Food Policy 67 (2017) 93–105 Contents lists available at ScienceDirect Food Policy journal homepage: www.elsevier.com/locate/foodpol Agricultural input credit in Sub-Saharan Africa: Telling myth from facts Serge G. Adjognon a, Lenis Saweda O. Liverpool-Tasie b,⇑, Thomas A. Reardon b a World Bank Group (WBG), Development Impact Evaluation Unit (DIME), Development Research Group (DECRG), USA b Department of Agricultural, Food, and Resource Economics, Michigan State University, East Lansing, MI, USA a r t i c l e i n f o a b s t r a c t Article history: Recent evidence shows that many Sub-Saharan African farmers use modern inputs, but there is limited Available online 24 October 2016 information on how these inputs are financed. We use recent nationally representative data from four countries to explore input financing and the role of credit therein. A number of our results contradict Keywords: ‘‘conventional wisdom” found in the literature. Our results consistently show that traditional credit Africa use, formal or informal, is extremely low (across credit type, country, crop and farm size categories). Farm inputs Instead, farmers primarily finance modern input purchases with cash from nonfarm activities and crop Credit sales. Tied output-labor arrangements (which have received little empirical treatment in the literature) Rural nonfarm employment appear to be the only form of credit relatively widely used for farming. Ó 2016 The World Bank. Published by Elsevier Ltd. This is an open access article under the CC BY IGO license (http://creativecommons.org/licenses/by/3.0/igo/). 1. Introduction were generally dismantled in the 1990s and 2000s during Struc- tural Adjustment. We hypothesize that few farmers use govern- It is generally accepted that Sub-Saharan Africa (SSA) farmers ment credit now. often have low yields which could be increased, all else equal if they Second, government subsidies to farmers to buy fertilizer were bought more ‘‘external inputs” (chemical fertilizer, pesticides, and common before Structural Adjustment. The subsidy was adminis- seeds). Moreover, it is often asserted after liberalization and priva- tered as a reduction of fertilizer price, or as a coupon to farmers tization dismantled many government farm credit programs in the (as a direct transfer). Many input subsidy programs were elimi- 1990s (Kherallah et al., 2002), that small farmers face severe credit nated by Structural Adjustment. However, in several SSA countries constraints and that this is a cause of low use of external inputs they were partially revived in the mid-2000s on the heels of con- (Kelly et al., 2003; Morris et al., 2007; Poulton et al., 1998, 2006). cerns that fertilizer use had dropped since Structural Adjustment. Yet Sheahan and Barrett (2014) find that SSA farmers now pur- Malawi and Tanzania governments provide many farmers a cou- chase more external inputs than in the 1990s, and much more than pon for fertilizer sufficient for an acre. The Nigerian government is generally asserted in the debate. Farmers are thus financing had a subsidy scheme during our study period (2010À2012) but inputs somehow. Is it by credit? If so what kind? Is it by own cash our analysis showed only 5% of the farmers bought fertilizer from sources from crop sales and labor sales? These issues lead us to the government sources that disbursed the subsidy. three research questions we address here: (1) how do farmers Third, private-sector banks tend, according to much of the liter- finance input purchases? (2) Is there a correlation between finance ature, to lend little to farmers (Poulton et al., 1998, 2010). The rea- source and farm size and thus ‘‘inclusiveness” of the financial sons given are that banks face high transaction costs in rural areas, arrangement used? (3) Is there a relation with crop type and thus farmers tend to lack collateral, and lending is risky because recov- relation to cash crop versus food crops? ery rates are low (Dorward et al., 2009). We hypothesize that few To derive hypotheses for these questions, we briefly review the farmers obtain bank credit, but those that do are larger farmers literature concerning the potential finance sources for inputs. (based on work by Zeller and Sharma (1998) in Cameroon, Ghana, First, government credit was common before the 1990s for both Madagascar, and Malawi). farmers producing cereals as well as export cash crops. The Fourth, informal credit from friends and family and local schemes generated fiscal deficits and suffered frequent non- moneylenders is often presented as a significant source of funds recovery, considered ‘‘strategic default” used by farmers as de facto for farmers to buy inputs and consumption items (Poulton et al., insurance after bad harvests (Poulton et al., 1998). These schemes 2006; Zeller and Sharma, 1998). Our hypothesis is thus that infor- mal credit is important to all strata of farmers. ⇑ Corresponding author. Fifth, finance from ‘‘tied output-credit” or ‘‘interlinked credit” E-mail address: lliverp@anr.msu.edu (L.S.O. Liverpool-Tasie). arrangements (Bardhan, 1980; Poulton et al., 1998) involve an out- http://dx.doi.org/10.1016/j.foodpol.2016.09.014 0306-9192/Ó 2016 The World Bank. Published by Elsevier Ltd. This is an open access article under the CC BY IGO license (http://creativecommons.org/licenses/by/3.0/igo/). 94 S.G. Adjognon et al. / Food Policy 67 (2017) 93–105 put buyer or input seller advancing the farmer cash for inputs or The paper proceeds as follows. Section 2 discusses data and inputs in kind at the start of the season, and being reimbursed from sampling. Section 3 descriptively examines the purchase of ‘‘exter- the farmer’s harvest. The literature presents this in two categories. nal inputs” and use of credit sources for those purchases, as well as The first category is tied output credit from processing or export cash income sources. The analysis stratifies by country, farm size, companies for traditional export cash crops as well as for non- and crop type (using the triad of crop categories in the SSA litera- traditional crops like horticulture. The literature is ambiguous as ture: traditional export crops, non-traditional commercial crops to the occurrence of this. On the one hand, a number of studies such as horticulture, and staple food grains). Section 3 focuses on especially of particular schemes document this arrangement. On Nigeria to econometrically test for the effects of different cash the other hand, some studies note that processing and export com- sources on fertilizer demand. The analysis uses panel data estima- panies may not use this arrangement frequently or apply it to all tion techniques to more consistently identify the effect of RNFI on farmers because they fear farmers will ‘‘side sell” (to other buyers) fertilizer demand by accounting for unobserved time invariant or because there is a dearth of effective farmer cooperatives to household characteristics likely to affect participation in non- enforce repayment among their members (Shepherd and Farolfi, farm activities and fertilizer demand. As far as we are aware, there 1999; Poulton et al., 1998, 2010; Chao-Béroff, 2014). are no other studies that have used nationally representative panel The second category is interlinked credit from grain wholesalers data to explore the effect of non-farm activities on input demand. and input dealers. This is commonly posited to be important in Most of the older literature (cited above) focused on qualitative Asia (Bardhan, 1980; Conning and Udry, 2007) and in some reports analysis, comparison of means and ordinary least square (OLS) hypothesized to be common in SSA (Pearce, 2003; Zeller and estimations that are potentially biased (e.g., Ellis and Freeman, Sharma, 1998). 