 0 of bias in the estimation of treatment effects. These include the ini-
tial differences between non-farm work participants and non-par-
where li is the random disturbance term. ticipants in the absence of treatment, and the difference between
To analyze the relationship between participation in non-farm the two groups in the potential effect of the treatment (Winship
work and outcomes such as household income and food security and Morgan, 1999).
status, we start from the linear function
Y i ¼ ki þ ai X 0i þ di Li þ ei ð5Þ The propensity score and treatment effects

where Yi is household income or food security status, L denotes a The propensity score p(Zi) is defined as the conditional probabil-
dummy variable representing one, if an individual participates in ity of participating in non-farm work given pre-participation
non-farm work and zero otherwise; X 0i is a vector of personal and characteristics:
household characteristics and assets, and location characteristics,
ki is a vector of unknown parameters, and ei is a random error term. pðZ i Þ  Pr½Li ¼ jZ i  ¼ E½Li jZ i ; pðZ i Þ ¼ FfhðZ i Þg ð6Þ
The specification above in Eq. (5) treats participation in non-
where Li = (0, 1) is the indicator of non-farm work participation and
farm work as an exogenous variable on the premise that house-
Zi denotes a vector of pre-participation characteristics, and F{  } can
holds participate in non-farm work to increase their income or im-
be a normal or logistic cumulative distribution. The propensity
prove their food security status. However, this need not be the
score can be predicted with either the logit or probit model (Hujer
case, since wealthier households may be better disposed to partic-
et al., 2004; Sianesi, 2004; Faltermeier and Abdulai, 2009).5
ipate in non-farm work. Moreover, the decision to participate or
The predicted propensity scores can then be used to estimate
not to participate in non-farm employment may be dependent
treatment effects. The most common treatment effects in the eval-
on the benefits from participation itself. Thus, participation in
uation literature include the Average Treatment Effect (ATE) which
non-farm work is not random, with the group of participants being
captures the treatment effect for the whole sample, Average Treat-
systematically different. Specifically, selection bias occurs if unob-
ment Effect on the Treated (ATT) or the participation effect, and the
servable factors influence both the error term (l) of the participa-
Average Treatment Effect on the Untreated (ATU). Becker and Ich-
tion Eq. (4), and the error term (e) of the income Eq. (5), thus
ino (2002) indicate that the parameter of interest in the estimation
resulting in correlation of the error terms.
of the propensity score is the Average Treatment Effect on the Trea-
When correlation between the error terms occurs, estimating
ted (ATT). However, under the assumption of unconfoundedness,
Eq. (5) with ordinary least squares method results in biased esti-
Titus (2007) and Millimet and Tchernis (2008) point out that an
mates. Heckman two-step approach has been mostly employed
appropriate approach of evaluating policy-relevant outcomes in a
to address this selection bias. However, the approach depends on
counterfactual framework is to examine not only the ATT, but also
the restrictive assumption of normally distributed errors. An alter-
the Average Treatment Effect on the Untreated (ATU), which cap-
native way of controlling for selection bias is to use instrumental
tures the effect of treatment on individuals who do not participate
variable (IV) method. A major limitation of the approach is the dif-
in non-farm work.
ficulty in finding identifying instruments in the estimation. In addi-
Given the propensity score p(Zi), the three effects are evaluated
tion to these limitations, both OLS and IV procedures tend to
as:
impose a linear functional form assumption, implying that the
coefficients on the control variables are similar for participants ATE ¼ E½EfY 1i jLi ¼ 1; pðZ i Þg  EfY 0i jLi ¼ 0; PðZ i Þg ð7Þ
and non-participants. However, this assumption may not hold,
since the coefficients could differ (Jalan and Ravallion, 2003). ATT ¼ E½EfY 1i jLi ¼ 1; pðZ i Þg  EfY 0i jLi ¼ 0; PðZ i ÞgjLi ¼ 1 ð8Þ
Unlike the parametric methods mentioned above, propensity
score matching, which is employed in the present study, requires
ATU ¼ E½EfY 1i jLi ¼ 1; pðZ i Þg  EfY 0i jLi ¼ 0; PðZ i ÞgjLi ¼ 0 ð9Þ
no assumption about the functional form in specifying the rela-
tionship between outcomes and predictors of outcome. The draw- where Y 1i
and Y 0i
are the two counterfactual outcomes of participa-
back of the approach is the Conditional Independence Assumption tion and non-participation in non-farm employment.
(CIA), which states that for a given set of covariates, participation is A number of matching algorithms have been suggested in the
independent of potential outcomes. Smith and Todd (2005) rightly literature to match participants and non-participants of similar
note that there may be systematic differences between the out- propensity scores. The most widely employed algorithms include
comes of participants and non-participants, even after condition- the nearest neighbor matching, caliper matching, and the kernel
ing on observables. Such differences may arise because of matching methods. The nearest neighbor, which matches each
selection into treatment based on unmeasured characteristics. participant with its closest neighbor with similar observed

4 5
The reservation wage for non-farm work is the marginal value of the individual’s Other studies have also employed the logistic model in the estimation of the
time when all of it is allocated to farm and leisure. propensity scores (Hirano and Imbens, 2001; Caliendo and Kopeinig, 2005).
V. Owusu et al. / Food Policy 36 (2011) 108–118 111

