Drivers of and barriers to adoption of improved sunflower varieties amongst smallholder farmers in Singida, Tanzania: A double-hurdle approach

Abstract

1. Introduction

Yield-enhancing technologies such as improved crop varieties are essential in agricultural and rural development, because they improve agricultural productivity, technical efficiency, income from crop production, household welfare, and food security.¹ Increasing adoption of improved technologies is thus important for achieving the said positive outcomes, particularly in developing countries. Public and private agencies as well as policy makers have widely promoted the use of yield-enhancing technologies such as improved crop varieties. However, the adoption of improved crop varieties among smallholder farmers in sub-Saharan Africa (SSA) remains low (e.g. Asfaw et al. 2012b; Schroeder et al. 2013; Wineman et al. 2020). Fostering the adoption of improved crop varieties requires knowledge of the barriers to and drivers of the adoption.

To obtain this knowledge, our study analyses barriers to and drivers of the adoption of improved sunflower varieties by smallholder sunflower farmers in Singida region, Tanzania. We distinguish two dimensions of the adoption: (a) whether or not a farm household adopts the improved sunflower varieties and (b) the extent of the adoption. This paper shows to which extent various factors are related to these two dimensions of the adoption of improved sunflower varieties.

Existing studies on adoption of improved crop varieties in SSA reveal that adoption is related to numerous factors and barriers. The factors include household characteristics such as household size, education, age, gender, and farming experience, as well as economic circumstances such as the prices of seeds and fertilizers, off-farm employment, and household income.² The barriers include lack of awareness of or information about the availability of improved crop varieties, inadequate supply of seeds of improved crop varieties in the market, risk aversion, liquidity constraints, and limited access to credit, high prices of hybrid seeds, negative perceptions about the cultivation of improved varieties, misidentification of seed type, poor development of market infrastructure, and limited access to agricultural extension.³

Most of the factors and barriers were studied in the context of cereals such as maize and rice as well as grain legumes.⁴ The adoption of improved oilseed crop varieties has been analyzed in only a few studies such as the adoption of improved groundnut varieties in Malawi (Simtowe et al. 2010, 2011) and Uganda (Kassie et al. 2011),⁵ and improved oil palm varieties in Cameroon (Assoumou Mezui et al. 2013). In developing countries, the value chains of oilseeds such as sunflower usually largely differ from the value chains of cereals and grain legumes. Most oilseeds are processed by companies or cooperatives before they are sold to households, whereas most cereals and grain legumes are frequently sold to consumers without prior processing as households usually process them at home or do not process them at all. As oil mills have different requirements regarding the purchased oilseeds (e.g. homogenous product, varieties with high oil content) than consumers have regarding the purchase of cereals and grain legumes (e.g. taste, smell, appearance in terms of colour and size, preference for traditional varieties), adoption decisions regarding improved oilseed varieties likely differ from adoption decisions regarding improved varieties of other crops such as cereals and grain legumes because farmers base their adoption decisions on different characteristics of the crops (i.e. objective characteristics preferred by oil mills instead of subjective characteristics preferred by their own household and by consumers).

Tanzania’s government and private sector are actively promoting the cultivation of improved sunflower varieties through contract farming and stockists. Seeds of improved sunflower varieties are sold to smallholder farmers by stockists, sometimes at subsidized prices. The promotion of improved sunflower varieties is partly induced by a growing demand for sunflower edible oil in local, domestic, and international markets. For example, the expansion of the processing capacity and the awareness of health advantages of sunflower oil (Adam Smith International 2014; Ministry of Industry, Trade and Investment 2016) have contributed to the increased demand for sunflower oilseeds in Tanzania. There is also a growing demand for oilseeds and its by-products in foreign markets (Food and Agriculture Organization 2016; Ministry of Industry, Trade and Investment 2016). The growing demand for sunflower oil and its by-products (e.g. sunflower seed cake) provide opportunities to sunflower farmers to expand production without lowering prices. Thus, the cultivation of improved sunflower varieties is an opportunity for smallholder sunflower farmers to gain income from the growing demand for sunflower oil and by-products on the domestic and foreign markets.

In spite of a plethora of studies that analyze adoption of improved crop varieties, to the best of our knowledge our study is among few studies of adoption of improved oilseed varieties and the first on the adoption of improved sunflower varieties among smallholder farmers in Africa.⁶ Besides this empirical contribution, our study contributes to the literature in two further aspects. Firstly, we analyze the relationship between farmers’ adoption decisions and their liquidity constraints as well as the availability of seeds of improved varieties, which was largely ignored in previous studies. Secondly, we suggest a microeconomic model for adoption decisions, where households can choose between different extents of adoption, i.e. adoption is not a binary variable but a continuous variable that is left-censored at zero. The results of this analysis have implications for policy and practice in the public and private sector as they can be used to improve policies, programmes, and business decisions that increase the adoption of improved varieties and ultimately improve agricultural productivity and household welfare.