2004). More recent empirical work such as Oseni and Winters In both cases farmers enter these ‘‘tied” arrangements princi- (2009) use cross sectional data while Smale et al. (2016) use panel pally because formal credit markets idiosyncratically fail for them, data but do not use a nationally representative sample (they focus and thus these are ‘‘second best” arrangements (Binswanger and on one maize producing region of Kenya). Rosenzweig, 1986). We hypothesize that empirical analysis will show that such arrangements are common in SSA, perhaps with 2. Data a bias toward traditional cash crops. A variant on the above is a tied output-labor market arrange- We use data from the Living Standard Measurement Study ment where farm workers advance labor in exchange for payment (LSMS) household panel surveys in four countries. The most recent (typically in kind but can be in cash) at harvest (Bardhan, 1984). years of the panels are used for the descriptive analysis in all the While discussion of this was common in the South Asia literature countries, and the most recent two years for the econometrics in the 1970s/1980s, to our knowledge it has not been examined analysis in Nigeria. The sets are as follows: (a) the Malawi Inte- empirically in SSA. We hypothesize that it exists in SSA. One justi- grated Household Panel Survey (IHPS) of 2012/2013, with 3219 fication for this expectation is that labor by one household provided farm households; (b) the second wave of the Nigeria Living Stan- to another is monitored and upheld by local norms/customs and dard Measurement Study – Integrated Survey on Agriculture social pressure. (LSMS-ISA) Panel for two years, 2010/2011 and 2012/2013, cover- Sixth, household retained earnings such as from rural nonfarm ing 3000 farm households; (c) the Tanzania National Panel Survey employment and crop sales are in principle candidates for poten- 2012/2013, covering 3047 farm households; and (d) the Uganda tial liquidity sources for farmers to buy inputs. Indeed, National Panel Survey 2010/2011 covering 2109 farm households. Haggblade et al. (2010) note that rural nonfarm income (RNFI) is The surveys differ somewhat in the specific questions they use to a main cash source of rural households in SSA, and Reardon et al. elicit information on the variables of interest. We treat the survey (1994) and Davis et al. (2009) hypothesize that RNFI is a key cash datasets as uniformly as possible to ensure that the information is source and determinant for input purchases, especially in the face comparable. Where one set or the other lacks some information we of idiosyncratic failure of credit markets. Yet the empirical litera- note that in the table notes. ture rarely compares household own-cash sources with credit as In general, the surveys used a two-stage sample design. In the potential liquidity sources for farmers to buy inputs. Zeller and first stage, enumeration areas were selected in each district of Sharma (1998) note that the literature on farm credit is largely the country. Within each enumeration area a listing of households independent of the literature on farm household income sources. was done for the sample frame. A random sample of households However, several studies in SSA provide evidence of the role of was drawn from that frame. We selected only households doing RNFI as a finance source for investments of rural households. any farming. In the analyses, we use sampling weights from the Aryeetey (1997) provides evidence of the latter for Ghana for rural datasets to account for the survey design and construct nationally microenterprises but not for agriculture. Some work has shown the representative statistics. The weight for each household is the impact of RNFI on external input use by African farmers (e.g., inverse of the probability of being selected based on the sample Savadogo et al. (1994) for animal traction in Burkina Faso; Clay frame structure. et al. (1998) and Oseni and Winters (2009) for fertilizer in Nigeria The data used are on farm households’ use of inputs and cash and Rwanda), and for Asia (e.g., Stampini and Davis (2009) for and in-kind arrangements to pay for them. The analysis is done purchased seeds in Vietnam); some work has shown the effects by crop, household, and plot. The data also have characteristics of off-farm income on farm productivity (such as Rozelle et al., of the farm households such as nonfarm income, crops sales, loans 1999 for China). We thus hypothesize that own cash sources are received, and farm size. a significant determinant of input purchases. The aim of this paper is to examine the above hypotheses and thereby ‘‘update the landscape” of knowledge of SSA farm house- 3. Descriptive analysis of cropping and input purchases holds’ sources of finance for external inputs. To our knowledge, there has been no such survey-based analysis especially over coun- 3.1. Patterns in cropping tries using recent and nationally representative data. We analyze recent (2010À2012) LSMS data sets comprising 11,000 farm Table 1 shows crop composition by country and farm size households in Malawi, Nigeria, Tanzania, and Uganda. strata. Crops are classified into sets: crops traditionally called ‘‘food S.G. Adjognon et al. / Food Policy 67 (2017) 93–105 95 Table 1 Table 2 Share of households producing key cash and food crops across farm size strata. Source: Share of farm households who purchase external inputs. Source: Authors from LSMS Authors from LSMS data. data Crop types Farm size strata (ha) Share of farms with crop (%) Countries Farm Households buying Farm Households (%) by type of external inputs (%) external inputs purchased Malawi Nigeria Tanzania Uganda Fertilizers Pesticides Seeds Cash crops 0–0.49 4 10 4 26 Malawi 70 49 4 51 0.5–0.99 18 10 8 34 Nigeria 71 42 38 29 1–1.99 39 11 11 39 Tanzania 18 8 13 NA 2–4.99 49 20 14 46 Uganda 16 5 14 NA 5+ 28 14 18 54 All 17 11 11 37 Note: NA means information is unavailable in the dataset. Food crops Grains 3.2. Patterns in input purchases 0–0.49 98 69 61 70 0.5–0.99 99 87 74 83 1–1.99 99 86 79 86 Table 2 shows farmers’ purchases of ‘‘external inputs” – variable 2–4.99 99 84 83 81 inputs apart from labor, including inorganic fertilizer, seeds, and 5+ 100 88 85 81 pesticides. All 99 77 76 80 First, there is a marked contrast between Nigeria and Malawi, Horticulture with a high share of farmers buying external inputs (70%), com- 0–0.49 29 33 22 55 pared to Uganda and Tanzania (16% and 18% respectively). The 0.5–0.99 31 21 13 50 1–1.99 37 22 12 48 Malawi-Nigeria results are at odds with the traditional notion that 2–4.99 32 23 9 46 few farmers in SSA use external inputs but consistent with the 5+ 43 17 7 63 findings of Sheahan and Barrett (2014).1 All 31 28 13 51 One might say that the Nigeria and Malawi results are driven by Legumes the fertilizer subsidy program. While that might be true in Malawi 0–0.49 62 29 12 76 where about 60% of households receive subsidized fertilizer 0.5–0.99 76 56 10 75 1–1.99 79 60 12 77 (Chirwa and Dorward, 2013), this is unlikely in the case of Nigeria. 2–4.99 77 53 16 82 While the Nigeria data show persistently high fertilizer use rates 5+ 93 54 16 82 across both survey years rounds, in 2010, when subsidized fertil- All 71 42 13 78 izer was only channeled through the government, fewer than 5% Tubers of the households who purchased fertilizer bought it from govern- 0–0.49 8 61 16 74 ment sources (the channel by which the subsidy was delivered).2 0.5–0.99 9 30 19 79 Second, among farm households buying external inputs, fertil- 1–1.99 14 34 19 74 2–4.99 16 39 18 76 izer and seeds are common purchases. The results are mixed for 5+ 0 49 20 71 pesticides. Many farmers buy pesticides in Nigeria, but not in All 10 48 18 75 Malawi. Only about a half and a third of the farmers who buy All food crops external inputs in Tanzania and Uganda buy fertilizer, yet a larger 0–0.49 100 98 95 100 share buy pesticides; this appears surprising, but is consistent with 0.5–0.99 100 98 97 99 Sheahan and Barrett (2014) for Uganda. 1–1.99 100 99 96 100 2–4.99 100 98 95 100 Table 3 disaggregates input purchases over five strata, very 5+ 100 99 97 99 small farmers (with less than 0.5 ha) to larger farmers with more All 100 98 96 100 than 5 ha. Several points are salient. First, across the countries and contrary to conventional percep- tions, farmland is concentrated. We find 65–75% of the land but only 20–25% of the farms in the medium and large farm strata crops” (although they are often also sold for cash), including grains, (above 2 ha). Small farmers of less than 2 ha have only 25–35% of horticulture products, legumes, and tubers (grown as a staple), and the land but 75–80% of the farms in Nigeria, Tanzania, and Uganda; crops traditionally called ‘‘cash crops”, including tobacco, cotton, in Malawi the farms above two hectares are only 4% of the farms tea/coffee, and edible oil crops. but nearly 40% of the land. Several points stand out. First, as expected, grain farming dom- Second, surprisingly, the shares of farmers buying external inates, but is not ubiquitous, as it is practiced by only about three- inputs do not differ much over small (up to 2 ha) versus med- quarters of the farms in Nigeria, Tanzania, and Uganda, being near ium/large (above 2 ha): in Malawi, 71% versus 88%, Nigeria, 78 ver- 100% only in Malawi. There is little farm size bias in participation sus 83%, Tanzania, 15% versus 23%, and for Uganda, 14% versus 24%. in food cropping. Over the countries on average nearly a third of But this masks differences in rates, or level of external input use the farms grow horticultural crops, half grow beans/pulses, and a third grow tubers. Food cropping is thus fairly diversified on 1 average. They covered Ethiopia, Malawi, Niger, Nigeria, Tanzania, and Uganda. 