'
OEWEVi)
•""""
MIT LIBRARIES
ff-6S\
/V^^lS
no
3 9080 03317 5826
^- ^
(
Massachusetts Institute of Technology
Department of Economics
Working Paper Series
Nudging Farmers
to
Use
Fertilizer:
Theory and
Experimental Evidence from Kenya
Esther Duflo
Michael Kremer
Jonathan Robinson
Working Paper 09-19
June 26, 2009
Room
E52-251
50 Memorial Drive
Cambridge,
MA 02142
This paper can be downloaded without charge from the
Social Science Research Network Paper Collection at
http://ssrn.com/abstract=1 427446
Nudging Farmers
to
Use
Fertilizer:
Theory and Experimental Evidence from Kenya
Esther Duflo. Michael Kremer, and Jonathan Robinson'"
June 26, 2009
While many developing-country policymakers see heax'y fertilizer subsidies as critical to
raismg agricultural productivity, most economists see them as distortionary, regressive,
environmentally unsound, and argue that they result in politicized, inefficient distribution of
fertilizer supply. We model farmers as facing small fixed costs of purchasing fertilizer, and
assume some are stochastically present-biased and not fully sophisticated about this bias.
Even when
relatively patient, such fanners
until later periods,
many
when
they
may
may
procrastinate, postponing fertilizer purchases
be too impatient to purchase
fertilizer.
Consistent with the
Western Kenya fail to take advantage of apparently profitable
do invest in response to small, time-limited discounts on the
cost of acquiring fertilizer (free delivery) just after har\'est. Later discounts have a smaller
impact, and when given a choice of price schedules, many farmers choose schedules that
model,
farmers
in
fertilizer investments, but they
induce advance purchase. Calibration suggests such small, time-limited discounts yield
higher welfare than either laissez faire or heavy subsidies by helping present-biased fanners
commit
to fertilizer
overuse
fertilizer.
'
The authors
use without inducing those with standard preferences to substantially
are respectively
from
MIT
(Department of Economics and Abdul Latif Jameel Poverty Action
Lab (J-PAL)); Harvard, Brookings, CGD, J-PAL, and NBER; and
Ikoluot,
Edward Masinde, Chris Namulundu,
Evite Ochiel.
UC
Santa Cruz and .T-PAL.
We
thank John
Andrew Wabwire, and Samwel Wandera,
outstanding fieldwork, Jessica Cohen, Anthony Keats, Jessica Lemo,
Owen
Ozier, and Ian
Tomb
for
for excellent
research assistance, and International Child Support (Kenya), Elizabeth Beasley and David Evans for work
setting
up the
project.
We
thank Abhijit Banerjee for his persistent encouragement and
many extremely
helpful
conversations, as well as Orley Ashenfelter, Pascaline Dupas, Rachel Glennerster, and seminar participants
Arizona, Berkeley, Brown,
and the
"New
Products
in
EBRD, LSE, Nonhwestem, Pompeu
Microfmance" conference
at
UCSD
Fabra,
for
UCLA, UCSD, USC,
comments.
the
at
World Bank,
Digitized by the Internet Archive
in
2011 with funding from
Boston Library Consortium IVIember Libraries
http://www.archive.org/details/nudgingfarmerstoOOdufl
"The
of the world
rest
to
it
is
fed because of the use of good seed and inorganic fertilizer, full
has not been used
stop. This technolog}'
get access
is
give
it
in
most of Africa. The only way you can help farmers
away free or subsidize
it
"
heavily.
-
<-
•
--
Stephen Can; former World Bank specialist on Sub-Saharan Afi'ican agriculture, quoted
Many
-...•.-,
.-..,.,..-
Dugger,2007.
agricultural experts see the use of
,,,..
modem
.
,
,
.,,
,.-........'
,
,
inputs, in particular fertilizer, as the
agricultural productivit}'. Pointing to the strong relationship
in test plots,
......
between
fertilizer
in
key
to
use and yields
they argue that fertilizer generates high returns and that dramatic growth in
agricultural yields in Asia
and the stagnation of yields
in
'
;
Africa can largely be explained
by
increased fertilizer use in Asia and continued low use in Africa (Morris, Kelly, Kopicki, and
Byerlee, 2007). Based on this logic, ElHs
subsidies.
Many governments
amounted
fertilizer subsidies
( 1
992) and Sachs (2004) argue for
have heavily subsidized
to 0.75
percent of
fertilizer
fertilizer. In India, for
GDP m
•'
example,
•
1999-2000 (Gulati and Narayanan,
2003). In Zambia, fertilizer subsidies consume almost 2 percent of the government's budget
(World Development Report. 2008).
In contrast, the
presumption
that
levels.
'
.
tradition associated with Schultz (1964) starts with the
farmers are rational profit maximizers, so subsidies will distort
away from optimal
beyond
Chicago
'
Others have argued that
these Harberger triangles.
They
fertilizer
use
fertilizer subsidies create large costs
are r)'pica!!y regressive as wealthier farmers and
those with more land often benefit most from subsidies (Donovan, 2004), and loans for
fertilizer often
moderate
go
to the politically
fertilizer
use
is
may
rates.
Moreover, while
environmentally appropriate, overuse of fertilizer induced by
subsidies can cause environmental
subsidies
connected and have low repayment
damage (World Bank, 2007). Furthermore,
fertilizer
lead to government involvement in fertilizer distribution, politicization, and
very costly failures
to
Partly due to the
supply the right kind of fertilizer
at the right time.
dominance of the anti-subsidy view among economists and
international financial institutions, fertilizer subsidies have been rolled back in recent
decades. Recently, however, they have seen a resurgence. For example, after Malawi's
removal of fertilizer subsidies was followed by
subsidy on
fertilizer.
a
famine, the country' reinstated a two-thirds
This was followed by an agricultural
boom which many,
including
Jeffrey Sachs, attribute to the restoration of the fertilizer subsidies (Dugger, 2007).
'•
A
key assumption
would use
in the
Chicago
tradition case against fertilizer subsidies
the privately optimal quantity of fertilizer without subsidies.
from
To
is
that
reconcile low
use found in agricultural research
fertilizer
use with the large increases
stations,
economists often note that conditions on these stations differ from those on
may
world farms, and returns
be
in yield
much lower
fertilizer
in real conditions,
other inputs optimally. There
is
irrigation, greater attention to
weeding, and other changes
may have
trials
evidence that
difficulty in implementing.
with farmers on their
low. Those
trials
showed
of 36 percent over
own farms
that
when
in a
used
we implemented
Kenya where
in limited quantities,
season on average, which translates
a
in agricultural practice that
region of Western
fertilizer is
where farmers cannot use
previous work
in
real-
complementary with improved seed,
fertilizer is
However,
fanners
a series
fertilizer
it
fanners
use
of
is
generates returns
70 percent on an annualized basis
to
(Duflo, Kremer, and Robinson, 2008), even without other changes in agricultural practices.
Low
is
investment rates
in the face
well-known and long-used
m
of such high returns are particularly puzzling since fertilizer
the area.
Moreover, since
fertilizer is divisible,
standard
theory would not predict credit constraints would lead to low investment traps in this
context.' There could of course be fixed costs in
example, making a
important role
in
trip to the store).
buying or learning
use fertilizer (for
to
Indeed, small fixed costs of this type will play an
our model. However, such costs would have to be implausibly large to
justify the lack of fertilizer investment in the standard model."
In this paper
we
argue that just as behavioral biases limit investment
financial investments in pension plans
and Madrian, 2008), they
developing countries.
We
may
by workers
in the
United States
simple model of biases
In the
model some farmers are
when
odds
'
As discussed below,
profits are
fertilizer to
literature (see
that they will
they are patient today. Going
and perhaps deciding what type of
by farmers
in
O'Donoghue
(stochastically) present-biased and at least
partially naive, systematically underestimating the
future, at least in the case
Choi, Laibson
farmer decision-making inspired
in
by models of procrastination from the psycholog\' and economics
and Rabin, 1999).
(e.g.,
limit profitable investments in fertilizer
set out a
in attractive
use and
concave rather than convex
to the store,
how much
in fertilizer
since farmers always have the option of applying fertilizer intensely on
be impatient
to
buying
in the
fertilizer,
buy, involves a
utility
use per unit of land area. Moreover,
some
land while
leavmg other pieces of
land unfertilized, returns must be non-increasmg.
"
For instance, consider
fertilizer takes
a
farmer with an hourly wage of SO.
one hour and
who
can only
initially afford $1
1
3
over for
worth of
whom
round
fertilizer.
trip travel to
town
to
buy
Since half a teaspoon of top
Even
cost.
if this
cost
is
small, so long as farmers discount future utility, even farmers
plan to use fertilizer will choose to defer incurring the cost until the
they expect to
still
being impatient
be willing
to
purchase the
period
in the last
in
fertilizer later.
which buying
is
last
moment
who
possible, if
However, farmers who end up
possible will then
fail to
invest in
fertilizer altogether.
Under
the model,
heavy subsidies could induce
fertilizer
use by stochastically hyperbolic
farmers, but they also could lead to overuse by farmers without time consistency problems.
The model implies
that if offered just after har\'est
(when fanners have money)
small, time-
limited discounts on fertilizer could induce sizeable changes in fertilizer use. In particular,
early discounts of the
same order of magnitude
purchase can induce the same increase
m
fertilizer
of magnihide of the out-of-pocket costs of
(before the harvest)
to
have an option
to
as the psychic costs associated with fertilizer
to
Moreover, ex ante
to fertilizer use.
(NGO), we designed and
tested a
randomized design, we compared the program
fertilizer subsidies or
larger discounts of the order
be eligible for the discount early on, so as
(Kenya) a non-government
In collaboration with International Child Support
organization
much
fertilizer later in the season.
some farmers would choose
commit
use as
reminders
to
use
program based on these predictions. Using
to alternative inter\'entions,
fertilizer.
The
such as standard
results are consistent with the
model.
Specifically, offering free delivery to farmers early in the season increases fertilizer use
to
60 percent. This effect
subsidy on
is
a
by 46
greater than that of offering free delivery, even with a 50 percent
fertilizer, later in the season.
Following an approach similar
to
O'Donoghue and Rabin
(2006).
we
use the model to
analyze the impact of different policies depending on the distribution of patient, impatient,
and stochastically present-biased farmers. Calibrations based on our empirical
that 71 percent
of farmers are stochastically present-biased, 16 percent are always patient,
and 13 percent are always impatient. This yields
farmers should never use
fertilizer in the three
a prediction that
seasons
percent of comparison farmers do not use fertilizer
have
data.
The
calibrated
proportion of farmers
like to
results suggest
be offered free
in
we
take up fertilizer
follow them. Empirically, 52
any of the three seasons for which
model matches other moments
who
roughly 55 percent of
when given
we
in the data, in particular the
the choice of
which date they would
fertilizer delivery.
dressing fertilizer yields returns of 36 percent over a season, netting out the
lost
wages would leave
the famier
with a 23 percent rate of return over a few months.
3
The
calibration suggests that a "paternalistic libertarian" (Thaler and Sunstein, 2008)
approach of small, time-limited discounts could yield higher welfare than either laissez faire
policies or
heavy subsidies, by helping stochastically hyperbolic farmers commit themselves
to invest in fertilizer
m fertilizer use
while avoiding large distortions
among
time-consistent
farmers, and the fiscal costs of heav>' subsidies.
The
rest
of the paper
on agriculture and
is
strucmred as follows: Section 2 presents background information
fertilizer in
Western Kenya. Section
testable predictions. Sections 4 lays out the
results,
presents the model and derives
3
program used
to test the
model; Section 5 reports
and Section 6 calibrates the model and then uses the calibrated model
welfare under laissez
faire,
compare
heavy subsidies, and small time-limited subsidies. Section
examines alternative hypotheses, and Section
for realistically scaling
to
8
7
concludes with a discussion of the potential
up small, time-limited subsidies
in a
way
that
would not involve
excessive administrative costs.
2.
Background on
Our study
area
is
Fertilizer use in
Western Kenya
a relatively poor, low-soil fertility area in
Western Kenya where most
farmers grow maize, the staple food, predominantly for subsistence. Most farmers buy and
sell
maize on the market, and store
it
at
home. There
are
two
agricultural seasons each year,
the "long rains" from March/April to July/August, and the less productive "short rains"
July/ August until December/Januar}'.
;,
.
from
;.
Based on evidence from experimental model fanns (see Kenyan Agricultural Research
Institute, 1994), the
seeds.
Di-Ammonium Phosphate (DAP)
Nitrate
one
to
Kenyan Ministry of Agriculture recommends
(CAN)
fertilizer at top
two months
(and occasionally
typical farmer
is
in local
a
and Calcium
fertilizer at planting,
dressing (when the maize plant
is
Ammonium
knee-high, approximately
after planting). Fertilizer is available in small quantities at
would need
Although there
that farmers use hybrid
market centers
shops outside of market centers). Our rough estimate
to
walk for roughly 30 minutes
market for reselling
fertilizer,
to
is
that the
reach the nearest market center.
not very' liquid and resale involves
it is
substantial transaction costs.''
Experiments on actual farmer plots suggest low, even negative returns
combination of hybrid seeds and
"
fertilizer at planting
and top dressing, (Duflo, Kremer, and
Discussions with people familiar with the area suggest reselHng
approximately 20 percent of the cost of
to the
fertilizer typically
fertilizer in addition to the search costs
involves a discount of
of finding a buyer.
4
Robinson, 2008), although
it
is
plausible that returns might be higher if farmers changed
other fanning practices. Similarly, the use of a
dressing
is
full
teaspoon of
not profitable, because farmers realize large losses
and seeds do not germinate. However,
teaspoon of
fertilizer
just
more conservative
per plant as top dressing, after
it is
when
per plant as top
rains fail or are
delayed
strategy of using only one half
clear that seeds have germinated,
much of the downside
yields a high return and eliminates
sample plants
a
fertilizer
risk.
The average farmer
in
our
under one acre of maize. Using one half teaspoon of fertilizer per plant
increases the yield by about $54 per acre and costs $40 per acre, a 36 percent return over the
several
basis)
months between
the application of fertilizer and harvest (70 percent
on real-world farms even
on an annualized
absence of other complementary changes
in the
in
farmer
behavior.
The incremental
valued
approximately $18 per acre, corresponding to a negative return of around -55
percent
at
at full price,
return to the
first
yield associated with the second half teaspoon of fertilizer
but a 30 percent return under a two-thirds subsidy, very close to the
half teaspoon at
full price.
However, despite these large potential returns
as top dressing, only
40 percent of farmers
and only 29 percent report using
program.''
When
asked
why
in
in at least
it
they do not use
unprofitable, unsuitable for their
our sample report ever having used fertilizer
one of the two growing seasons before the
farmers rarely say fertilizer
trials
we
money
to
purchase
it.
