166
(3) The incidence of agricultural development indicators is higher where households claim to be
the owners of the farmland.
These hypotheses were tested using regression analysis. Due to relationships among the mode
of land acquisition, land rights, and landownership perceptions, the hypotheses were tested in
alternative regression models. In each model, other nontenure factors were included and were selected
on the basis of their likely links to investment. These include size of farm, size of household, age of
household head, education level of household head, frequency of extension visits, incidence of
livestock, quality of residential house, location of household within study site, and distance to roads,
markets, and agricultural offices. These variables were selected because they are hypothesized to affect
the agricultural decisions of farmers. A greater likelihood of investment is expected on larger farms,
those with more labor, among middle-aged or elderly household heads, among the more educated,
those who are in more contact with extension agents, those with greater wealth (livestock, quality of
house), and those who are located closer to markets and/or major roads.
It is certain that decisions regarding the various productivity measures are mutually dependent.
However, this decision-making process is quite difficult to handle from a statistical point of view. The
inclusion of several dependent variables in a simultaneous system requires a large number of
observations, while only 100 per region are available in these samples.'' One method is to use all
possible investment combinations or bundles as distinct outcomes.' However, this too requires
significant numbers of observations to assure that the number of cases in each investment combination
is sufficiently large. An alternative method was used instead. The 11 different improvements
enumerated were grouped into 5 categories: organic inputs, chemical inputs, earthwork structures,
water structures, and fencing (see table 5.8). Nearly all farmers used chemical inputs (mainly
fertilizer) so there was insufficient variation for explanatory analysis. Single equation logit regressions
were made using credit use, presence of oxen, use of organic inputs, incidence of earthwork
structures, water structures, fencing, fruit tree plantings, and multipurpose tree (i.e., nonfruit)
plantings as dependent variables. This grouping does not eliminate the problems of dependent decision-
making, but reduces them to some extent since the more important interlinked decisions are likely
within groups (e.g., wells and irrigation in water structures).'
Before proceeding to the empirical results, it should be stressed that the regression analyses
indicate associations with variables but should not be taken to imply causal relationships. The reason
for this has to do with the aforementioned time frames involved. The actual date of investment is
unknown and, therefore, one cannot be certain that any of the independent variables actually caused
the outcome (except for variables which are fixed at the time of land acquisition such as how the land
was acquired). Furthermore, it is important to point out that because the independent variables reflect
current household characteristics (e.g., current size of household), their relationships to land
improvement variables can become obscured. Unfortunately, this problem is very difficult to correct
since it would require superlative recall powers on the part of respondents (i.e., to provide dates and
amounts for investments).
u Perhaps more constraining is the lack of software capability to handle large simultaneous systems given that the
dependent variable is not continuous.
u In a multinomial logit framework, for instance.
26 On the other hand, this method is sensible only if the combined improvements are alike and respond to other factors
in the same manner.