characteristics, can be done either with replacement or without Data description


replacement. Matching with replacement results in bias reduction
since each treatment group can be matched to the nearest compar- The data employed in this study come from a recent household
ison group as a result of a reduction in the propensity score dis- survey conducted during 2007 in 10 rural communities – Bung-
tance.6 Kernel matching tends to use more non-participants for lung, Challam, Kpendua, Fazihini, Kadia, Kanshegu, Laligu, Tari-
each participant, thereby reducing the variance but possibly increas- kpaa, Zion and Tibale in the Savelugu-Nanton district of Northern
ing the bias. The caliper matching, besides using the nearest neigh- Ghana. The district is mainly an agriculture-based economy
bor within each maximum propensity score distance, can also use employing about 97% of the economically active population. Save-
all the comparison members within the caliper. This method is em- lugu-Nanton district consists of 143 communities out of which 10
ployed in the present study because of this advantage. communities were randomly selected for the survey. Through
For robust estimation of the propensity score, a balancing prop- informal focus group meetings with community leaders and mem-
erty which is a function of the relevant observed covariates must bers of farmer-based organizations in the 10 survey sites, 744
be satisfied.7 This property implies that conditional on the propen- male-headed and 300 female-headed farm households were iden-
sity score, each individual should have the same probability of par- tified. A random sampling approach was then used to select 150
ticipating in non-farm work as in a randomized experiment. As households from each of the male and female-headed households
noted by Hujer et al. (2004), the distribution of Zi is expected to be making a total sample size of 300 households. Information from
balanced in the two samples if this balancing property is satisfied. the households was gathered through interviews with question-
Another relevant assumption for robust estimation of the p-score naires. The data covered information on farm and cash-oriented
is to satisfy the common support condition (Heckman et al., 1999). non-farm activities, demographic characteristics, as well as market
This requires that individuals with the same covariates Z should and resource access.
have positive probabilities of being both participants and non-partic- The savannah vegetation of the region coupled with the single
ipants in non-farm work, implying that all individuals in the com- rainy season and long dry season often result in shortfalls in agri-
mon support region can actually participate in all states cultural production. This condition compels farmers to search for
(0 < PðL ¼ 1jZÞ < 1). Smith and Todd (2005) point out that if there additional sources of income to supplement their unstable agricul-
are regions where the support of Z does not overlap for the differ- tural incomes. The predominant cash-oriented non-farm work en-
ent groups, matching is only justified when performed over the gaged by the sampled individuals includes cotton ginnery, soap
common support region and the estimated treatment effect must manufacturing, trading in foodstuffs, and small-scale agro-process-
be redefined as the treatment impact for participants whose prob- ing of sheanuts and groundnuts. Females mostly engaged in small-
abilities lie within the overlapping support region. Given the pro- scale processing of sheanuts and groundnuts.
pensity score, the Conditional Independence Assumption (CIA) The dependent variable used in the study is a dummy variable
must also be satisfied.8 This states that once the set of observable that takes the value of one, if the household participated in cash-
characteristics, Zi, are controlled for, the treatment variable (non- oriented non-farm work, and the value zero, if no participation
farm work participation), and the outcome variables (household in- was recorded. The outcome variables used in this study are total
come and food security) should be uncorrelated. household income and two binary indicators of household food
For the standardized bias approach proposed by Rosenbaum security status. The household income was measured as the total
and Rubin (1985), the quality of matches is assessed by comparing income from cash-oriented non-farm work, farm work, livestock
the situation before and after matching to check if there remain sales and transfers to the household in 2007. Two food security
any differences after conditioning on the propensity score.9 Sianesi variables that have been suggested in the literature are employed
(2004) has also recommended an additional assessment of the in the analysis. Both measures are used to ascertain whether the
matching quality which requires the re-estimation of p-score of results are sensitive to the indicator employed (Stamoulis and Zez-
matched participants and matched non-participants. The matching za, 2003; Smith and Wiesmann, 2007; Devereux, 2008). One of the
procedure balances the covariates very well if the pseudo-R2 after food security indicators captured food availability and was mea-
matching is fairly low, while the probability of the F-statistics is sured as a binary indicator that takes the value of one, if the house-
not significantly different from zero. As observed by Hujer et al. hold does not mortgage its standing field crops for current
(2004), a possible hidden bias might occur if there are unobserved consumption (MSFC) during the farming season in 2007, and zero
variables that tend to influence simultaneously non-farm work par- otherwise.11 The other food security indicator captures food avail-
ticipation and household income and food security status. Given that ability and access by the household. It is actually termed as food
it is not possible to estimate the magnitude of selection bias with insecurity, and was measured as an indicator that takes the value
non-experimental data, Rosenbaum (2002) has suggested the use one, if the household’s harvested food stock declines during critical
of the bounding approach to examine the influence of unmeasured periods of food shortages (DHC16) in 2007, and zero otherwise.12
variables on the selection process.10 As a livelihood strategy, most rural farm households in Northern
Ghana participate in non-farm activities as alternative sources of
income generation. This tends to cushion them against the effects
of food crop shortages particularly during the first sixth months
6
When there are few comparison groups similar to treated groups, one may be of the year, that normally arises as a result of low productivity
compelled to match treated groups to comparison groups that are not even similar in
terms of p-scores, when matching is done without replacements. There is also the
possibility of encountering bad matches, if the closest neighbor is far away (Dehejia 11
An individual facing acute food shortages in Northern Ghana mortgages his or her
and Wahba, 2002). standing field crops for current consumption by using the standing field crops as
7
The balancing property is specified as L ? ZjpðZÞ. collateral to obtain food supplies from food endowed individuals with the promise of
8
This is often referred to as unconfoundedness expressed as Y 1 ; Y 0 ? LjZ ) settling the food deficit when standing field crops are harvested.
Y ; Y 0 ? LjpðZÞ.
1 12
The critical periods of household food declines and shortages in Northern Ghana
9
As observed by Smith and Todd (2005), there is either failure of conditional occur in the first sixth month of every year. Another food security indicator is the
independence assumption (CIA) or mis-specification in the model used for the p- ability of the household to meet nutritional requirements (Stamoulis and Zezza,
score-estimation if after conditioning on the propensity score, there is still depen- 2003). Household-level food security also implies household daily food energy
dence on Z. availability per capita, whether a household is food energy deficient, household diet
10
This method is well documented in Rosenbaum (2002) and Imbens and diversity, whether a household has low diet diversity, and the percent of available
Wooldridge (2009). food energy derived from staples (Smith and Wiesmann, 2007).
112 V. Owusu et al. / Food Policy 36 (2011) 108–118

Table 1
Descriptive statistics for male’s non-farm employment participation.