Of farmers
conducted, only 9 percent said that
unprofitable while 79 percent reported not having enough money. At
to take at face value: fertilizer
can be bought
proceeds would eventually yield sufficient money
Even poor farmers could presumably
from consumption
One way
to fertilizer
to reconcile farmers"
is
first this
interviewed
was
fertilizer
seems
difficult
small quantities (as small as one kilogram)
in
and with annualized returns of 70 percent, purchasing
har\'esl
fertilizer
or too risky: instead, they overwhelmingly reply that
soil,
before the small-scale agricultural
applying limited quantities of
to
fertilizer,
they want to use fertilizer but do not have the
entire plot.
is
to
a small
amount and
in\'esting the
generate sufficient funds to fertilize an
reallocate
some of the proceeds of their
investment per acre.
claims that they do not have
money
to
buy
fertilizer
with
the fact that even poor fanners have resources available at the time of han'est
is
may
follow through
initially intend to
on those
save in order
to
purchase
plans. In fact, 97.7 percent of fanners
These figures
differ slightly
from those
ir.
fertilizer later but
who
then
fail to
that farmers
participated in the demonstration plot
Duflo, Kremer. and Robinson (2008) because the sample of famiers
differs.
5
program reported
that they
planned
to
use
fertilizer in the
following season. However, only
36.8 percent of them actually followed through on their plans and used fertilizer in the season
which they said they would. Thus,
in
use
have no money
fertilizer often
appears that even those
it
to invest in fertilizer at the
planting or top dressing, several months
who
time
are initially planning to
it
needs
to
be applied, for
later.
Model
3.
Below we propose
of many workers
in
a
model of procrastination similar to those advanced
developed countries
to take
(O'Donoghue and Rabin, 1999) and derive
to
explain the failure
advantage of profitable financial investments
testable predictions. In the model,
are present-biased, with a rate of time preference that
is
When
all
they are very present-biased, fanners
consume
moderately present-biased, farmers make plans
to
use
some farmers
realized stochastically each period.
they have.
fertilizer.
When
But early
they are
in the season,
patient farmers overestimate the probabilit)' that they will be patient again, and thus they
postpone the purchase of fertilizer
to
a
be impatient, they consume
all
until later,
in cash instead. Later, if they turn
of their savings instead of investing in
lower usage of fertilizer than the farmer
3.1
and save
in the early
out
fertilizer, resulting in
period would have wanted.
Assumptions
Preferences and Beliefs
Suppose
that
some
fraction of farmers
exponentially discount the future
A
proportion
d) is
').
are patient.
They
are time consistent
and
at rate pi^.
(stochastically) present-biased, and systematically understate the
extent of this present bias. In particular suppose that in period k\ these farmers discount every
future period at a stochastic rate pk (for simplicity
between future periods).
=
In
each period
-
/.-,
we assume
that there
is
with some probability p, the fanner
fanner
is
quite impatient {6k
Furthermore, while fanners do recognize that there
is
a
(/3i.
pfi),
and with probability
(
1
no discounting
p), the
=
is
fairly patient
Pl).
chance that they will be impatient
in
the future, they overestimate the probability that they will be patient. Specifically, the
probability that a patient farmer believes that she will
There are several ways
still
to interpret this stochastic rate
be patient
in the future is
of discount.
One
p
>
p.
interpretation
is
that farmers are literally partially naive about their hyperbolic discounting, as in the original
O'Donoghue and Rabin (1999) framework. An
Banerjee and Mullainathan (2008a),
(e.g., a party) that is
farmer
A
tempting
that a
is
farmer
to the
of
alternative interpretation, along the lines
consumption opportunity occasionally
in that period, but
which
arises
not valued by the
is
in other periods.
final
proportion
one of these three
i"
are alwcn's impatient so that
tj^pes so
Finally, for simplicity',
^
+
+
(p
=
V'
we assume
(3k
=
Pl
in all periods. All
farmers are
1-
per-period
in that period, less a small utility cost associated
utility' in
any period
is
simply consumption
with shopping for fertilizer and the time cost
associated with deciding what quantity of fertilizer to buy, which will be described below.
Timing and Production
There are four periods. Period
to save,
which
consume
pre-commit
fertilizer in
period
period
those
who have
will later consider a situation
fertilizer pricing in this period.
make
m
We
choice of a pnce
a
0.
between consumption, purchase of fertilizer
that yields liquid returns
of
we
but later allow the fanner .to
the farmer harvests maize, receives
7,
The farmer does not plan
to the har\'est.
to different patterns
from period
will initially abstract
In
immediately prior
or purchase fertilizer in this period, but
the farmer can
schedule for
is
by the time
income
.t
>
and can allocate income
2,
for the next season, and a short-run investment
fertilizer
needs
Some
be applied.
to
farmers, such as
shops where they can use more working capital, will have high return
investments that yield liquid returns over
return investment opportunities.
proportion A of farmers, and low
We
a short period,
therefore
{R >
assume
0) for the rest.
whereas others
R
the net return
Farmers know
will
is
have lower
high (/?) for a
their rate
of return with
certainty.
Fanners can choose
fertilizer
investment
to
to use zero,
keep the analysis tractable and
work, which examined the returns
However,
to zero, half or
to parallel
assume discreteness of
our previous empirical
one teaspoon of fertilizer per
plant.
the discreteness does not drive our results.
Let pfi denote the price of fertilizer
small
We
one or two units of fertilizer.
utilitv'
cost
period
1.
Purchasing any
/ (encompassing the time cost of going
well as deciding what type to use and
amount of fertilizer purchased. Note
assumption
in
that there
is
some
how much
to buy).
that while fertilizer
to the
shop
This cost
is
fertilizer also entails a
to
is
buy
the fertilizer, as
independent of the
a divisible technology, the
fixed cost of shopping for fertilizer
is
consistent with our
-
finding that few farmers use very small amounts of fertilizer
the beginning of period 2,
season, those
who have
if
invested in period
receive
1
1
+
i? for
Borrowing
cost of producing fertilizer
=
laissez-faire, pji
of the
each unit invested. Farmers
which
in
=
1.
We
<
7:1^1
1
and pf2
Define the incremental yield
assume
that the cost
cash
to
be easily traded
at
=
income
Y'iz),
of reselling
so.
=
pfo
where
=
=
|,
is
the
heavy government subsidies
as well as a small, time-limited
z
V'(l)
—
amount of fertilizer
i'(0)
and
Maize, on the other hand
is
is
=
j/(2)
fertilizer is sufficiently large to
any time. Empirically, maize
at local
at
1.
to fertilizer as j/(l)
impatient farmers from doing
converted
their
not possible.
is
will also consider the impact of
In period 3, farmers receive
We
consume by using
be one, so that under competition and
to
adopted by Malawi, under which p/i
r>'pe
subsidy
p/2
assumed
is
next
for the
they have sufficient wealth, purchase either one or two units of fertilizer
price p/2 per unit incurring co^r/if they do so.
The
use no
which can be thought of as the time of planting
receive no additional income during this period: farmers can only
savings and,
to either
of their crop.^
fertilizer or fertilize a significant fraction
At
— they tend
used.
Y{2)
-
V'(l).
discourage even
completely liquid and can be
much more
liquid than fertilizer and can
markets.
Assumptions on Parameters
;
^ye assume:
;....:
.,
.•:.-.
'^=
/3Hy{l)>l
+
1
''
<y(2)<i
.
'
'.
f
+ /3^f>0^yil)>l +
,.•
;
-
-
(1)
;
f
(2)
-
(3)
•
•
,
1
>y(2)
The
'
-
first
has
it
to
be purchased right away. The second implies that an
impatient farmer will prefer to consume
^
price
is
For instance,
it
now
not heavily subsidized, even
among
farmers
who were
and 30 percent use top dressing
use
:(4)
condition ensures that a patient farmer prefers using one unit of fertilizer to zero
units of fertilizer, even if
if the
.'
on their entire
if
rather than to save in order to invest in fertilizer
it
is
possible to delay the decision and shopping
not offered free delivery or subsidized
fertilizer in a
fertilizer,
given season, but over 75 percent of those
'
plot.
between 20 percent
who do
use fertilizer
costs of purchasing fertilizer to a future period, and even if the rate of return to the period
investment
is
fertilizer if
it
their period
The second condition
high.
is
heavily subsidized
at
use more than one unit
two-thirds the cost of fenilizer, whatever the return to
market price (and
that therefore
no farmers will want
and also implies that patient farmers will prefer
at full price),
heavy subsidy of two-thirds of the cost of fertilizer (note
at a
condition does not include the shopping cost / because the cost
any
buy
investment opportunity. The third condition implies that the second unit of
1
fertilizer is not profitable at the full
two units
also ensures that impatient farmers will
1
is
to
to
use
that the third
incurred
if
the farmer uses
and does not depend on the quantity used). The fourth condition implies that
fertilizer
impatient farmers will not use a second unit of fertilizer even with a heavy subsidy of two-
,,,,._.,
thirds the cost of fertilizer.
These conditions match our empirical evidence on the
et al.,
2008) since we find
incremental return
to the
that the return to the first unit
second unit
is
negative
at
of
rates of return to fertilizer (Duflo,
similar to the return to the
(who use
fertilizer
first
/5h(1
+
R) >
the period
1,
we assume
which implies
=z
Pf2
=
I:
/?'//
(1 -h /?)
1
that patient farmers
at
subsidized prices,
> IJdl{1 + R) <
1,
farmers with high returns always
pj2
=
i;
Malawi
time-limited discounts under which pji
<
1
Farmer Behavior Under
c)
laissez-faire,
consider farmer behavior under laissez-faire,
m
3.2
and
we
heavy subsidies of the type adopted
traditional
=
Laissez-faire (p/i
by assumption
(1), the
every period will always use one unit of
fertilizer in
that
that patient period
and 3.4
p^j
Under
which suggests
two-thirds subsidy
at a
make
investment while impatient fanners never do.
1
In subsections 3.2, 3.3,
Pf.^
unit at market prices,
without a subsidy) would be likely to use two units
Finally, for completeness,
that the
market prices. The assumptions are also
consistent with evidence that the incremental return to the second unit
is
and
fertilizer is high,
penod
2.
By assumption
impatient will never use
fertilizer.
p/2
=
fertilizer.
and pf2
who
All will save at rate
proportion
w
which
which
=
1.
1)
proportion n of fanners
(2), the
By
=
in
in
of farmers
R
are always patient in
in
who
period
1
and buy
are always
our other assumptions, they will not avail themselves of
the investment opponunit)', whatever the return.
Now
consider the problem of a stochastically present-biased farmer deciding whether
(and when)
to
buy
fertilizer.
problem of a farmer
in
To
period
2,
solve the model,
we work backwards, beginning
who must choose between consuming one
with the
unit, or investing
9
it
in fertilizer.
Assumption
in period 2 will
2 will not use
use
(1) implies that a
fertilizer.
Assumption
impatient in period
consumer, and
will
1
from saving
1
is at
in
to
purchase
period
1
.
First,
best Piyfl)
(2), is
fertilizer in
-
Now consider a
farmer
who
consuming everything today:
today.
the rest
By
do
it
in
in
period
period
2"!
—
x
is
assumption
Now.
who
then
2).
If a
1
—
is
smaller than
period
patient in period
1
is
it
be optimal, note
(the loss in
we assume
/
+
dnvi'^), while her utility
from buying
who
is
that /3l(1
utility cost
fertilizer
is
is
if
.?•
<
/?)
away, or plan
to
buying
\
+R
p),
2.
To
1
see that postponing
period
2,
her utility
is
in
investment and
+
/S/^.
Thus, waiting
is
optimal if
X
-
-i^^+1
Rearranging,
we
and thus purchases
'"
+/3«(y(T)-/)
=
0,
+ p/3«(y(l) -/) + (!- v)^H >
the right
is
(5)
1
-
p,
.T
-
1
-
/
+
/3Hy(l)
(6)
find that the farmer will wait if
cost of using fertilizer
'
•
,
.
/(1-P/3h) +
p
may
.•
1
•'
wait to
fertilizer in
ends up being impatient (which she believes will happen with probability
r -
and
she consumes everything
fertilizer right
farmer waits, ends up being patient
^
utility is
1.
higher than not buying.
of buying fertilizer until period
-
When
+
patient today has a sufficiently high subjective
'-
her
consumption
Investing in fertilizer today dominates
(which she believes will happen with probability
If she
from
farmer with a high return
best to wait, and thus realize the return on the period
that if the
the gain
1,
is
farmer ends up being patient
p^
2, for a
.
should a patient farmer buy the
farmer
who
the farmer's utility if she purchases one unit of fertilizer
(1), utility
1.
postpone paying the
fertilizer
patient
impatient in period
is
from period
probabilit}' of being patient again (and therefore a high probability' of
period
is
observe that a farmer
/u./ (if the
saving opportunity). This farmer will also not save since
consumes
farmer
will not save: seen
which, by assumption
fertilizer),
period
has sufficient wealth and
(2) implies that a
problem of a farmer
investing in one unit of fertilizer
in
who
fertilizer.
Now consider the
and buys
farmer
hand side
is
•
y^ >/3//(y(l)-l)(l-p)
equal
to /3//(y(l)
small enough that /?//(y(l)
-
-
1).
If
1) is larger
(7)
we assume
than /
+
that the utilir\'
-~-^, then the right
10
hand side of
the inequality
decline with p. but the right hand side
hand side (which
the right
(0,1) such that for every
invest in the
easy
the
first
to see that
more
For p
steeper.
=
Both sides of the inequality
1,
the left
equal to zero). Thus, for each R, there
is
p > p*iR),
a
p'(R]
it
is
is
farmer
who
hand
side
a p'(/?) in the interval
is
intending to use fertilizer later prefers to
is
fanners will not save
it
is
2. It is
investment,
1
for the farmer to wait.
in
we assume
any case,
it
that
p
>
p*{R). Note that since impatient
not necessan,' that they believe they will be
is
patient in the future than they are in the present for this procrastination
Instead,
larger than
is
decreasing with R: the higher the return to the period
For the remainder of the model,
1
is
side.
period investment opportunity, and plans to buy fertilizer in period
more valuable
period
hand
larger than the left
is
problem
to arise.
only necessary that patient farmers overestimate the probability that they will
continue to be patient in the fuaire. This tendency to believe that future tastes will more
closely resemble current tastes than they actually will, termed "projection bias," has found
considerable empirical support (Loewenstein. O'Donoghue. and Rabin, 2003).
3.3
Farmer Behavior Under Malawian-StA le Heavy Subsidies
One
potential
way
buy two
pjo
=
j)
address underinvestment in fertilizer would be through heavy,
to
Malawian-sty'le subsidies. Under heavty subsidies, by assumption
patient will
=
(p/i
units of fertilizer,
and by assumption
(2),
farmers
(3),
who
farmers
who
are
always
always
are
impatient will buy one unit.
To
solve for the behavior of the stochastically impatient farmers
work backwards from period
in
2.
Assumption
period 2 will use one unit of fertilizer
(2) implies that
if ;>/2
=
|,
impatient farmers will not want to use two units of
penod
2 will thus purchase exactly
purchased
Now
purchase
it
again
even fanners who are impatient
while assumption (4) implies that
fertilizer.
