Analyzing the Determinants and Extent of Crop Diversification among Smallholder Coffee Farmers in Kirinyaga Central and East Sub-Counties, Kirinyaga County, Kenya

Theorization of crop diversification in the context of smallholder coffee farmers was guided by utility maximization theory. A household's decision to diversify was considered a major economic decision which had a strong bearing on the livelihoods of farmers. Utility maximization theory was used in this study because farmers base their decision to diversify crops on the agricultural household (Taylor and Adelman, 2003). Crop diversification is said to involve choices on production strategy that increase resilience and improves economic benefits. Using the Random Utility Model (RUM), the choice of smallholder farmer for crop diversification strategies can be envisioned using producer and consumer theory (Singh et al., 1986). This is due to the fact that it is excellent at modeling discrete choice decisions. Subject to uncertainty brought on by climate change, consumption costs and output prices, a farmer's production decisions, including optimal allocation, are thus dependent on their attitude towards risk and output produced, which maximizes farm household utility. Farmers in developing countries tend to be risk-averse and crop diversification may be a better strategy for them to insure against production risks (Hitayezu et al., 2016).

U_ij represents the utility farmer i receives for participating in the j^th crop diversification process, j assumes the value of 0 and 1 for no diversification and complete diversification, respectively. Thus, the utility derived from crop diversification by the i^th farmer is subject to X, which is a vector of farm and farmer attributes of the diversifier and a vector of factors related to crop diversification. Therefore, the relation between the utility derivarable from the j^th diversification process is hypothesized to be a function of the vector of observed farm, farmer and crop diversification specific characteristics with zero mean random error as expressed in Equation 1 and Equation 2.

(1) {Uij=f(Xijβi)+εij} j=1; and i=1,2,3…, n \{{U_{i{\rm{j}}}} = f({X_{i{\rm{j}}}}{\beta _i}) + {\varepsilon _{i{\rm{j}}}}\} \,j = 1;\,{\rm{and}}\,{\rm{i}} = 1,2,3 \ldots ,\,{\rm{n}}

(2) {Uij=f(Xijβi)+εij} j=0; and i=1,2,3…, n \{{U_{i{\rm{j}}}} = f({X_{i{\rm{j}}}}{\beta _i}) + {\varepsilon _{i{\rm{j}}}}\} \,j = 0;\,{\rm{and}}\,{\rm{i}} = 1,2,3 \ldots ,\,{\rm{n}}

Farmer i will choose to participate in crop diversification only if the utility derived from participating is greater than the utility from not participating. That is, a smallholder farmer adopts crop diversification if (U_i1 > U_i0). The only thing that can be observed is characteristics of household and attributes of the alternatives as faced by the decision-maker, since utilities are not observable. The following latent structure model for participation in crop diversification can express the utilities. (3) Zi*=γXi+εi Z_i^* = \gamma {X_i} + {\varepsilon _i} where:

• Zi* Z_i^* is a binary variable that has a value of 1 for engagement in crop diversification and commercialization and a value of 0 for non-participation. Explanatory variables chosen based on the literature are X_i.

The study was conducted in Kirinyaga East and Central Sub-Counties located in Kenya's Kirinyaga County, which is bordered to the north and east by Nyeri County, to the west by Murang'a County, and to the east and south by Embu County. It is located between Longitude 37°10′0″ E and 37°30′0″ E and latitudes 0°10′0″ S and 0°40′0″ S. The county, which has three agro-ecological zones (lowlands, midlands, and highlands) and is situated between 1,158 meters and 5,380 meters above sea level, is situated on the south-eastern slopes of Mount Kenya. There is a bimodal pattern in rainfall, with heavy rains (averages 2,146 mm) from March to May and short rains (averages 1,212 mm) from October to November. There is also variation in food production and consumption. During the hot season, the mean temperature varies from 18.4°C in the upper zones to 30.3°C in the lower zones. The majority of people live in rural areas, where agriculture is the main source of income (County Government of Kirinyaga, 2018).