The remainder of this paper is organized as follows. The next section presents an overview of sunflower farming and sunflower varieties in Tanzania as well as the microeconomic background of our analysis. The third section presents the data and empirical methods. The subsequent section presents and discusses the results. Finally, the last section concludes and presents policy implications.

2. Empirical and theoretical contexts

2.1 Sunflower farming and sunflower varieties in Tanzania

Sunflower was introduced in Tanzania during the colonial era (Food and Agriculture Organization 2012; Ministry of Industry, Trade and Investment 2016) and it is mostly grown in the Eastern, Central, Northern, and Southern Highlands of Tanzania (Ministry of Agriculture, Food Security and Cooperatives 2011; Ministry of Industry, Trade and Investment 2016). About 61% of the sunflower production in Tanzania is located in the highlands of the central corridor of Tanzania, which is located in the administration regions Dodoma and Singida (Salisali 2012).

Sunflower production in Tanzania increased from 1,083,000 tons in 2013 to 3,112,500 tons in 2017 (Food and Agriculture Organization 2016; Ministry of Agriculture Livestock and Fisheries 2016; Ministry of Agriculture, Livestock and Fisheries 2018) and accounts for 2.4% of global production (Ministry of Industry, Trade and Investment 2016). Tanzania provides 35% of all sunflower production in Africa, which makes it the largest sunflower producer in East Africa and the second largest sunflower producer in Africa after South Africa (Kombe et al. 2017). Sunflower is the leading oilseed crop in Tanzania, followed by groundnuts, sesame, palm oil, cotton oil, and soya (Ministry of Agriculture Livestock and Fisheries 2016; Ministry of Industry, Trade and Investment 2016). Tanzania exports sunflower seeds for crushing, crude sunflower oil, refined sunflower oil, and seed cake, which in total account for about 74.8 million US dollars of Tanzania’s export earnings (Ministry of Industry, Trade and Investment 2016). The largest export shares among all sunflower products have sunflower seed cakes and seed meal of which almost half the production is exported, mainly to India and Kenya (Ministry of Industry, Trade and Investment 2016).

Sunflower is grown by farm households individually or in farmer groups using both monoculture and mixed cropping systems that include cassava, maize, sorghum, and cowpeas (RLDC 2008; Ministry of Industry, Trade and Investment 2016). All sunflower varieties available in Tanzania are bred for oil production, while varieties that are specifically bred for direct consumption of so-called ‘confectionary seeds’ are not yet available in Tanzania (Ministry of Industry, Trade and Investment 2016). Hence in Tanzania smallholder sunflower farmers use only a very small proportion of the produced sunflower seeds for direct household consumption, while selling the vast majority of their produce to oil mills or to traders who sell it to oil mills. About 95% of sunflower farmers in Tanzania are smallholders who cultivate less than two ha of land with sunflower (Ministry of Industry, Trade and Investment 2016). Due to their small size of production, smallholder farmers often encounter a multitude of internal and external constraints including limited access to input and output markets, credit, and technologies such as improved sunflower varieties (e.g. Adam Smith International 2014).

Public and private organizations have made ubiquitous efforts to address these constraints, e.g. through breeding, certifying, promoting, and distributing improved sunflower varieties. In Tanzania, seed production and certification are regulated under the Seed Act of 2003 as amended in 2007 (Kombe et al. 2017) and in 2014 (Westengen et al. 2019). The Seed Act in Tanzania stipulates the formation of a National Seeds Committee to advise the government on the development of the seed industry in Tanzania (Ministry of Industry, Trade and Investment 2016). It also states the minimum standards for seeds and requires proper labelling of seeds in order to curb counterfeit seeds in the market (Ministry of Industry, Trade and Investment 2016; Kombe et al. 2017). Moreover, seed quality control and certification in Tanzania is done by the Tanzania Official Seed Certification Institute (TOSCI). Since the 1950s, TOSCI has been collaborating with the Agricultural Research Institute (ARI) and private seed companies within and outside the country in developing and marketing sunflower seeds of both open-pollinated varieties (OPVs) and hybrid varieties. However, only 8 of the 1,058 seed companies that are registered with TOSCI breed new sunflower varieties and certify them with TOSCI (Tanzania Official Seed Certification Institute 2020). According to the Tanzania Official Seed Certification Institute (2020), only seventeen sunflower varieties have been approved, released, and disseminated in Tanzania (see Table A1 in the Appendix).