A
unit if he has the wealth
farmer
do
to
who
it
is
impatient in
and has not already
consider the case of a stochastically hyperbolic farmer deciding whether to
fertilizer in
units. Recall that
penod
we
earlier.
period
(3) implies that a patient
two
one
in this case,
1
.
First consider a
farmer wants
it is
farmer
to either
who
patient
is
m period
1
.
.Assumption
purchase two units, or save enough
efficient for farmers to
purchase
all
of their
fertilizer
m
to
buy
a single
since by doing so they only need to pay the shopping cost of fenilizer once.
If a
fanner buys two units of
fertilizer
immediately, her
utility is:
11
.
.T.-^-f +
If the
farmer instead plans
she
if
impatient in period 2 she will purchase only
-
Thus, she will prefer
save and plan
+
3^^^;^^^
By
if
>
;;
reasoning similar
p*'(R). farmers
unit.
1
to the
who
/(I
-
to
R
at return
py(2)
+
however, she
If,
> pH{y{2)
-
(1
fertilizer later
utility
from waiting
P)^]
is:
(9)
if:
(10)
jfiR) such
a threshold
is
will wait until period 2 to
I
is
-Iki-p)
case without a subsidy, there
are patient in period
for future fertilizer use,
Thus, her expected
-/+
buy
M
(8)
patient in period 2.
is
^(^-^ + /3//b(i)
to
yil))
use fertilizer and saves
to
she will purchase two units of fertilizer
^
+
B„{y{2)
purchase
that
(it is
also easy to see that the threshold decreases with R, so those with higher returns to
investment
in
period
1
more
will be
p'(B). However,
values, f>"{R) could be smaller or larger than
the second unit of fertilizer at the subsidized price
on the
first unit
than p'{B].
of
Below we assume
scenario for hea\7 subsidy;
who
are patient in period
overusing
if
that
p
is
if
the
mcremental renim of
greater or equal to the incremental return
(i.e.,
3y(2)
>
v/(l)),
above both thresholds. Note
then p**(/?)
is
larger
that this is the best case
p was lower than p*'{R), the stochastically impatient farmers
would
1
is
an unsubsidized price
fertilizer at
Depending on parameter
likely to defer purchases).
all
buy two
units in period
1,
and thus would
all
end up
fertilizer.
Now, consider
a stochastically patient fanner
Given our assumptions, she wants
to
who happens
and she will postpone
this plan.
buymg
be impatient
use one and only one unit of fertilizer
subsidized price. If she saves, she will thus save enough
always follow through on
to
Therefore, there
is
fertilizer until period 2,
purchase one
to
at the
unit,
period
in
1
heavily
and she will
no time inconsistency issue for
and will buy exactly one
her,
unit,
Overall, a heavy subsidy will induce 100 percent fertilizer usage, but will cause the
always-patient farmers and the stochastically impatient farmers
both periods to overuse
3.4
(pfi
Consider the impact of a small discount on
to the
to
be patient
in
fertilizer.
Impact of Time-Limited Discount
corresponds
who happen
case in which pfi
<
1
<
1
and p/2
=
fertilizer, valid in
and pf2
=
1).
1)
period
Consider
a
1
only (which
discount that
is
not large
12
enough
make purchasing two
to
will see that this
To make
purchase
to
m period
1
even
farmer (we
for a patient
assumption since the necessar>' discount will be small).
farmer prefer purchasing
2, the
penod
fertilizer in
period
waiting to
to
1
price needs to be such that:
1
;
x-—^ + piy{\)-f) + {l-p)0H<x-pfi-f + l3Hy{l)'
.
we
a reasonable
a patient period
fertilizer
-
If
is
units of fertilizer profitable,
(11)
define p}i(R) as the price that just satisfies this condition for a farmer with return
investment
i?,
then /yyj(/?)
.—
Note
that
p*^(/?)
when p
is
IS
given by:
^_,
_
,
..
= /(p/3^_l) + /3^(l-p)(y(l)_l) +
close to
the price
1,
p*^j
{R) differs from
the utility cost/ plus the foregone return to investment
additional costs that a farmer
who
patient in period
is
choosing between investing one unit
fertilizer are the utility cost
in the
period
of purchasing the
1
1
1
(p^)- The
-^.
by
.
(12)
a term proportional to
intuition
that the
is
has to immediately bear
only
when
investment and buying one unit of
fertilizer,
and the foregone investment
opportunity. Thus, the fanner just needs to be compensated for incurring the decision and
shopping cost / up
investment. If the returns to the period
reduction
in the utility cost
induce the farmer
to
period
2.
An
to the
(such as free delivery
in
period
1,
1
may
)
then be sufficient to
instead of relying on her period 2
'
the impact of a subsidy
m
period
1
to
an unanticipated subsidy in
unanticipated period 2 subsidy will not affect the period
1
f -
p*j.-,-
induce
3Ly(l).
fertilizer
We
denote the p/n. which just
purchase, the discount
now needs
satisfies this inequality as
to
m which
fertilizer price
of
u(l)
is
close to
1
+
/
incumng
(so that the return to fertilizer
from an ex ante perspective), the discount
1
is
is
the utility cost /:
just positive at a
approximately the cost of
delaying one unit of consumption for one period for an impatient person. Thus,
discount
m
penod
discount
in
period
1
will
In order to
be large enough to compensate an
impatient fanner for postponing consumption of py2, not only for
in the case
An
decision.
if
<
1
investment are low, even a small discount, or a
impatient period 2 farmer with sufficient wealth will decide to use fertilizer
Pf2
period
,
compare
useful to
1
switch to buying fertilizer in period
self to purchase fertilizer.
It is
foregone returns
front, rather than later, as well as for the
have as large of an effect on ultimate
fertilizer
use as
a
small
a large
2.
13
who
In each case, the farmers affected will be those
are patient in period
1,
but impatient
in period 2.
3.5 Choice of
Finally,
us examine what will happen
let
which she
Timing of Discount
gets a small subsidy. Specifically,
induce patient period
farmers
1
induce impatient farmers
to
buy
=
farmer can choose either p/j
Consider
first the
farmers
investment opportunity
request the subsidy
is
in the
fertilizer in period 2.
purchase
fertilizer.
1
-
(5
who
we
1
large
enough
some
to
to
fixed discount 5 and the
price in the other period remains
Because the return
to the period
1.
I
it is
second period.
they will save in anticipation of buying
In period
and will follow through on
who
penod
1,
low, those farmers will always
that plan.
always impatient. They are not planning
are
probabilit)' that they will
the small discount in
is
The
(5.
are always patient.
they are in principle indifferent on
even some small
-
is
to the date at
immediately but not large enough
Suppose there
=
period
consider a subsidy that
fertilizer
or pjo
in
always positive even when
Next, consider farmers
fertilizer, so
to
farmer can commit
if the
2, rather
when
be patient
to get the return.
to
save or use
However,
in the future, they will
if
choose
there
is
to receive
than refuse the program.
Finally, consider the case of the stochastically impatient farmers. If the discount does not
reduce the price of fertilizer below
discount
in
period
immediately
in
2,
p'f-^{R),
because the discount
period
1
so the only
way
patient in both periods. In what follows,
then farmers will always choose to take the
not big enough to induce them to buy
is
buy
that they will
we
fertilizer is if
consider the case in which S
In this case, if a farmer chooses to receive the discount in period
is:
•,
.
,
chooses
If she
to receive the
"
^k[x
•
Note
first that
discount
they happen to be
•
.^
..
,
period
in
1,
>
I
—
p'j^{R).
her expected utility
.
2,
her expected
utility
is:
+ p\y{l)-^^-f]+p{l-p)Pf2Y^]
(14)
current impatience does not affect this decision (since farmers discount
future periods at the
same
rate in period 0).
Second, observe that when
long as p does not equal zero, the farmer will chose the discount
farmer does not care whether the period
gain to delaying the decision
is
1
in
R
close to zero, so
is
period
all
1
:
since the period
or period 2 fanner pays the utility cost, the only
the return to the period
1
investment opportunity'. However,
.14
.
as
R
increases, the value of delaying the discount to period 2 increases, and if j'?
is
high
Thus, depending on
enough, the farmer will choose
to receive the
whether the returns
opportunity are high or low. the farmers will choose
the returns in period
3.6
period
to
I
or in period
1
discount in period
2.
receive
to
2.
Summary
To summarize,
Some
1
the
model gives
rise to the
make
farmers will
following predictions.
plans to use fertilizer but will not subsequently follow
through on their plans.
2.
Farmers will switch
3.
A
fertilizer
2.
in
m
small reduction
and out of fertilizer use.
the cost of using fertilizer offered in period
purchases and usage more than a similar but unexpected reduction offered
The subsidy only needs
to
be large enough
the period
1
investment.
A
the
same discount
Therefore,
buying
period 2.
in
4.
if there are
some farmers
it.
period
for incurring the
well as for the foregone returns to
in
period 2 to induce the same
I.
will
choose the discount
would always prefer
who choose
to
period
1.
Recall that
receive the discount in period
the discount in period
some fanners
in
or
1
I
and follow through by
are time inconsistent,
and have
at least
some
.
.
Testing the Model
As noted above,
of farmers
made
We
farmers
suggests that
fertilizer, this
awareness of
needed
in
farmers are offered an ex-ante choice betVi-een a small discount in period
for a positive R, time-consistent farmers
2.
later, as
larger subsidy will be
increase in usage as a small subsidy in period
When
compensate the farmer
to
decision and shopping cost up front, rather than
4.
will increase
1
who
there
is
some empirical evidence
in favor
of predictions
1
and
2: in a
participated in the demonstration plot program, two-thirds of those
sample
who had
plans to use fertilizer do not end up carrying through with these plans (prediction
also find significant switching
between using and not using
1).
fertilizer (prediction 2): a
regression of usage during the main growing season on usage in the main growing season
previous year (as well as a
full
vector of controls) gives an R" of only 0.25. Suri (2007)
similarly finds considerable switching in and out of fertilizer use in a nationally
representative sample.
15
Of course, we may
not want to attach
much weight
to the declared intentions
and therefore discount the evidence on prediction one. Similarly, other
switching
in
and out of fertilizer
use.
We
Dutch
We
fertilizer use.
stories could generate
therefore focus on predictions 3 and 4 below.
Predictions 3 and 4 of the model suggest that
impacts on
of farmers
some simple
interventions could have large
collaborated with International Child Support (ICS)
NGO that has had a long-lasting presence in
Western Kenya, and
well
is
-
Africa, a
known and
respected by farmers, to design and evaluate a program using a randomized design that
encourage
fertilizer
use
if
predictions of the model,
them with
farmers did indeed behave according
we implemented
to the
model. To
would
test the
multiple versions of the program, and compared
alternative inten.'entions. such as a fertilizer subsidy and reminder visits.
4.1.TheSAFI Program
The main program was
called the Savings and Fertilizer Initiative (SAFI) program.
program was
first
piloted with
with farmers
who
participated in the on-farm trials described in Duflo, et
pilot
programs,
we
minor variations over several seasons on
focused on acceptance of the program and willingness
2003 and 2004, the program was implemented on
determine
Basic
In
its
impact on
its
a
fertilizer
a larger scale,
The
very small scale
al.
to
(2008). In these
buy from ICS. In
and we followed fanners
to
usage.
SAFI
simplest form, the
SAFI program was
any combination of planting or top dressing
offered
at
fertilizer.
harvest, and offered free delivery of
The basic SAFI program worked
as
follows: a field officer visited farmers immediately after han'est, and offered them an
opportunity' to
farmer had
to
buy
a
voucher for
decide during the
fertilizer, at the
visit
regular price, but with free delivery.
whether or not
buy any amount of fertilizer. To ensure
to participate in the
The
program, and could
that short-term liquidity constraints did not
prevent
farmers from making a decision on the spot, farmers were offered the option of paying either
in
cash or
in
by offering
maize (valued
free
market
price).
To avoid
distorting farmers' decision-making
maize marketing services, farmers also had the option of selling maize
without purchasing
who purchased
at the
fertilizer.
fertilizer
was not overvalued
m
Across the various seasons, the majonty (66 percent) of those
through the program bought with cash, which suggests that maize
the program. Participating farmers chose a delivery date and received a
voucher specifying the quantity purchased and the delivery
date.
Choosing
late delivery
16
would provide somewhat stronger commitment
be re-sold
The
(at
some
cost)
and the vouchers themselves were non-transferable.
SAFI program could have reduced
basic
reduced procrastination,
is
use fertilizer since fertilizer can potentially
to
two ways.
in
typically about a 30 minute tnp
First,
it
the utility cost of fertilizer use, and thus
can save a tnp
town
to
to
buy
from the farmers' residences. Suri (2007) argues
distance to a fertilizer provider accounts for her surprising finding that those
had the highest return
typically available in
to
using fertilizer are some of the least likely to use
major market centers around the time
maize crops. Since most farmers
travel to
errands, they could pick up fertilizer
it
is
it.
that
who would have
Fertilizer
and offering
a
market centers occasionally for shopping or other
when
they go to town for other reasons.^
simple option, the program
thinking through which t^'pe of fertilizer to use, and
SAFI with
To
test
to
this visit,
earlier, in a
fertilizer
would
naive, they
purchase
in
use fertilizer would then invest
to
penod
would
lead to
to invest in
low ultimate adoption.
investment opportunit}' has
remms
for transport.
On
a
would take
who makes
time, or about
1
1,
and be
fertilizer in
period
2.
m
the
This
our model, stochastically hyperbolic farmers whose
high return also choose a
the
who have
SAFI program
who bought
fertilizer
of 135 Kenyan shillings), and only
the average fanner roughly an hour to
SI a
and purchase
1
of the investment, but those
average, fanners
fertilizer (at a total cost
It
period
farmers were present-biased but completely
period
In
Most farmers who bought fenihzer through
pay
2. If
in
also have chosen a late delivery date, since they expect to be patient
and would then plan
forgoins the
told that,
standard exponential model, farmers would be expected to
who want
future,
1
SAFI program: farmers were
when
decide
to
they would have the opportunity to pay for fertilizer and to choose a deliver)'
a late visit: those
prepared for
penod
quantit}'.
our model) and offered the opportunity'
in
be visited again later to receive a
As discussed
choose
what
reduced time spent
prediction 4 of the model, in the second season of the experiment, farmers were
they wanted
date.
m
may have
field
ex-ante choice of timing
visited before the harvest (period
during
is
needed for application for
Second, and more speculatively, by requiring an immediate decision dunng the
officer's visit
which
fertilizer,
walk
to
1
a
d)d not
through the
avoid
late deliver)' date to
low return investment
buy enough
that they
would have had
to
SAFI program bought SJ kilograms of
percent of farmers bought more than 10 kilograms.
town, buy
fertilizer,
and walk back. For
day over an eight-hour workday, the SAFI program would save her about SO.
a
farmer
13 in lost
work
percent of the cost of the fertilizer bought by the average farmer. This cost would be
substantially smaller if the farmer
were gomg
to
town anNavay and so would not miss any work time.
17
opportunity in period
eventually use
4.2
1
will
chose an early delivery date,
to increase the probability that
they
fertilizer.