Kirinyaga Central Sub-County and Kirinyaga East Sub-County hold significant importance in comprehending the consequences of crop diversification adopted by smallholder coffee farmers. They are characterized by high population density, a high coffee production level, favorable agricultural potential, several agro-ecological zones; the majority of farmers are small-scale, owning less than two hectares of land and also with high number of agricultural markets (Jaetzold et al., 2006). The kinds of crops planted are influenced by the various agro-ecological zones. The region is home to a variety of annual and perennial crops, including sweet potatoes (Ipomea batatas L.), vegetables, fruits, coffee (Coffee spp.), tea (Camellia sinensis L.), maize (Zea mays), beans (Phaseolus vulgaris L.), bananas (Musa spp.), and banana plants. These two sub-counties were chosen because of their capacity to produce a wide variety of food crops and their comparatively prominent involvement in agricultural and commercial operations (Fig. 1) (Jaetzold et al., 2006].

Fig. 1.

Thematic map of Kirinyaga Central Sub-County and Kirinyaga East Sub-County

Source: GeoCurrents (2015).

The study used descriptive survey design. It focused more on making specific predictions, narrating facts and features about people, groups or circumstances. This study design was appropriate, since it was of help in examining, characterizing and explaining the factors influencing smallholder coffee farmers on making decisions to diversify their crops and also to what extent.

The target population was 18420 smallholder coffee farmers in Kirinyaga Central (6777) and Kirinyaga East (11643) Sub-Counties (County Government of Kirinyaga, 2019), where ecological zones. The study employed the Cochran (2007) formula to calculate the sample size, which was estimated to be 408 households, and were chosen based on the population size of each strata selected in order to produce a sample for each agro-ecological zone. According to Minai et al. (2014), approximately 25% of the coffee farmers are in UM₁, 50% in UM₂ and about 25% in UM₃. (4) n1=n⋅p1 {n_1} = n \cdot {p_1} where:

• p₁ – represents the proportionate of population included in stratum 1, n – denotes the size of the entire sample and n·p₁ – denotes the number of smallholder coffee farmers chosen from stratum 1, UM₁ – is the coffee-tea zone, UM₂ – is the main coffee zone and UM₃ – is the marginal coffee zone. As a result, population size and sample size adopted a proportional allocation for each stratum in region of the study (Table 1). The study used a sample size of 408 smallholder farmers drawn from a population of 18420, which is divided into two Sub-Counties and three strata.

A multistage sampling technique was used, where in the first stage, Kirinyaga central and Kirinyaga East Sub-Counties were purposively selected due to their prominence as significant coffee-growing regions in Kirinyaga County. In the second stage, three agro-ecological zones suitable for coffee production were chosen through stratification, and they included the coffee - tea zone (Upper Midland one – ), the main coffee zone (Upper Midland two – ), and the marginal coffee zone (Upper Midland three –. In the third stage, 12 sub-locations were randomly chosen from each of the 12 locations in the three AEZs. Next, smallholder coffee farmers were randomly selected from the 12 sub-locations. Finally, the sample size needed for each stratum was used to determine how many smallholder coffee farmer households were chosen.

Before actual data collection, a pilot study was conducted in Runyenjes Sub-County, Embu County, where 38 questionnaires (10% of the sample size) were pre-tested before the actual survey among smallholder coffee producers. Face-, content- and construct validity were examined, and piloted data was also used to assess the instrument's reliability through use of the Cronbach's coefficient alpha. The questionnaire items were found to have a reliability value of 0.820, which is above 0.7 (acceptable level). Data from smallholder coffee producers were gathered using structured questionnaires administered on Android phones and tablets using the free open-source program Kobo Toolbox (https://www.kobotoolbox.org/). The structured questionnaire was used to capture socio-economic and demographic information of smallholder coffee farmers, which was used to capture individual-level heterogeneity such as gender, age, education level, extension services, land size, distance to market, cooperative membership and credit access. The survey took between 30 and 50 minutes to complete. The survey obtained information on the factors influencing farmers to diversify and extent of diversification.