2 There is no explicit question in the Nigeria LSMS for whether a household got a Second, by contrast, cropping of traditional cash crops is more fertilizer subsidy. However, until recently, only the government sold subsidized concentrated in every country. On average, only a fifth of farmers fertilizer; thus we assume that farmers buying from government sources are the only grow traditional cash crops, and that is but a tenth if one excludes ones getting a subsidy (based on Takeshima and Nkonya, 2014). While this might be Uganda. There is a marked correlation of the share of farms pro- an underestimate in 2012 (since it was possible starting in 2012 for farmers to ducing any cash crop and farm size. The crop focus differs over purchase subsidized fertilizer from dealers in the market with a coupon) this is unlikely since the new program was still very new (launched in 2012). We find the countries, with tea/coffee and oil crops standing out in Uganda, very low numbers (and a tiny share) of farmers purchasing fertilizer from government cotton and oil crops in Tanzania, oil crops in Nigeria, and tobacco sources in 2012 to be similar to those in 2010 (when the government was the sole and cotton in Malawi. distributor of fertilizer). 96 S.G. Adjognon et al. / Food Policy 67 (2017) 93–105 Table 3 Purchase of external inputs by farm size strata. Source: LSMS data. Farm Farms in Farmland in Farms buying Fertilizer bought by Pesticides bought by Seed bought by Total inputs bought by strata (ha) stratum (%) stratum (%) external inputs (%) stratum (%) stratum (%) stratum (%) stratum (%) Malawi 0–0.49 45 13 65 30 12 28 30 0.5–0.99 33 24 69 21 11 34 22 1–1.99 18 24 79 29 40 23 29 2–4.99 4 11 91 19 30 13 19 5+ 0 27 84 1 7 2 1 Overall 100 100 100 100 100 100 Nigeria 0–0.49 53 8 62 30 19 55 30 0.5–0.99 20 12 78 25 20 17 23 1–1.99 15 16 83 23 24 13 22 2–4.99 9 22 82 16 21 8 16 5+ 3 43 85 5 16 7 8 Overall 100 100 100 100 100 100 Tanzania 0–0.49 20 2 13 5 5 NA 5 0.5–0.99 19 5 14 9 7 NA 9 1–1.99 24 14 17 20 13 NA 19 2–4.99 26 32 22 41 46 NA 42 5+ 11 47 24 25 29 NA 26 Overall 100 100 100 100 NA 100 Uganda 0–0.49 26 4 6 6 5 NA 5 0.5–0.99 24 10 16 9 10 NA 10 1–1.99 26 20 20 35 48 NA 44 2–4.99 19 30 20 34 25 NA 28 5+ 6 37 28 16 12 NA 14 Note: NA, information unavailable in the dataset. External inputs include fertilizer, seeds and pesticides. per hectare. Binswanger and Ruttan (1978) note that one should near absence of the use of any credit, formal or informal, tied with expect smaller farms to use more external inputs as substitute input or output traders, in kind or in cash. The converse is that 94% for land. Our data indeed show smaller farmers using more exter- use only their own cash to buy external inputs. This can be from nal inputs per hectare than do the medium/large farms: while sales of crops and employment earnings (farm wage labor, migra- medium/large farmers crop 70% of total farmland, they constitute tion, and RNFI), as discussed in more detail below. only 35% of the external input purchase ‘‘pie”. This finding varies Moreover, among the tiny share of farmers buying external little over input types. It also holds true across Malawi, Nigeria, inputs on credit, there is sharp variation over input types. There and Uganda. The outlier is Tanzania, where medium/large farms tends to be 2–3 times more households getting some kind of credit use external inputs almost as intensively as small farms. for fertilizer compared to seeds or pesticides. Table 5a shows the shares of the farm size strata in all credit- 3.3. Farm input finance by farm size strata and crop categories based input expenditures. In Malawi, Tanzania, and Uganda, input credit is roughly correlated with farm size – most of the credit- Table 4 shows consistent evidence across countries of very low based external input expenditures are concentrated outside the use of any form of credit to buy external inputs. On average, among below-one-hectare group. These results do not differ much over farm households who bought external inputs, only about 6% used input types. Nigeria has the lowest share of farmers purchasing any form of credit. As noted in the introduction, there has been a external inputs on credit (3%); it differs somewhat from the other presumption in the literature that to the extent farmers buy exter- countries in that the great majority of the input credit is taken by nal inputs, they do it at least with informal credit or trader credit. the ‘‘under 1 ha” group; however, this is still taken by merely a tiny But the analysis here shows that conventional wisdom is not sup- share of the smallest farmers. ported empirically, and it is not just a lack of formal credit, but a Table 5b shows the share of each external input’s expenditure that a given stratum buys with credit. Input credit tends to be much more important for the middle to higher farm size strata, Table 4 Share of households purchasing external inputs that finance the purchase on credit. and extremely little for the smaller strata. It is also mainly in fertil- Source: Authors from LSMS data. izer and very little in pesticides and seeds. In Malawi, Tanzania, and Uganda, input credit is relatively substantial only for fertilizer. Of those who bought Of those who bought the noted external inputs, share input, share who bought on credit It averages 9% of fertilizer input outlay in Malawi but is concen- buying on credit (%) by input type trated in the upper-small and medium farmers (1–5 ha) where it Fertilizers Pesticides Seeds averages a fifth of external input expenditure. In Tanzania, the share of input expenditure done on credit is correlated with land Malawi 5 5 7 3 Nigeria 3 2 NA 3 size, with about 10% for smaller farmers and about a quarter and Tanzania 11 14 7 3 a half for medium and larger farmers. For Uganda, it is only rela- Uganda 6 14 4 NA tively important for the 1–5 ha group, where it reaches 40–50% Note: NA implies information unavailable in the dataset. of fertilizer expenditure. In Nigeria, the share is low for all types Column 2 is the share among households who purchased at least one external of external inputs, with about 3% on average, differing little over input. strata. S.G. Adjognon et al. / Food Policy 67 (2017) 93–105 97 Table 5a Credit-based expenditure on external inputs, by shares of strata. Source: from LSMS data. Countries Farm size strata Buying on In all credit-based In all credit-based In all credit-based In all credit-based credit (%) fertilizer outlay pesticide outlay seed outlay input outlay Malawi 0–0.49 3 4 11 13 4 0.5–0.99 3 4 15 16 4 1–1.99 10 61 38 44 60 2–4.99 10 32 36 27 32 5+ 14 0 0 0 0 Overall 100 100 100 100 Nigeria 0–0.49 3 49 NA 13 45 0.5–0.99 5 22 NA 22 22 1–1.99 4 11 NA 62 16 2–4.99 1 2 NA 0 2 5+ 6 16 NA 3 14 Overall 100 NA 100 100 Tanzania 0–0.49 2 0 0 NA 0 0.5–0.99 6 4 3 NA 4 1–1.99 8 10 15 NA 10 2–4.99 20 36 69 NA 38 5+ 24 50 12 NA 48 Overall 100 100 NA 100 Uganda 0–0.49 0 0 0 NA 0 0.5–0.99 2 3 17 NA 5 1–1.99 11 57 54 NA 56 2–4.99 11 40 28 NA 39 5+ 0 0 0 NA 0 Overall 100 100 NA 100 Note: NA implies information unavailable from dataset. Column 3 pertains to farm households buying at least one external input. Table 5b Share of credit-based outlay in overall outlay per external input. Source: from LSMS data. Countries Farm size strata Credit-based outlay in Credit-based outlay in Credit-based outlay Credit-based outlay in total fertilizer outlay (%) total pesticide outlay (%) in total seed outlay (%) total ext. input outlay (%) Malawi 0–0.49 1 3 2 1 0.5–0.99 2 5 2 2 1–1.99 22 4 8 21 2–4.99 18 4 8 17 5+ 0 0 0 0 Nigeria 0–0.49 6 NA 1 4 0.5–0.99 3 NA 3 3 1–1.99 2 NA 12 2 2–4.99 1 NA 0 0 5+ 11 NA 1 5 Tanzania 0–0.49 2 0 NA 2 0.5–0.99 12 4 NA 11 1–1.99 15 10 NA 14 2–4.99 26 12 NA 23 5+ 58 3 NA 48 Uganda 0–0.49 0 0 NA 0 0.5–0.99 12 3 NA 6 1–1.99 53 2 NA 17 2–4.99 40 2 NA 19 5+ 0 0 NA 0 Note: NA implies information unavailable in the dataset. Column 3 is among households who purchased at least one external input. 3.4. Finance by crop type with an added focus on interlinked credit ular from processors in interlinked credit arrangements; food crop producers also may access such interlinked credit from traders. To Conventional perceptions from the literature, as discussed in test this, we explore the shares (by crop type) of farm plots on the introduction, suggest that farmers growing traditional cash which inputs purchased on credit in interlinked relations are crops would commonly access external inputs on credit, in partic- shown (Table 6). The findings are surprising. 98 S.G. Adjognon et al. / Food Policy 67 (2017) 93–105 Table 6 Table 7 Share of plots on which external inputs purchased with interlinked credit, by crop Shares of farmers using harvest to reimburse input credit. Source: Generated by type. Source: from LSMS data. authors using LSMS data. Malawi Nigeria Tanzania Uganda Countries Farm size strata Share of farmers Share of farmers using using their harvest their harvest to repay Cash crops to repay labor external inputs Tobacco 16 NA 87 81 received on credit received on credit (%) Cotton 11 8 11 0 (%) Tea/coffee NA NA 22 1 Oil crops 6 3 4 11 Malawi All cash crops 14 4 26 8 0–0.49 37 1 0.5–0.99 45 3 Food crops 1–1.99 50 2 Grains 5 3 11 7 2–4.99 47 1 Horticulture 4 3 0 4 5+ 24 0 Legumes 5 2 11 6 All 42 1.8 Tubers 7 3 4 5 All food crops 5 3 10 6 Nigeria 0–0.49 26 1 NA implies information unavailable in dataset. 