Experimental Design
Two
SAFI programs were implemented
versions of the
experiment, allowing for a
test
as part
of a randomized
field
of the model. Farmers were randomly selected from
frame consisting of parents of fifth and
sixth grade children in sixteen schools in
Busia distnct. The program was offered to individuals, but data was collected on
fanned by the households. And
a farmer
was considered
The experiment took place over two
seasons. In the
on any plot
in the
to
use fertilizer
sample of farmers was randomly selected
randomization took place
to set
at
plots
all
was used
season (beginning after the 2003
first
in the
2004 long
form of "starter
kits'"
(a
to test the
2004
program
to
by school,
class,
provide farmers with small
they could use on their
short rains), the
rains season), a
SAFI program. The
receive the basic
own
-
up demonstration plots on the school property).
enriched design
Kenya's
if fertilizer
the individual farmer level after stratification
In the following season (the
some
to
two prior agricultural programs
in
amounts of fertilizer
program
sample
household.
short rain harvest, in order to facilitate fertilizer purchase for the
and participation
a
farm, and a
.
;
;
,/,.•
program was repeated, but with an
main empirical predictions of
the
model
in
Section 3 as well as
predictions of alternative models. All treatment groups were randomized
at the
individual level after stratification for school, class, previous program participation, and 2003
treatment status.
First, a
new
set
of farmers was randomly selected to receive a basic SAFI
another group of farmer was offered
SAFI with
visit.
Second,
ex ante choice of timing (as described above).
Third, to test the h^'pothesis that small reductions in the utility cost of fertilizer have a
bigger effect
fertilizer
if
needs
offered in period
to
1.
another group of farmers was visited close to the time
be applied for top dressing (approximately 2 to 4 months after the previous
season's harx'est, the equivalent of period 2
fertilizer
was
with free delivery.
visited during the
To
in
our model), and offered the option to buy
calibrate the effect of a discount, a fourth
same period, and offered
us to compare the effect of a 50 percent subsidy
the
SAFI program.
fertilizer at a
to the effect
In all of these programs, farmers could
planting, top dressing, or both.
However, one caveat
to
group of farmers
50 percent discount. This allows
of the small discount offered by
choose
to
bear in mind
buy
is
either fertilizer for
that in the late visits
18
many
farmers had already planted and could only use top dressing
farmers preferred using
fertilizer for
.
fertilizer at planting,
however, they could have bought planting
use in the next season, so a standard model would suggest that these farmers
should have taken advantage of the discount for
later use.
subset of farmers was offered the option to
the field officer before the
to test the alternative
way
cash
m
program took
hypothesis that the
to protect their
who
thus arguably easier to waste. If the
is
-
•
comparison group,
a
random
quantity of maize at a favorable price to
The objective of this
additional treatment
SAFI program was just seen by
accepted the offer, which
SAFI
to take
is
more
up.
to
was
the farmers as a
some
liquid than maize,
main reason why farmers purchased
because of an aversion
have encouraged them
place.
,
.
,
savings than available alternatives. The purchase of maize put
the hands of the farmers
SAFI program
sell a set
,
.
Finally, in each of the intervention groups as well as in the
safer
If
fertilizer in that season.
fertilizer
and
under the
holding liquidity, the purchase of maize should
Under our model,
this
would make no
difference,
however.
Appendix
4.3
figure
1
summarizes
the experimental design for this second season.
Data and Pre-Intervention Summary
The main outcome of interest
an intermediate outcome.
We
is
Statistics
fertilizer use,
with
fertilizer
have administrative data from ICS on
the program. Data on fertilizer use
was
collected
at
harvest) for that season and for the previous season.
fertilizer
in
in
fertilizer
purchase under
baseline (before the 2003 short rains
We
usage data for the three seasons following the
2004 and one season
purchase through the program as
later visited
first
farmers
SAFI program
to collect
(i.e.,
both seasons
2005). The baseline data also included demographic information
and some wealth characteristics of the sampled households.
In
members farm
polygamous households), we
asked each
different plots (which
member
is
individually about fertilizer use on her
own
plot,
use on each
plot.
The data
of the household (the husband) about
household
Table
fertilizer
and
we
is
asked the head
aggregated
at
the
level.
1
shows descriptive
participate in the basic
statistics. In
eligible for the
season one, 21
SAFI program, and 713 farmers
season two, 228 fanners were eligible
retail price
tvpically the case in
households where different
1
farmers were eligible
constituted a comparison group. In
to participate in the basic S.A.FI
SAFI with ex ante choice of timing; 160 were
with free deliver)'
at
to
program; 235 were
offered fertilizer
at
the normal
top-dressing time; and 160 were offered fertilizer
at
half
19
price with free delivery at top dressing time.
An
additional 141 farmers served as a
comparison group.
There were some relatively minor pre-treatment differences between groups
However, there were some pre-treatment differences
fertilizer.
each
SAFI and comparison groups had previously ever
season. In season one 43 percent of both
used
in
comparison group farmers had 0.6 more years of education
percent level), and were about 5 percent less likely to live
in other observables:
(a difference significant at the
home
in a
mud
with
mud
floors,
mud
walls, or a thatch roof (though only the difference in the probability of having a
10
floor is
statistically significant, at 10 percent).'
In season two, the
program
(table
1
,
comparison group was more
likely to
have used
fertilizer prior to the
panel B). The point estimate for previous fertilizer usage
is
5
1
percent for
the comparison group, but only betvv'een 38 percent and 44 percent for the various treatment
groups.
Many
of these differences are significant
10 percent level (the difference
at the
comparison group
significant at 5 percent for the 50 percent subsidy group). In addition, the
has significantly
(at
10 percent)
more years of education than
the
is
group offered SAFI with
the ex ante timing choice.
These pre-treatment differences are
bias our estimated effects
downwards.
in general relatively
We
minor and would,
anything,
if
present results with and without controls for
variables with significant differences prior to treatment
—
inclusion of these
in all cases, the
controls does not substantially affect our results.
5.
Results
..:'.,:',
'•,-
.;r„.
'"''''
>....;„,
,,•;,;
-^^
''.-.'
'
'
.
;,,
'
'
'
5.1
The SAFI Program
The SAFI program was popular with
were offered SAFI bought
fertilizer
offered the basic version of
those offered
fertilizer is
Appendix
appearance
sampled
SAFI with
table
at
percent of the farmers
who
to the
through the program, as did 41 percent of
fertilizer
The
fraction of farmers
who purchase
impact of the program on use: some program farmers
suggests that attrition patterns were similar across groups. Regressions of indicators for
1
pre-treatment background and post-treatment fertilizer adoption questionnaires on being
for treatment yield
do not appear
1
through the program. In season two. 39 percent of those
SAFI bought
of course not equal
in the
fanners. In season one. 3
ex ante choice of timing.
no significant differences between groups. Overall,
and we obtained adoption data
not
\^,
^^.
i^'
in the dataset
for
925 of them (75.1 percent). There were few
were not known by other parents
home when ICS enumerators
visited their
in the
1
,232 farmers were sampled,
refusals.
Nearly
all
of those
who
school and so could not be traced, or were
homes.
20
1
who were going
to
use
fertilizer
anyu'ay presumably bought
may
advantage of the free delivery. In addition, some fanners
purchased through SAFI on
their
fertilizer
through SAFI,
take
to
not have used fertilizer
maize crop: they could have kept
sold
it.
used
it,
it
some
on
other crop, or the fertilizer could have been spoiled. In the 2005 adoption questionnaire 76.6
percent of the farmers
who purchased
on the plot of
plot, 7.3 percent
under SAFI reported using
wife or husband, and
their
8.1
it
on
their
own
percent reported saving the
use in another season. The remainder reported that they had used the fertilizer on
fertilizer for
a different crop
( 1
percent) or that the fertilizer had been spoiled.
.6
Overall, in both seasons, the
on
fertilizer
fertilizer use. In
SAFI program had
compared
percentage point difference
is
34 percent of those
to
significant at the
m the
fairly sizeable
SAFI program
season one 45 percent of farmers offered the
fertilizer in that season,
and
a significant
report using
comparison group. ^ The
percent level (see table
1
impact
1
panel A). In
1,
season two (the 2004 short rains), the basic SAFI program increased adoption by 10.5 percent
(table
panel B).
1,
.
Table 2 confirms these results
in a
regression framework. For season one,
we
run
regressions of the following form:
y,
where
a
y, is a
dummy
dummy
= Q+
/3:T,^«
+ A77 +
/'
was offered
the
'/
is
SAFI program
of control variables for the primary respondent
a vector
-
indicating whether the household of farmer
indicating whether farmer
(15)
•
£-
in the
using
in
fertilizer,
Ti^^
season one, and
is
A", is
household, including the
school and class from which the parent was sampled, educational attainment, previous
fertilizer
usage, gender, incom.e, and whether the farmer's
home
has
mud
walls, a
or a thatch roof, and whether the farmer had received a starter kit in the past.
The
mud
floor,
table
presents fertilizer usage statistics for the season of the program and the two subsequent
seasons.
Both specifications suggest
adoption
to an
1
m
1.4
a positive
this
paper,
we
in fertilizer
adoption, while one with fuller controls
focus on usage of fertilizer rather than the quantity of fertilizer used because there
substantial underlying vanation in the quantity of fertilizer used by farmers,
pick up effects
in
average quantities. The standard deviation
fanners that bought
'
The
fertilizer
season one: the specification with sparser controls suggests that the program led
percentage point increase
Throughout
and significant program impact on
starter kit
fertilizer
was an
through the
inter>.'ention
conducted
involved distributing a small quantity of
in
kilograms of
SAFI program bought only
in a
fanners to
let
fertilizer
3.7 kilograms,
previous season, which
fertilizer to
which would make
we
used
is
it
is
difficult to
54, whereas
on average.
discuss in a
them experiment with
companion paper.
It
fertilizer.
21
suggests a 14.3 percentage point increase. Both are significant
baseline usage rate of 22.8 to 24.7 percent (shown on the last
at
the
percent level. Given a
1
row of table
2),
these effects
represent a 46 to 63 percent increase relative to the comparison group.
The remaining columns show
the
two subsequent seasons
usage drops back
that the
SAFl program does
2004 and
(the short rains of
to the level
buy
the state variables of wealth and
Panel
B shows
SAFl
knowledge and thus
the impact of the
regression has the
the other
all
in this
program
induce the farmer
is to
the fertilizer early in the season, rather than later. In contrast, in learning
models, and models of credit constraints, inducing use
The
the long rains of 2005), fertilizer
of the comparison group. This lack of persistence would be
expected under our model since the only role of SAFl
to
not have persistent impacts: in
same form
SAFl program
as for the season
in
in general affect
future behavior.
in the
second season on
fertilizer usage.
one regression, but includes dummies for
dummy
treatments, and controls for a
one period would
by doing
for long rains treatment status
4
Vi
= a + J] ^f^'Tl,^ +
(5,Bf^
+ p.Bf^T^^^ +
^,'7
+
(16)
e,:
k=i
In this regression, Tif ^ represents the basic
SAFl program, and T^f " through T]^^
represent the other treatment groups, respectively,
visit at
top-dressing time that offered fertilizer
that offered fertilizer with a
the farmer
harvest
season
the
was offered
visit.
1
As
we
(the season before the
first
row
in
maize
3
we saw
was
earlier,
in the
to 18.1
season
it
was
show
offered.
adoption was slightly greater
a
dummy
full sets
stratified
1
if
of controls, for
is
the impact of the basic
which
SAFl program
Without control variables, the point
even larger than
in the
in the first season. Since,
comparison group before the program
when
effects represent proportional increases of
by prior treatment
equal to
(the season during
percentage points. Given a baseline usage rate of 29.7
comparison group, these
Treatment was
is
top-dressing time
an above-market price during the post-
introduced, the point estimate of the effect increases slightly
adoption
'°
at
Bf'^
visit at
the
(one season after the programs were offered).
estimate for the effect (16.5 percentage points)
as
dummy
programs were offered), season 2
panel B, columns (3) and (4)
on adoption of fertilizer
and the
present regressions with and without
programs were offered), and season
The
at full price;
50 percent subsidy. The
the opportunity to sell
before,
SAFl with ex ante choice of timing;
56
controlling for past
to
to
30 percent
in the
60 percent.
status.
22
Columns
groups
the
(1)
in the
first
and
(2)
show
that, reassuringly, there is
season before
it
was
Columns
SAFI program
season: the impact of the
These
offered.
no difference
(5)
and
(6) replicate the results
in the utility cost
can substantially increase adoption. Free delivery saves the fanner a
town
to get the fertilizer and,
type and quantity of fenilizer
cost of time spent
costs.
It is
smce taking advantage of free
to
order during the
inconsistency and procrastination as posited
The model
remm
period
1
visit,
making these decisions and thus
why
therefore plausible that the reason
the
the
of using
fertilizer
nearest
market
deliver,' required deciding
on the
trip to the
program may have reduced
the
chance of procrastmation on those
program increased adoption
this
in the
found for
not persistent.
is
timed reduction
results suggest that a properly
SAFI
adoption across
in
is
time
model.
predicts that those stochastically hyperbolic farmers
who do
not have a high
investment opportunit>' will request early delivery. The results for the SAFI
with ex ante timing choice are consistent with the idea that a sizeable fraction of farmers have
a preference for
commitment. .Almost half of
timing choice asked the field officer
of those actually bought
fertilizer.
delivery and 39 percent of those
come back immediately
to
Of the
who
the farmers (44 percent) offered
may want
to
induce their period
of free delivery early unless they have
In contrast, time consistent farmers
choose
a
penod
If the
1
SAFI program. These
1
a high return to their period
who
is
0,
even quite naive
I
investment opportunity.
attach any probability to using fertilizer
would never
discount (so long as the returns to investment are positive).
our model predicts that fewer farmers should end up using
in
which
free deliver)'
fertilizer
is
under SAFI
restricted to period
because the stochastically hyperbolic farmers with high returns to the period
investment opportunity buy fenilizer
in
penod
1
under the basic SAFI. but choose
discount under S,A.FI with timing choice, and some of those choosing a
wind up impatient
in
period 2 and do not buy fenilizer. Empirically,
of the "S.A.FI with ex ante timing choice"" on
the basic
results are very
selves to purchase fertilizer by requesting the offer
with choice of timing then under the basic SAFI,
This
>
fertilizer;
parameters are such that farmers with high return investment opportunities prefer
late delivery,
I.
late
requested late delivery eventually purchased
consistent with the model, which predicts that as long as p
farmers
and 46 percent
remaining fanners, 52 percent requested
the remaining 4 percent declined to participate in the
much
after harvest,
SAFI with
SAFI program.
fertilizer
use
is if
we
to
period 2
late deliver,' date
find that the impact
anything slightly larger than
Overall, 41 percent of farmers purchased fertilizer under
ex ante timing choice (compared
a
1
SAFI with
39 percent without timing choice), and more farmers
23
reported using fertilizer under
SAFI with
ex ante timing choice (47 percent versus 38
percent), although these differences are not significant (see the second
row of panel B,
table
2)."
