Impact of cooperatives on smallholder dairy farmers’ income in Kenya

Abstract

Apart from generating Gross Domestic Product (GDP), the dairy industry in Kenya contributes to the employment of over a million people along the milk value chain. The sub-sector, however, still experiences low productivity and extensive informal marketing. Farmers have, therefore, formed cooperative societies that have received support in cooling plants and access to dairy services, animal health and artificial insemination. However, participation by members of these cooperative societies is still low, and participation’s impact on income has not been determined. This study sought to assess the impact of the choice of cooperatives on smallholder dairy farmers’ incomes. Data was collected using a semi-structured questionnaire from a sample of 277 Nyamira and Homa Bay Counties farmers. An endogenous switching regression model was used to analyze data. The results indicated that cooperative market participation increased farmers’ incomes by about 10%. The results also revealed that distance to market, milk payment period, number of lactating cows, farm size, and farmer location influenced farmers’ decision to participate in the cooperatives. Furthermore, the level of education, number of lactating cows, farm size, access to off-farm income and access to extension services were significant in determining farmers’ income. This paper suggests policies that increase cooperative market participation for smallholder dairy farmers in Kenya.

Keywords:

• income effect
• endogenous switching regression
• cooperatives participation
• smallholder farmer

1. Introduction

The dairy industry in Kenya is the fastest-growing sub-sector and the largest contributor to the agricultural gross domestic product (GDP). The sub-sector is dominated by smallholder dairy farmers who own 2–15 lactating cows and produce more than 80% of the total milk production in the country (Rademaker et al., 2016).The sub-sector also provides a means of livelihood to about 1.7 million Kenyans (Ministry of Agriculture, Livestock, Fisheries and Irrigation, 2019). Regarding milk consumption, Kenya leads the East African region with the highest per capita consumption, currently at 125 litres annually (Merem et al., 2022). Uganda has a per capita milk consumption of 62 litres, while Tanzania is 47. The average milk consumption per capita is projected to double by 2030. To meet the rising demand for milk driven by rapid urbanization and the growing middle-class population, domestic production has been rising by approximately 5.8% annually (Bebe et al., 2017).

Despite the potential, the dairy sub-sector still experiences an extensive informal marketing and trading system alongside low productivity challenges (Kwamboka et al., 2022). According to Rao et al. (2016), low productivity is due to limited access to quality, affordable inputs and services, and output markets. Farmers have formed collective action groups such as dairy cooperative societies to address the challenges affecting dairy farming. Smallholder dairy farmers collaborating can reduce transaction costs involved in searching for markets, securing new technologies, tapping into better markets and increasing their bargaining power (Fischer et al., 2012). Moreover, the cooperatives ensure the quality and safety of agro-food and improve farm performance and household welfare, especially in rural areas (Ma & Abdulai, 2016; Ma et al., 2018, 2022; Wossen et al., 2017)

According to the Kenya Dairy Board (2019), the government has distributed over 350 milk coolers to dairy cooperatives through private partnerships. These coolers assist farmers in increasing the shelf life of milk as they await buyers, such as processors, to purchase from them. Some cooperatives have developed systems that allow their members to access advances against future milk payments. This process is facilitated through the collection and bulking enterprises affiliated with Saving and Credit Cooperative Organizations. These collection and bulking enterprises represent the most significant source of loans for smallholder dairy farmers (Makoni et al., 2014). Cooperative membership also enables access to improve farming technologies and capacity building (Abdul-Rahaman et al., 2021)

Several studies have been carried out on different dimensions of cooperative societies. Some of the previous studies (for example, Kumar et al., 2018, Kumar et al., 2018, Mojo et al., 2017, Twumasi et al., 2021Dan et al. (2021))) have assessed the impacts of cooperative membership on welfare. The results showed that indeed cooperative membership increases household incomes and yield produced. Kumar et al. (2018) used the ESR model to determine the impact of cooperative membership on milk yield, net returns and compliance with food safety measures. The results indicated that dairy cooperative membership positively and significantly affected the three outcome variables. The year dummy and share of dairy income positively impacted members’ income. The age of the household head, years of experience in dairy farming, years of schooling and household size did not significantly affect farmers’ income.