Between 1950 and 2015, only four sunflower varieties, two open-pollinated varieties (OPV), and two hybrid varieties were certified and made available in the market for sale to farmers (see Table A1 in the Appendix). During this period, seeds of the OPV ‘RECORD’ and later on also the OPV ‘KENYA FEDHA’ were most widely available in the central corridor of Tanzania. However, most smallholder farmers very rarely buy certified seeds but usually sow seeds that they retained from the seeds that they harvested in the previous growing season (e.g. Kosmowski et al. 2018; Maredia et al. 2019). These seeds are usually called recycled seeds, local seeds, or traditional seeds and they usually give lower yields than certified seeds, because they are of low-yielding varieties, crossbred with lower-yielding varieties, or affected by genetic deterioration (see, e.g. Morris et al. 1999), particularly in case of continued recycling over many years or decades. For example, due to the use of recycled seeds and other yield-limiting farming practices, the average sunflower yield in Tanzania is 0.69 t/ha, which is much lower than the average potential yield of 3 t/ha (e.g. Adam Smith International 2014; Ministry of Industry, Trade and Investment 2016; Kombe et al. 2017).

In order to encourage farmers to reduce seed recycling and thus to achieve higher yields, so-called ‘quality declared seeds’ (QDS) were introduced. The production of sunflower QDS is conducted by farmers who sow foundation seeds⁷ of an OPV (usually the ‘RECORD’ variety), follow certain production rules and are monitored by TOSCI in order to guarantee a high quality and purity of the harvested seeds (for details see Table A3 in the Appendix). The harvested seeds are distributed to other farmers for a much lower price than the certified seeds. As QDS are direct descendants of foundation seeds and are produced under regulated conditions, yields obtained from QDS are expected to be almost as high as yields obtained from certified seeds and much higher than yields obtained from recycled seeds. In the Central sunflower corridor of Tanzania, QDS production started with two selected farmers in 2007. While QDS production initially was unrelated to sunflower processors or contract farming, in 2009 sunflower processors were given the right to be involved in QDS production and marketing. Since then, several sunflower processors have used this opportunity and purchased foundation seeds of a sunflower OPV and contracted selected farmers to produce QDS for them. In addition to ‘independent’ sunflower farmers and sunflower farmers who are contracted by sunflower processors, QDS are sometimes also produced by Tanzania’s Agricultural Seed Agency (ASA).

In order to guarantee a sufficient utilization of their oil mills, many sunflower processors in Tanzania make contract farming arrangements with farmers’ associations and individual smallholder farmers. These contracts often specify that processors provide QDS (particularly of the ‘RECORD’ variety) or certified seeds of hybrid varieties (particularly ‘HYSUN 33’) to contracted farmers at a price determined by the processor, either for immediate payment or on credit. This improves the access of contracted farmers to seeds that are expected to give high yields (Henningsen et al. 2015). However, although these contract farming arrangements seem to benefit farmers, participation in contract farming is still low among smallholder sunflower farmers (Ministry of Industry, Trade and Investment 2016).

In summary, sunflower farmers in Tanzania can choose among three main categories of sunflower seeds for sowing: recycled seeds, QDS (of OPV), and certified seeds (of OPV or hybrid varieties). As the demand for sunflower oil and its by-products is growing rapidly in Tanzania and in the global markets (Adam Smith International 2014; Ministry of Industry, Trade and Investment 2016), there could be a great potential for Tanzanian sunflower farmers to gain from the increasing demand by switching to high-yielding sunflower varieties. However, private and public programmes for distributing seeds of improved sunflower varieties are challenged by low demand for QDS and certified seeds because the majority of the sunflower farmers still mostly sow recycled sunflower seeds.

2.2 Microeconomic background

Microeconomic theory provides insights into understanding decisions about the adoption of new technologies as well as into associated factors and barriers. Most of the studies on adoption of new technologies such as improved crop varieties by farm households (e.g. Asfaw et al. 2012b; Mariano et al. 2012; Wang et al. 2012; Jaleta et al. 2015) apply the random utility framework. This framework assumes that a decision maker faces a discrete set of alternative choices and decides to adopt the technology that gives the highest expected utility. However, in many situations, a decision maker who decides to adopt an improved technology must also decide on the extent of the adoption. We present an extension of the random utility framework that does not only explain whether a new technology is adopted but also the extent of adoption. In our microeconomic model for adoption decisions, households can choose between different extents of adoption, i.e. adoption is not a binary variable but a continuous variable that is left-censored at zero.

3. Data and methods

3.1 Data collection

The data for our empirical study were collected in Iramba and Mkalama districts in Singida region, Tanzania. Singida region was selected because the agro-climatic conditions in this region are favourable for sunflower farming (Business Care Services Limited and Centre For Sustainable Development Initiatives 2012; Adam Smith International 2014) and because sunflower contract farming is practised in this region (RLDC 2008). A cross-sectional data set was collected between November 2015 and January 2016.