Note, however, the fact that the effect of the SAFI with ex ante choice of timing
large as the effect of the basic
SAFI
helps rule out an alternative explanation for the
popularity of basic SAFI: an "impulse purchase" effect in which
when
fertilizer at harvest,
they have
as
is
money and maize,
when farmers
are offered
they feel "flush" and buy
it
without
thinking, as an impulsive purchase (under this hypothesis, if the field officers had offered
beer or dresses
at that point,
the pre-harvest season
not offer to
sell the
is
they would have bought those). This seems reasonable given that
known
as the
"hungry season"
farmer anything immediately
Instead, the field officer offers an opportunity to
on when
to call the field officer
back
is
fertilizer
who
Under
well as
some maize
interaction with the
were the
SAFI when
case,
decision
at a
SAFI dummy,
simply an
fertilizer is not
farmers were no more likely
test this,
we
purchase
to
ran a small test in which
favorable price before offering SAFI.
while 50.7 percent of farmers sold maize, 36 percent
out the possible alternative explanation that
if this
purchase of
under SAFI, and thus the effects of the "bought maize"
its
choice.
fertilizer in the future: thus, the
under SAFI when they had cash on hand. To
this condition,
fertilizer
that the
feel "flush" is that
the field officer offered to purchase
does
unlikely to be an impulsive decision.
Another piece of evidence suggesting
impulse purchase of farmers
field officer
SAFI with ex ante timing
in the
buy
Kenya, and the
in
dummy
still
on
purchased
fertilizer use, as
are insignificant and small. This also helps rule
SAFI
is
one would have expected them
used by farmers as a safe savings option;
to
be more likely
to take
advantage of
they had cash on hand.
Thus, the impact of the two versions of the SAFI suggest that time inconsistency and
procrastination
A
may
play a role in explaining low fertilizer use.
fanners
may
fertilizer
and
differ
m
their discount rates. In the
model,
we assume
if the
purchase
in
period
I
only
intermediately patient in period
return to that investment
1.
if their
period
1
is
high).
investment has
These fanners will only use
intermediately patient) in period 2, and so will request a late
SAFI but may buy
we
that stochastically hyperbolic
that impatient fanners will
However,
be (stochastically) intermediately patient (with a discount rate between
fertilizer
is
rule out alternative
in
the
SAFI with timing
choice.
give them a bit more time to be ready with cash
Another
when
a
low
may
and
1
be that some fanners
,'J/v)
return, if they
fertilizer if
SAFI
/3l
it
'
never use
that all patient farmers value the return to fertilizer higher than their alternative period
investment opportunitv' (even
may
SAFI with timing choice
possible interpretation for the larger effect of
To
and will commit
happen
to
to
be
they end up being patient (or
date and will never buy fertilizer in the basic
possibility
is
that
by warning fanners
in
advance,
the field officer arrives.
•
24
explanation of the role
SAFI played
in
inducing farmers
use
to
programs with random subsets of farmers, which allow us
alternative
we
fertilizer,
tried
two
to test alternative
hypotheses and additional predictions of the model.
Free Delivery, Free Delivery with Subsidy
5.2.
Both versions of the SAFI program offered
resulting decrease in the
unlikely
to
utilit}'
cost of using fertilizer
induce large changes
an alternative explanation
in fertilizer
period
to
2).
randomly selected group of farmers
As shown
in table
1
,
use
m
a
is
Our
interpretation
small enough that
it
that the
is
would be
purely time-consistent model. However,
that the free delivery is a substantial cost reduction.
is
hypothesis, and to test prediction three
season (corresponding
free deliver)'.
m our model, we offered free deliver)'
We
50 percent subsidy
also offered a
at the
same point
To
test this
later in the
to a separate,
in the season.
panel B. free delivery later in the season did not lead to fertilizer
purchases from ICS as often as under the SAFI program (20 percent under free deliver)' vs.
39 percent
fertilizer
1
in the S.A.FI).
The difference between
under free deliver)'
percent level, while
all
late in the
the fraction of farmers
season and any of the other groups
3
shows very consistent
fertilizer
use by 9
SAFI program
the
late in the
up
results: the offer
offered a 50
fertilizer.'"
of free delivery
late in the
fertilizer
season increased
10 percentage points (not significant), less than half the increase due to
(or
SAFI with
ex ante timing choice).
Our model
season will have no adoption impact, since those farmers
this offer
significant at the
and 4) presents the impacts of the different programs on
use, and
to
is
purchase
When
the other groups have similar levels of adoption.
percent subsidy late in the season, 46 percent of farmers bought
Table 2 (columns
who
would have bought
fertilizer
on
their
own anyway
predicts that free delivery
who
are patient
and take
(so purchase with free
delivery would entire crowd out purchases that would have happened anyway). Indeed,
cannot reject the hypothesis that the program had no
As mentioned
earlier,
one issue when interpreting these
or at top dressing (when the plant
visited after planting,
ver\'
few of
it
the farmers
was too
is
results
is
them
to
buy planting
offered fertilizer at
full
although the positive point
that fertilizer can be
knee high), or both. Since fanners
late for
who were
effect,
in
fertilizer for
price
at
in a
future season.
By
contrast,
used either
at
planting
the subsidy and full price groups
use
in that
were
season (however, while
top dressing bought planting fertilizer, 17
percent of the farmers offered the subsidy actually bought planting fertilizer
it
we
SAFI farmers could choose berween
—presumably
to either sell
it
or use
planting and top dressing fertilizer, or
could get both. This would complicate interpretation of the comparison berween the programs
if fertilizer at
top
dressing were not effective. However, our earlier estimates (Duflo, Kremer, and Robinson, 2008) suggest that
the average rate of return to using fenilizer at top dressing only
using
fertilizer at
is
70 percent.
We
planting rather than top dressing as a timing decision similar to
view the decision between
when
to buy.
25
estimate
may
patience, for
suggest that there
whom
may
free delivery'
is
exist
some people who
sufficient to induce fertilizer use. Importantly,
between the percentage point increase due
the difference
increase due to free delivery
are at an intermediate level of
SAFl and
to
the percentage point
significant at the 8 percent level. Thus,
is
however,
we
can reject that the
timing of the offer does not matter.
Interestingly, a
to 14
50 percent subsidy
percentage points), which
is
in
period 2 significantly increases fertilizer use (by 13
very similar
time (and statistically undistinguishable). This
to the
is
impact of the free delivery
at harvest
consistent with prediction three of the
model.
6 Calibration
and Welfare Comparisons
In this section
we
calibrate the
model
to
who
determine the fraction of farmers
are
stochastically hyperbolic, the probability that they are patient each period, and the proportion
of stochastically hyperbolic farmers
choose
SAFI
to take
at a later date.
who have
We
predictions for the fraction of farmers
usage among fanners
who choose
under SAFI. Finally,
we
then
who
a high return to the period
show
that the calibrated
never use
fertilizer
early and late delivery
investment and so
1
model yields reasonable
and for ultimate
when given
fertilizer
ex ante timing choice
use the calibrated model to compare welfare between laissez
faire,
heav7 Malawian-st)'le subsidies, and small, time-limited discounts.
"
'
\'
'
6.1
'
'
'
,
,
Calibrating the Model
Recall that a fraction 7 of farmers are always patient and always use fertilizer and a fraction
ijj
1
of farmers are always impatient and never use
—7—
't/'
=
(6
of farmers are stochastically hyperbolic
any period with probability
To
fertilizer.'"'
The remaining
fraction
(as described above),
and patient
in
p.
solve for the parameters of the model, note that the model implies that the fraction of
farmers using
fertilizer
without the
fanners use fertilizer only
if
SAFI program
is c/j/r
patient in both periods
group usage from the two SAFI seasons
in
Table
1
and
+
7 (since stochastically hyperbolic
2).
Taking the average comparison
2, this quantits' is
about 0.27 (Columns
1-4).
26
Under SAFI,
stochastically hyperbolic farmers
all
fertilizer, as will all
+
will be (pp
percentage
A
is
Hence
time-consistent farmers.
about 0.44
7
+
cplp'^)^,
and
p
=
0.40,
=
0.13
+
0.71,
=
7
0.16, and
of non-program farmers that
we
—
m
50 percent subsidy.
who were
fertilizer
is
given by
=
0.13,
we would
predict that
i'
for three years
+
—
4>{l
p-)^
=
were derived solely from data on average use with and without
at
the correlation in fertilizer use over time, so this provides a first
fit
of the calibradon.
Another check of the model
have seen,
V'
which we observe them (we followed farmers
that these estimates
of farmers using
is
expect to
0.55 of farmers would not use fertilizer in those three seasons.
piece of evidence on the
farmers
we would
follow them. This percentage
SAFI). Given the parameters above,
SAFI, not from looking
the
controls in table 2, this
with our finding that 52 percent of comparison farmers do not
in line
any season
0.71 * .84^
Note
full
equal to 0.16 in our dataset. Solving these equations gives us that
is
These estimates are
after the first
the proportion of farmers using fertilizer
our dataset.
in
third equation gives the percentage
fertilizer in
are patient in the first period will use
Using the regression-adjusted estimates with
7.
find using fertilizer in the three seasons that
use
who
If a
is
50 percent subsidy
patient in period
fertilizer
who end up
the fraction of farmers
1,
enough
is
to
induce stochastically impatient
but impatient in period
under a 50 percent subsidy
using fertilizer under
use
2, to
fertilizer, the fraction
period 2 should be
in
7 which, we
(/^p -I-
44 percent (since the same formula gives us the fraction of farmers who use
under SAFI). Empirically, the fraction
is
m
exactly 44 percent
our data set (see table
2).
To
calibrate A, the proportion of farmers with a high-return period
opportunit)', note that
under the model,
if the
investments
those with low-return period
1
with high-return investments
to
do
value of the discount
choose early
to
so. then a
proportion
delivery and the remainder ask for late delivery.
'"
An
alternative Interpretation
this interpretation,
use
fertilizer
'"*
It
even
is
that these fanners
hea\7 subsidies would be
if the social
We
planner would not do
is
investment
large
enough
deliven,' but not to
—
A)(p
=
fertilizer.
One
possible reason for this
is
induce
induce those
0.35, since
96 percent
not suitable for fenilizer. Note that under
because such subsidies could lead these farmers
to
so.
should be noted, however, that the model cannot match the large percentage of farmers
never used
to
of farmers choose early
therefore set A
have land that
less attractive,
(1
is
1
that farmers
may
who
report having
forget if they had used fenilizer long
in
the
past.
27
of those offered SAFI with timing choice accepted
it,
and 44 percent of those offered
it
chose
early delivery.
The model implies
that
~
^^^o+v'
"^^"^
of those choosing
late delivery
would end up
actually buying fertilizer. In reality, 39 percent did. This again suggests that the
reasonably well
matching
in
Similarly, since the
were not used
statistics that
model
predicts that the only farmers
will be the stochastically hyperbolic farmers
we would
expect that a proportion p
eventually purchase. This
The model does
timing choice.
is
= 40%
is
We would predict that 4>{Xp +
Finally,
further
percentage of 46 percent.
(1
adoption impact of the
—
A)p-)
+
than the basic SAFI), whereas
7
= 38%
in reality
from our calibrated estimate than the other
\s
(5l[\
—
/)
>
y.
Since the
is
fii.
mean
similar to an estimate from Laibson, et
6.2 Laissez Faire,
The
calibrated
laissez fair,
Heavy
47 percent did (more
figures.
whether
is
The condition
of return
rate
a fraction
We
assume
that the
rational).
For
to fertilizer is
the experiments described in Duflo. et
Warlters and Aunol (2005) estimate
a
l3i
<
al.
0.73. This
(2006), to evaluate optimal taxes
this calculation,
we assume
in
Malawi.
We
is
that/is small
20 percent'^ and
from
also use estimates
(2008). which imply that the incremental return to a
marginal cost of public funds of 17 percent
Kleven and Kreiner (2006) report similar estimates
for
OECD
estimates.
in
The marginal
Sub-Saharan Africa.
cost of public funds
could be substantially higher, depending on the choice of taxes implemented and other parameters
andFullerton, 1992).
36 percent
estimate a B around 0.7.
marginal cost of government funds
consider a two-thirds subsidy similar to that adopted
'^
who
fanner
for an impatient
(this is similar in spirit to the exercise carried
O'Donoghue and Rabin
of agents are not fully
(effectively zero).
(2007),
implies
provide a rough companson of the welfare impacts of
heavy subsidies, and small nudges
out in Similar to the approach in
when
to
al.
it
Nudges?
Subsidies, or
model can be used
ex ante
would end up using
(Duflo, Kremer. and Robinson, 2008), this implies that for / close to 0,
estimate
SAFI with
confidence interval of our point
lies in the
one other check on the plausibility of the estimation
use fertilizer
will request early delivery
prefer committing immediately to saving,
implausibly low discount rate of impatient farmers,
to not
who
it.
of farmers requesting early deliver)' will
a bit less well predicting the
than the basic SAFI). Although 38 percent
it
who
ver\' close to the actual
fertilizer in this variant (less
estimate,
to calibrate
model does
.
.
(i.e.,
Ballard
;.
28
second unit of fertilizer
is
-55 percent
market prices, but about 30 percent under
at
a
two-
thirds subsidy.
Under
the model, a two-thirds subsidy will induce
cause patient farmers
to
use two units of
prediction directly: farmers
resell
who do
might take time for fanners
farmers to use
Unfortunately,
fertilizer.
to adjust to
fertilizer but will
cannot
buy
test this
fertilizer
sufficient to cover the transaction costs.
using two units of fertilizer
and then
Moreover,
it
they need to build up
if
need time
assets gradually over time due to credit constraints or they
to a
we
not intend to use fertilizer might
smce heav'y subsidies would be
it
all
to learn
about the return
second unit of fertilizer.
We
assume
that only patient farmers (the
h>T3erbolic farmers
a two-thirds
who end up
being patient
subsidy (as discussed below,
if
always patient farmers and those stochastically
in
both periods) will use two units of fertilizer at
even impatient farmers use two units of
fertilizer
under a two-thirds subsidy, heavy subsidies would yield even lower welfare). These
categories comprise a proportion 0.16
percent use one unit of
we
first
= 27%
0.71 * .4^
of farmers. The remaining 73
fertilizer.
To compare welfare under
subsidies,
+
laissez faire,
heavy subsidies, and small, time-limited
normalize welfare under laissez
faire to zero,
and then calculate the costs
and benefits of heavy subsidies and small, time-limited subsidies relative
With
20 percent marginal cost of funds, the deadweight loss cost of financing a two-thirds
a
fertilizer
subsidy will be 0.2*0.67*[2*0.27+0.73] = 0.170. The deadweight loss from farmers
inefficiently using a
therefore cost 0.3
The
would not use
=
0,
1
second unit of fertilizer
we
first
is
(y
—
1
—
/)('(/'
+
(t>{l
—
P')],
unit of fertilizer and the second term
fertilizer
without the subsidy. If
we
is
where the
first
hea\7 subsidies
use 1.36 for y (Duflo,
will clearly be sensitive to
contrast, the
relative to laissez faire
exceed
term
is
the
the proportion of farmers
et
who
al, 2008). then for
get a benefit of 0.36*0.73=0.26. For the particular parameter values
the costs of
By
0.27*0.55 = 0.149. Overall, heavy subsidies
is
9 relative to laissez faire.
benefit of this subsidy
benefit from the
/
to laissez faire.
we examine,
their benefits, but this
conclusion
assumptions on parameters.