In this study, STATA version 15 was employed to analyze data where descriptive statistics such as mean, standard deviations, percentages and frequencies were used to address the first objective, which profiled the socio-economics characteristics of smallholder coffee farmers. Inferential statistics were obtained from Cragg's double-hurdle model to examine the determinants of household decisions and the extent of crop diversification among smallholder coffee farmers in Kirinyaga Central and Kirinyaga East Sub-Counties.

Empirical model specification

The Herfindahl Index (HI) was employed to measure crop diversification, since it is one of the most generally employed indexes in the crop diversification literature (Asante et al., 2017; Mengistu et al., 2021 and Appiah-Twumasi and Asale, 2022). It is a measure of how evenly a farmer spreads their land and resources across different crops. The crop diversification index (CDI) is a concentration index. As a result, CDI was used to assess crop diversification. The formula devised by Hirschman (1964) in Equations 5 and 6 was used to compute HI, which is the sum of squares of all n proportions: (5) Pi=Ai∑i=1nAi {P_i} = \matrix{{{A_i}}\cr{\sum\nolimits_{i = 1}^{\rm{n}} {{A_i}}}\cr} where:

• P_i = proportion of i^th crop, A_i = Area under i^th crop (ha), Σⁿ_i=1A_i is the total cropped land (ha), and 1 is the number of crop i.e. 1, 2, 3, …, n

(6) Herfindahl Index−HI=∑i=1nPi2 {\rm{Herfindahl}}\,{\rm{Index}} - {\rm{HI}} = \sum\nolimits_{i = 1}^{\rm{n}} {P_i^2}

Subtracting the HI from 1 and 0 yielded the CDI values (equation 7). Furthermore, a crop diversification index value of 0 implies perfect specialization, while a trend towards 1 indicates a rise in the extent of agricultural diversification (Gniza and Loa, 2023).

(7) Crop diversification Index−CDI=1−HI {\rm{Crop}}\,{\rm{diversification}}\,{\rm{Index}} - {\rm{CDI}} = 1 - {\rm{HI}}

In general, the CDI value rises when diversification increases and falls to zero when producers plant and nurture a sole crop. The producers in this research study mostly grow maize, beans, cabbage, kale, bananas, avocado, cowpeas, sweet potatoes, potatoes, tomatoes, French beans, arrow roots, cassava, and pumpkin. To calculate the Herfindahl index, total cropped land (ha) for the diversifiers and the ratio of land allotted for producing each crop per ha in the harvest season of 2021/2022 were used.

Cragg's Double-Hurdle Model

Model specification

To determine the choice and level of crop diversification, Cragg's double-hurdle model was employed for this study because two decisions were taken at several stages, and similar factors had varying effects on the two decisions (Asfaw et al., 2022). According to Tura et al. (2016), the Heckman, Tobit and double-hurdle models are the most common approaches for modeling the choice and extent of crop diversification circumstances. The double-hurdle procedure is a more flexible parametric generalization of the Tobit and Heckman models that enables two distinct stochastic processes to influence one's choice to participate and the crop diversification level. The 1^st stage (probit) and the 2^nd stage (truncated regression) models must be applied jointly, either sequentially or concurrently to provide asymptotically efficient and consistent estimates for all of its parameters (Derso et al., 2022). Following Gniza and Lao (2023), the theoretical foundation for the Cragg's (1971) double-hurdle estimating framework is the probit model, in which the likelihood of crop diversification at observation t, denoted by P(E_t), is given by: (8) P(Et)=∫−∞X,β(2π)12exp{−z2/2}dz P({E_t}) = \int_{- \infty}^{X,\beta} {{{(2\pi)}^{{1 \over 2}}}\exp \{- {z^2}/2\} {\rm{dz}}} where:

• X_t – is a vector K x 1 of variables exogenous to observation t and β represents a vector of parameter estimates. C(z) is the cumulative unit normal distribution as shown in Equation 9.