In their study, Mojo et al. (2017) found that cooperative membership positively influenced farmers’ income. Education of the household head, family size, land size and distance to the collection centre positively impacted the choice of cooperatives. Dan et al. (2021) found that family status and village terrain negatively and significantly impacted cooperative members’ income in China. The authors suggested that family status in the form of increased off-farm income-generating activities reduced the dependence on agricultural income. The land terrain in the hilly areas increased transportation costs thus reducing farmers’ income. Farmland area, access to loans and labour input did not significantly influence income. Contrary to their findings, Twumasi et al. (2021) found that off-farm occupation negatively impacted farmers’ income. However, these past studies did not consider the income difference between the active members of the cooperatives and those who have decided to sell to other marketing outlets.

In Borabu and Rachuonyo South Sub-counties in Kenya, the farmers have formed collective action groups (cooperatives) to sell milk at low transaction costs. These cooperatives own milk cooling plants with a capacity of 5,000 litres. Feed the Future Accelerated Value Chain Development (AVCD) project, implemented by the International Livestock Research Institute (ILRI), recently supported the Kasbondo dairy cooperative. Farmers in Nyamira have also benefited from the Smallholder Dairy Commercialization Program. The societies also provided the member farmers access to services such as artificial insemination, animal health services and feed access. However, participation by members of these cooperative societies is still low, as farmers prefer other marketing outlets.

Furthermore, the welfare impacts of participating in these cooperatives as a marketing outlet choice have not been determined. This study, therefore, fills the gap in the literature by determining the impact of participating in dairy cooperatives by smallholder members on their income in Kenya. There is a need to establish whether the cooling plants benefit the farmers. This study tests the hypothesis that there is no statistically significant difference between the incomes of participants and non-participants of the cooperatives. This study will contribute toward realizing Sustainable Development Goal 1 (ending poverty) through increased incomes. To the management of the cooperative, this study will provide appropriate policies for increased participation by farmers and, hence, the viability of the joint milk marketing business.

2. Materials and methods

2.1. Study area

The study was conducted in Homa Bay and Nyamira counties, Kenya. Homa Bay County is in Kenya’s Western region along Lake Victoria’s south shore. It lies between latitudes 0^°5 and 0^°2 South and between longitudes 34 °and 35^° East. The county covers an area of 3152.5 km² with a population of 1,131,950 persons (KNBS, 2019). There are two rainy seasons in Homa Bay County, with the long rains from late ranging from 800 mm to 1800 mm and the short rains ranging from 250 mm to 700 mm. Agriculture is a significant source of livelihood for many households. Livestock breeds reared for milk and meat production comprise the East African zebu, dairy goats and exotic dairy cows.

Nyamira is also located in the western region of the country. It shares its northern border with Homabay County, borders Kisii County to the west, Bomet County to the southeast, and Kericho County to the east. The county encompasses an area of 899.4 with a population of 605,576 people. It lies between latitude 0^°30’ and 0^°45’south and between longitude 34^°45’ and 35^° 00’ east. The annual rainfall ranges between 1200 mm and 200 mm per annum. The area is suitable for both crop and animal production. The two regions were selected because they can potentially increase milk production in the western part of the country. In addition, investments have been made to facilitate milk production and cooling.

2.2. Sampling procedure

This study collected data from 277 dairy farmers from Borabu and Rachuonyo sub-counties. The selection process employed a multistage sampling procedure. First, the two sub-counties were purposely chosen because of their potential to improve milk production. In the second stage, dairy cooperatives with a cooling capacity of 5,000 litres within the two sub-counties were selected. A stratified random sampling of respondents from a list of all registered cooperative members was done in each stratum (sub-county). The number of respondents drawn from Borabu was 153, and 124 were from Rachuonyo South relative to their population. The number of dairy farmers actively participating in the cooperatives over the past year was 116. The sample size was determined using Yamane’s (1967) formula.