0.5–0.99 29 1 1–1.99 26 3 2–4.99 21 2 First, while there is a lot of variation over countries, the average 5+ 22 3 over all traditional cash crops is only 13%, lower than what would All 26 1.4 be expected from the literature on cash crops that suggests a wide Tanzania distribution of interlinked credit for cash croppers. This average 0–0.49 NA 0 masks variation over countries. Malawi and Tanzania average 0.5–0.99 NA 1 20% (the share, among all plots for this crop category, that receiv- 1–1.99 NA 1 2–4.99 NA 4 ing inputs purchased in interlinked credit arrangements). Nigeria 5+ NA 5 and Uganda average only 6%. All NA 1.9 The difference between the pairs of countries is mainly driven Uganda by tobacco in Tanzania and Uganda, where four-fifths of the plots 0–0.49 54 NA are grown with inputs bought on credit from the processors. Also, 0.5–0.99 63 NA that outlier is composed of a tiny group of tobacco farmers in the 1–1.99 74 NA sample for each country, about 1% of the total sample. The limited 2–4.99 78 NA 5+ 81 NA and ‘‘enclave” nature of tobacco farming and its correlation with All 68 NA farm size in those countries could explain why these are the main cases where the conventional image of contract-farming related Notes: NA implies that information is unavailable in the dataset used. credit is manifest. Removing the tobacco outlier (for just Tanzania and Uganda) puts the overall credit share for cash crops about 6% – Malawi it is in an inverted-U shape relation with farm size. Thus very close to that for food crops as noted below. one cannot say that this traditional-tying of labor and harvest is Second, only 6% of all plots of food crops receive inputs pur- more a phenomenon of the smallest farmers holding on to an old chased in interlinked credit arrangements. This is the first time practice, as one might expect, given our hypothesis that larger farms in the literature this has been tested and demonstrated, and we are more apt to use monetized labor relations only. Moreover, Table 8 consider this a key contribution of this paper. shows that use of harvest repayment for labor is very minor for cash To triangulate the above results on output/input credit arrange- crops (except for oil crops in Uganda where it is a quarter of farmers ments, we examined the data in another section of the LSMS sur- using it), but is significant in food crops across the countries, such as vey questionnaire, the management of crop harvests. We used about a third in horticulture and a quarter in grains. There is only a farmers’ responses concerning use of part of their harvests to repay single situation (crop plus country) where this arrangement (using advances for inputs from input or output traders and processors harvest to pay for inputs) is important for external inputs, and that (especially for cash crops) for external inputs, and for labor. is for tobacco in Tanzania. This corroborates the results from above. Table 7 shows the share of farmers using part of their harvests We conjecture that this high prevalence of the use of harvest to for these ends. The main finding is that such ‘‘tied credit” is very reimburse for external inputs received on credit to produce rare for external inputs (fewer than 2% of the farmers) across all tobacco in Tanzania is related to a widespread use of contract study countries. This corroborates the results from above. For har- farming arrangement over tobacco production in Tanzania. If our vest payment for external inputs, the shares are so small that there conjecture is true, we should expect to see more contract farming are no interesting inter-strata differences. When we consider the (outgrower) arrangements over tobacco compared to cotton, tea/- ‘‘reimbursement of credit with the harvest” by type of crop, it is coffee, and oil crops. But in the Tanzania data set,4 we found that very minor or zero for the other cash crops (except tobacco in Tan- only 1.8% of farmers are involved in outgrower schemes. In this tiny zania, discussed above), and all of the food crops (Table 8). set, tobacco farmers dominate (as 78% of the plots in outgrower By contrast, and reported for the first time in the SSA literature schemes are under tobacco, followed by cotton with 19%. using cross-country surveys for comparison, we find that labor- Overall our results indicate that there is much less tied credit output tying is much more common, with as many as 42% of the arrangement to finance external input than expected from the con- farmers in Malawi, 26% of Nigerian, and 68% of Tanzanian farmers jectures in the literature. Even though those arrangements appear doing this practice. (The dataset for Uganda did not allow this cal- to be more formal (from contract farming arrangements) and more culation.)3 The patterns over strata differ by country. In Uganda, the likely for cash crops, we still see far less than expected (except for share rises with farm size, in Nigeria it slightly declines, and in tobacco). 3 Of interest (but not reported in Table 7) is that tying land access and output 4 markets was not found to be common. That is, the land tenure section of the surveys There is no information about contract farming in the datasets of the other showed that sharecropping was extremely limited. countries to allow us to compare this pattern across countries. S.G. Adjognon et al. / Food Policy 67 (2017) 93–105 99 Table 8 Financing inputs on credit with harvest across key cash and food crops. Source: Generated by authors using LSMS data. Crops types Share of plots where harvest is used to repay (advanced) labor (%) Share of plots where harvest is used to repay external inputs (%) Nigeria Malawi Uganda Tanzania Nigeria Malawi Uganda Tanzania Cash crops Tobacco 0 2 0 NA 0 2 NA 79 Cotton 10 0 0 NA 0 1 NA 6 Tea/coffee NA NA 1 NA NA NA NA 3 Oil crops 8 0 25 NA 0 0 NA 0 Food crops Grains 17 22 27 NA 1 1 NA 1 Horticulture 18 32 36 NA 1 0 NA 0 Legumes 9 21 25 NA 1 1 NA 0 Tubers 5 29 30 NA 1 1 NA 0 Notes: NA implies that information is unavailable in the dataset used. 3.5. Households’ use of loans not specifically linked to input Integrated Survey on Agriculture (LSMS-ISA); the panel version of transactions the nationally representative dataset used in previous sections. We use the term ‘‘loans” for credit unconnected directly and 4.1. Conceptual and empirical framework specifically to transactions of outputs or inputs. Loans can come from formal sources (banks), semi-formal sources (micro- The fertilizer purchase decision follows a standard input finance), and informal sources (friends, relatives, cooperatives, demand function derived from a constrained household utility etc.). The survey data show that households take loans, but rarely maximization problem (Sadoulet and de Janvry, 1995). Fertilizer for agriculture. In Nigeria 38% of the farmers took loans (but there demand can be expressed as a function of output and input prices, is no information in the survey on the purpose of the loan). In risk proxies, complementary and substitute farm capital, and rele- Malawi, 23% of the households took a loan, but only 5% of them vant shifter variables such as crop type. We consider the decision did so for farming. In Tanzania, only 11% took loans, of which 2% to purchase fertilizer and then the intensity of use. for farming. This is striking because one would expect credit- In each case constrained farmers to use these loans to finance farm input pur- chases. For Uganda the survey did not report loans. Y it ¼ f ðX it ; uit Þ Instead, the data show that the loans were taken mainly for nonfarm business startups and non-farm enterprise inputs (40% where Y it refers to the binary input use variable or the quantity of in Malawi, 24% in Tanzania) and for food consumption (31% in fertilizer purchased (in kg), while X it refers to a vector of controls Malawi and Tanzania).5 As our regressions show below, a key factor that explain fertilizer demand. uit ¼ eijit þ ci is a composite error that determines fertilizer purchase is engaging in nonfarm enter- term comprising time invariant unobservable heterogeneity (ci ) prises. Thus it appears that farmers prefer to use loans to finance and time varying unobserved characteristics eit of our input demand the set up/expansion of their nonfarm enterprises but use the gener- function. We model the farmer’s fertilizer purchase decisions using ated cash from these nonfarm enterprises to finance external input the standard unobserved effects binary dependent variable model purchases for their farms. (Green, 2000; Wooldridge, 2010). The intensity of fertilizer use is modeled using the unobserved effects Tobit model to account for the corner solution nature of the dependent variable (Wooldridge, 4. Determinants of fertilizer purchases in Nigeria 2010). In both models, ci represents the unobserved effect parame- ter, modeled using the Mundlak (1978) special case of the