Variable name Variable definition Participants Non-participants Difference


N = 90 (30%) N = 210 (70%) in means
Mean S.d Mean S.d
Treatment variable
MNFARM 1 if male participates in non-farm employment, 0 otherwise
Outcome variables
HHINC Household income (¢/million) 10.85 8.93 6.09 5.43 4.76**
MSFC Food security (1 if household does not mortgage its standing field 0.89 0.32 0.68 0.47 0.21**
crops for current consumption, 0 otherwise).
DHC16 Food insecurity (1 if household harvested food stock decline during 0.42 0.49 0.38 0.48 0.04***
critical periods of food shortages, 0 otherwise).
Independent variables
Household characteristics
MAGE Age of male (in years) 50.42 11.99 55.04 15.17 4.62
MEDUC Years of formal education of male 2.71 4.23 0.75 2.61 1.96***
GPMEMB 1 if individual is a member of village group, 0 otherwise 0.71 0.46 0.53 0.50 0.18
PRCLS3YR 1 if there is presence of children less than 3 years old, 0 otherwise 0.80 0.40 0.72 0.45 0.08*
HHSIZE Household size 10.27 4.95 9.06 4.53 1.21**
DPRATIO Dependency ratio 0.99 0.53 0.95 0.56 0.04
MIGRATN Number of migrated household members 1.07 1.50 1.05 1.40 0.02
CRACCES 1 if individual has access to credit, 0 otherwise 0.44 0.50 0.12 0.50 0.32**

Household assets
IFBIKE 1if individual possess bicycle, 0 otherwise 0.99 0.15 0.98 0.14 0.01
TRACTA 1 if household possesses tractor, 0 otherwise 0.16 0.37 0.04 0.19 0.12*
IFCATTLE 1 if individual possesses cattle, 0 otherwise 0.60 0.50 0.30 0.46 0.30***
IFGOAT 1 if individual possesses goat, 0 otherwise 0.89 0.32 0.80 0.40 0.09**
IFGFWL 1 if individual possesses fowl, 0 otherwise 0.64 0.48 0.58 0.49 0.06

Location dummies
DBUNGLUN 1 if individual is located at Bunglung, 0 otherwise 0.16 0.37 0.08 0.27
DTARIPKA 1 if individual is located at Taripka, 0 otherwise 0.11 0.32 0.10 0.29
DFAZIHIN 1 if individual is located at Fazihini, 0 otherwise 0.16 0.37 0.08 0.27
DLALIGU 1 if individual is located at Laligu, 0 otherwise 0.02 0.15 0.13 0.34
DKPENDUA 1 if individual is located at Kpendua, 0 otherwise 0.07 0.25 0.11 0.32
DKANSHEGU 1 if individual is located at Kanshegu, 0 otherwise 0.16 0.37 0.08 0.27
DCHALLAM 1 if individual is located at Challam, 0 otherwise 0.04 0.21 0.12 0.33
DTIBALE 1 if individual is located at Tibale, 0 otherwise 0.04 0.21 0.12 0.33
DKADIA 1 if individual is located at Kadia, 0 otherwise 0.11 0.32 0.10 0.29
DZION 1 if individual is located at Zion, 0 otherwise 0.13 0.34 0.09 0.28

Household income in millions of Old Ghanaian Cedi (¢). Exchange rate: US $1 = ¢9457 in 2007.
*
Denotes significant at 10%.
**
Denotes significant at 5%.
***
Denotes significant at 1%.

and crop harvest failures. Income from non-farm work assists rural variables, there appear to be statistically significant differences in
households’ to smoothen their consumption when self-produced household income and food security status between male partici-
food crops fail to suffice for the whole year and to ensure food secu- pants and non-participants. Although there are differences between
rity. Participation in non-farm work therefore helps rural house- female participants and non-participants, there is only a statistically
holds in Ghana to enhance their food security status, or reduce significant difference in food insecurity status. There are also signif-
their food insecurity. icant differences in education, household size and access to credit, as
Tables 1 and 2 present the definitions and sample statistics of well as household assets.
the variables used in the analysis for males and females, respec- Quite interesting is the significant difference between partici-
tively. The observed mean difference of 0.48 in the effects of treat- pants and non-participants regarding access to credit for both
ment for males (0.30) and females (0.78) is statistically significant males and females. For males, 44% of participants and 12% of
at 1% level indicating the presence of gender heterogeneous treat- non-participants had access to formal credit, while the correspond-
ment effects. The difference in rates of participation between males ing figures for females were 47% and 27%, respectively, revealing
and females reflects the fact that males in the area are more en- the low levels of credit availability to rural farm households. As ar-
gaged in farming activities, while females are predominantly en- gued by Reardon et al. (1994), poor households who lack access to
gaged in non-farm activities.13 Also presented in the Tables are credit may be handicapped by limited access to non-farm income.
differences in means of the variables used in the matching analyses There are also significant differences between participants and
for both male and female participants, alongside their significance non-participants in terms of education. On average, participants
levels. The significance levels suggest that there are some differences tend to have a higher number of years of formal education than
between participants and non-participants with respect to house- non-participants. Reardon (1997) points out that education tends
hold and farm-level characteristics. With regards to the outcome to be correlated with ability to mobilize capital through non-farm
work. Although it may appear that basic technical skills are nor-
mally acquired outside the formal school system, the broader skills
13
As noted by an anonymous reviewer, another empirical strategy of testing the
of reading, writing and numeracy are acquired within it, making
presence of heterogeneous treatment effects is to fit a regression model which schooling an important determinant of participation in non-farm
depends on household characteristics including gender. work (Norcliffe, 1983).
V. Owusu et al. / Food Policy 36 (2011) 108–118 113

Table 2
Descriptive statistics for female’s non-farm employment participation.