We applied a three-stage sampling technique. First, we purposely selected two neighbouring districts, Iramba and Mkalama, because there is more sunflower production in these two districts than in other districts. Second, we purposely selected all villages in the two districts, in which sunflower contract farming was available, as the presented study is a part of a research project on contract farming. This resulted in the selection of twelve of the seventy-eight villages in Iramba district and of twelve of the fifty villages in Mkalama district. In each of the twenty-four villages, we obtained lists of farmers, grouped into three strata: sunflower contract farmers, sunflower non-contract farmers, and non-sunflower farmers. Third, we used non-proportional stratified random sampling to select farmers in each stratum. The non-proportional stratified random sampling was used because it allowed us to have a sufficient number of sunflower contract farmers in our sample in spite of only a small proportion of sunflower contract farmers in each village. We randomly selected eight farmers from the list of sunflower contract farmers in villages with more than eight contract farmers and all sunflower contract farmers in villages with eight or less sunflower contract farmers. In each village, we selected nine further farmers from the combined list of the two strata of sunflower non-contract farmers and non-sunflower farmers, where we chose the proportions of these two strata in our sample to be the same as in the population.

This should have resulted in a sample of 404 smallholder farmers, consisting of 188 sunflower contract farmers (46.5%), 186 non-contract sunflower farmers (46%), and 30 non-sunflower farmers (7.5%).⁸ However, 13 of the 404 farmers (3%) refused to take part in the survey. These thirteen farmers were replaced by other farmers from the same stratum if possible, while some other farmers were accidentally interviewed due to unclear information. This resulted in a total sample of 416 farmers.

While our survey covers the 2014/15 growing season, the stratification of farmers was based on the farmers intentions for the 2015/16 growing season, because our research aimed at obtaining a two-year panel data set including both the 2014/15 growing season and the 2015/16 growing season. However, due to flooding and a drought in the 2015/16 growing season, many sunflower fields could not be harvested so that the data collection for the 2015/16 growing season was dropped. In the 2014/15 growing that is covered in our data set, 8 of 416 farmers in our sample were sunflower contract farmers (2%), 383 were non-contract sunflower farmers (92%), and 25 were non-sunflower farmers (6%).

The interviews in our farm household survey were done by the main author of this paper together with six trained enumerators. Data collection was done through face-to-face interviews with the selected smallholder farmers—usually with the heads of each household—using a structured questionnaire. The questionnaire used in the survey underwent a series of reviews by experts in the field and was then pre-tested with purposely selected smallholder farmers in order to assess the relevance and clarity of the questions and thus to improve this survey instrument (Sekaran and Bougie 2016). The survey collected data on socio-economic factors of the household, agricultural production, and institutional factors.

Two questions regarding the use of observations in our empirical analysis arise: (i) whether non-sunflower farmers should be included in the analysis or excluded from the analysis and (ii) whether only initially selected farmers should be included in the analysis or also farmers selected for replacement and accidentally interviewed farmers should be included in the analysis. While previous studies of adoption of improved crop varieties only included farmers who actually cultivated the respective crop, one could argue in our case that also farmers who did not cultivate sunflower in the 2014/15 growing season should be included because also these farmers had the opportunity to buy and sow seeds of improved sunflower varieties in this growing season (and indeed many farmers change their decision to cultivate sunflower or not to cultivate sunflower from year to year). Although there is no obvious reason that including farmers who were not initially selected for the survey could somehow bias our sample, one could argue that these observations should be excluded as a precautionary measure. However, including all observations and, thus, using the largest possible number of observations increases the statistical power of the empirical analysis and the precision of our estimates. In order to assess the robustness of our results regarding the choice of observations for the empirical analysis, we present results obtained from four different samples: (i) with all observations included, (ii) without observations from non-sunflower farmers, (iii) including initially selected farmers only, and (iv) including initially selected farmers only and excluding non-sunflower farmers.

3.2 Categorization of sunflower varieties

The collected data include information on the use of four different types of seeds for sunflower production, i.e. certified seeds of hybrid varieties, certified seeds of OPV, QDS of OPV, and recycled seeds. The majority of smallholder farmers in our data set (88.5%) used only recycled sunflower seeds (Table 1). Given the very low proportions of farmers who use certified seeds of hybrid varieties, certified seeds of OPV and QDS of OPV, we subsume these three categories under the category of ‘improved seeds’. Hence, our empirical analysis distinguishes two types of varieties: ‘improved varieties’ and ‘non-improved varieties’.⁹

3.3 Econometric specification

Most empirical studies that analyze the adoption of technologies estimate probit or logit models, while Tobit-type models are also applied if the analyzed technology is divisible (Shiferaw et al. 2015). As an improved crop variety is a divisible technology, we could analyze its adoption with a Tobit model. However, Tobit models assume that explanatory variables affect the binary adoption/non-adoption decision in exactly the same way as they affect the decision on the extent of the adoption. As it is questionable whether this very restrictive assumption is fulfilled in our empirical application, we apply a double-hurdle model (DHM) as suggested by Cragg (1971).¹⁰

Cragg (1971) developed several different specifications of DHMs that can be used for analyzing censored dependent variables that are the outcomes of sequential decision processes. We apply the DHM that is specified in Equations (7) and (9) in Cragg (1971) because this specification of the DHM is in line with our microeconomic model and corresponds well with farmers’ real-world adoption decisions in our empirical study. In the context of the adoption of improved sunflower varieties, this specification assumes that the farmers first decide to adopt (or not to adopt) improved sunflower varieties and in case of deciding to adopt, afterwards decide on the extent of adoption.