approach This section infers how farmers finance their input purchase by of Chamberlain (1982) called correlated random effects (CRE): estimating the determinants of fertilizer purchases by Nigerian farmers. Our analysis emphasizes the roles of the main cash ci ¼ w þ X i n þ ai ; sources of farm households, including RNFI (from both wage and self-employment), crop sales, and loans, in rapidly descending order of importance. We also control for agricultural productivity ai jX i $ Normalð0; r2 aÞ risks (captured by zone rainfall variability), as well as regional dif- where X i represents time averages of the explanatory variables. The ferences (north versus south) in decisions on fertilizer purchase. CRE model is preferred over alternative methods such as the fixed We focus on the Nigeria case to abstract from possible issues of effects (FE) and random effects (RE) models in the case of non- the fertilizer subsidy directly driving fertilizer purchase, which linear models (Wooldridge, 2010). However, for comparison, we could be an issue if we were to do the analysis on Malawi and Tan- estimate the linear model with household FE given its suggested zania as noted above.6 We use the two available waves (for 2010 conceptual robustness over nonlinear models such as the Probit and 2012) of The Nigeria Living Standard Measurement Study- and Tobit (Angrist and Pischke, 2008). 5 Consistent with the CRE model, the determinants of the fertil- Zeller and Sharma (1998) also found that 50–90% of formal and informal loans in izer purchase decision and the level of use are estimated using their African study countries went to consumption-related purchases. Poulton et al. (2010) also make this point in a general way. pooled Probit and pooled Tobit regressions, respectively. Each 6 Even in 2012, when it was possible that farmers purchased subsidized fertilizer regression equation includes a set of explanatory variables as well from agro-dealers in the private market less than 5% of farmers could have done so. as the time averages of the explanatory variables. A Wald test of According to the Federal Ministry of Agriculture and Rural Development about 1.6 joint significance of the time average variables is performed to test million farmers participated in the government subsidy program in 2012 (FMARD, 2015). According to the LSMS-ISA surveys, there were over 32.5 million households in whether a traditional random effects model would be appropriate. Nigeria in 2012. Even if we assume the program only allowed 1 participant per A dummy variable for the time period is included to account for household, this would amount to about 5% of farmers. time-specific factors that affect fertilizer demand. 100 S.G. Adjognon et al. / Food Policy 67 (2017) 93–105 Although the use of the FE and CRE models address potential to purchase fertilizer, participate in non-farm self-employment, biases due to time invariant unobserved heterogeneity, conditional participate in non-farm wage employment, take a loan, and sell strict exogeneity implies there is no endogeneity after controlling crops). for the time-invariant unobservables. If this assumption fails our As in the single equation estimations, we control for specific estimates might be biased. To minimize any remaining bias from time-invariant unobservable heterogeneity and include time- time-varying unobservables, we include various observable char- varying covariates. acteristics to proxy for a number of unobservables. Conditional The explanatory variables used in the models and their levels on the covariates used, the likely major source of endogeneity are reported in Table 9. The variable sets and key descriptive points should be time invariant and thus addressed by the CRE approach are as follows. (corroborated with FE results). However, since it is not possible to First, three (potential) sources of input finance are included in completely rule out endogeneity due to time-varying unobserv- the model: (1) a dummy variable for RNFI, including self- ables, these results are interpreted as correlates rather than causal employment and wage employment; (2) crop sales per hectare of effects. land; and (3) a dummy variable for any member of the household As an alternative specification, we also consider the likelihood having taken a loan the year before the survey period. Table 9 that the decision to work in RNFI, sell crops, or take a loan might shows that around 60% of households have at least one member be jointly made with the decision to use fertilizer. A farmer may in RNFI self-employment and around 20% with a member with decide to engage in non-farm activities (or take a loan or sell some wage employment. The RNFI patterns are similar in the North of his crops) to get cash to purchase inputs including fertilizer. Fur- and South. Table 10 shows that together they are about three- thermore, the joint decision process could be due to unobserved quarters of rural household cash income in 2012. Crop sales in characteristics that determine both non-farm participation (taking the South were more than double those in the North. In both a loan and selling crops) and input use such as labor availability regions they average about a quarter of cash incomes. Note from and networks. Consequently we estimate a seemingly unrelated Table 10 that livestock sales and remittances are tiny compared multivariate Probit regression. Given the non-linearity of our out- with these other sources. Also note that the cash levels of the credit come variables and the recursive structure of the model, we do not transactions for external inputs are very low compared to cash face the classical identification issue common in linear SUR (Wilde, incomes. 2000; Smale et al., 2016). This system approach offers an efficiency Second, we included several socio-economic variables (gender, gain by taking into account correlations among the residuals of the age, and education of the household head, as well as the depen- equations in a system of equations capturing the binary decisions dency ratio and distance from the market) to proxy for systematic Table 9 Summary statistics of variables used in the regression analysis, Nigeria, South, North. Source: Generated by authors using LSMS data. Variables Nigeria South North 2010 2012 2010 2012 2010 2012 Household head is Male (0/1) 88 87 76 74 96 96 Age of the household head (years) 51 52 56 57 47 49 Household dependency ratio 1.1 1 0.9 0.8 1.2 1.1 Household head has formal education (0/1) 60 60 71 70 52 53 Household resides in an urban area (0/1) 13 12 18 17 10 8 Land holding size (hectares) 0.9 0.8 0.5 0.4 1.2 1 Agricultural assets index 0.3 0.2 0.4 0.1 0.2 0.3 A household member is engaged in Non-Farm self-employment (0/1) 56 60 51 57 60 62 A household member is engaged in off Farm wage employment (0/1) 23 18 24 23 23 15 Household received any loan (0/1) 39 40 36 42 42 39 Household received loan from formal source (0/1) 3 5 3 8 3 3 Household received loan from informal source (0/1) 18 19 18 24 18 16 Household received loan from friends or relatives (0/1) 28 29 22 26 33 30 Value of sales per ha of land cultivated (in 000 Naira) 43 43 65 69 30 30 Use fertilizer (0/1) 45 45 25 21 59 61 Purchase Fertilizer (0/1) 41 42 23 20 55 56 Fertilizer price (in Naira per kg) 85 103 93 106 80 100 Distance to Nearest Market (km) 71 70 66 66 75 73 Coefficient of variation of rainfall 94 95 68 68 111 112 Share of land cultivated allocated to grains crops 43 44 15 16 59 59 Share of land cultivated allocated to legumes crops 16 17 1 1 25 25 Share of land cultivated allocated to tubers crops 28 25 58 53 10 10 Share of land cultivated allocated to oil crops 3 3 5 6 2 2 Share of land cultivated allocated to horticulture crops 7 8 15 18 3 3 Share of land cultivated allocated to cotton 0 0 0 0 0 0 Share of land cultivated allocated to tobacco 0 0 0 0 0 0 Share of land cultivated allocated to tea/coffee 0 0 0 0 0 0 Share of land cultivated allocated to other crops 3 3 7 7 1 0 Geographic zones North central 17 17 0 0 29 27 North east 20 20 0 0 34 34 North west 22 24 0 0 37 40 South east 20 19 49 49 0 0 South south 13 13 31 32 0 0 South west 9 7 21 19 0 0 Note: Means of binary variables are expressed in percentage. S.G. Adjognon et al. / Food Policy 67 (2017) 93–105 101 Table 10 Sources of cash income in Nigeria, North, and South, 2012. Source: Generated by authors using LSMS data Income sources Household cash sources (000 naira) Share of cash from each source (%) Nigeria South North Nigeria South North Cash income Profit from household enterprises 119.9 110.2 127 46.2 38.0 53.5 Wage income 77.7 105.4 57.3 29.9 36.3 24.1 Crop sales (gross) 60 71.4 51.7 23.1 24.6 21.8 Livestock net sales 1 1.1 0.9 0.4 0.4 0.4 Remittances 1.1 2 0.4 0.4 0.7 0.2 Total cash 259.6 290.1 237.3 100.0 100.0 100.0 Inputs credit transactions 0.4 0.1 0.6 0.2 0.03 0.3 Inputs non credit transactions 10.8 4.2 15.6 4.2 1.4 6.6 Total input purchase 11.1 4.2 16.2 4.3 1.4 6.8 Hired