Variable name Variable definition Participants Non-participants Difference


N = 234 (78%) N = 66 (22%) in means
Mean S.d Mean S.d
Treatment variable
FNFARM 1 if female participates in non-farm employment, 0 otherwise

Outcome variables
HHINC Household income (¢/million) 10.84 6.72 8.93 5.43 0.91
MSFC Food security (1 if household does not mortgage its standing field 0.73 0.45 0.79 0.42 0.06
crops for current consumption, 0 otherwise).
DHC16 Food insecurity (1 if household harvested food stock decline 0.40 0.49 0.36 0.49 0.04**
during critical periods of food shortages, 0 otherwise).

Independent variables
Household characteristics
FMAGE Age of female (in years) 40.03 9.56 43.00 12.24 2.97*
FMEDUC Years of formal education of female 0.73 2.32 0.18 1.04 0.55***
GPMEMB 1 if individual is a member of village group, 0 otherwise 0.60 0.49 0.55 0.51 0.05
PRCLS3YR 1 if there is presence of children less than 3 years old, 0 otherwise 0.76 0.43 0.70 0.47 0.06
HHSIZE Household size 9.97 4.74 7.48 3.94 2.49**
DPRATIO Dependency ratio 0.91 0.49 1.14 0.71 0.23**
MIGRATN Number of migrated household members 1.03 1.19 1.15 2.08 0.12
CRACCES 1 if individual has access to credit, 0 otherwise 0.47 0.50 0.27 0.45 0.20***

Household assets
IFBIKE 1if individual possess bicycle, 0 otherwise 0.98 0.13 0.97 0.17 0.01
TRACTA 1 if household possesses tractor, 0 otherwise 0.09 0.28 0.03 0.17 0.06
IFCATTLE 1 if individual possesses cattle, 0 otherwise 0.40 0.49 0.36 0.49 0.04
IFGOAT 1 if individual possesses goat, 0 otherwise 0.81 0.39 0.86 0.33 0.05
IFGFWL 1 if individual possesses fowl, 0 otherwise 0.61 0.49 0.58 0.50 0.03

Location dummies
DBUNGLUN 1 if individual is located at Bunglung, 0 otherwise 0.10 0.30 0.09 0.29
DTARIPKA 1 if individual is located at Taripka, 0 otherwise 0.10 0.30 0.09 0.29
DFAZIHIN 1 if individual is located at Fazihini, 0 otherwise 0.11 0.32 0.06 0.24
DLALIGU 1 if individual is located at Laligu, 0 otherwise 0.12 0.33 0.03 0.17
DKPENDUA 1 if individual is located at Kpendua, 0 otherwise 0.06 0.24 0.24 0.44
DKANSHEGU 1 if individual is located at Kanshegu, 0 otherwise 0.12 0.33 0.03 0.17
DCHALLAM 1 if individual is located at Challam, 0 otherwise 0.09 0.29 0.12 0.33
DTIBALE 1 if individual is located at Tibale, 0 otherwise 0.11 0.32 0.06 0.24
DKADIA 1 if individual is located at Kadia, 0 otherwise 0.08 0.27 0.18 0.39
DZION 1 if individual is located at Zion, 0 otherwise 0.10 0.30 0.09 0.29

Household income in millions of Old Ghanaian Cedi (¢). Exchange rate: US $1 = ¢9457 in 2007.
*
Denotes significant at 10%.
**
Denotes significant at 5%.
***
Denotes significant at 1%.

The discussion of the differences between participants and non- Empirical results
participants from Tables 1 and 2 centered on mean differences in
the outcome variables and other household and farm-level vari- The propensity scores were estimated with a probit model.14
ables. However, these comparisons of mean differences do not ac- The results of the propensity score estimates for male and female
count for the effect of other characteristics of farm households, and participants are reported in Tables A1 and A2 in the appendix. The
as such may confound the impact of participation in non-farm propensity scores only serve as a device to balance the observed dis-
work on household income and food security status with the influ- tribution of covariates across the treated and the untreated groups
ence of other characteristics. Multivariate approaches that account (Lee, 2008). A detailed interpretation of the propensity score esti-
for selection bias arising from the fact that participants and non- mates is not undertaken in this study. However, as the results show,
participants may be systematically different are essential in pro- most of the variables included in the estimations have the expected
viding sound estimates of the impact of participation on household signs. In particular, education and access to credit are found to be
income and food security. positively and significantly related to participation in non-farm
The independent variables used in the probit regression models work. The common support condition was imposed and the balanc-
to predict the propensity scores were based on past research on ing property was satisfied in all the estimated regression models.
determinants of participation in non-farm employment (Abdulai The distribution of propensity scores before and after matching as
and Delgado, 1999; Barrett et al., 2001). These variables include shown in Fig. 1 clearly indicate that estimating the p-score appears
household characteristics such as age to capture experience, and to balance the treated and untreated groups extremely well, a result
education of the household head to present productivity potential, which underscores the relevance of the propensity score matching
household size, access to credit to capture liquidity constraints, approach.
dependency ratio, number of migrants from household, household
assets, presence of children, and location characteristics to capture
community fixed effects. 14
All specifications were estimated using the econometric software package STATA.
114 V. Owusu et al. / Food Policy 36 (2011) 108–118

Male’s non-farm employment participation


Unmatched Matched Unmatched Matched

0 .2 .4 .6 .8 1 0 .2 .4 .6 .8
0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 Propensity Score Propensity Score
Propensity Score Propensity Score
Untreated Treated: On support Untreated Treated: On support
Untreated Treated Untreated Treated Treated: Off support Treated: Off support

Impact on MSFC Impact on DHC16


Female’s non-farm employment participation
Unmatched Matched Unmatched Matched

0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 1 .2 .4 .6 .8 1
Propensity Score Propensity Score Propensity Score Propensity Score
Untreated Treated: On support Untreated Treated: On support Untreated Treated: On support Untreated Treated: On support

Treated: Off support Treated: Off support Treated: Off support Treated: Off support

Impact on MSFC Impact on DHC16


MSFC is food security where the household does not mortgage its standing field crops for current consumption.
DHC16 denotes food insecurity where household harvested crops declines during critical periods of food shortages.
Fig. 1. Distribution of propensity scores for unmatched and matched samples.