SAFI program described
in this
paper provided farmers a
much
smaller,
time-limited discount, arguably worth less than 10 percent of the cost of fertilizer. Since
SAFI would be taken up by
the 16 percent of farmers
stochastically hyperbolic farmers
who
incurred in financing these subsidies
is
who always
are patient in period
1,
therefore 0.1 * 0.2(7
use fertilizer and the
the total
-f-
pep)
=
deadweight cost
0.009. In addition, there
29
is
a further loss
of (-)
+ p(t>){XR +
investment opportunity. This
likely to
one
and the time
fertilizer is
some farmers have very high
take up S.A.FI.
in this specific
The
example, SAFI
is
many
that yield liquid returns
over the short period between
needed for top dressing (only
(y
—
1
—
f){cp{p
—
a
which
p')),
1
farmers, as few farmers are
few weeks). Also note
rates of return investment opportunity, they
would be
benefit
farmers inefficiently forgoing the period
unlikely to be large for
is
have high return investments
har\'est
that if
- X)R) from
(1
is
would not
equal to 0,06. Overall,
likely to yield higher welfare than either a laissez faire or
heavy subsidy approach under reasonable assumptions about R.
Note
that the
favorable to
parameter values
heaw
subsidies.
we have chosen
The impact of heav>'
marginal cost of public funds
is
for this calculation are ones that are
subsidies
would look worse:
most
(1) if the
higher than 20 percent in developing countries or
if
providing
subsidies encourages costly rent-seeking, (2) if subsidies induce impatient farmers to overuse
fertilizer, (3) if the
fertilizer
never patient farmers
such that the returns
overusing
fertilizer
our model actually have land that
to fertilizer are
to
important to note that
our calibrated model.
We
adopt
is
unsuitable to
lower for them than for other farmers,
has additional environmental costs, or (5)
induce the never patient
It is
in
if
(4) if
even heavy subsidies do not
fertilizer.
we have
not considered the whole spectrum of potential policies in
have obviously examined only one particular level of heavy
subsidy, and other levels might perform better. In our simple two-type model a perfectly
informed policymaker could potentially choose a level of subsidy just sufficient
to
induce
impatient farmers to use one unit of fertilizer while not inducing patient farmers to overuse
fertilizer.
However,
this result
would not be robust
to
more complicated heterogeneity
in
patience or continuous choice of fertilizer quantity'. Time-limited discounts are likely
generically
more robust than heavy
subsidies, in that the losses
from time-limited subsidies
are likely to be limited, while heavy subsidies run the risk of seriously distorting fertilizer use
and incurring large deadweight losses from taxation.
Another conclusion
that
seems
likely to be generic
is
that small, time-limited subsidies
are likely to be preferable to a laissez-faire policy for a wide range of parameter values, so
long as there exists even a small proportion of procrastinating fanners.
with sufficiently
many
stochastically hyperbolic farmers, and sufficiently
farmers, heavy fertilizer subsidies
become more
each fanner. Doing
this
the other hand,
few always-patient
attractive.
The "heavy subsidy" policy could be made more
fertilizer available to
On
attractive
would help avoid
by limiting the quantity of
o\'eruse of fertilizer and
30
would
also help address the
would use lower
quantities.
problem of high
tie
up
their
of heav'y subsidies because people
Another potential policy would be
accounts that could allow them
completely
fiscal costs
money,
to "soft
commit"
for instance
to fertilizer but
—
far less liquid than
accounts could be preferable
case of other
an intermediate
that liquidit)'
is
valuable, these r>'pes of
to a targeted discount.'
The empirical
•
We now
results in this section are consistent with the predictions of the
model
data.
7.1
Farmers are Time-Consistent but the
An
alternative explanation for the large impact of the free delivery of fertilizer
Utility
Cost of Using Fertilizer
are time-consistent, but the fixed cost of acquinng fertilizer
purchasing
fertilizer
much
in large
from
Under
enough quantities
a trusted
source
may
deliver}' later
m
farms (see Duflo, Kremer, and Robinson (2008) for
its
away
own comparison
(as in the basic
a description
to these are
of
advance. Prior
SAFI program).
being implemented
In the
in
a
as
to
number of small
of the demonstration
in different
seasons
in
of the free delivery differed across
SAFI program farmers were asked
about the program, asked whether they wanted
Accounts similar
only worth
group. Farmers were always informed about
after harvest, but the timing
years. In the first vanant, pilot
is
farmers
participated in demonstration plots on
Three randomly assigned variants were conducted
program immediately
in
SAFI program described above. ICS conducted
their
different villages, each with
that
would increase usage
were announced
SAFI programs with farmers who had previously
right
high, so fertilizer
the season
pilot
plot programs).
is
that credit constraints bind. In this case, free delivery
as free deliver.' at harvest if the free deliveiy
full scale
is
Large
is
increase purchase substantially.
model, free
this alternative
implementing the
the
in section
review three alternative models that could have similar qualitative predictions,
and report additional evidence on whether these models can explain the
°
bank
Alternative Explanations
7
2.
fall in
in
force farmers to
holding cash on hand, but more liquid than reselling fertilizer
been purchased. To the extent
that has already
would not
by making money available
emergencies. The transactions costs of such accounts would
category
provide farmers with bank
to
to
to
pay for the
fertilizer
second variant, farmers were informed
order
fertilizer,
Malawi by Xavier Gine,
and given
a
few days before
Jessica Goldberg, and
Dean
Yang.
31
the field officer returned to collect the
went back
similar, but the field officer only
For the three
pilot
money and provide
SAFI programs,
data
money just
to collect the
is
the voucher.
3.
program, between 60 percent and 70 percent of the farmers
initially
because
SAFI, the
in the pilot
place
week following
in the
In all the versions
In the full-scale version
the harvest.
to pay, the fraction
when
percent;
who
fertilizer.
These
because
likely
with the farmer and SAFI was offered
When
of the SAFI program, the
visit
took
the field officer did not immediately collect
the payment, fertilizer purchase falls significantly: from table 3,
few days
ordered
of the
had been working intensely for several months and
field officer harvested
on the very day of the harvest.
before planting.
SAFI program, most
rates are substantially higher than under the full-scale
whom ICS
was
third variant
available only on purchase under the program,
not on evenmal fertilizer use. Results are presented in table
these were farmers with
The
when farmers
from 64 percent
actually purchase fertilizer falls
they are given a few months, purchase
are given a
falls to 17 percent.
to
30
These differences
in
purchase rates remain significant when controlling for various background characteristics.
These
SAFI
SAFI programs were conducted
different
52 farmers
Though
in the
the sample size
who had
to
pay
same schools were offered
is
day
after the han^'est.
that the
to pay,
50 percent
extreme
this
result
initially
is
fertilizer.
made an
stark pattern:
among
47 percent purchased
farmers
fertilizer.
Among
farmers
who had
to
is
is
fertilizer,
fertilizer.
in a larger
evident.
Farmers are Fully Sophisticated, but Resale of
alternative hypothesis
Among
wait several
order but none eventually purchased
probably not representative of what would happen
sample, the sharp decline across the options
Another
same
same season.
wait a few days to pay, 47 percent of farmers initially ordered
to
but only 29 percent eventually purchased
months
the three options in the
small, the results follow the
for fertilizer the
who had
farmers
7.2
To confirm
options themselves, rather than other differences, explain the differential take-up
results,
While
in different villages.
Fertilizer
is
Possible
that farmers are stochastically hyperbolic (as in our
model)
but fully sophisticated. Since these farmers fully anticipate the probability that they will be
impatient
m
the future, they
would
particular, these farmers could
needed. However,
farmers
if
may end up
buy
like to tie their
hands even
fertilizer at the harvest
resale of fertilizer
being so impatient
is
on
in the
their
—
in
until
it
absence of SAFI
own and
hold
it
is
possible, with reasonably high probability these
in the future that
they will
increase consumption. If these fanners buy fertilizer, they
sell the fertilizer to
would pay
a
purchase cost /
in
32
period
and
1.
sophisticated farmers
this, fully
would
a resale cost in period 2, but
fertilizer until
period 2 to see
incurrmg resale
if
who
still
are patient in
they are
still
end up without
penod
1
may
fertilizer.
Anticipating
prefer to delay buying
patient, rather than to
buy
penod
in
1
and risk
costs.
Data on the choice of delivery time under the basic SAFI program provides some
evidence against
this hypothesis.
Recall that
when farmers purchased vouchers through SAFI.
NGO.
they chose a date on which the fertilizer would be delivered by the
could only receive
fertilizer at the
pre-chosen delivery
date.'"'
This feature was introduced
Under
precisely to be useful to farmers needing a strong commitment.
them from impatient period
that fenilizer
needs
almost immediate
to
the hypothesis above,
would take advantage of the SAFI program
patient sophisticated farmers
to protect
2 farmers
fertilizer delivery (this
could be because they thought there was some
to
keep the
farmers
may have
officers
were
felt
fertilizer after
bad admitting
ver\' careful to
were
that the farmers
seem
free to
Of farmers
that
buying
it.
While our data
is
from
to
be strong
self-reports,
emphasize
to
farmers that
this
was not
do whatever they wanted with the
a subsidy
fertilizer
had been spoiled. Thus, unless farmers
the fraction re-sold
in theor}', this is
fi.xed costs
is
bought under
fertilizer they
it
in
still
had the
another season, and 1.6 percent of farmers reported that the
and planned
use
program, and
bought through the SAFI program, 84 percent of the farmers
fertilizer
to
and the
to the field officer that they re-sold the fertilizer, field
report having used the fertilizer on their plot or that of a spouse, 8.1 percent
and
of selling back
flexibility'
guard against resale by future selves). Furthermore, the evidence suggests that
almost nobody sold the
the program.
lock up resources
be applied. In practice, however, about 90 percent of farmers requested
the fertilizer in case of a serious problem, but in any case, there does not
to
to
by requesting delivery just before the time
hazard rate of ICS bankruptcy or because they wanted
motivation
Therefore, farmers
lied
about
fertilizer use. the
probably 6 percent. This suggests that while selling
probably sufficiently costly
in practice,
upper bound on
fertilizer
is
possible
and involves sufficient time delays
of searching for buyers that even impatient period 2 farmers do not think
it
is
worthwhile.
Further evidence against the hypothesis that the main benefit of the program for farmers
was
the opportunity of strong
ICS
field officer offered to
the
SAFI
visit (this
commitment
it
offered
purchase some maize
at a
comes from
premium
program was described above). Since cash
the farmers
from
whom
an
price at the very beginning of
is
more
liquid than maize,
33
farmers might particularly want to get strong commitment
However,
as discussed above, farmers
likely to take
who were
asked
when
they have cash on hand.
to sell their
maize were no more
up the SAFI program than other farmers.
Farmers are Absent-Minded
7.3
Another possible alternative explanation
is
many competing
inconsistency problems, they deal with so
simply do not remember
to
buy
that while farmers are
pressures and issues that they
season even
fertilizer early in the
aware of their own time
when
should (see, for instance Banerjee and Mullainathan, 2008b). Under
officer's visit acts as a
in period
A
1
buy
to
reminder
fertilizer
to stochastically
while they are
'"reminder" inten'ention provides
of post-treatment adoption data
in
impatient farmers
intervention
that fertilizer
many
was
farmers
who happen
manage
to
after the initial
SAFI
treatment, and
(at the
plans
to
use
field officer then
fertilizer,
one
same time the
reminding them
a script,
we had met
available at nearby shops and in small quantities, and that
implement them. The
be patient
support for this explanation. During collection
little
2005 (two seasons
who had made
to
patient.
still
would normally be conducted), and read farmers
in the area
they
this hypothesis, the field
season after the second), field officers visited farmers right after harvest
SAFI
know
they
but subsequently did not
urged the farmers
to
buy
fertilizer early if
they thought they were likely to have this problem (note that this intervention would also
increase fertilizer take up under our model
time inconsistency problem).
surveyed farmers
purchased
at the
fertilizer or
To measure
if
raised p,
making farmers more aware of their
the impact of the inter\'ention, field officers
time of top dressing for the following season to determine
planned
to.
The reminder
whether the fanners either bought or planned
were surveyed (see table
8.
it
to
if
they had
intervention did not significantly affect
buy top dressing
by the time they
fertilizer
...
^
4).
-
.
.
Conclusion
In earlier
'
work (Duflo,
et al.,
2008)
we
presented evidence that fertilizer
Western Kenya but many farmers do not use
this,'
the
model and evidence
in this
it.
Though
is
profitable in
several factors likely contribute to
paper suggests behavioral factors likely play an
important role.
'^
Farmers could also come
'*
In particular,
to the
ICS
office if they lived near town, but in practice very
demonstration plot experiments suggested that
do not know how much
to use,
many
farmers
and would likely have low or negative return
who do
if
few farmers did
this.
not currently use fertilizer
they used as
much
as they think
34
Our model suggests
small, time-limited discounts can potentially help present-biased
farmers overcome procrastination problems, while minimally distorting the investment
decisions of farmers
who do
not suffer from such problems. Empirically, small, time-limited
reductions in the cost of purchasing fertilizer
at the
time of har\'est induce substantial
much
increases in fertilizer use. comparable to those induced by
in the season.
A
'
.
.
may
policy of small, time-limited subsidies
fertilizer
farmers
therefore be attractive.
use for present-biased farmers, but would create
who were
not present-biased.
subsidies might.
not achieve the
who
worth noting
One important
first
caveat
would
is
mmimal
It
would mcrease
distortions in behavior
thus presumably be environmentally
that this policy
heaw
of
more
rent-seeking as large
of small, time-limited discounts does
best from the perspective of a hypothetical period zero farmer, since
are impatient in period
while the
that
It
and would not encourage
attractive than hea\'>' subsidies,
farmers
larger price reductions later
S.A.FI
1
will not take
program boosted
advantage of such a discount. Indeed,
fertilizer
is
it
use substantially from pre-
existing levels, take-up remained quite low.
Calibration suggests that small, time-limited subsidies are likely to yield higher welfare
than either heavy subsidies or laissez faire. However, that calibration ignored the
administrative and staff costs of implementing either type of program. With those costs
figured
the
in.