The DHM defined in Equations (4)–(7) is estimated by the maximum likelihood method based on the assumption that the error terms in Equations (4) and (7) are independent i.e. COV(u_i, ϵ_i) = 0.

Unlike the Tobit model, the DHM allows for two different sets of explanatory variables (i.e. Z_i and X_i) and two different sets of parameters (i.e. θ and β) for the two stages of the decision process. As such, the DHM relaxes some of the restrictive assumptions of the Tobit model and thus can provide consistent parameter estimates even if the restrictive assumptions of the Tobit model are not fulfilled (Cragg, 1971; Shiferaw et al. 2008). In addition to the DHM, we conduct our analysis with the Tobit model in order to compare the results and as a robustness check.¹¹

In many cases, it is reasonable to assume that the same variables affect both stages of the decision process, i.e. Z_i = X_i, but to allow for different effects of the explanatory variables on the two stages of the decision process, i.e. θ ≠ β. If the explanatory variables have the same effects on both stages of the decision process, i.e. θ = β, the specification of the DHM that we use in our analysis collapses to the Tobit model (Cragg 1971). Hence, we can easily test the Tobit model against the DHM in our empirical application (Cragg 1971, p. 831).

Given our sampling strategy, the observations in our data set are ‘clustered’ in twenty-four villages (Abadie et al. 2017). To account for the ‘clustered’ observations, we calculate and present cluster-robust standard errors of the estimated coefficients using the sandwich method and the ‘HC1’/ ‘CR1’ specification in the terminology introduced by MacKinnon and White (1985) and extended to cluster-robust standard errors by Cameron and Miller (2015).¹²

As coefficients of DHMs and Tobit models do not have meaningful real-world interpretations, we calculate marginal effects of the explanatory variables for each hurdle of the DHM as well as for the Tobit model (e.g. Drichoutis 2011; Wooldridge 2019). These calculations are done at the mean values of all explanatory variables. The marginal effects of binary explanatory variables are calculated as the effect of a discrete change from zero to one.¹³ We calculate the approximate standard errors of marginal effects with the Delta method based on the cluster-robust variance-covariance matrices of the estimated coefficients.

3.4 Empirical specification

In our empirical application, the dummy variable D_i indicates whether a household has adopted improved seeds, i.e. whether the household cultivated at least a part of its sunflower area with an improved variety by sowing certified seeds of a hybrid variety, certified seeds of an OPV or QDS of an OPV (see Table 1).

The extent of adoption of improved crop varieties can be operationalized in different ways. One option could be to measure the extent of the adoption as the proportion of improved sunflower seeds in the total sunflower seeds sowed. However, forty four of the forty six adopters in our data set use 100% improved sunflower seeds. Hence, we cannot use this variable to measure the extent of adoption because this variable has too little variation.

Other measures of the extent of adoption could be the land area cultivated with improved sunflower varieties, the proportion of this land area in the total land area cultivated with sunflower or the proportion of this land area in the total land area cultivated by the household (with any crop). However, land areas cultivated with sunflower are often difficult to measure because farmers frequently intercrop sunflower with cassava, sorghum, cow peas, or other crops so that the actual land area used for sunflowers cultivation is often unknown.

We initially also used the quantity of improved sunflower seeds as measure of the extent of adoption but this variable has a very right-skewed distribution, which makes the econometric analysis vulnerable to outliers and heteroscedasticity.¹⁴ Furthermore, this variable is almost entirely driven by the farm size (e.g., measured as total land area cultivated), indicating that it could be beneficial to normalize this variable in order to obtain a measure of the extent of adoption that is more comparable across farm sizes.

Finally, we decided to measure the extent of adopting improved sunflower varieties, i.e. variable |$\widetilde{A}_i$|⁠, as the quantity of seeds of improved sunflower varieties (in kg) divided by the total land area that the household cultivates (in acres; including all crops). This is a proxy variable for the proportion of the household’s agricultural land area that is cultivated with improved sunflower varieties, assuming a roughly constant seed rate (i.e. amount of seed sown per land area). This measure of the extent of adoption has a sufficient variation between farms, avoids difficulties with measuring areas cultivated with sunflowers due to intercropping and is comparable across different farm sizes.

Tables 2a, 2b, and 2c define and explain the two dependent variables as well all explanatory variables that we use in our empirical analysis. According to our microeconomic model, the explanatory variables include variables that indicate exogenous conditions z, preferences w, and provision of information M that could be related to the adoption of improved sunflower varieties, i.e. Z_i = X_i = (z, w, M) (see Equation 3).