(9) C(z)=∫−∞z(2π)12exp{−t2/2}dt C(z) = \int_{- \infty}^z {{{(2\pi)}^{{1 \over 2}}}\exp \{- {t^2}/2\} {\rm{dt}}}

The first stage of the probit model is used to estimate the probability of a producer to diversify crops or not. The second stage determines how much diversification should be applied to land acreage. Only non-zero values in the first stage can be employed to estimate the level of crop diversification in stage two by use of the truncated regression model. The model presupposes that at least one explanatory variable from the first equation must be absent from the second stage of identification because it expects the inclusion of one or more variables in the selection rather than the output equation. This means that the choice and extent of crop diversification were not determined by a closely related set of explained variables. In this study, group membership was the only identifier variable that influenced just the first stage (probability of diversification) but not the second stage (extent of diversification) of the selected diversification strategies. A double-hurdle model was therefore adopted for this study, which is specified in Equation 10 as: (10) CDi*=Zi*α+εi (Diversification decision)QCD**=Xi′β+μi (Intensity of diversification)n(εiμi)~N[(00)(101∂2)] \matrix{{{CD}_i^* = Z_i^*\alpha+ {\varepsilon _i}\,({\rm{Diversification}}\,{\rm{decision}})}\cr{{Q^{CD**}} = X_i^{'}\beta+ {\mu _i}\,({\rm{Intensity}}\,{\rm{of}}\,{\rm{diversification}})}\cr{{\rm{n}}\left({\matrix{{{\varepsilon _i}}\cr{{\mu _i}}\cr}} \right)\sim {\rm{N}}\left[ {\left({\matrix{0\cr0\cr}} \right)\left({\matrix{1 & 0\cr1 & {{\partial ^2}}\cr}} \right)} \right]}\cr} where:

• CD_i^* is latent variable that denotes binary censoring or the choice in our study

Taking value of 1 for diversity of crops and 0 suggests the opposite. Q^CD** is the latent variable that represents the number of diverse crops. Zi* Z_i^* and X_i' are explanatory variables vectors, and β are parameter estimates and, ɛ_i and μ_i are the error terms for the decision and level of crop diversification, respectively. The model assured that the error terms were normally and independently distributed, since each farmer decides to participate in both crop diversification decisions independently. The producer is faced with two decisions, where the first decision to diversify crops or not is handled by a probit model as shown in Equation 11.

(11) CDi*=1 if CDi*>0, CDi*=0 if CDi*≤0 or CDi*=={1;CDi*>00;otherwise \matrix{{{CD}_i^* = 1\,if\,{CD}_i^* > 0,\,{CD}_i^* = 0\,if\,CD_{\rm{i}}^* \le 0\,{\rm{or}}\,{CD}_i^* =}\cr{= \left\{{\matrix{{1;} \hfill & {{CD}_i^* > 0} \hfill\cr{0;} \hfill & {{\rm{otherwise}}} \hfill\cr}} \right.}\cr}

The second decision on the extent of diversification of crop is elaborated in Equation 12 as: (12) QCD*=max(Q**, Q, 0) {Q^{CD*}} = \max ({Q^{**}},\,Q,\,0) Q^CD* is normally written as y_i and can be calculated as: (13) QCD*=CDi*QiCD* {Q^{CD*}} = {CD}_i^*Q_i^{CD*}

The double-hurdle log-likelihood function, which nests the probit model and truncated regression is denoted as: (14) Log L=Σ0 ln[1− Φ(zi′α) Φ(Xi′βσ)]+Σ+ ln[Φ(zi′α)σ1∅(yi−Xi′βσ)] \matrix{{{\rm{Log}}\,L} \hfill & {= {\Sigma _0}\,\ln [1 - {\,^\Phi}(z_i^{'}\alpha){\,^\Phi}({{X_i^{'}\beta} \over \sigma})]} \hfill\cr{} \hfill & {+ {\Sigma _ +}\,\ln {[^\Phi}(z_i^{'}\alpha)_\sigma ^1\emptyset \left({{{{y_i} - X_i^{'}\beta} \over \sigma}} \right)]} \hfill\cr} where:

• ^Φ is the standard normal probability function and ∅ as the density function. Z_i and X_i represent probit and truncated model explanatory variables respectively. α, σ, and β were parameters to be predicted for each model.