Table 3
Treatment effects and sensitivity analysis.

Outcome indicators Caliper PSM Critical value of Treated Control


C (hidden bias)
ATE ATU ATT On-support Off-support On-support Off-support
Males HHINC 0.5 3.2461** 3.3652*** 3.4679** 1.65 90 – 210 
(2.11) (2.60) (2.13)
MSFC 0.5 0.1839** 0.1722** 0.2115*** 1.75 90 – 210 
(2.04) (2.27) (3.065)
DHC16 0.002 0.0825* 0.0417** 0.1717** 2.05 26 64 210 
(1.67) (2.02) (2.05)
Females HHINC 0.5 2.4479** 1.7010*** 2.6586** 2.95 234 – 66 
(2.10) (2.46) (2.15)
MSFC 0.005 0.0657** 0.0599** 0.0673** 2.00 44 190 66 
(1.98) (2.00) (1.99)
DHC16 0.002 0.3000 0.2500 0.3333** 1.25 20 214 66 
(0.88) (0.95) (2.07)

Notes: t-statistics are in parentheses.


Critical values of C are presented for average treatment effects on the treated (ATT).
Outcome indicators:
HHINC denotes household income.
MSFC is food security where the household does not mortgage its standing field crops for current consumption.
DHC16 denotes food insecurity where household harvested crops declines during critical periods of food shortages.
*
Denotes significant at 10%.
**
Denotes significant at 5%.
***
Denotes significant at 1%.

The results of the treatment effects (ATE, ATT and ATU) for both robust to the matching algorithm employed in the analysis, we also
male-headed and female-headed households, all estimated by the employed nearest neighbor and kernel based matching models. The
caliper matching method are presented in Table 3, while the indi- results from these estimators, which are presented in Table 3A in
cators of matching quality are provided in Table 4. All the matching the appendix, generally indicate that our results are robust to the
techniques produced consistent estimates of the treatment effects matching algorithm employed. Focusing first on the ATT in Table 3,
of participating in non-farm work. To ascertain if the results are the matching estimates reported in Table 3 shows that for both
V. Owusu et al. / Food Policy 36 (2011) 108–118 115

Table 4
Indicators of matching quality before matching and after matching.

Outcome indicator Caliper Pseudo-R2 Pseudo-R2 p-Valuea p-Valuea Mean absolute bias Mean absolute Absolute bias
(unmatched) (matched) (unmatched) (matched) (unmatched) bias (matched) reduction
Males HHINC 0.5 0.3171 0.0521 0.0000 0.1794 18.2501 5.3513 70.6781
MSFC 0.5 0.3184 0.0525 0.0000 0.1907 18.3530 6.0132 67.2359
DHC16 0.002 0.3161 0.0422 0.0000 0.1654 14.5946 7.5696 48.1341
Females HHINC 0.5 0.2884 0.0237 0.0006 0.4416 21.7627 4.8733 77.6069
MSFC 0.005 0.2885 0.0067 0.0009 0.2440 22.3500 10.8076 51.6438
DHC16 0.002 0.2960 0.0469 0.0004 0.1893 21.7247 7.5860 65.0811

Note: Pseudo-R2 from probit estimation indicates the goodness of fit or how well the regressors explain the probability to participate in an employment activity.
Standardized bias:
Unmatched, MABUM ¼ 100  ðZ ~ 0 Þð1 ðV 1 ðZÞ þ V 0 ðZÞÞÞ12 ; Matched, MABM ¼ 100  ðZ
~1  Z ~ 0 Þð1 ðV 1M ðZÞ þ V 0M ðZÞÞÞ12 .
~1  Z
2 2

Total bias reduction: BRT = 100(1  MABM/MABUM).


a
p-Value of likelihood ratio test (Pr > v2).