labor value for harvest only 12.9 7.5 16.9 5.0 2.6 7.1 Imputed value of own crop output 140.5 88.7 178.6 54.1 30.6 75.3 Note: The numbers in the left panel are zero-in averages. The shares on the right are based on ratio of number on the left to the total cash value. Inputs include fertilizer, seeds, and pesticides. For each value in the table, instead of deleting outliers we winsorized them i.e. replace top 10% values by the highest value within 90% of the distributions, thus creating a pile up at the top without changing the distribution (Cox, 2006). For imputation of value of own crop output method, we estimate unit prices of crops for crops that were sold, and then we use the median price in the local governments and multiply by harvest quantities to get the value of crop sales. The harvest labor for planting activities is missing in the 2010 dataset, and therefore we focus on the harvest labor only in both years. The values reported for the cash sources are nominal values for each year. differences in resource access, transaction costs, productive struc- results are generally consistent and in line with the literature on ture, and the number of years of experience in farming (Feder et al., fertilizer demand. However, they reveal substantial differences 1985). between northern and southern Nigeria. Most relevant determi- Third, we included household-level asset variables, in particu- nants of fertilizer purchases show higher significance in the North lar, farm size and agricultural quasi-fixed assets (tractor, plow, irri- compared to the South. This possibly reflects the North using more gation pump, and so on). The latter were captured using an asset fertilizer and therefore is more responsive to various determinants index computed using the principal component analysis approach than the South. (Filmer and Pritchett, 2001). Note from Table 9 that the farms in We find that participation in non-farm self-employment has the North are roughly double the size of those in the South. positive and significant effects on fertilizer purchases. The esti- Fourth, we included shares of crop types in the cropped area of mated APE (Average Partial Effects) indicates that participation in the farm. In general, there is much more grain cropping and much it raises the likelihood of purchasing fertilizer by about 7%. This less tuber and horticulture cropping in the North compared with result is consistent across both the South (10% increase) and the the South. This is roughly correlated with rainfall levels. North (5%). These findings coincide generally with the descriptive To account for zone and region effects, we include the following findings above, and corroborate earlier findings of nonfarm income sets of variables. on input purchase, such as Adesina (1996) for Ivory Coast and First, we have dummy variables representing the six main zones Oseni and Winters (2009) for Nigeria. However, contrary to Oseni (Northeast, Northwest, Southeast, Southcentral, and Southwest), and Winters (2009), we find that wage employment did not appear reflecting different infrastructural and growing conditions. Also as a significant determinant of fertilizer purchase and even has a at a broad level, we have a dummy for urban versus rural areas negative coefficient, perhaps due to wage employment drawing (as there is farming by households classed in urban areas). In addi- members away from the farm area and thus competing with farm- tion to the overall (country level) analysis, we estimate regional- ing (as Smale et al., 2016 hypothesizes)7. Moreover, neither RNFI level parameters for the Northern and Southern regions (the sub- variable is a significant determinant of the amounts of fertilizer pur- regions as noted above). As mentioned by Oseni and Winters chased, according to the Tobit results. The balance between farm and (2009), there are important cultural and socio-economic differ- non-farm competition for resources on one hand, and the relaxation ences between the two regions which can affect the way farmers of cash constraints to allow financing of agriculture inputs on the respond to changes in determinants of inputs use. Table 9 shows other hand, determine the observed effects of non-farm employ- that compared to the South, the north of Nigeria is more rural ment. In our case, they seem to cancel each other out, especially and traditional, with larger household sizes, greater poverty, and when we look at the effect on fertilizer amounts. less education. While lagged access to loans positively affects fertilizer pur- Second, we have several variables at a more disaggregate level, chase, the effect is significant only in the North. A closer investiga- the LGA (the ‘‘local government area”). These include the price of tion of the types of loans taken by farmers shows that loans from fertilizer, and agricultural productivity risk. The latter is captured friends and relatives (rather than loan from formal and semi- by the coefficient of variation of rainfall in the LGA, hypothesized formal institutions) drive most of these results (with the regres- to reduce the demand for fertilizer, especially in the absence of sions using different kinds of loans as explanatory variables not ex-post risk mitigation opportunities and lack of credit and insur- shown in the tables). This could illustrate the fact that loans, and ance mechanisms (Dercon and Christiaensen, 2011). in particular loans from formal and semi-formal institutions, are limited for agricultural investment. Given the risks related to agri- 4.2. Regression results cultural activities, formal and semi-formal credit suppliers are Tables 11a and 11b present the average partial effects of the 7 This is supported by our data as non-farm wage employment participants supply determinants of fertilizer purchase overall in Nigeria and by region statistically significantly lower amount of labor a week, on average to agriculture from the pooled Probit and pooled Tobit estimates. The CRE and FE (11.12 h) than non-participants (18.24) with a p value of 0.000. 102 S.G. Adjognon et al. / Food Policy 67 (2017) 93–105 Table 11a Estimation results of determinants of fertilizer purchase (0/1) decision. Source: Generated by authors using LSMS data. Variables Nigeria South North CRE Probit Linear FE CRE Probit Linear FE CRE Probit Linear FE Household head is Male (0/1) 0.050⁄ 0.115 0.026 À0.010 0.131⁄⁄⁄ 0.637⁄⁄ [0.058] [0.177] [0.355] [0.861] [0.004] [0.049] Age of the household head (years) 0.000 À0.001 0.000 À0.001 0.000 À0.001 [0.740] [0.725] [0.654] [0.784] [0.876] [0.716] Household dependency ratio À0.006 À0.012 0.030 0.028 À0.030⁄ À0.034⁄ [0.669] [0.446] [0.169] [0.225] [0.052] [0.081] Household head has formal education (0/1) 0.081⁄⁄⁄ 0.051⁄⁄ 0.056⁄⁄ 0.040 0.094⁄⁄⁄ 0.057⁄⁄ [0.000] [0.041] [0.046] [0.521] [0.000] [0.035] Land holding size (hectares) 0.017 0.026⁄ À0.033 À0.013 0.031⁄⁄ 0.037⁄⁄ [0.220] [0.057] [0.260] [0.603] [0.036] [0.020] Agricultural asset index 0.002 0.002 0.003 0.001 0.002 0.002 [0.267] [0.323] [0.403] [0.576] [0.349] [0.315] Total Livestock Units index 0.652⁄⁄ 0.413 À0.649 0.355 0.651⁄ 0.349 [0.049] [0.158] [0.656] [0.536] [0.080] [0.257] LOG of crop sales in naira per ha of harvested land 0.001+ 0.001⁄ 0.000 0.000 0.001⁄ 0.001⁄⁄ [0.114] [0.071] [0.623] [0.597] [0.083] [0.049] A household member is engaged in Non-Farm Self-employment (1/0) 0.070⁄⁄ 0.078⁄⁄⁄ 0.106⁄⁄ 0.135⁄⁄⁄ 0.052+ 0.062⁄ [0.012] [0.009] [0.020] [0.009] [0.149] [0.078] A household member is engaged in wage employment (1/0) À0.022 À0.019 À0.025 À0.022 À0.006 À0.004 [0.418] [0.545] [0.505] [0.595] [0.876] [0.918] A household member took a loan (0/1) 0.056⁄⁄⁄ 0.060⁄⁄⁄ 0.045 0.047 0.069⁄⁄⁄ 0.076⁄⁄⁄ [0.009] [0.001] [0.203] [0.163] [0.009] [0.001] Coefficient of variation of rainfall À0.003⁄⁄⁄ À0.003 À0.004⁄⁄⁄ [0.001] [0.623] [0.000] LOG of fertilizer price in Naira per kg À0.017 À0.019 À0.015 À0.003 À0.017 À0.033 [0.616] [0.481] [0.729] [0.920] [0.737] [0.438] Share of total land cultivated allocated to grains crops 0.002 0.003⁄ 0.000 À0.000 0.004⁄⁄ 0.003⁄⁄ [0.171] [0.061] [0.813] [0.711] [0.047] [0.047] Share of total land cultivated allocated to legumes crops 0.001 0.002 0.001 0.000 0.003+ 0.002 [0.492] [0.173] [0.478] [0.829] [0.138] [0.199] Share of total land cultivated allocated to tubers crops 0.001 0.002 À0.000 À0.001 0.004⁄ 0.002 [0.560] [0.207] [0.766] [0.217] [0.072] [0.179] Share of total land cultivated allocated to oil crops 0.001 0.002 0.000 À0.000 0.002 0.000 [0.558] [0.267] [0.809] [0.979] [0.560] [0.981] Share of total land cultivated allocated to horticulture crops 0.002⁄ 0.003⁄ 0.001 0.001 0.004⁄ 0.003+ [0.092] [0.052] [0.360] [0.380] [0.091] [0.123] Share of total land cultivated allocated to cotton À0.002 0.001 [0.387] [0.774] Urban dummy variable (0/1) 0.089⁄⁄⁄ À0.096 0.092⁄⁄⁄ À0.118 0.050 À0.143 [0.005] [0.701] [0.008] [0.189] [0.252] [0.607] Household Distance in (KMs) to Nearest Market À0.002⁄⁄⁄ À0.003 À0.000 À0.000 À0.003⁄⁄⁄ À0.006 [0.000] [0.771] [0.454] [0.990] [0.000] [0.259] Year 2010 (0/1) 0.017 0.017 0.029 0.054⁄⁄⁄ 0.005 À0.005 [0.237] [0.191] [0.154] [0.004] [0.784] [0.767] Zone dummies North east 0.095⁄ 0.071 [0.057] [0.185] North west 0.324⁄⁄⁄ 0.310⁄⁄⁄ [0.000] [0.000] South east À0.165⁄⁄⁄ 0.255⁄⁄⁄ [0.006] [0.000] South center À0.229⁄⁄⁄ 0.169⁄⁄ [0.002] [0.011] South west À0.400⁄⁄⁄ [0.000] Constant 0.339 0.168 0.231 [0.662] [0.875] [0.669] Number of observations 4843 4843 1670 1670 3173 3173 R-squared 0.027 0.045 0.037 Number of households 2730 995 1735 Note: ⁄⁄⁄, ⁄⁄, ⁄, and + indicate that the corresponding regression coefficients are statistically significant at the 1%, 5%, 10%, and 15% levels, respectively. Model estimated using partial MLE estimation method. P values based on clustered standard errors between brackets. CRE stands for Correlated Random Effects while FE stands for Fixed Effects. reluctant to provide loans for agricultural purposes, as they fear dominant sources of loans in each region. Friends and relatives higher risk of default. Although we could not test specifically this seem to be a dominant source of loans taken by households in hypothesis in Nigeria due to data limitations, as we noted above, the North compared with the South. Our analysis of the loan data the data for Malawi and Tanzania show that consumption and in Nigeria from the LSMS (not shown in a table) provides some evi- investment in business start-ups are by far the primary purposes dence for this. There are 22–26% of households reporting loans of the loans taken by households. Besides, the fact that the effect from friends and relatives in the South, compared to 30–33% in of loan is significant only in the North could be explained by the the North. S.G. Adjognon et al. / Food Policy 67 (2017) 93–105 103 Table 11b Estimation results of determinants of quantity of fertilizers purchased by farmers in Nigeria. Source: Generated by authors using LSMS data. Variables Nigeria South North CRE Tobit Linear FE CRE Tobit Linear FE CRE Tobit Linear FE Household head is Male (0/1) 65.803⁄⁄⁄ 139.042⁄⁄ 20.022 11.096 109.283⁄⁄⁄ 346.443⁄⁄ [0.003] [0.032] [0.333] [0.870] [0.007] [0.038] Age of the household head (years) 0.086 À1.871 0.220 0.828 0.319 À2.984 [0.802] [0.297] [0.698] [0.619] [0.462] [0.200] Household dependency ratio 0.941 4.056 24.422 10.978 À11.982 À0.741 [0.935] [0.847] [0.184] [0.727] [0.414] [0.978] Household head has formal education (0/1) 31.943⁄⁄ 1.798 43.532⁄⁄ 8.492 29.495⁄ 9.200 [0.014] [0.953] [0.032] [0.910] [0.079] [0.782] Land holding size (hectares) À49.135⁄⁄⁄ À111.516⁄⁄⁄ À44.826⁄ À81.532⁄⁄⁄ À59.393⁄⁄⁄ À117.974⁄⁄⁄ [0.000] [0.000] [0.051] [0.001] [0.000] [0.000] Agricultural asset index 1.779⁄ 1.372 1.948 0.838 1.721+ 0.915 [0.063] [0.228] [0.405] [0.511] [0.113] [0.599] Total Livestock Units index 794.160⁄⁄⁄ 890.156⁄⁄ À254.360 174.590 843.723⁄⁄⁄ 833.816⁄⁄ [0.001] [0.012] [0.832] [0.812] [0.002] [0.022] LOG of crop sales in naira per ha of harvested land 0.277 0.046 0.048 À0.164 0.531 0.433 [0.536] [0.940] [0.936] [0.859] [0.402] [0.571] A household member is engaged in Non-Farm Self-employment (1/0) 16.582 À12.894 57.657+ 2.836 7.946 À5.306 [0.484] [0.750] [0.144] [0.965] [0.798] [0.917] A household member is engaged in wage employment (1/0) 85.647 12.824 À12.627 16.299 16.685 31.760 [0.796] [0.690] [0.702] [0.779] [0.490] [0.404] A household member took a loan (0/1) 15.764 3.137 32.709 44.667 14.671 À4.289 [0.341] [0.893] [0.218] [0.234] [0.479] [0.884] Coefficient of variation of rainfall À2.147⁄⁄⁄ À4.164 À2.568⁄⁄⁄ [0.000] [0.356] [0.000] LOG of fertilizer price in Naira per kg À31.291 À56.164⁄⁄ À35.938 À60.673⁄⁄ À28.953 À64.421+ [0.208] [0.024] [0.293] [0.041] [0.437] [0.114] Share of total land cultivated allocated to grains crops 1.088 18.904⁄⁄⁄ À0.008 À1.120+ 2.802⁄ 18.990⁄⁄⁄ [0.224] [0.005] [0.993] [0.147] [0.074] [0.004] Share of total land cultivated allocated to legumes crops 1.019 19.196⁄⁄⁄ 0.926 À0.400 2.504+ 18.899⁄⁄⁄ [0.301] [0.005] [0.488] [0.685] [0.111] [0.005] Share of total land cultivated allocated to tubers crops 0.809 19.181⁄⁄⁄ 0.001 À0.121 2.074 18.070⁄⁄⁄ [0.375] [0.004] [0.999] [0.805] [0.167] [0.006] Share of total land cultivated allocated to oil crops 1.286 19.935⁄⁄⁄ 0.665 1.016 1.557 18.326⁄⁄⁄ [0.228] [0.003] [0.526] [0.250] [0.413] [0.005] Share of total land cultivated allocated to horticulture crops 1.623⁄ 19.315⁄⁄⁄ 0.956 0.116 2.441 18.318⁄⁄⁄ [0.078] [0.004] [0.318] [0.802] [0.161] [0.006] Share of total land cultivated allocated to cotton À7.072⁄⁄⁄ À7.280⁄⁄ [0.009] [0.041] Urban dummy variable (0/1) 48.053⁄⁄ À229.742 57.901⁄⁄ À99.588+ 18.113 À255.808 [0.012] [0.339] [0.019] [0.146] [0.426] [0.344] Household Distance in (KMs) to Nearest Market À1.074⁄⁄⁄ 1.853 À0.435 3.524 À1.208⁄⁄⁄ 2.646 [0.000] [0.538] [0.302] [0.326] [0.000] [0.544] Year 2010 (0/1) 20.286⁄ 10.804 39.974⁄⁄⁄ 92.170⁄⁄⁄ 1.296 À30.360 [0.070] [0.559] [0.009] [0.000] [0.938] [0.265] Zone dummies 72.740⁄⁄⁄ 54.593⁄ [0.009] [0.072] North east 179.017⁄⁄⁄ 172.165⁄⁄⁄ [0.000] [0.000] North west À32.435 211.842⁄⁄⁄ [0.425] [0.000] South east À155.953⁄⁄⁄ 105.926⁄⁄ [0.003] [0.021] South center À241.589⁄⁄⁄ [0.000] South west À1551.668⁄⁄ 89.799 À1654.857⁄⁄ [0.031] [0.755] [0.031] Number of observations 4843 4843 1670 1670 3173 3173 R-squared 0.059 0.050 0.083 Number of households 2730 995 1735 Note: ⁄⁄⁄, ⁄⁄, ⁄, and + indicate that the corresponding regression coefficients are statistically significant at the 1%, 5%, 10%, and 15% levels, respectively. Model estimated using partial MLE estimation method. P values based on clustered standard errors between brackets. The coefficient of variation of rainfall has, as expected, a and significant effect in both the North and the South. The farm strongly negative effect on fertilizer purchase, but this is only sig- size effect is significant and positive only in the North, while it is nificant in the North. This result is important as investments in negative but not significant in the South. Crop sales affect posi- modern input use though generally profitable, are costly and can tively, but not significantly, the fertilizer purchase decision. yield very low (or even negative) returns in case of negative The results of the seemingly unrelated regressions (available weather shocks. upon request from authors, given space limitation) are also consis- Other factors that significantly affect fertilizer purchase are as tent with the single equation results. Both the household and geo- expected such as education of the household head with a positive graphical factors that affect demand and more importantly the 104 S.G. Adjognon et al. / Food Policy 67 (2017) 93–105 positive effect of non-farm self-employment and loans on fertilizer nonfarm sector, in manufacture and services, could benefit farm use are maintained. However, the unexplained portions of the fer- input purchase and thus productivity and food security, and cer- tilizer purchase equation and the other sources of cash (including tainly be an important complement to credit policies and self-employment, crop sales and taking a loan) were not correlated programs. for the most part suggesting that these decisions are not necessar- ily made jointly and thus appropriately modeled using the single Acknowledgement equation CRE and FE. The authors acknowledge and appreciate financial support for this work from the Bill and Melinda Gates Foundation, MSU AgBio 5. Conclusions Research, and USAID via the Food Policy Innovation Lab Program at MSU. The authors are most grateful to Luc Christiaensen, Robert Many believe that SSA farmers’ increasing their purchase of Myers, Nicole Mason, Mywish Maredia, Robert Shupp, Jeffrey external inputs such as fertilizer, seed, and pesticides can bring a Wooldridge and two anonymous referees for their comments on welcome increase in yields. It has also been observed (Sheahan earlier versions. Any views expressed or remaining errors are and Barrett, 2014), and echoed in our paper, that the purchase of solely the responsibility of the authors. these external inputs is definitely no longer absent in SSA and is even very prevalent in some countries, contrary to the common perception. There had not been a systematic exploration of how References farmers are paying for these inputs – in particular, what were the relative roles of two sources of cash to pay for inputs (inter alia) Adesina, A.A., 1996. Factors affecting the adoption of fertilizers by rice farmers in – credit (informal and formal) and own cash income. This paper Côte d’Ivoire. Nutr. Cycl. Agroecosyst. 