1. The Question
Maize is Kenya's staple and smallholders are its principal producers. The puzzle that motivates this paper is the size and persistence of the yield gap. In a recent county-level study, average maize yield was about 2.2 tonnes per hectare against a stated agronomic potential of 5.5 tonnes, with fertilizer application still below county-recommended rates of 200 kg/ha of NPK and 150 kg/ha of CAN.
The standard diagnosis is that fertilizer is too expensive, or that farmers lack the cash to buy it at planting. The standard remedy follows directly: lower the price. Kenya has done this on a large scale and for a long time, through the National Accelerated Agricultural Inputs Access Programme (NAAIAP), through the NCPB scheme running since 2008, and since 2022 through the NFSP, a broad programme providing subsidized fertilizer to all registered farmers with quantities based on registered land size.
Earlier papers in this series have repeatedly found that Kenyan development problems diagnosed as "lack of access" turn out, on inspection, to be problems of instrument design: mobile money expanded financial access without touching the compliance costs that govern formalization; venture capital flooded into Kenya without reaching the median youth-led firm because the instrument was shaped for a different kind of business. This paper asks whether fertilizer subsidies belong to the same family of mismatches.
The question is therefore not "does fertilizer raise yields?", it plainly does, but: is the price of fertilizer at planting time the constraint that actually binds for the farmers Kenya is trying to reach, and if not, what is a universal price subsidy actually buying?
2. Four Mechanisms, Only One of Which a Price Subsidy Addresses
A price subsidy is a precise instrument. It lowers the cost of a unit of fertilizer at the moment of purchase. It is therefore effective if, and only if, the farmer's problem is that the price at the moment of purchase is too high relative to the expected return. Four distinct mechanisms can produce low fertilizer use in Kenya, and the subsidy addresses one of them.
Mechanism 1: Intertemporal mismatch and present bias. Smallholder income is lumpy. Cash arrives at harvest; fertilizer is needed months later at planting. A farmer who fully intends at harvest to buy fertilizer may find, by planting, that the money has been absorbed by school fees, medical costs, social obligations or simple consumption. Formally, this is a savings and commitment problem rather than a price problem. Duflo, Kremer and Robinson model it as farmers who are stochastically present-biased and partially naive: patient enough at harvest to intend to invest, impatient at the moment the decision must actually be executed, and insufficiently aware of their own future impatience to take precautions. Under this model, a price cut at planting arrives too late to help: the farmer's problem is not that fertilizer costs too much at planting but that there is no money at planting. Crucially, the model predicts that a small intervention delivered at the right moment can outperform a large intervention delivered at the wrong one.
Mechanism 2: Heterogeneous agronomic returns and soil quality. The assumption that subsidized fertilizer converts into yield at a stable rate is agronomically false. Marenya and Barrett document a von Liebig-type relationship between soil organic matter and maize response to nitrogen: below a threshold of soil organic matter, additional nitrogen produces little additional yield, because the limiting factor is not nitrogen. Soil acidity compounds this. Where pH falls below roughly 5.0, exchangeable aluminium damages root systems and impairs uptake of nitrogen, phosphorus and potassium, so fertilizer applied to such soils is partly wasted regardless of price. If returns vary this much across plots, a uniform price subsidy applies the same incentive to plots where fertilizer pays handsomely and to plots where it does not pay at all.
Mechanism 3: Inframarginality and market displacement. A subsidy only changes behaviour for farmers who would not otherwise have bought. For farmers already purchasing at market prices, the inframarginal group, the subsidy is a pure income transfer. It may be defensible as welfare policy, but it buys no additional yield. Worse, subsidized fertilizer distributed through state channels competes directly with commercial agrodealers, so some of the subsidized volume substitutes for commercial volume rather than adding to total use, and dealers facing delayed reimbursement may reduce stocking, thinning the private distribution network that non-recipients depend on.
Mechanism 4: Delivery-system and quality risk. Any subsidy that moves physical inputs through a state distribution monopoly inherits that channel's administrative and quality-assurance risks. A farmer who receives adulterated or substandard fertilizer suffers worse than a farmer who received nothing: they have paid, foregone the alternative, and applied a product that may damage the season's crop. This is a failure mode that a cash transfer or a voucher redeemable at competing private dealers does not share in the same way.