female-headed and male-headed households, participation in non- Table A2


farm employment exerts positive and significant impacts on Probit estimates of propensity score for female’s non-farm employment participation.
household income and food security status, indicating that
Variable Coefficient Standard error z-Value Probability
participation in non-farm work helped raise households’ income
and thereby increasing their probability of not mortgaging Household characteristics
FMAGE 0.1231 0.097 1.27 0.205
their standing crops for current consumption (MSFC). Specifi- FMAGE2/100 0.1727 0.109 1.58 0.113
cally, the ATT effects of 3.46 for males and 2.65 for females FMEDUC 0.2321 0.137 1.70 0.090
indicate that males’ participation result in an increase in household PRCLS3YR 0.7467 0.387 1.93 0.054
income by about ¢3467,900 (US$367), while the females’ participa- HHSIZE 0.1178 0.046 2.57 0.010
DEPRATIO 0.6659 0.298 2.24 0.025
tion tend to increase household income by about ¢2658,600
MIGRATN 0.3978 0.143 2.78 0.005
(US$281). GPMEMB 0.0671 0.392 0.17 0.864
The effect of non-farm employment participation on the decline CDACCES 0.3960 0.237 1.67 0.093
of household food stocks during critical periods of food shortages
(DHC16) was found to be negative and statistically significant, Household assets
again implying that the increased household income from non- TRACTA 0.3058 0.799 0.38 0.702
farm work helped in lowering the probability of household food IFCATTLE 0.0624 0.383 0.16 0.871
IFGOAT 0.5747 0.459 1.25 0.211
IFGNFWL 0.2027 0.329 0.62 0.538
IFBIKE 0.2366 1.330 0.18 0.859
Table A1
Probit estimates of propensity score for male’s non-farm employment participation. Location characteristics
DBUNGLUN 1.0952 0.843 1.30 0.194
Variable Coefficient Standard error z-Value Probability
DTARIPKA 0.1858 0.682 0.27 0.785
Household characteristics DFAZIHIN 0.8312 0.696 1.19 0.232
MAGE 0.0937 0.092 1.02 0.307 DLALLIGU 1.7368 0.810 2.14 0.032
MAGE2/100 0.1076 0.085 1.27 0.205 DKANSHEGU 2.4083 0.843 2.86 0.004
MEDUC 0.1217 0.044 2.78 0.005 DCHALLAM 0.2750 0.670 0.41 0.681
PRCLS3YR 0.3276 0.377 0.87 0.385 DTIBALI 1.0949 0.711 1.54 0.124
HHSIZE 0.0417 0.035 1.20 0.230 DKADIA 0.2388 0.699 0.34 0.733
DPRATIO 0.1304 0.270 0.48 0.629 DZION 0.9787 0.753 1.30 0.194
MIGRATN 0.2776 0.130 2.14 0.032 CONSTANT 1.9570 2.595 0.75 0.451
GPMEMB 0.5806 0.334 1.74 0.082
Pseudo-R2 0.3332
CDACCES 0.5200 0.315 1.65 0.094
Log likelihood 52.7050
Observations 300
Household assets
TRACTA 1.2874 0.642 2.01 0.045
IFCATTLE 0.9548 0.388 2.46 0.014
IFGOAT 0.2352 0.395 0.60 0.551
IFGNFWL 0.4329 0.339 1.28 0.202 stocks declining during periods of critical food shortages. The
IFBIKE 0.7021 0.926 0.76 0.448 ATT estimate of 0.212 for food security indicator MSFC suggests
that males’ participation in non-farm employment increases the
Location characteristics probability at which the households would not mortgage their
DTARIPKA 0.7173 0.738 0.97 0.331
standing field crops for current consumption by 21.2%. Thus,
DFAZIHIN 0.2732 0.740 0.37 0.712
DLALLIGU 1.7598 0.839 2.10 0.036 non-farm work appears to be crucial in raising the incomes of farm
DKPENDUA 1.7031 0.876 1.95 0.052 households and improving their food security status.
DKANSHEGU 0.0563 0.664 0.08 0.932 It is also worth mentioning that the estimates of ATT effects of
DCHALLAM 1.3495 0.801 1.68 0.092 participation in non-farm employment on all the outcome vari-
DTIBALI 1.4944 0.830 1.80 0.072
DKADIA 0.4178 0.682 0.61 0.540
ables for males were slightly higher than those of the females, sug-
DZION 0.0533 0.657 0.08 0.935 gesting that male participation tends to increase household
CONSTANT 3.5701 2.539 1.41 0.160 incomes higher than female participation. However, the fact that
2
Pseudo-R 0.3184 female participation results in increased household income and
Log likelihood 62.4525
improved food security status confirms the relevance of female’s
Observations 300
contribution in ensuring stable household food supplies in the
116 V. Owusu et al. / Food Policy 36 (2011) 108–118

Table A3
Sensitivity of matching algorithms on food security outcome variables.

Matching algorithms Males Females


ATT SE Off-support ATT SE Off-support
1. HHINC
Nearest-neighbor (1)a
norep common 3.3630* 1.851 16 0.2285 1.293 168
norep nocommon 3.8009** 1.649 – 0.2285 1.292 168
Kernel matchingb
epan nocommon trim (10) 3.2201* 1.942 14 1.9993* 1.190 34
epan common trim (10) 3.2201* 1.942 14 1.9993* 1.197 34
Radius matchingc
Caliper 0.5 3.4679** 1.627 – 2.6587** 1.237 –
2. MSFC
Nearest-neighbor (1)
norep common 0.1842** 0.092 16 0.1932** 0.099 168
norep nocommon 0.1778** 0.088 – 0.1932** 0.099 168
Kernel matching
epan nocommon trim(10) 0.1877* 0.098 14 0.1881* 0.100 34
epan common trim(10) 0.1877* 0.098 14 0.1881* 0.100 34
Radius matching
Caliper 0.5A 0.2115*** 0.069 – 0.0673** 0.034 190
3. DHC16
Nearest-neighbor (1)
norep common 0.0571 0.124 16 0.2658* 0.161 168
norep nocommon 0.0888 0.106 – 0.2658* 0.161 168
Kernel matching
epan nocommon trim (10) 0.0527 0.207 14 0.2535* 0.131 34
epan common trim (10) 0.0526 0.199 14 0.2535* 0.131 34
Radius matching
Caliper 0.002 0.1717** 0.084 64 0.3333** 0.161 214

Note:
Standard errors (S.E) are based on 200 bootstrap replications.
Off support indicates treated individuals discarded due to missing common support.
a
Nearest-neighbor (1-to-1 matching) algorithms are implemented with and without common support (common and nocommon), and without replacement (norep).
b
Kernel matching algorithms are implemented with kernel functions (epanechnikov), with and without common support (common and nocommon) and with the
additional imposition of a trimming level of 10%.
c
Radius matching with caliper levels (0.5 and 0.002) is implemented.
A
For female non-farm work participation, caliper level of 0.005 is implemented for outcome variable MSFC.
*
Significant at 10% level.
**
Significant 5% level.
***
Significant 1% level.