(1) $n=\frac{N}{1+N{\left(e\right)}^2}$(1)

$=\frac{906}{1+906\left({0.05}^2\right)}$

$n=277$

where: n = desired sample sizee = Margin of errorN= the population

2.3. Analytical framework

This study employed a combination of descriptive and inferential statistical methods. Descriptive statistics, including means, percentages, and standard deviations, were utilized to examine the difference in farmer characteristics based on their participation. Meanwhile, inferential statistics encompassed the use of the t-test and the Endogenous Switching Regression (ESR) model.

The t-test was applied to ascertain statistical differences in the cooperatives’ mean characteristics between participants and non-participants. On the other hand, the ESR model was utilized to assess the impact of participation.

The ESR model addresses unobserved characteristics that influence both the choice of the market and are likely to be correlated with unobserved factors affecting farmers’ income. Neglecting the endogeneity problem related to the selection of market outlets can lead to biased estimates. The model involves a two-stage framework. The farmers’ decision to sell to the market outlet is estimated using a Probit model in the initial stage. In the subsequent stage, the impact of this choice on income is evaluated.

The household’s choice of milk marketing outlet is assumed to be driven by the maximization of utility (expected income benefits) based on random utility theory. The decision of a household i to sell to a cooperative society can be represented by Equation 2

(2) ${\mathrm{P}}_i^{\ast }=\gamma {\mathrm{Z}}_i+{\mu }_i$(2)

where ${\mathrm{P}}_i^{\ast }$ is the latent variable indicating the choice of cooperative, ${\mathrm{Z}}_i$is a vector of exogenous variables influencing the choice of cooperatives, $\gamma$is a vector of parameters to be estimated (income) and ${\mu }_i$is the random error term associated with the choice of cooperatives. Equation 2(2) ${\mathrm{P}}_i^{\ast }=\gamma {\mathrm{Z}}_i+{\mu }_i$(2) is also known as the selection equation.

From equation 2(2) ${\mathrm{P}}_i^{\ast }=\gamma {\mathrm{Z}}_i+{\mu }_i$(2) , we can generate two separate regimes showing the income levels for farmers who participated in cooperative marketing and those who did not. The generated equations 3(3) $Regime1(cooperativeparticipants){\mathrm{{\rm Y}}}_{1i}={\beta }_1{\chi }_{1i}+{\epsilon }_{1i}\mathrm{i}\mathrm{f}{\mathrm{P}}_i=1$(3) and 4(4) $Regime2(non-participants){\mathrm{{\rm Y}}}_{2i}={\beta }_2{\chi }_{2i}+{\epsilon }_{2i}\mathrm{i}\mathrm{f}{\mathrm{P}}_i=0$(4) form the second stage of the ESR model (Tambo & Wünscher, 2017). The full information maximum likelihood (FIML) is used to estimate the selection equation (2)(2) ${\mathrm{P}}_i^{\ast }=\gamma {\mathrm{Z}}_i+{\mu }_i$(2) and the outcome equations 3(3) $Regime1(cooperativeparticipants){\mathrm{{\rm Y}}}_{1i}={\beta }_1{\chi }_{1i}+{\epsilon }_{1i}\mathrm{i}\mathrm{f}{\mathrm{P}}_i=1$(3) and 4(4) $Regime2(non-participants){\mathrm{{\rm Y}}}_{2i}={\beta }_2{\chi }_{2i}+{\epsilon }_{2i}\mathrm{i}\mathrm{f}{\mathrm{P}}_i=0$(4) simultaneously (Lokshin & Sajaia, 2004).