Only Mechanism 3's income-transfer component and the narrow price channel are addressed by a price subsidy. Mechanisms 1, 2 and 4 are untouched by it, and Mechanism 4 is actively created by the delivery design.
3. Evidence
3.1 The yield associations are real but are not causal estimates of subsidy effect
A study of 300 maize farmers in Malava and Mumias East sub-counties of Kakamega County found that subsidy participants applied 85.6 kg/ha against 74.9 kg/ha for non-participants, and achieved 2.46 t/ha against 1.97 t/ha. Controlling for fertilizer quantity, seed quality and household characteristics, participation was associated with 32.3% higher maize productivity. Subsidized fertilizer accounted for about 59% of participants' total inorganic fertilizer use.
An evaluation of NAAIAP (Mason et al., 2017, as summarized by UC Davis's BASIS programme) found productivity gains of up to 30% and an 11% reduction in poverty severity. Tegemeo Institute survey data show fertilizer users out-yielding non-users by roughly 50%.
These are genuine and consistent findings. But note what they measure. The Kakamega comparison is cross-sectional between participants and non-participants, and participation is not randomly assigned; farmers who enrol may differ in access to information, proximity to depots, or land quality. The Tegemeo comparison is between users and non-users of fertilizer generally, not between subsidized and unsubsidized farmers. Neither isolates the causal effect of the subsidy as distinct from the effect of fertilizer, and the policy question is precisely about the former.
3.2 The experimental evidence says timing beats price
The most rigorous evidence on Kenyan maize farmers comes from Duflo, Kremer and Robinson's work in Western Province, covering 1,230 small-scale farmers between 2000 and 2004. Three findings are decisive for this paper.
First, returns were high and adoption was nonetheless low. Top-dressing generated roughly a 70% annualized return in Western Province. Yet only 43% of sampled farmers in Busia district reported ever having used fertilizer, and in the study's own sample usage was 24% in one season and 26% in the next. A price-constraint story struggles here: fertilizer was already profitable at prevailing prices, and most farmers still did not buy it.
Second, a small, well-timed offer matched a large, badly-timed one. The Savings and Fertilizer Initiative (SAFI) offered farmers free delivery of fertilizer if they purchased immediately after harvest. Basic SAFI raised usage by 14 percentage points in the first season and 18 points in the second, a 69% increase over the base rate. A version allowing farmers to choose their delivery timing in advance raised usage by 22 points. By comparison, a 50% price subsidy offered two to four months after harvest raised usage by 13 points. A free-delivery nudge at the right moment did as much as, or more than, cutting the price in half at the wrong one.
Third, the timing, not the sweetener, was doing the work. A free-delivery visit offered later in the season, rather than right after harvest, produced no significant effect on usage. The same inducement, moved a few months, stopped working. This is difficult to reconcile with any pure price or transaction-cost story and is strong evidence for the commitment-and-timing mechanism. The authors also note that SAFI enrolment did not raise usage in later seasons, which argues against a learning explanation: farmers were not discovering that fertilizer works, they were being helped to act on an intention they already had.
The authors' calibration concludes that small, time-limited discounts deliver higher welfare than either laissez-faire or heavy subsidies, because they help present-biased farmers commit without pushing farmers with standard preferences into over-application.
3.3 Soil quality caps the return, and does so regressively
Marenya and Barrett, using plot-level data from western Kenya, find that fertilizer is profitable on average in their sample but unprofitable on roughly one third of plots, where depleted soil organic matter suppresses the marginal productivity of nitrogen below what prevailing prices justify. They further observe that poorer farmers disproportionately operate soil-organic-matter-deficient plots, and draw the uncomfortable conclusion that fertilizer-centred interventions may be less pro-poor than commonly assumed and could reinforce existing inequality.
This finding deserves more weight in Kenyan policy debate than it receives. It means a universal subsidy is not distributionally neutral in real terms. Two farmers receive the same nominal benefit; the one with better soil converts it into grain, the one with degraded soil converts less of it, and the second farmer is systematically more likely to be poor.