income diversification process of rural households in sub-Saharan neity would not alter the inference about the estimated treatment
Africa (Canagarajah et al., 2001; Gladwin et al., 2001). effects, suggesting that the findings are generally insensitive to
The magnitudes of the coefficients of the treatment effects indi- hidden bias.15 The indicators of matching quality presented in Table
cate that the Average Treatment Effects for the treated (ATT) are 4 shows substantial reduction in absolute bias for all the outcome
higher than the Average Treatment Effects for the Entire sample variables for both males and females. As indicated in the last column
(ATE) and the Average Treatment Effects for the Untreated (ATU). of the Table, the mean bias in the covariates Z after matching lies be-
This finding demonstrates that households that have a higher low the 20% level of bias reduction suggested by Rosenbaum and Ru-
probability of participating in non-farm work are able to obtain bin (1985). This indicates that the covariates were significantly
higher incomes and improve their food security status over and balanced as a result of the propensity score matching procedure. In
above those that are less inclined to participate in non-farm work. addition, the pseudo-R2’s after matching are fairly low with none
The caliper matching estimates of the ATU effect indicate that if of the F-statistics being significantly different from zero, suggesting
individuals who did not participate in non-farm work had actually that the overall results from the matching procedure are satisfactory
participated, their household income and food security status in balancing the covariates between the participants and non-partic-
would on average be lower than that of those who participated. ipants (Sianesi, 2004).
The implication here is that the income and food security gains
from participation in non-farm employment are slightly higher
for households with a higher probability of participating than Conclusion
households with slightly lower chances of participating in non-
farm employment. This paper has examined the impacts of non-farm work on farm
Also presented in Table 3 are the critical levels of gamma (C), at household income and food security status, using a sample of farm
which the causal inference of significant participation impact may households from ten communities in Northern Ghana. A propen-
be questioned. For example, the value of 1.65 for male participa- sity score matching model was employed to account for selection
tion implies that if households that have the same Z-vector differ bias that normally occurs when unobservable factors influence
in their odds of participation by a factor of 65%, the significance both participation in non-farm employment and outcomes such
of the participation on income may be questionable. The lowest
critical value of C = 1.25 and the highest critical value C = 2.95 15
DiPrete and Gangl (2004) have indicated that the Rosenbaum bounds are a worst-
clearly indicate that even large amounts of unobserved heteroge- case scenario.
V. Owusu et al. / Food Policy 36 (2011) 108–118 117

as household income and food security status. The paper addresses Davis, J., 2004. The Rural Non-farm Economy, Livelihoods and their Diversification:
Issues and Options. Natural Resource Institute, Chatham, UK.
gender heterogeneity by explicitly providing separate estimates for
Dehejia, R.H., Wahba, S., 2002. Propensity score matching methods for non
males and females. experimental causal studies. Review of Economics and Statistics 84 (1), 151–
Results of the propensity score matching show that non-farm 161.
work exerts a positive and statistically significant effect on house- Deininger, K., Olinto, P., 2001. Rural nonfarm employment and income
diversification in Columbia. World Development 29 (3), 455–465.
hold income and food security status. These findings are generally Devereux, S., 2008. Linking social protection and support to small farmer
consistent with the widely held view that income from non-farm development. Report of a Workshop Held at Food and Agriculture
work is crucial to food security and poverty alleviation in rural Organization of the United Nations (FAO), Rome, on 17–18 January,
2008.
areas of developing countries. The estimates revealed that not- Dickson, K.B., Benneh, G., 1988. A New Geography of Ghana. Longman, London.
withstanding the fact that participation rates of females was higher DiPrete, T., Gangl, M., 2004. Assessing bias in the estimation of causal
than males, participation of males did contribute to higher incomes effects: rosenbaum bounds on matching estimators and instrumental
variables estimation with imperfect instruments. Sociological Methodology
and better food security status of households compared to their fe- 35.
male counterparts. As argued by Reardon (1997), women generally Faltermeier, L., Abdulai, A., 2009. The impact of water conservation and
appear to be limited to the low-wage activities in the non-farm intensification technologies: empirical evidence for rice farmers in Ghana.
Agricultural Economics 40 (3), 279–365.
sector, resulting in lower earnings that tend to have lower impacts FAO, 1998. The State of Food and Agriculture: Rural Non-farm Income in
on improving food security. Developing Countries. Food and Agriculture Organization of the United
The results from this study indicate that the growing interest of Nations, Rome.
Gladwin, C.H., Thomson, A.M., Paterson, J.S., Anderson, A.S., 2001. Addressing food
policy makers in promoting non-farm activities, particularly in rur-
security in Africa via multiple livelihood strategies of women farmers. Food
al areas of developing countries is in the right direction. Besides Policy 26 (2), 177–207.
being a valuable source of income for rural households in develop- Haggblade, S., Hazell, P., Reardon, T., 2002. Strategies for stimulating
ing countries, non-farm work also helps in smoothing incomes, poverty-alleviating growth in the rural non-farm economy in developing
countries. EPTD Discussion Paper 92. International Food Policy Research
which in turn smoothens consumption over long periods of time. Institute, Environment and Production Technology Division, Washington,
Given that women normally face entry barriers to participation DC.
in non-farm work, which exerts a positive and robust effect on Heckman, J., LaLonde, R., Smith, J., 1999. The economics and econometrics of active
labor market programs. In: Ashenfelter, O., Card, D. (Eds.), Handbook of Labor
household income and food security, policy measures could target Economics, vol. III. Elsevier, Amsterdam, pp. 1865–2097.
them to lower the entry barriers into non-farm activities. In partic- Hirano, K., Imbens, G.W., 2001. Estimation of causal effects using propensity score
ular, food safety net policies should generally pay more attention weighting: an application to data on right heart catheterization. Health Services
and Outcomes Research Methodology 2 (3–4), 259–278.
to the factors that allow increased participation in higher earning Holden, S., Shiferaw, B., Pender, J., 2004. Non-farm income, household welfare, and
activities by women, since they could then build on existing pat- sustainable land management in a less-favoured area in the Ethiopian
tern of addressing food security. Highlands. Food Policy 29, 369–392.
Huffman, W.E., 1991. Multiple jobholding among farm families. In: Findeis, J.L.,
Food security policies go beyond just food production measures, Hallberg, M.C., Lass, D.L. (Ed.), Agricultural Households Survey and Critique.
and also include measures that help in generating adequate levels Iowa State University Press, Ames IA, USA (Chapter 5).
of effective demand via income growth or transfers policies. Partic- Hujer, R., Caliendo, M., Thomsen, S.L., 2004. New evidence on the job creation
schemes in Germany – a matching approach with threefold heterogeneity.
ularly in rural areas, enhancing the ability of farm households to
Research in Economics 58, 257–302.
participate in non-farm work is a reasonable strategy to improve Imbens, G.W., Wooldridge, J.M., 2009. Recent developments in the econometrics of
food security in these areas. Promising policy measures that can program evaluation. Journal of Economic Literature 47 (1), 5–86.
help boost non-farm work availability include increasing the ac- Jalan, J., Ravallion, M., 2003. Estimating the benefit incidence of an antipoverty
program by propensity-score matching. Journal of Business and Economic
cess of rural households to assets such as financial capital and Statistics 21 (1), 19–30.
non-price factors such as education and infrastructure. However, Lanjouw, P., 2001. Nonfarm employment and poverty in rural El salvador. World
it is significant to mention that development of non-farm activities Development 29 (3), 529–547.
Lanjouw, P., Feder, G., 2001. Rural Nonfarm Activities and Rural Development: from
should actually complement the effort to develop agriculture, since Experience Towards Strategy. Rural Development Strategy Background Paper
activities in the former depend directly or indirectly on the latter. Number 4. The World Bank, Washington, DC.
Lee, W., 2008. Propensity score matching and variations on the balancing test. In:
3rd Conference on Policy Evaluation, ZEW, Mannheim (Germany), 27–28
Appendix October.
Millimet, D. L., Tchernis, R. (2008). Minimizing bias in selection on observables
estimators when unconfoundness fails. In: The Center for Applied Economics
See Tables A1–A3. and Policy Research (CAEPR) Working Paper No. 2008-008.
MoFaEP (2003). The Growth and Poverty Reduction Strategy Paper II (GPRSPII).
Norcliffe, G., 1983. Operating characteristics of rural nonfarm enterprises in central
References province, Kenya. World Development 11 (11), 981–994.
Profile, District., 2006. District profile of Savelugu-Nanton district of Northern
Ghana. District Assembly, Savelugu-Nanton district.
Abdulai, A., CroleRees, A., 2001. Determinants of income diversification amongst
Reardon, T., 1997. Using evidence of household income diversification to inform
rural households in Southern Mali. Food Policy 26, 437–452.
study of the rural nonfarm labor market in Africa. World Development 25, 735–
Abdulai, A., Delgado, C.L., 1999. Determinants of non-farm earnings of farm based
747.
husbands and wives in Northern Ghana. American Journal of Agricultural
Reardon, T., Crawford, E., Kelly, V., 1994. Links between nonfarm income
Economics 81, 117–130.
and farm investment in African households: adding the capital market
Ashong, K., Smith, D.R., 2001. Livelihoods of the poor in Ghana: a contextual review
perspective. American Journal of Agricultural Economics 76, 1172–
of Ghana-wide definitions and trends of poverty and the poor with those of
1176.
peri-urban Kumasi. Kumasi and London, Centre for Development of People and
Reardon, T., Stamoulis, K., Balisacan, A., Cruz, M.E., Berdegue, J., Banks, B., 1998.
University of Greenwich, Natural Resources Institute. <www.livelihoods.org>.
Rural nonfarm income in developing countries. In: The State of Food and
Barrett, C.B., Reardon, T., Webb, P., 2001. Nonfarm income diversification and
Agriculture 1998. Food and Agricultural Organization of the United Nations,
household livelihood strategies in rural Africa: concepts, dynamics and policy
Rome.
implications. Food Policy 26, 315–331.
Reardon, T., Berdegué, J., Escobar, G., 2001. Rural nonfarm employment and incomes
Becker, S.O., Ichino, A., 2002. Estimation of average treatment effects based on
in Latin America: overview and policy implications. World Development 29 (3),
propensity scores. Stata Journal 2, 358–377.
411–425.
Caliendo, M., Kopeinig, S., 2005. Some practical guidance for the implementation of
Rosenbaum, P.R., 2002. Observational Studies. Springer, New York.
propensity score matching. Journal of Economic Surveys 22 (1), 31–72.
Rosenbaum, P.R., Rubin, D., 1983. The central role of the propensity score in
Canagarajah, S., Newman, C., Bhattanmishra, R., 2001. Non-farm income, gender,
observational studies for causal effects. Biometrika 70, 41–45.
and inequality: evidence from rural Ghana and Uganda. Food Policy 26 (4), 405–
Rosenbaum, P.R., Rubin, D., 1985. Constructing a control group using multivariate
420.
matched sampling methods that incorporate the propensity score. The
Chang, H.H., Mishra, A., 2008. Impact of off-farm labor supply on food expenditures
American Statistican 39, 33–38.
of farm households. Food Policy 33, 657–664.
118 V. Owusu et al. / Food Policy 36 (2011) 108–118