SAFI program
farmers by field officers,
is
itself,
with
its
too expensive
delivery of small quantities of fertilizer to
(m terms of staff costs)
therefore could not be directly adopted as policy.
program designed
expensive
visits to
to
to
be cost effective and
However, preliminary
mimic key elements of SAFI without individual
results
farms) suggest that time-limited coupons for small discounts on fertilizer
shillings (17 percent) in the price of
94 parents
(there
up
to 5
kilograms of
was no comparison group). Coupons had
identified shops in the region within 10 days,
and
to
1
percent of farmers
they should. Furthermore,
it
seems
who
to
a
reduction of 6
were distnbuted
at a set
field officers obser\'ed fertilizer sales in
redeemed by fanners.
likely that there are important barriers to social learning in this
in
to
of
received the coupon purchased fertilizer (most of them
since demonstration plots led to significant increases
no spillover
fertilizer
be redeemed
these selected location to ensure that the coupons were actually
Overall, 3
a pilot
free deliver)' (and thus
could cost effectively increase take-up. During school meetings, coupons for
Kenyan
from
adoption (presumably due
to
at
environment,
informational effects) but
geographical neighbors or agricultural contacts.
35
.
Though
the end of the ten-day period).
know whether
the
program increased
the absence of a control group
actual fertilizer usage, '^
it
makes
it
impossible to
striking that a
is
reduction in the price of fertilizer immediately after han'est, which
1
7 percent
required a visit to the
still
shop, potentially led to almost as large an increase in fertilizer purchases as a 50 percent
reduction in the cost of fertilizer with free delivery
Since
we
much
of
did not monitor farmers to see
this
was
possibility of
smce
who
offset in reduced purchases
some
resale of fertilizer, but
we
at the
time fertilizer needs
actually used fertilizer,
We
from other sources.
believe
it is
to
be used.
we cannot know how
also cannot rule out the
unlikely there
was much
resale
prices on the resale market typically involve substantial. discounts, and the total
discount farmers received for
5
kilograms of
was only Ksh 30 (about US
fertilizer
This makes a strategy of purchase for resale, therefore, seem unattractive. Overall,
version of the time-limited subsidy
program on
may
fertilizer
is
$0.50).
this pilot
thus encouraging that a time-limited small discount
be an effective, easy
to scale up,
policy to encourage fertilizer use
without distorting decision making and inducing excessive use of
fertilizer.
References
Anderson, Jock, Robert Herdt and Grant Scobie (1985). "The Contribution of International
Agricultural Research to World Agriculture." American Journal of Agricultural
"
Economics 67
Ballard, Charles and
(5):
Don
1080-1084.
Fullerton (1992). "Distortionary Taxes and the Provision of Public
Goods." Journal of Economic Perspectives 6(3):
11
7-13
1
Banerjee, Abhijit and Sendhil Mullainathan (2008a). "Climbing Out of Poverty:
Long Term
Decisions under Income Stress." mimeo. Harvard and MIT.
Banerjee, Abhijit and Sendhil Mullainathan (2008b). "Limited Attention and Income
Distribution."
American Economic Review 98
(2):
489—493.
Choi, James, David Laibson and Brigitte Madrian (2008). "$100 Bills on the Sidewalk:
Suboptimal Investment in 401(k) Plans." NBER Working Paper #1 1554.
Donovan, G. (2004). "Fertilizer Subsidies in Sub-Saharan Afnca: A Policy Note." Working
Paper. World Bank, Washington, DC.
Duflo, Esther, Michael Kremer and Jonathan Robinson (2008). "How High are Rates of
Return to Fertilizer? Evidence from Field Experiments in Kenya." American
Economic Review Papers (Papers and Proceedings Issue) 9S (2): 4?,2^SS.
Dugger, Celia W. (2007). "Ending Famine, Simply by Ignoring the Experts." 777^ New York
Times,
Ellis,
2, sec.
International/ Africa.
Frank (1992). Agricultural Policies
Gulati,
"
December
in
Developing Countries. Cambridge. UK:
Cambridge University Press.
Ashok, and Sudha Narayanan (2003). The Subsidy Svndrome
New Delhi: Oxford University Press.
In fact,
purchases through the program were similar
in the paper,
though there
-
may have been
to control
group adoption
in
Indian Agriculture.
in the
programs reported earher
seasonal differences between those years and the coupon program.
36
Loewenstein, George. Ted O'Donoghue, and Matthew Rabin (2003). "Projection Bias in
Predicting Future Utility'." Quarterly Journal of Economics 18 (4): 1209-1248.
1
Kleven, Henrik Jacobsen. and Claus Thustrup Kreiner (2006). "The Marginal Cost of Public
Funds in OECD Countries: Hours of Work Versus Labor Force Participation."
Journalof Public Economics 90 {\Q-\\): 1955-1973.
Laibson, David, Andrea Repetto, and Jeremy
Tobacman
(2007). "Estimating Discount
Functions with Consumption Choices over the Lifecycle." Working Paper, Harvard.
Morris, Michael, Valerie A. Kelly,
in
Ron
J.
Kopicki, and Derek Byerlee (2007). Ferlilizer
African Agriculture: Lessons Learned and
Good Practice
Guidelines.
Use
Washington
DC: World Bank.
O'Donoghue, Ted and Matthew Rabin (1999). "Doing it Now or Doing it Later." American
Economic Review: '&9{\)AQl>-\2'\.
O'Donoghue, Ted and Matthew Rabin (2006) "Optimal Sin Taxes", Journal of Public
Economics: 9Q{W-\\). 1825-1849.
Sachs, Jeffrey (2004). "The Case for Fertilizer Subsidies for Subsistence Farmers." Working
Paper, the Earth Institute at Columbia University.
Schultz, Theodore (1964), Transforming Traditional Agriculture. New Haven, CT: "^'ale
University Press.
Suri,
Tavneet (2007). "Selection and Comparative Advantage
working paper, MIT Sloan School.
in
Technology Adoption."
Thaler, Richard and Cass Sunstein (2008). Nudge: Improving Decisions
About Health,
and Happiness. New Haven, CT: Yale University' Press.
Warlters, Michael and Emmanuelle Auriol (2005). "The Marginal Cost of Public Funds
Africa," World Bank Policy Research Working Paper 3679.
World Bank (2007). World Development Report 2008: Agriculture For Development.
Washington, DC: The World Bank
Wealth,
m
37
Table
1.
SAFI & Subsidy Programs
Panel A. SAFI for Season
SAFI Season
Income
sfiiliings)
Years Education Household Head
Household had Used
to
Season
Home
Home
Home
has
has
Fertilizer Prior
1
Mud
Mud
Walls
Floor
has Thatch Roof
Observations
Post Treatment Behavior
Household bought fertilizer through
program
Observations
Adoption
in
SAFI
Comparison
Difference
(1J
(2)
(3)
1
1,000 Kenyan
(in
1
Season
of
Program
2.10
2.86
-0.77
(5.51)
(6.70)
(0.52)
6.62
7.20
-0.58
(3.96)
(4.13)
(0.321)*
0.43
0.43
0.01
(0.50)
(0.50)
(0.04)
0.91
0.87
0.04
(0.29)
(0.33)
(0.03)
0.90
0.85
0.05
(0.31)
(0.36)
(0.027)*
0.56
0.52
0.05
(0.50)
(0.50)
(0.04)
211
713
924
.
.
(0.46)
-
-
242
-
0.31
0.45
0.34
0.11
(0.50)
(0.47)
(0.038)'*
204
673
each Panel, means and standard deviations for each variable are presented, along with
differences (and standard errors of the differences) between each treatment group and the comparison
they had
group. The comparison group in Panel A consists of those not sampled for both SAFI, even
been sampled for other treatments (see text and Table 2).
'
Exchange rate was roughly 70 Kenyan shillings to US $1 during the study period. '
'**
'*
*
5%;
10%;
significant at 1%
significant at
significant at
Observations
Note:
In
if
'
'
.'
Table
1
(continued). SAFI & Subsidy Programs
SAFI
(1)
SAFI Season
SAFI with
Subsidy
at
Full Price at
Timing Choice
Top Dressing
Top Dressing
(2)
{3)
(A)
Comparison
(5)
2
Means
Baseline Characteristics
Income (in 1.000 Kenyan shillings)
Years Education Household Head
Household had Used Fertilizer
to Season 1
Home has Mud Walls
Home
Home
Prior
has Mud Floor
has Thatch Roof
Obsen^ations
2.84
2.86
2.29
2.81
2.40
(7.53)
(7.36)
(4.01)
(6.68)
(4.47)
6.99
6.84
7.13
6.99
7.58
(3.98)
(4.12)
(4.13)
(4.02)
(4.30)
0.42
0.41
0.38
0.44
0.51
(0.49)
(0.49)
(0.49)
(0.50)
(0.50)
0.88
0.89
0.86
0.91
0.87
(0.33)
(0.32)
(0.35)
(0.29)
(0.34)
0.83
0.88
0.85
0.89
0.86
(0.38)
(0.33)
(0.36)
(0.32)
(0.35)
0.53
0.53
0.49
0.55
0.53
(0.50)
(0.50)
(0.50)
(0.50)
(0.50)
228
235
160
160
141
0.39
0.41
0,46
0.20
_
(0.49)
(0.49)
(0.50)
(0.40)
-
208
207
145
143
-
0.38
0.47
0.41
0.33
0.28
(0.49)
(0.50)
(0.49)
(0.47)
(0.45)
179
208
133
135
102
Post Treatment Behavior
HH
bought
fertilizer
through program
Observations
Adoption
in
Season
of
Program
Observations
Differences Between Treatment and Comparison
Baseline Characteristics
Income
Years Education Household Head
Household had Used Fertilizer
to Season
Home has Mud Walls
Prior
1
Home
Home
has
Mud
Floor
has Thatch Roof
Observations
Post Treatment Behavior
Adoption in Season of Program
0.440
0.456
-0.110
0.402
(0.727)
(0,714)
(0.514)
(0,692)
-0.595
-0.740
-0.456
-0.588
(0.440)
(0.446)*
(0.487)
(0.479)
-0.094
-0.100
-0.129
-0.073
(0.053)'
(0.053)*
(0.057)**
(0.058)
0.005
0.013
-0.012
0.034
(0.035)
(0.035)
(0.040)
(0.036)
-0.034
0.018
-0.010
0.029
(0.040)
(0.036)
(0.041)
(0,038)
0.006
0.003
-0,031
0.025
(0.054)
(0.053)
(0.058)
(0.058)
228
235
160
160
0.105
0.197
0.139
0.051
(0.059)*
(0.059)***
(0.063)**
(0.060)
179
208
133
135
each Panel, means and standard deviations for each variable are presented, along with
differences (and standard errors of the differences) between each treatment group and the comparison
group. The companson group consists of those not sampled for both SAFI, even
they had
been sampled for other treatments (see text and Table 2).
The number of obser^'ations is the number of farmers in each group with non-missing adoption data in the
season of the program.
Exchange rate was roughly 70 Kenyan shillings to US $1 during the study pehod.
"' significant at 1%
* significant
at 10%; " significant at 5%;
ObsePv-ations
Note:
In
if
141
.
Table
2.
Adoption for Parents Sampled for SAFI & Subsidy Programs
Used Fertilizer
Season 1
SAFI Season
Starter Kit
Starter Kit
1
Farmer
Farmer
Demonstration
'
Plot
School
Demonstration Plot School
Household had Used
to
Season
Fertilizer Prior
1
has
mud
(1)
(2)
(4)
(5)
(6)
0.143
0.007
0.007
0.006
0.01
(0.035)***
(0.038)***
(0.041)
(0.044)
(0.037)
(0.041)
0.059
0.080
0.024
0.005
-0,009
-0.027
(0.042)
(0.046)*
(0.047)
(0.051)
(0.043)
(0.048)
-0.026
-0.061
0.024
-0.005
0.004
-0.031
(0.060)
(0.066)
(0.068)
(0.075)
(0.063)
(0.070)
0.006
0.441
0.362
0.464
0.362
0.437
(0.314)
(0.435)
(0.460)
(0.463)
(0.335)
(0.465)
0.251
(0.037)'"
0.369
0.315
0.319
0.284
0.281
(0.031)**'
(0.035)***
(0.035)**'
(0.040)*"
(0.033)"'
walls
Education primary respondent
0.012
0.014
0.026
(0.033)
(0.037)
(0.034)
-0.193
-0.183
-0.021
(0.081)**
(0.091)"
(0.085)
0.004
-0.004
(0.004)
(0.005)
0.015
(0.005)"'
past month
0.004
0.006
0.002
1,000 Kenyan shillings)
(0.003)
(0.003)"
(0.003)
Income
(in
In
716
876
Observations
Used Fertilizer
Season
1
Panel B. 2004 Season 2 Treatments
SAFI Season 2
SAFI Season 2
with
Choice
on Date of Return
Half Price Subsidy Visit
Full Price Visit at
at
Top Dressing
Top Dressing
Bought Maize
-.'
Bought Maize
'
SAFI Season 2
Household had Used
Season
Male
to
Fertilizer Prior
1
Home
has
mud
in
Mean Usage
Used Fertilizer
Season 3
(3)
(4)
(5)
(6)
0.165
0.181
-0.024
-0.005
(0.053)
(0.057)
(0.061)***
(0,066)***
(0.056)
(0.061)
-0.014
0.03
0.207
-0.027
0,003
(0.048)
(0.053)
(0.055)***
0.216
(0.060)"*
(0.050)
(0.056)
-0.035
-0.039
0.142
0.127
0.023
0.041
(0.052)
(0,057)
(0.059)**
(0.065)*
(0.054)
(0.061)
-0.065
-0.034
0.096
0.104
-0.053
-0.031
(0.052)
(0.058)
(0.059)
(0.066)
(0.054)
(0.061)
-0.002
-0.011
-0.042
0.079
0.002
-0.014
(0.043)
(0.048)
(0.049)
(0.054)
(0.046)
(0.050)
-0.048
-0.073
-0.085
-0.057
0.005
-0.011
(0.075)
(0.082)
(0.087)
(0.096)
(0.080)
(0.087)
0.248
(0.037)'"
0.369
0.316
0.325
0.283
0.278
(0.031)***
(0.035)***
(0.035)***
(0.040)*'*
(0.033)*"
.
shillings)
Comparison Group
734
(2)
past month
1,000 Kenyan
Used Fertilizer
Season 2
902
0.042
Education primary respondent
Income
626
(1)
walls
•'
756
-0.009
--'""-"
(In
M
Used Fertilizer
Season 3
0.114
Male
Home
Used Fertilizer
Season 2
0.01
0.014
0.028
(0.033)
(0.037)
(0.035)
-0.197
-0.197
-0.017
(0.081)**
(0.091)*'
-0.086
0.004
-0.003
(0.004)
(0.005)
0.015
(0.005)*"
0.004
0.006
0.003
(0.003)
(0.003)"
(0.003)
0.297
0.300
0.392
0.397
716
756
626
902
734
the farmer adopted planting or top dressing fertilizer In the
Note: Dependent variable is an indicator equal to 1
given season. All regressions control for school, and whether the farmer was a parent of a Standard 5 or 6
child (see text). Panel B also Include controls for the Season 1 Treatments listed in Panel A.
The comparison group means listed In the bottom of the table are for Individuals that did not participate In either
SAFI, were not offered fertilizer at any price, and did not participate in the starter kit program. This accounts
for the difference in mean usage between this Table and Table 1
There are fewer observations than In Table 1 because of missing values for previous usage. For all programs
listed above, respondents were allowed to buy either DAP (for planting) or CAN (for top dressing) fertilizer.