Soil acidity points the same way. Acid soils cover roughly 13% of Kenya's land area, and in acid conditions fertilizer efficiency falls because aluminium toxicity impairs root function. Trial data presented by KALRO and IFDC from western Kenya illustrate how large the non-fertilizer margin can be: maize yields of 0.53 t/ha with no inputs rose to 1.59 t/ha with 2 t/ha of lime and 2.84 t/ha with 4 t/ha of lime, with lime plus phosphorus performing comparably to phosphorus alone in that particular trial. (The source presentation flags ambiguity in its own table headings, so these figures should be verified against the underlying studies before external citation.) The qualitative point survives the caveat: on acid soils, liming can move yields by more than a fertilizer subsidy does, and Kenya does not run a national lime subsidy of comparable ambition.
3.4 Most of the money is an inframarginal transfer
The displacement evidence is the clearest indictment of the universal design. Adoption had already reached roughly 70% nationally through ordinary retail markets before the subsidies began, helped by a 27% real decline in fertilizer prices after the input market reforms of the 1990s. In a Tegemeo sample, 85% of NAAIAP recipients had purchased fertilizer consistently in 2000, 2004 and 2007, meaning the targeted programme largely reached people who were already buyers. Makau et al. estimate that each kilogram of NCPB-subsidized fertilizer displaced about 0.2 kg of commercial fertilizer, and a 2013 World Bank assessment found that delayed government reimbursement to agrodealers reduced their confidence and delayed restocking.
Against this, NAAIAP reached 537,218 farmers by 2016 at a cost of KSh 517 million, while the NCPB scheme cost KSh 31 billion through 2018. The NFSP, by providing to all registered farmers, is broader still by design.
3.5 Thin margins and the risk problem
A review by Jayne and co-authors estimates value-cost ratios for Kenyan fertilizer ranging from 1.05 in western and central Kenya to 3.7 in the eastern lowlands, and notes that ratios below 2.0, while profitable on average, generally do not compensate farmers for the risk of drought or flood losses. A farmer deciding whether to spend scarce cash on an input with a value-cost ratio near 1.05 and a meaningful probability of total crop failure is not behaving irrationally by declining.
This produces an uncomfortable juxtaposition. The Kakamega study, in western Kenya, reports a 32.3% productivity association. The Jayne review places western Kenya at the bottom of the national profitability range. Both can be true, a large proportional yield gain on a low base, with input costs consuming most of the additional output value, but together they suggest that in exactly the region generating the headline yield result, the net income gain to farmers may be small. This is our inference from combining two sources, not a finding either reports.
3.6 The delivery channel failed in 2024
In March 2024, the Kenya Bureau of Standards found that fertilizer distributed by KEL Chemicals failed required test parameters, covering consignments distributed between 5 and 10 March 2024 under brands including Kelphos Plus, Kelphos Gold and NPK 10:26:10. KEBS reported withdrawing 5,840 bags, with 560 bags seized at the NCPB depot in Molo, Nakuru County, on 23 March 2024. The government suspended KEL Chemicals' operations and paused NCPB distribution pending investigation, and advised farmers to stop using the product. Published reporting did not quantify how many farmers applied the affected product before the withdrawal.
The economic significance is not the scandal itself but what it reveals about the design. Channelling a national input programme through a single state distribution network concentrates quality risk at one point. A farmer who applies substandard fertilizer loses the input cost, the season's yield potential, and, critically for a programme whose justification is behavioural, confidence in the institution the following season.
4. Argument
Read together, the evidence supports a conclusion stronger than "subsidies are poorly targeted." It is that the universal price subsidy is the wrong instrument for three of the four constraints that actually limit Kenyan maize yields, and is the direct cause of the fourth.
The experimental evidence is the pivot. If fertilizer already returned roughly 70% annualized in Western Province while adoption sat between 24% and 43%, then a substantial share of non-adoption was never about price being too high relative to return. It was about the money not existing at the moment the purchase had to be made. The decisive result is the comparison between a free-delivery offer at harvest (+14 to +18 points, rising to +22 with advance timing choice) and a 50% price subsidy offered months later (+13 points). Halving the price is an expensive intervention; free delivery at the right moment is a cheap one. That the cheap, well-timed intervention performed at least as well is a direct measurement of how much of the problem is timing rather than price. The finding that the same free-delivery offer produced no significant effect when moved later in the season closes off the alternative reading that the delivery subsidy was simply a disguised price cut.
The soil evidence then explains why even the price channel underdelivers where it does operate. If fertilizer is unprofitable on about a third of plots because of degraded organic matter, and if acid soils covering roughly 13% of Kenya's land area suppress nutrient uptake until limed, then subsidizing nitrogen uniformly pushes input onto land that cannot convert it. The distributional sting is that the plots least able to convert fertilizer belong disproportionately to the poorest farmers, so the programme's real benefit is more concentrated among better-off farmers than its nominal universality suggests.