Rukuni, M., 2002. Africa: addressing growing threats to food security. Symposium Stamoulis, K., Zezza, A. (2003). A conceptual framework for national agricultural,
on feeding the world in the coming decades. The Journal of Nutrition 132, rural development, and food security strategies and policies. ESA Working
3443S–3448S. Paper No. 03–17, November 2003. Agricultural and Development Economics
Shepherd, A., Gyimah-Boadi, E., Gariba, S., Plagerson, S., Musa, A.W., 2004. Bridging Division, The Food and Agriculture Organization of the United Nations.
the North–south Divide in Ghana. Background Paper for WDR 2006 on Equity <www.fao.org/es/esa>.
and Development. Titus, M.A., 2007. Detecting selection bias using propensity score matching, and
Sianesi, B., 2004. Evaluation of the active labor market programmes in Sweden. The estimating treatment effects: an application to the private returns to a Master’s
Review of Economics and Statistics 86 (1), 133–155. degree. Research in Higher Education 48 (4), 487–521.
Smith, J., Todd, P., 2005. Does matching overcome LaLonde’s critique of non- Whitehead, A., 2006. Persistent poverty in North East Ghana. Journal of
experimental estimators? Journal of Econometrics 1251, 1252. Development Studies 42 (2), 278–300.
Smith, L. C., Wiesmann, D., 2007. Is food insecurity more severe in South Asia or Winship, C., Morgan, S.L., 1999. The estimation of causal effects from observational
Sub-Saharan Africa? A comparative analysis using household expenditure data. Annual Review of Sociology 25, 659–706.
survey data. International Food Policy Research Institute (IFPRI) Discussion
Paper 00712, August 2007.

S-ar putea să vă placă și