Exchange rate was roughly 70 Kenyan shillings to US $1 during the study period.
Standard errors In parentheses. * significant at 10%; " significant at 5%; *** significant at 1%
Observations
in
0.247
876
if
0.228
•
S
"
CO
ocoojS'^'—
,^~
ocD'3-S'^ej)O^CD^
'r-
CO
OOt-CO'-o'^^O
O
CD
ddd99d'9ddddd
f,l
TT
^
to
d
CO
§ll^
to
OJ
cu
Sou
Ferr/i
c
d_
O
d d P
d d
o g o
S
o
CO
B :^5i
m
O.
CO
d
Initi.
Acce
^
"cr
;
CM
.
C31
'
°
*-
S
CM
do
o
QJ
J
^
o
?j
d_
"O
^m ^
o
J
•^
CNJ
C3^
r^
00
^
cjo
d d d d
Q.
i_
o
r^ irT r^ in o
OJ
m
d d d d d d
in
5_^
CD CD
O
O
CO
00 00
o CD
^ d d
Ccl
Cvl
tl
II
;
^
u
=1
LO
to
:
o
CM
in
—
H?
m
^ CO o
CD
5^
o
o
° d 9 ^
CO
en
0^
'^
^ ^
5
o
CO o
O
.
^
"
o o T g
9
d d> d> '^ ^
CO
01 eg JM
T^ "5 CO
CD
CO en
C3
-.r\i
1
r^
CO
^
*_^
qT
C\J
T- T- -^ CO "D
0)
KH
o o -
1
^.^ CM
*
e- CO
^
d o d o
d o
d
d_
d
c
.
.
CD
^
"tB
•^
r--
LO
Ln
CO
CO
3g
—
d
OJ
E
"C
Q)
r^
CO
<D
q.
^
^
un h-
CD
r^
00
CO CD
00
•^
^
CO en
in r^
0)
_2
.£
CO _^
X3
.52
o o
—
hm
o
o
d o
d d d_ d d
&
= 8
"5:
(T"
c
-DO
~ en
0)
0)
n.
to
CD
<"
e"
"D
e^
d
—
=1^0)2
^
P
en
g
Poo
9 d d
r~~
^0)
*-
CD
n3
9 c S
9 o d
1
i
-D
=
to
oi
.
"
c
o
CO
x:
is
g
!=
o
to
CO
CD
en
in
N
=
t:
—
CD
>.
:3
—CO ^
en
_in
to
<^ rr^
cn cp
CM
-D
E o
o
5 O
^ oo ^ o ^
g
(0
to
o
.
.
.
_aj
^'
0=^
°§9
-2
=
en
> ^ CD
<—
r- CO c3) _,
CD •- CO JJ
CO r^ CM S>J
o
O
o>
.
ra
"^
i
CD
CO
IS)
> -B
'cD
3
E ^
CJl
CO i
a> a> ID CO CD en
un N. to oo in CO
O)
i
S
9
9 g
o o o o o d
to
TO
CO
f—
t/1
"c^
.=
nj
g 5
en
dI
;
T
^
CO CM CO in
oi
CM
.
o •_^ •^
CO ^ o CO CO CO
to
CO CO
m
o
o
d o
d d d d d
tb
^ S^
'^
>
^
„
ECO
^
s
*
00
«
m S "
9~
o
o
d q
d d d d d
2^
_o
<
V)
CD
CM
in
re
:
1
-§? =
o
OJ
in in CD
in
in
in to in in
_
_
:i-l
"cn
to
ra
||
1
D
O
a
_ra
~ Q
tn
1
-rr
•
:
o
>
f^
r^
99^0
*
TD
<U
"^
r- r^
LL.
in
S
D
:=
d
CO CO Ci^
T- C=
o ^
r^ CD f^ o
^ Ol CM a> o O
d o d d
d o
d
d
T—
t^
r^
San
d
S -
r~
CT)
O ^
^
GO
^o
in
CO OJ CO
(O CO CNJ
t-^
to
*
.-
XJ
=
CD
0)
CD
.0
N d p
= C H
o
ai
?^
o =
i=
m
f/^
tr
to
CO
a?
o-^
re
i - ^
O) ^
=
.y
.
S^
O
-^
^
^
"
en
"
:^
£ 2
OJ
<P
Q)
°
_3 _3 _D
ra
ra
CO
"o
> > >
CO
5
QJ
w
-
<
m
Q.
o
§ ^
"*-
QJ
-3
i
^
ro
^
Q-
k.
o
"S
>
o
C\J
«
=O
cj
03
*-
CO
—
CJ
=
0)
^ c ^ o
Q.
o
— .—
o
±:
^
ci>
=>
^ ^
- C C C
"D
u
en
~
s
C
r;
1
E
£
en
Q.
.
c
.9
U
S 3
OXO
CL
Q)
"o
w
en
_w
ffi
OJ
*J
to
CO
5
Z.
S
-Q
.^
li ul ul
ra
X)
O
55
C
0)
I
=!
D.
D
-0 CO
g
to
c
"to
"D -a
en
CD
CO
E c
a.
re
a>
CD
CO
Q
q
c
CD
Q.
Q.
01
(11
E
a)
CD
ro
CO
to
E
Q)
.0
c
CD
53
c
re
CO
CD
(J
en
5
to
0)
CC
c
CD
^2
CD
en
OJ
^
-77.
<
"re
CO
>^
™ 5 O en
S j; g- 9
>
en
sen
ro
£ S
>
CO
c o o o S
5223
C=
CM
en
CD
O)
Li.
g
Q.
0^
o
2
CO CO
>
<2
^
t
cu
to
5
Q. Q. Q.
CNJ
c
-D _§ ">
Q.
"P
en
a;
m m
en
lo
1=
'—
p £
"O
0)
^
"
•c
CO
"D
ro
"tc
c
CD
£
C/3
[0
Z
'
3
o
o
TO
CO
CD
T3
c
c
o
r^cDLntDLnoocMCM'd-^-rococncocn
'^fcO'^T-coiDLn'^OLOroLficoLnaDLn
oocNJcoqqooqqqqoqqq o
o — dooooooooooooo'ocJ o
6 2.
c
g
o_S O
E
o
CJ
O^ CM
-^
o
TO
o
CD
Vi
-a
<
O
c
O
TO
C
c
o
r
Cl
p
TO
r^ccLn<X)LnoocMC\j'd-r--rococxioocnoco
^n^-^coLnLn-^otocoLnooLncoLnon
oociooodo'oooooooooo
^O
Eg
TO
,"^
CO tP
r~~
C
C
g
O
CO
CM
to
CD
3
in
o
o
<
CD
CNJ
TO
CD
O
^
e
8
o
'to
^
o
O
^
"O
c
o
'-^
cn.
^
S
E
S O
:
o CO ro
O
o o vt
o o ^
o ^
o
d d 9 o o o o o
OT 05 r: OT -^
T- LO LO
CO
<^ CO
^
n
o
o
o
^
O O
o o o
o o d d —
CM cr> O)
CD
LO CO LO
CD CNJ 55
•
^§
o 2
TO
00
TO
^-
.;=
^
o
o
o^
CD
TO
a>
O
CD
CD
TO
^
CD
V)
c/5
'•
OV)
c
o
a
op
-
3
CT
-a-
JD
O
.2
CD
=
x:
o
o
5? t3 "o
LO TO !D
03
^ ^
-S
4—
TO
a>
g
Q.
CC
CO
«
(/)
o
o
o
C
'E
o
En
D-
E
»<
TO
UJ
LL
c
o
c
o
E
i5
^
C
o
F
E
TO
U)
Q
O
O
TO
CD
CO
l/J
^
CD
^
CD
C
00
O
c
c
cn
TO
CD
CO
CD
TO
CD
CO
c/:)
CO
C/5
u.
LL
LL
c
o
<
00
CM
CM
in
TO
CU
re
CM
i
C/5
o
<
w
o
<
W
TO
>
CD
N
n w
CD
O)
o.|
—O >
c;i
3
o
Ol
^3
Cl.
CD
Q.
—
^—
TO
CD
C/5
LL
CO
5 °
8 ^
"a
2
>
•a
_.
o
o
CD
TO
'co
^
C
O
O
r-
*
LL
03
CD
CJ
Q
TO
0)
i—
H
c
c
o
g
^
:
<
'to
CO
CD
o
-
en
TO
CO
U
CO
TO
TO
c/5
CD
CO
J3
CO
CD
9
8
^
CO
to
CD
^
O
^ O Z
CD
CD
CC
TO
CD
>
O
'
•D
c
O
E
o
m
*
f^ -- 00
CO
';^ CO
a> a>
CO un CO CD
C^ CO -vt- CO CO c\i
CO
CT,
o
q
o
o
q
>CO >- ,—
CO CO
1^ 1^ 00
Ln
CD •* CM
5j 1^
i< eg r^ cvj cri CO
m
Z
m
in
o „
in in
CO
-^
CO CD O) CD
M- 00 CD
CO
CD CO
CO CM -^ Ln
O
O
1—
o ©
£
=:
c w so ~^
a> ~
E
o o
o
CD
oj
CO
^'
^
^
uo CM in a>
-^ h- r- in
r^ CD CO CO CO
cJ T-' c\j c\i in
o _
in CM
CD CO CM ct;
CO CO CM CO
D.
T3
O
to
ro
^
0)
E
o
"o
o
O
CO CD
CD o m
^ en en
o o_ o o_ d
O
^U ^
—
CD
OJ CD
y: •^
cr
IT)
TO
I -C
H
en
ro
sz
0)
E
o
TT-
in
o
a.
„
in
CM
in CO CT) CO
53CD C3^ CD
_a)
D.
E
o_
TO
CD
en
a.
m
b
o
O
^
-D
"9-
o o
CM
CM
5
CO CO CO
o o o o O o
s
1-^
13
t3)
^ „
C
in in
CD CO 00 CO
in CD CD CD
O —
iZ STO
CD O
O
CO
m -^
rC ^ r^
O 0_ CD
TO
0)
5
P T3
O 3
I 5
03 CO CO
T- CO 03
CO CO
'a0_ C3 O,
<
o
CO —
^ °
° ^
in o 03
o
CD CO M- CO
CO
o
b
TO
C
c
u>
.9
(5
<D
03
>
CvJ
o o
^ CO
CO in ^
5 o ^ CM o
l<
CO CD '^
CO
CM
o
3
"D
LU
CT)
'a-
'*'
i
CM [^_ in
CD f^ r^ CM
•^
CM
CM
CD CD CD
O ^
TO
00
O
?
c
-^
O
o m
CD
k-
O
* O^
o O
^ in CO ^
O O^ C3 0_
CD
IT)
N
ro
•a
o
'c
o
Q)
C3)
O)
2
O
"3 LL
ID "D
r--.
CM Tin CO
CO =!•
CD 0_
m
O
r^
en
CD
tn
t^ CM „, S
rj- CO JJ
_^
03 r'at3
CD t3 CD
W
CO
TO
CD
S
^.
"(5
CD
_2 _2
N
E
cco
TO
TO
CD
^
o o
_Q.
TO
LL
CD
>
^
TO
>
n:
TO
S
0^
^
s
^
TO
TO
O
.t^
cn
C
C
CD
C ^ c
^ O ^ o
^ E 3 E
£ o "S o
a
o
^
TO
i>
JC
TO
TO
> ^c
FT"
o
<
a.
C/)
.
.
1
o o
o
CM
_Q)
C O
O
CO
^
o
o
g o o o
Q.
Q.
o
O
II
CO
Q. Cl D.
O O O
to CO W
CD
V
CD
Q)
=
TO
CD
CO
"V -^
LL U. LL
"V
cn
TO
TO
>,
A
Q-
Q)
> C
o
—
^
C
TO
TO
CO. '^.
-D
^?a. :S^
TO
CD
tn
CD
in
^
„
o o
o £
c
^
CD
CM
o o
cn
T-
Q. Cl
;=:
-^
CD
o o
1
v>
^_
.
^
t:
CM CO CO
-;;
</>
00
a. a. a.
o
E
S
TO
Q)
TO
en
O)
> > >
--
T-
^ S>t
LL C
cn
C/3
(D
p
B 5
o o S
TO
o
Q)
ro
^ Sl3cM
^.E
3
C
H ^
^
=J
-i^
O
LLI
TO
O
g
=5
£
5
> "3
S2'
=
—
O
r^
>
ra
•
CD
T-
—
O p TO
^
C
CO
ai CD E '~
TO
O
J. TO t; <D O
=
m O E
r XJ o !=
TO
O)
CJ
CD
o CD .C
£ 00
CD
OZ ^
—
'
— Q
w
c
0)
F
o
CO
CO
0)
CO
CL
E
o
o
x:
0)
CO
C31
y
8
N ;r
^
^
CD
^
m
c
^
c
m
in
ni
tn
re
CD
U
V)
c
a.
o
N
>-
^re o
O
:5
tD
m
TO
CD
'"^
re
CM
^
CD
g'
O)
CJ
00
00
CD
P
fTi
re
•
•C
-"^
r-
CD
00
-!
re
D
U c
n
m ^
r
(
CD
CO
w
E
)
re
o
>, CD
'to
c
C/3
O 5
S^ .N "^
^ ^
CQ
> E
CD
^
CD
-n
c:
p
00
C/D
CD
w
CM
c
£^O
CM
CO
_
CD
CT
O
re
i-
0
?
CD
00
r^
cz
•^
F
TO
CD
CT
O"
W
3
CD
Q.
^^
(1>
C/)
l!^
Q)
r
cc
^
?-
^
C
CO
>•
CD
N
-—
O
o c
^
CD
"a TCD
(/)
CD
0)
i
a3
00
>.
CO
re
D
c E
en
^t=
2
CD
re
"O
IT
re
CD
E
00
re
00
W s
?
-!J
^
0-''
4)
i_
O
5
re
CO
^
CD
«
Tj
rn
CD
CD
a.
F
a
^
Q
CD
<n
re
ro
r
CD
c
u
CD
1-
CD
CD
^
re
re
CD
CD
LL
F
re
CD
0
CD
"^
c:
CD
CO
1
ID
o
CO
re
JD
0)
TO
0)
"in
CD
T)
C/)
re
(D
c-j
^
CD
o
N S"
re
Q.
—
'
0)
^
CM
D
CM
S
SI
CO
tr
u
CO
3
1—
CD
Q)
x:
>
_C
CE
!^
—J
TO
E
CO
C/5
CM
CD
:"
CO
C
CU
>-
N
2'
O
-C
CO
c/5
CD
- S
™ 5 ^
^
Q (g
u
^"
QJ
p
^
CM_
CO
r
iZ
<
S
00
CO
ICS
Bought
Maize
i
3
O)
1_
CD
00
=z
lU
X3
u.
X
T3
'
2
<
U)
Q-
F
Q.
c
m
CD
<D
(62)
iZ
J-
CO
C
CA)
i
Ljy
^
re
CO
CD
CD
2
re
05