The inframarginality evidence completes the picture on cost. With roughly 70% pre-subsidy adoption, 85% of earlier-programme recipients already consistent buyers, and 0.2 kg of commercial displacement per subsidized kilogram, the share of each shilling that produces additional fertilizer on a plot that can use it is considerably smaller than the headline distribution volumes suggest.
Finally, the delivery design converts a fiscal inefficiency into an operational hazard. The 2024 episode is the predictable tail risk of concentrating national input distribution in a single channel.
This places the paper squarely in the series' central theme. In the M-Pesa paper, financial access expanded without touching the regulatory cost that governs formalization. In the youth entrepreneurship paper, capital was abundant in aggregate but structurally mismatched to the median firm. Here, fertilizer is subsidized heavily while the constraints that actually bind, the inability to carry cash from harvest to planting, the agronomic ceiling imposed by depleted and acidic soils, and the thinness of the margin once risk is priced, go largely unaddressed. In each case the policy relieves a constraint that is visible and measurable rather than the one that binds.
5. A Testable Alternative: Harvest-Time Commitment Vouchers
The most useful contribution this series can make is not to criticize an existing programme but to specify an alternative precisely enough that it can be tested. The evidence above points to one with unusual clarity.
The Duflo-Kremer-Robinson result identifies the binding constraint as the farmer's inability to carry harvest cash to planting, and shows that a commitment device deployed at harvest resolves a large part of it. Separately, this series' paper on chamas found that Kenyan savings groups persist despite near-universal mobile money precisely because they supply a commitment technology, public, socially enforced, and therefore effective against exactly the present-bias problem that defeats individual saving. Kenya thus already possesses both the behavioural diagnosis and a culturally embedded institutional form that addresses it.
The proposal is a harvest-time input commitment product: at harvest, a farmer commits a chosen sum into a locked mobile-money sub-account, redeemable only as an input voucher at a registered agrodealer during a defined planting window, with the subsidy delivered as a top-up on the committed amount rather than as a discount at the point of sale. Four features follow directly from the evidence. The commitment occurs at harvest, when both intention and cash are present. The lock substitutes for the social enforcement a chama provides. Redemption at competing private dealers preserves the agrodealer network and avoids the single-channel quality risk that materialized in 2024. And because the benefit accrues only to farmers who commit, it is self-targeting toward the marginal group: farmers already purchasing every season have less need to commit in advance, which mechanically reduces the inframarginal transfer that consumes much of the current programme's budget.
This design is directly testable against the status quo, and the comparison that matters is cost per additional kilogram of fertilizer applied on responsive soil, not cost per bag distributed.
6. Implications
- For programme design: The case for shifting from a universal planting-time price discount toward a harvest-time commitment mechanism rests on experimental evidence, not on theory alone. At minimum, the two should be trialled against each other before the NFSP's structure is further entrenched.
- For targeting: The combination of ~70% pre-existing adoption and 85% inframarginality among earlier recipients implies that self-targeting mechanisms are likely to outperform administrative targeting, which has historically failed to exclude existing buyers.
- For agronomic complements: Soil testing and lime deserve policy attention proportionate to their measured effect. On acid soils, liming moved trial yields by margins comparable to fertilizer application. A fertilizer subsidy without a soil-amendment programme subsidizes an input onto land that cannot use it.
- For market structure: Voucher-based redemption through competing agrodealers preserves the private distribution network that delivered 70% adoption before subsidies existed, and distributes quality risk rather than concentrating it.
- For measurement: Programme evaluation should report cost per additional tonne of maize, not farmers reached or bags distributed. The former is the policy-relevant quantity; the latter two are insensitive to the inframarginality and soil-heterogeneity problems documented here.
7. Original Research Proposals
Three projects follow directly from the gaps identified above, in ascending order of difficulty.
(a) Measuring the inframarginal share under the NFSP. A short structured survey asking NFSP recipients whether, and in what quantity, they purchased fertilizer in seasons before the programme, and whether they would have purchased this season absent the subsidy. This directly estimates the single number the current literature leaves least precisely pinned for the current programme. Sample requirements are modest and the instrument is short.