The findings indicates that households head in the study area had an average age of 48.6 years, with a standard deviation of 13.31 (Table 3). This shows that the vast majority of the survey participants were within the active labor force, had extensive farming experience and were risk-averse. It is probable that older farmers have greater access to production resources and information may impact their decision to diversify. This is in line with the findings of Dembele et al. (2018), who stated that producers in Mali are likely to adopt crop diversification as they age. Regarding the education level, household heads had completed 10.8 years in school, which was the minimum required educational level (Table 3). This may have implied that farmers were more likely to have more knowledge and skills, resources and also be better informed about agricultural instructions and information provided by extension officers. This would help them to identify new crops that are in demand and adapt more diverse farming practices accordingly. Mulwa and Visser (2020) reported that highly educated heads of household diversify crop farming more compared to their counterparts, as farmers with a higher education level may be more likely to engage in research and experimentation, which would allow them to explore and adopt new crop varieties and farming methods.

The study further revealed that the average membership size of a household was 4 persons, which reflects the human capital availability for agronomic activities (Table 3). It is possible that a larger household may provide more hands to help with farming activities, which can make it easier to manage a diverse range of crops. Appiah-Twumasi and Asale (2022) noted that larger household sizes have more individuals available to contribute to agricultural activities, which allows for the cultivation and management of multiple crops. The average distance between the sampled households and the closest marketplace was 2.4 km, with a range of 0.5 km to 8 km (Table 3). These findings may imply that the majority of the farmers were near the market area, and access to markets may boost farmers' incentives to diversify crops and grow surplus food crop that can be conveniently carried to market, thereby increasing their income. This study's findings are consistent with those of Maru et al. (2022), who noted that proximity to markets provides a way of exchanging and sharing information among small-scale producers and providers of services; thus, producers closer to marketplaces are highly inclined to diversify crops.

The findings showed that the average land size (1.59 ha) of the respondents was less than 2 hectares (Table 3). It is possible that smaller land sizes limit ability to diversify, but through implementation of crop diversification strategies like intercropping, farmers can obtain more yields. Lv et al. (2023) noted that farmers need to implement measures such as crop rotation and intercropping to boost soil fertility, which may result in increased crop yields regardless of land size. Further, during this study it was observed that the mean total cropped land was 0.91 ha per household, which meant that more than half of the total land area was used for growing a variety of food crops (Table 3).

About 68.83% of the households surveyed were headed by men, whereas the remaining 31.17% were headed by women (Table 4). This could imply that most of the households in the area studied are male-dominated and can cultivate various crops on a particular plot of land because men can access factors of production more easily than females. During the study, it was found that 37.5% of the respondents had diversified their crops in less than 10 years (Table 4), which implies that farmers in the research area had less experience in crop diversification, probably due to low years of carrying out the practices. Makate et al. (2023) noted that farmers who had acquired experience from earlier crop diversification had a better probability of intensifying subsequent diversified systems and of adapting to recurring rainfall shocks.

This study's findings showed that 72.07% of the heads of households accessed services provided by extension workers, where the majority of them accessed them more than thrice at 46% (Table 4). This may imply that farmers were capacitated through training by extension officers in good agronomic practices, marketing and business-oriented farming. Shangshon et al. (2023) noted that agricultural training in Bhutan helped farmers adopt production-enhancing technology and implement effective marketing techniques, such as the use of better varieties of crops, intercropping, optimum plant spacing, optimal fertilizer use and collective marketing.