The exogenous conditions z in our microeconomic model include the household size, off-farm income, total cultivated area, the district in which the household is located, market constraints regarding the availability of seeds of improved sunflower varieties, and liquidity constraints regarding the purchase of seeds of improved sunflower varieties.¹⁵

The preferences of the household w are operationalized as risk aversion as well as by including age, sex, and household size as proxy variables because these household characteristics are usually correlated with preferences.¹⁶ We measure risk aversion as self-reported risk attitudes on a Likert scale, where the respondents were asked to choose one of five statements that best describes their behaviour regarding decisions under risk. These five statements were: (1) ‘I always take chances,’ (2) ‘I often take chances,’ (3) ‘I sometimes take chances,’ (4) ‘I rarely take chances,’ and (5) ‘I never take chances.’ As only very few respondents chose statements (3), (4), and (5), we subsume these three levels of risk aversion under one level of risk aversion. Thus, the variable ‘risk aversion’ used in our empirical analysis has three categories: ‘not risk averse’ (1), ‘slightly risk averse’ (2), and ‘risk averse’ (3, 4, 5).¹⁷

Finally, we include ownership of radios, ownership of mobile phones, extension service, membership in farmers’ groups, and household size as variables that are likely related to the provision of information M. We selected the explanatory variables not only based on our theoretical microeconomic model but also based on the scientific literature (see Tables 2a, 2b, and 2c).

While numerous empirical studies found that adoption of improved crop varieties is related to various socioeconomic and institutional factors, these studies have largely ignored the role of the availability of seeds of improved varieties on the market and farmers’ liquidity constraints in the adoption decisions. These research gaps are addressed in our study.¹⁸

Several explanatory variables such as ownership of radios, ownership of mobile phones, off-farm income, and membership in farmers’ groups are not exogenously given but are endogenous decision variables. Hence, we cannot exclude that these variables are correlated with unobserved factors that affect the adoption of improved sunflower varieties (known as unobserved heterogeneity) or are even affected by reverse causality. Therefore, we cannot interpret the estimated relationships between the explanatory variables and the adoption of improved sunflower varieties as causal effects but we interpret them as associations.

4. Results and discussion

4.1 Descriptive results

Table 3 presents descriptive statistics of the variables that are included in the analysis, for the entire sample as well as separately for adopters and non-adopters of improved sunflower varieties. Adopters and non-adopters differ significantly in some of the variables. For example, on average, adopters have more household members and are less risk averse than non-adopters. Furthermore, adopters are on average more likely to own a radio, to receive government extension service, to be a member of a farmers’ group and to have off-farm income than non-adopters. Finally, adopters tend to perceive liquidity constraints for buying sunflower seeds to be less stringent than non-adopters.

4.2 Results of the double hurdle model

The estimated coefficients of the DHM and the corresponding marginal effects of the explanatory variables are summarized in Table 4, while detailed estimation results are presented in Tables A5–A12 in the Appendix.¹⁹

The upper panel of Table 4 presents the estimated coefficients and marginal effects of the explanatory variables on the probability to cultivate improved sunflower varieties, while the lower panel of this table presents the estimated coefficients and marginal effects on the extent of the cultivation of improved sunflower varieties. The results indicate that the decision about cultivating or not cultivating improved sunflower varieties is partly related to different factors than the decision about the extent of cultivating improved sunflower varieties. This is confirmed by a likelihood ratio test that rejects the Tobit model in favour of the DHM at 10% significance level when including all observations, and at 5% significance level when using each of the three subsets of observations (see Table A4 in the Appendix). Thus, the DHM appears to be more suitable for our analysis than a Tobit model.

Table 4 indicates that the estimation results are generally very robust to changing the sample that is used in the estimation. While excluding observations from farmers who were not initially selected for the survey has some minor effects on the results, excluding farmers that did not cultivate sunflowers in the 2014/15 growing season has only negligible effects on the estimation results.

We did not find any statistically significant relationship between the age and the sex of the household head and the adoption of improved sunflower varieties, neither for the binary adoption decision nor for the extent of adoption.

Our results indicate that slight risk aversion is not a substantial barrier to the adoption of improved sunflower varieties, while more risk-averse farmers are substantially less likely to adopt improved sunflower varieties (by 6–7 percentage points) than not risk-averse farmers. However, both slightly risk-averse farmers and more risk-averse farmers adopt improved sunflower varieties to a lesser extent than not risk-averse farmers (by around 76% and 61%, respectively, for an average adopter).²⁰ This concurs with our theoretical expectation that risk aversion is negatively related to the adoption of new technologies and confirms earlier studies (e.g. Liu 2013; Barham et al. 2014; Ward and Singh 2015; Abdoulaye et al. 2018; Bridle et al. 2019; Magnan et al. 2020).