(b) Joint mapping of subsidy uptake and soil response. Pairing programme uptake records with soil test results (organic matter and pH) at plot level in one or two counties would test whether subsidized fertilizer is disproportionately reaching plots in the unresponsive tail that Marenya and Barrett identify. If it is, the policy implication, pair the subsidy with soil amendment, or condition it on soil testing, is immediate and actionable.
(c) A small harvest-commitment pilot. The design in Section 5 could be piloted at small scale against a conventional subsidy arm, measuring take-up, redemption, applied quantity and yield. This replicates the Duflo-Kremer-Robinson logic in the contemporary mobile-money environment, which did not exist when the original experiments ran. Given that the original study used physical delivery vouchers in the early 2000s, testing whether the same mechanism works through M-Pesa rails is a genuine and publishable contribution rather than a replication.
8. Limitations
The Duflo-Kremer-Robinson results, while experimentally identified, come from Western Province between 2000 and 2004 and from a sample of 1,230 farmers; Kenya's input markets, mobile money infrastructure and subsidy architecture have all changed substantially since. Their external validity to the 2026 NFSP context is an assumption of this paper, not a demonstrated fact, and it is the single assumption on which the central argument most depends.
Several supporting figures are drawn from a secondary research summary (UC Davis BASIS) reporting results from studies this paper did not read in the original, including Mason et al. (2017), Sheahan et al. (2014), Makau et al. (2016), Ariga et al. (2009) and Jayne et al. (2018). These should be verified against the primary sources before external citation.
The Kakamega study's authors and publication year were not visible on the repository page consulted. It is a cross-sectional comparison of 300 farmers in two sub-counties and is vulnerable to selection on unobservables; its 32.3% figure is an association, not a causal estimate.
The lime trial figures come from a presentation whose own table headings the source flags as unclear, and the benefit-cost ratios it reports (95.02 and 24.51) are implausibly large on their face and almost certainly reflect a units or labelling issue. They are cited here only for the direction of the yield effect, not the magnitude of the returns.
The juxtaposition in Section 3.5, a large proportional yield gain in a low-profitability region implying small net income gains, combines results from separate studies with different samples and methods. It is our inference and has not been tested directly.
Most of the quantitative evidence concerns NAAIAP and the NCPB scheme rather than the 2022 NFSP, whose effects were not reported in the sources reviewed. The 2024 KEL Chemicals episode is documented through press reporting of KEBS findings; the number of farmers who applied affected product is not established.
Finally, this paper does not produce a cost-per-additional-tonne estimate for any Kenyan programme, which is the figure that would most cleanly adjudicate between the instruments it compares.
References (selected)
- Duflo, E., Kremer, M. & Robinson, J. Nudging Farmers to Use Fertilizer: Theory and Experimental Evidence from Kenya. NBER Working Paper 15131; published in American Economic Review. Study summary via J-PAL and Innovations for Poverty Action.
- Marenya, P. P. & Barrett, C. B. (2009). Soil quality and fertilizer use rates among smallholder farmers in western Kenya. American Journal of Agricultural Economics, 91(4), 991-1006.
- Esilaba, A. et al. Soil Acidity and Liming in Kenya. KALRO / IFDC / Ministry of Agriculture presentation (2018).
- Kakamega County study of fertilizer subsidy and maize productivity, Malava and Mumias East sub-counties (Maseno University repository; authors and year not shown on the page consulted).
- UC Davis BASIS. Research summary on Kenya's input subsidy programmes, citing Mason et al. (2017), Sheahan et al. (2014), Makau et al. (2016), Ariga et al. (2009), Jayne et al. (2018), Tegemeo Institute (2015) and World Bank (2013).
- Tiruneh, H. (2024, March 26). How Kenya can boost agricultural productivity with fertiliser subsidy. IFPRI / CGIAR, originally published in Business Daily Africa.
- Kenya Bureau of Standards findings on substandard fertilizer, March 2024, as reported in Nation and Capital FM.
- Michigan State University Food Security Policy. The Effects of Kenya's 'Smarter' Input Subsidy Program on Smallholder Behavior and Incomes (identified but not accessible for review).
Prepared as part of the East Africa Microeconomics Research Series. This is a student working paper that synthesises existing research and published data rather than presenting original fieldwork; feedback and data corrections are welcome via the contact page.