Destination or Distraction? Querying the Linkage between Off-farm Income and Farm Investments in Kenya
December 1, 2014 - Author: Melinda Smale, Yoko Kusunose, Mary K. Mathenge, and Didier Alia
IDWP 136. Melinda Smale, Yoko Kusunose, Mary K. Mathenge, and Didier Alia. 2014. Destination or Distraction? Querying the Linkage between Off-farm Income and Farm Investments in Kenya
Off-farm earnings account for a substantial and growing share of household income among
smallholder farmers in most of Sub-Saharan Africa, but evidence concerning the effects of
these earnings on investment in food production remains sparse. Conceptually, some
factors may push farm families to send members in search of cash to relieve expenditure
constraints or serve to meet consumption needs under duress; other factors pull members
of rural households toward the promise of steady, dependable income. Previously
published research suggests that the search for off-farm income has a negative impact on
In this study, we explore the relationships among three types of off-farm earnings (labor on
other farms, known as farm kibarua; income from self-employed businesses; and income
from salaries or wages, including remittances) and investment in fertilizer use in maize
production. We test the robustness of linkages by applying a range of econometric models,
utilizing panel data collected between 2000 and 2010 in four waves from a sample of
Kenyan smallholders. In particular, we hypothesis that as rural economies change with
economic development, family labor used in producing maize, the primary staple food,
could be drawn toward other sources of income because these are more remunerative—
diminishing its use in activities such as fertilizer application or weeding. Also, if income is
high enough in other activities, it may make more sense to buy than to produce maize.
Two features of the underlying relationship complicate the choice of econometric models.
First, not all farm households in the dataset earn income from off-farm sources, and many
farm households apply no fertilizer. Thus, both earnings and investment variables have a
large proportion of zeros. Second, there is reason to expect that farm families make their
decisions regarding labor allocation to farm and off-farm activities simultaneously. This
suggests the potential for endogeneity in the off-farm earnings and fertilizer use variables.
Various approaches have been recommended to address these problems, each with
advantages and disadvantages.
We explore and compare several of these methods, to gauge the robustness of findings.
Recent concerns about identification strategies and other shortcomings of non-linear
models lead us to estimate two-stage Fixed Effects Instrumental Variables (FEIV) as a
base case. We also estimate a seemingly unrelated, recursive probit model in which the
binary decisions to work off-farm and to apply fertilizer are simultaneously estimated. To
reflect the continuous nature of the variables of interest when values are observed above
zero, we then estimate a Tobit-Tobit specification in which off-farm income is first predicted
and then used to explain fertilizer application rates (an instrumented, Control Function
Approach or CFA). A Cragg model is also tested to reflect the notion that separate
underlying processes may shape the decisions to use fertilizer and the amount used. Finally,
we apply Generalized Propensity Score Matching (GPSM) to capture possible non-linearities
or threshold effects in the relationship between earned income levels and fertilizer applied.
In all three of the non-linear regression models, we employ the Mundlak-Chamberlain
technique (also known as Correlated Random Effects, or CRE model) to control for timeinvariant
unobserved effects that may be related to household decision-making. The outcome
of interest—fertilizer application rate—is measured in terms of N nutrient kgs per ha, which
has the double advantages of being a more precise measure of nitrogenous fertilizer
application and a universal measure that takes into account the many different combinations
(fertilizer formulae) through which nitrogen is applied by farmers surveyed.
The overall picture that emerges portrays the effects of non-farm income sources (business
and salary earnings) on fertilizer use in maize as consistently and strongly negative. This
suggests competition between farm family investments in maize production and nonfarm
sectors. At the same time, the relationship between fertilizer use in maize production and
earnings from labor on other farms (farm kibarua) is statistically weak though positive,
perhaps reflecting their minor importance in household income, their relative infrequency,
and the role they play in easing cash constraints for some households.
Comparisons also show sensitivity of some estimated parameters to modeling assumptions.
Application of the GPSM model adds to our understanding by demonstrating that the
magnitude of the marginal effects of non-farm income on fertilizer use rates varies as income