We find that the probability of adopting improved sunflower varieties is significantly positively related to household size, where one additional household member is associated with an increase of the probability of adopting improved sunflower varieties by around one percentage point. In contrast, the extent of adopting improved sunflower varieties are found to have no statistically significant relationship with household size. Our results regarding the binary adoption decision confirm our theoretical expectation given that larger households have more members who potentially get information about the availability of improved sunflower varieties. This result also confirms the findings of earlier studies (e.g. Yu et al. 2011; Khonje et al. 2015).

In contrast to some previous studies (e.g. Verkaart et al. 2017; Armel Nonvide 2020), our empirical analysis does not find a robust statistically significant association between off-farm income and the adoption of improved sunflower varieties, neither the binary adoption decision nor the extent of adoption.

We do not find a statistically significant relationship between farm size (measured as the total land area cultivated) and the binary adoption decision but we find that an increase of the farm size by one percent is statistically significantly associated with a decrease of the proportion of the land area used for cultivating improved sunflower varieties by 0.43–0.48 percentage points (for an average adopter). Given that this relative marginal effect is smaller than one (in absolute terms), larger farms tend to cultivate a smaller proportion of their land but a larger (absolute) land area with improved sunflower varieties than smaller farms.

Our study does not find a robust statistically significant association between ownership of radios and the probability of adopting improved sunflower varieties, while radio ownership is clearly associated with an increase of the extent of the adoption (by 41–57% for an average adopter). These findings confirm earlier studies (e.g. Simtowe et al. 2010) and are consistent with our theoretical expectation given that radios are an important channel for farmers for getting information about agricultural production. Indeed, farmers were informed about the availability of improved sunflower varieties through a rural radio programme called ‘INUKA’ (Salisali 2012). In contrast, we did not find a statistically significant relationship between ownership of mobile phones and the adoption of improved sunflower varieties, neither for the binary adoption decision nor for the extent of adoption.

We find that receiving government extension service is associated with an increase of the probability of adopting improved sunflower varieties by 18–21 percentage points, while we do not find a statistically significant association to the extent of adoption. Our finding regarding the binary adoption decision confirms earlier studies (e.g. Khonje et al. 2015; Seymour et al. 2016; Yigezu et al. 2018; Armel Nonvide 2020) and suggests that extension service is an important source of information about new technologies. Indeed, there are hardly any other organizations or companies that give advice and training to the farmers (e.g. about new technologies) in our study area as even private entities such as processors of sunflower seeds expect the extension service to be provided through the government extension officers (Kuzilwa and Mpeta 2017). As such, extension service is an institutional arrangement for solving the predominantly rural market imperfection in the provision of information.

We find a positive and statistically significant relationship between membership in a farmers’ group and the binary adoption decision in some of the regression analyses (12–17 percentage points) but we do not find a statistically significant relationship between membership in a farmers’ group and the extent of adoption of improved sunflower varieties. Our finding regarding the binary adoption decision confirms earlier studies (e.g. Khonje et al. 2015).

Contrary to our theoretical expectations, we do not find a robust statistically significant relationship between market constraints and the binary adoption decision and we find that farmers who perceive that they are somewhat market constrained adopt improved sunflower varieties to a larger extent than farmers who responded that they are not or only slightly constrained regarding the availability of seeds for improved sunflower varieties on the market. Our results suggest that limited availability of seeds of improved sunflower varieties on the market is not a substantial barrier to the adoption of improved sunflower varieties. Indeed, improved sunflower seeds can not only be purchased on the market but also obtained by farmers (on credit or for cash) through contract farming arrangements (e.g. Ministry of Industry, Trade and Investment 2016; Mpeta et al. 2017). Our results regarding the relationship between market constraints and the extent of the adoption can potentially be explained by reverse causality: the larger the proportion of their land area sown with improved sunflower seeds, the more likely that farmers experience that the availability of seeds of improved sunflower varieties on the market is at least somewhat constrained.

In line with our theoretical expectations, we find that somewhat and severely liquidity-constrained farmers have a lower probability of adopting improved sunflower varieties (around 5–10 percentage points) than not or only slightly liquidity-constrained farmers but this relationship is only statistically significant in some of our regressions analyses. The relationship between liquidity constraints and the extent of adoption is statistically insignificant for somewhat liquidity constrained farmers but statistically significant for severely liquidity constrained farmers: farmers who indicate to be severely liquidity constrained adopt improved sunflower varieties to a 46–72% smaller extent than farmers who indicate to be not or only slightly liquidity constrained. This result indicates that liquidity constraints are a much larger barrier to the adoption of improved sunflower varieties than market constraints.