46, 29–39. systematically delved into nationally representative datasets for Angrist, J.D., Pischke, J.S., 2008. Mostly harmless econometrics: An empiricist’s companion. Princeton University Press. four countries in SSA with widely varying characteristics (Malawi, Aryeetey, E., 1997. Rural Finance in Africa: Institutional Developments and Access Nigeria, Tanzania, and Uganda) and examined the roles of these for the Poor. In: Bruno, M., Pleskovic, B. (Eds.), Annual World Bank Conference sources. on Development Economics 1996. World Bank, Washington DC, pp. 149–173. Bardhan, P.K., 1980. Interlocking factor markets and agrarian development: a While the literature emphasized that with the reduction or review of issues. Oxford Econ. Pap. 32(1), March, 82–98. elimination of parastatal agrarian banks formal bank credit is sel- Bardhan, P.K., 1984. Land, Labor, and Rural Poverty: Essays in Development dom or never available to Sub-Saharan African farmers for inputs, Economics. Columbia University Press, New York. Binswanger, H.P., Rosenzweig, M., 1986. Behavioural and material determinants of there was explicitly or implicitly in the literature the working production relations in agriculture. J. Dev. Stud. 22 (3), 503–539. hypothesis that farmers used traditional tied credit with output Binswanger, H.P., Ruttan, V.W., 1978. Induced Innovation: Technology, Institutions, and input traders, and other sources of informal credit to finance and Development. Johns Hopkins University Press, Baltimore. their purchase of external inputs for non-contract farming situa- Chao-Béroff, R., 2014. Global dynamics in agricultural and rural economy, and its effects on rural finance. In: Finance for Food. Springer, pp. 3–21. tions. For cash contract-farming situations and cash cropping in Chamberlain, G., 1982. Multivariate regression models for panel data. J. Econom. 1, general, the working hypothesis in much of the literature is that 5–46. processors front inputs or cash for inputs to farmers. Chirwa, E., Dorward, A., 2013. Agricultural Input Subsidies: The Recent Malawi Experience. Oxford University Press, Oxford. By and large, our paper contradicted these ‘‘common wisdoms” Clay, D., Reardon, T., Kangasniemi, J., 1998. Sustainable intensification in the concerning the use and role of credit in input purchase. First, we highland tropics: Rwandan farmers’ investments in land conservation and soil found that very few farmers use any form of credit, formal or infor- fertility. Econ. Dev. Cult. Change 46 (2), 351–378 (January). Conning, J., Udry, C., 2007. Rural financial markets in developing countries. mal to finance external input purchase. Second, we found that Handbook Agric. Econ. 3, 2857–2908. ‘‘tied” credit-output relations are very rare and very minor in Cox, N.J., 2006. WINSOR: Stata module to Winsorize a variable. Statist. Softw. external inputs, but especially among smaller farmers in poorer Compo. Davis, B., Winters, P., Reardon, T., Stamoulis, K., 2009. Rural nonfarm employment places. What is still significant is tied labor-output markets where and farming: household-level linkages. Agric. Econ. 40 (2), 119–123. local workers advance labor and are paid at the harvest, largely Dercon, S., Christiaensen, L., 2011. Consumption risk, technology adoption and ignored in the literature. Third, we found that generally ‘‘tradi- poverty traps: evidence from Ethiopia. J. Dev. Econ. 96 (2), 159–173. Dorward, A.R., Kirsten, J.F., Omamo, S.W., Poulton, C., Vink, N., 2009. Institutions and tional cash crop farmers” rarely receive credit from processors, the agricultural development challenge in Africa. In: Kirsten, J.F., Dorward, A.F., except in a few enclaves like larger tobacco farmers in Tanzania. Poulton, C., Vink, N. (Eds.), Institutional Economics Perspectives on African Furthermore, we found econometrically that nonfarm self- Agricultural Development. IFPRI, Washington DC, pp. 3–34. employment (but not wage employment) plays a significant and Ellis, F., Freeman, H.A., 2004. Rural livelihoods and poverty reduction strategies in four African countries. J. Dev. Stud. 40 (4), 1–30. positive role in inputs purchase decision, especially given the lim- Feder, G., Just, R.E., Zilberman, D., 1985. Adoption of agricultural innovations in ited availability of credit for agricultural purposes. It appears that developing countries: a survey. Econ. Dev. Cult. Change 33 (2), 255–298. farmers use loans to start nonfarm enterprises (and finance con- FMARD, 2015. End of Program Report 2011–2014. Unpublished Document Produced by the Federal Ministry of Agriculture and Rural Development. sumption) and plow back the cash partly into their farm input Abuja, Nigeria. needs; an important observation worthy of further exploration. Filmer, D., Pritchett, L.H., 2001. Estimating wealth effects without expenditure These findings do not reflect on or test whether farmers face data—or tears: an application to educational enrollments in states of India. Demography 38 (1), 115–132. credit constraints; the fact that farmers use very little credit, infor- Green, W.H., 2000. Econometric Analysis. Prentice Hall International, New York. mal or formal, for farm inputs, does not inform researchers or pol- Haggblade, S., Hazell, P.B.R., Reardon, T., 2010. The rural nonfarm economy: icymakers whether the farmers have too little access to credit. prospects for growth and poverty reduction. World Dev. 38 (10), 1429–1441. Kelly, V., Adesina, A.A., Gordon, A., 2003. Expanding access to agricultural inputs in What we can say from the data is that nonfarm employment is pro- Africa: a review of recent market development experience. Food Pol. 28, 379– viding a major source of cash that currently far eclipses use of 404. credit for inputs purchases. When farmers take loans, they mainly Kherallah, M., Delgado, C.L., Gabre-Madhin, E., Minot, N., Johnson, M., 2002. Reforming Agricultural Markets in Africa. IFPRI and Johns Hopkins University use the funds to start nonfarm enterprises or finance consumption. Press. They then often use nonfarm income cash to buy farm inputs. That Morris, M., Kelly, V.A., Kopicki, R.J., Byerlee, D., 2007. Fertilizer Use in African appears to imply that farmers see that employment as a crucial Agriculture: Lessons Learned and Good Practice Guidelines. Directions in cash source to meet their farm needs. Further rigorous analysis is Development: Agriculture and Rural Development, Report 39037. The World Bank, Washington. necessary to confirm this but it implies that rural development Mundlak, Y., 1978. On the pooling of time series and cross section data. policies and programs that spur broad development of the rural Econometrica 46, 69–85. S.G. Adjognon et al. / Food Policy 67 (2017) 93–105 105 Oseni, G., Winters, P., 2009. Rural nonfarm activities and agricultural crop Sheahan, M., Barrett, C.B., 2014. Understanding the agricultural input landscape in production in Nigeria. Agric. Econ. 40 (2), 189–201. Sub-Saharan Africa: Recent plot, household, and community-level evidence. Pearce, D., 2003. Buyer and supplier credit to farmers: do donors have a role to Policy Research Working Paper 7014. World Bank, Africa Region. play? Paper presented at Paving the Way Forward for Rural Finance: An Shepherd, A., Farolfi, S., 1999. Export Crop Liberalization in Africa: A Review. FAO, International Conference on Best Practices. Washington DC, June 2–4. Rome. Poulton, C., Kydd, J., Dorward, A., 2006. Overcoming market constraints on pro-poor Smale, M., Kusunose, Y., Mathenge, M.K., Alia, D., 2016. Destination or distraction? agricultural growth in Sub-Saharan Africa. Dev. Pol. Rev. 24 (3), 243–277. Querying the linkage between off-farm work and food crop investments in Poulton, C., Dorward, A., Kydd, J., 1998. The revival of smallholder cash crops in Kenya. J. Afr. Econ. 25 (3), 388–417. Africa: public and private roles in the provision of finance. J. Int. Dev. 10 (1), 85– Stampini, M., Davis, B., 2009. Does nonagricultural labor relax farmers’ credit 103. constraints? Evidence from longitudinal data for Vietnam. Agric. Econ. 40 (2), Poulton, C., Dorward, A., Kydd, J., 2010. The future of small farms: new directions for 177–188. services, institutions, and intermediation. World Dev. 38 (10), 1413–1428. Takeshima, H., Nkonya, E., 2014. Government fertilizer subsidy and commercial Reardon, T., Crawford, E., Kelly, V., 1994. Links between nonfarm income and farm sector fertilizer demand: evidence from the Federal Market Stabilization investment in African households: adding the capital market perspective. Am. J. Program (FMSP) in Nigeria. Food Pol. 47, 1–12. Agric. Econ. 76 (5), 1172–1176. Wilde, J., 2000. Identification of multiple equation probit models with endogenous Rozelle, S., Taylor, J.E., DeBrauw, A., 1999. Am. Econ. Rev. 89 (2), 287–291. dummy regressors. Econ. lett. 69 (3), 309–312. Sadoulet, E., de Janvry, A., 1995. Quantitative Development Analysis. Johns Hopkins Wooldridge, J.M., 2010. Econometric Analysis of Cross Section and Panel Data. MIT University Press, Baltimore. Press. Savadogo, K., Reardon, T., Pietola, K., 1994. Farm productivity in Burkina Faso: Zeller, M., Sharma, M., 1998. Rural Finance and Poverty Alleviation. Food Policy effects of animal traction and nonfarm income. Am. J. Agric. Econ. 76 (3), 608– Report. International Food Policy Research Institute, Washington, D.C.. 612.