The findings of this study showed that social groupings were a common component of social ties in rural agricultural households in Kirinyaga County, where 95.01% of the respondents were members of them. The findings revealed that 47.24% of the respondents joined producer groups to obtain agricultural information (Table 4). These findings may indicate that farmer groups are an important catalyst for adopting innovation and upgrading farming systems as recommended by extension officers and trainers through efficient information flows. In agreement with the study's observations, Kehinde et al. (2022) revealed that membership of an agricultural organization increased the intensity of soil conservation practices such as mulching among small-scale farmers in Nigeria.

The study finding showed that marketing information was a benefit to the social group at 10.24% (Table 4). This indicated that farmers could access marketing information through other channels such as extension visits, radios and social media. In Uganda, Hill et al. (2021) noted that market information is the main factor inducing market participation, since households know what is demanded, when to make a sale, to whom to sell to, and the right price. It is possible that through marketing information farmers are able to increase their market access due to economies of scale in procuring inputs, bargaining power in making sales and marketing produce collectively.

According to the findings of this study, the majority of the households studied accessed credit (87.28%), while 12.72% did not. These findings may imply that financial literacy is high and farmers access credit for agricultural activities. Olutumise (2023) reported that increased access to credit may incline farmers to devote resources to more expensive but more rewarding production practices in order to enhance smallholder farmers' livelihoods. According to the findings of this survey, more than three-quarters (94.76%) of farmers were crop diversifiers, while the remaining households (5.24%) were non-diversifiers (Table 4). This indicated that the majority of farmers were diversifying their crops to lower the chance of yield loss and boost farm resilience to pests, diseases and adverse weather conditions, while more resources, such as labor and equipment, were necessary to manage the strategy.

The findings of this study indicated that the mean crop diversification index in the study area was 0.39, while 65.09% of the diversified farmers had a CDI that oscillated between 0.1 to 0.4 and 29.92% had a CDI range of 0.5 to 0.8 (Table 5; Fig. 2). This implies that there was a low degree of diversification among small-scale farmers. These findings are comparable to those of Kanyua et al. (2013) and Awiti et al. (2022), who found a CDI of 0.34 and 0.42 in Gatanga (Murang'a County) and Kisumu West Sub-County, respectively. The study's findings are in line with those of Kanyua et al. (2013), who postulated that more than three-quarters (70.08%) of respondents showed an index of 0.4 and below.

Fig. 2.

Crop diversification distribution in the study area

Source: field survey, 2022.

Based on the findings of this study, the crop diversification index household differentiation on small farms was influenced by farm and farmer characteristics, natural conditions, and area allocated to a crop. The findings indicated that households with low crop diversity were distinguished by small land holdings and a scarcity of labor as a result of small household sizes.

During the study period, it is possible that these households had low employment in agriculture and were also distinguished by a low level of adoption of crop diversification strategies such as intercropping, crop rotation, landscape heterogeneity, and crop species and varietal diversity. Kiryluk-Dryjska and Więckowska (2020) found that regions that had developed agricultural structures diversified more compared to areas with structural disadvantages, where diversification was needed more. Further, Kurdyś-Kujawska et al. (2021) noted that in Poland, regions with high crop diversity were characterized by low land and labor productivity caused by a predominance of poor soil fertility and agricultural culture. It was possible that greater diversity in crops was a rational behavior for smallholder farmers because they had to adapt to the improved crop diversification strategies to boost their crop diversity systems for food security and increased productivity.

The second stage of Cragg's double-hurdle model involved the use of truncated regression, which ensured that the coefficient estimates were interpreted in terms of the likelihood between the dependent variable (extent of diversification) and the independent variables. Thus, it was not necessary to generate marginal effects as in the first stage of the model. The key factors affecting the level of crop diversification include gender of household head, access to extension services, land size, education level, and household size (Table 7).