4.3 Tobit model

In addition to the DHM, we conduct our analysis with the Tobit model in order to compare the results and as a robustness check. Tables A13–A20 in the Appendix present the estimated coefficients and marginal effects of Tobit regressions with the entire sample as well as with the three subsets of observations descried above. The results of the Tobit models are generally in line with the results of the DHM, particularly with the results regarding the binary adoption decision. However, as the Tobit model does not distinguish between the two different dimensions of adoption, it does not find all the relationships that we find with the DHM. For instance, the DHM indicates that the extent of adoption is statistically significantly related to farmers identifying themselves as slightly risk averse, to the total cultivated land area and to experiencing to be somewhat market constrained but the Tobit models do not find these relationships.

These differences support the results of the likelihood ratio tests presented in Table A4 in the Appendix that the DHM is more suitable for our empirical analysis than the Tobit model.

5. Conclusion and policy implications

Although adoption of improved crop varieties by smallholder farmers in developing countries is considered to be a promising way of increasing food production and household welfare, the adoption rate of improved crop varieties by smallholder farmers is still very low in most developing countries. Increasing the adoption of improved crop varieties through policies, development programmes, and business decisions requires knowledge about drivers of adoption and barriers to adoption. In order to provide this information, this paper analyzed the adoption of improved sunflower varieties amongst smallholder sunflower farmers in Singida region, Tanzania. We used a double-hurdle model (DHM) for our empirical analysis, which allows us to separately analyze the (binary) decision of adopting or not adopting improved sunflower varieties as well as the extent of adoption. We found several factors that are significantly related to the binary adoption decision, to the extent of adoption, or to both of them.

The main limitation of our study is that we cannot interpret our estimated coefficients as causal effects because we cannot exclude that some of our results are driven by reverse causality or unobserved heterogeneity (omitted confounding factors). For instance, the positive and highly statistically significant relation that we found between receiving extension service and adoption of improved sunflower variety could—at least partly— also originate from reverse causality (e.g. if extension officers find it more interesting to visit farmers who cultivate improved sunflower varieties than farmers who only cultivate non-improved sunflower varieties or if farmers who cultivate improved sunflower varieties try harder to get in contact with an extension officer than farmers who only cultivate non-improved sunflower varieties) or from unobserved heterogeneity (e.g. if extension officers prefer to visit farmers who they know are more open to new technologies or if farmers who are more open to new technologies try harder to get in contact with extension officers than farmers who are less open to new technologies). In order to separate causal effects from other sources of correlation, we suggest that our most important policy-relevant findings be validated with randomized controlled trials (RCT).

In spite of this limitation, we can derive some recommendations for policies, development programmes, and business decisions. For instance, given that we found that adoption of improved sunflower varieties (including the extent of adoption) is negatively related to risk aversion, we suggest an investigation, e.g. with small pilot projects, of whether weather (index) insurances can alleviate this barrier to the adoption of improved crop varieties (see e.g. Karlan et al. 2014).

Our study also indicates that adoption of improved sunflower varieties is positively related to various ways of information provision such as ownership of radios and receiving extension service. Hence, we suggest an investigation of whether the benefits of providing and expanding these ways of providing information outweigh their costs.

While it seems that the availability of seeds of improved sunflower varieties on the market is not a major barrier, we found that liquidity constraints are significantly negatively related to the extent of adoption of improved sunflower varieties. Thus, we suggest that ways of alleviating liquidity constraints, e.g. loans with reasonable interest rates through Saving and Credit Cooperatives (SACCOs) and Village Community Banks (VICOBA), be investigated.

In addition to the suggestions for further investigations given above, future research could use longitudinal data in order to investigate long-term relationships between various factors, drivers and barriers, and the adoption of improved sunflower varieties. Furthermore, given the limited external validity of our study and other studies of adoption of new technologies, a study based on multiple-country data about the adoption of improved varieties of various crops could provide results with much higher external validity.

Finally, as our data set includes a too small number of contract farmers for analyzing the relationship between contract farming and the adoption of improved sunflower varieties, we suggest to study this relationship with data sets that are more suitable for analyzing this research question.

Acknowledgments

The authors would like to thank two anonymous reviewers and the associate editor Alan de Brauw for their valuable comments and suggestions on earlier drafts of this article. Of course, the authors take full responsibility for any remaining errors.

Funding

This research was conducted within the research project ‘Productivity, Market Access, and Incomes for Small Farming Businesses in Tanzania: Potentials and Limitations in Contract Farming’ (POLICOFA), which was funded by the Ministry of Foreign Affairs of Denmark and administered by Danida Fellowship Centre (grant number: 13-P03-TAN).

Author contributions

F.Y.T. and M.A.M. developed the idea for the empirical analysis and for the paper in general, F.Y.T. and M.A.M. reviewed the literature, A.H. developed the theoretical microeconomic model, F.Y.T., A.H., and other project participants developed the questionnaire, F.Y.T. collected the data, F.Y.T. and A.H. conducted the empirical analysis, A.H. programmed the calculation of the marginal effects and their standard errors, and F.Y.T., A.H., and M.A.M. wrote the manuscript.

Competing interests

All authors declare that they have no competing interest.

Footnotes

References