(3) $Regime1(cooperativeparticipants){\mathrm{{\rm Y}}}_{1i}={\beta }_1{\chi }_{1i}+{\epsilon }_{1i}\mathrm{i}\mathrm{f}{\mathrm{P}}_i=1$(3)

(4) $Regime2(non-participants){\mathrm{{\rm Y}}}_{2i}={\beta }_2{\chi }_{2i}+{\epsilon }_{2i}\mathrm{i}\mathrm{f}{\mathrm{P}}_i=0$(4)

where $y_{i1}$ and ${\mathrm{{\rm Y}}}_{2i}$ are latent variables (household income level) that define observed income levels, ${\mathrm{P}}_i=1$ or ${\mathrm{P}}_i=0$ (whether the household participated or did not). $\chi$ is a vector of exogenous variables determining the income level, $\beta$ are the vector of parameters to be estimated while ${\epsilon }_{1i}$ and ${\epsilon }_{2i}$ are the disturbance terms. $\chi$ contains factors affecting income levels that are observable only. The error term from the selection equation and the outcome equation (eq. 3(3) $Regime1(cooperativeparticipants){\mathrm{{\rm Y}}}_{1i}={\beta }_1{\chi }_{1i}+{\epsilon }_{1i}\mathrm{i}\mathrm{f}{\mathrm{P}}_i=1$(3) and 4(4) $Regime2(non-participants){\mathrm{{\rm Y}}}_{2i}={\beta }_2{\chi }_{2i}+{\epsilon }_{2i}\mathrm{i}\mathrm{f}{\mathrm{P}}_i=0$(4) ), according to Greene (2012), are assumed to be normally distributed with zero mean and a covariance matrix, as shown in equation 5(5) $Cov\left(\mu ,{\epsilon }_1,{\epsilon }_2\right)=\left(\begin{array}{ccc}{\sigma }_{\mu }^2& {\sigma }_{\mu {\epsilon }_1}& {\sigma }_{\mu {\epsilon }_2}\\ {\sigma }_{\mu {\epsilon }_1}& {\sigma }_{{\epsilon }_1}^2& .\\ {\sigma }_{\mu {\epsilon }_2}& .& {\sigma }_{{\epsilon }_2}^2\end{array}\right)$(5) . The covariance between ${\epsilon }_{1i}$and ${\epsilon }_{2i}$is indicated as dots because ${\mathrm{{\rm Y}}}_{1i}$and ${\mathrm{{\rm Y}}}_{2i}$are not simultaneously determined (Maddala, 1983).

(5) $Cov\left(\mu ,{\epsilon }_1,{\epsilon }_2\right)=\left(\begin{array}{ccc}{\sigma }_{\mu }^2& {\sigma }_{\mu {\epsilon }_1}& {\sigma }_{\mu {\epsilon }_2}\\ {\sigma }_{\mu {\epsilon }_1}& {\sigma }_{{\epsilon }_1}^2& .\\ {\sigma }_{\mu {\epsilon }_2}& .& {\sigma }_{{\epsilon }_2}^2\end{array}\right)$(5)

fwhere ${\sigma }_{\mu }^2$is the variance of the error term in the selection equation, ${\sigma }_{{\epsilon }_1}^2$and ${\sigma }_{{\epsilon }_2}^2$shows the variances of the error terms in the outcome equations, ${\sigma }_{\mu {\epsilon }_1}$and ${\sigma }_{\mu {\epsilon }_2}$the covariance of the error terms between the selection equation and that of the outcome equations (Maddala, 1983). According to Greene (2012), the error term is assumed to be non-zero.

As noted earlier, ESR accounts for selection bias by including the inverse mills’ ratio (${\lambda }_{i1}$ and ${\lambda }_{2i}$ for participants and non-participants, respectively) to the outcome (Greene, 2012; Wooldridge, 2002). Equations 6(6) ${\mathrm{{\rm Y}}}_{1i}={\beta }_1{\chi }_{1i}+{\sigma }_{{\mu }_1}{\lambda }_{i1}+{\epsilon }_1$(6) and 7(7) ${\mathrm{{\rm Y}}}_{2i}={\beta }_2{\chi }_{2i}+{\sigma }_{{\mu }_2}{\lambda }_{2i}+{\epsilon }_{2i}$(7) show the mill’s ratio added to the outcome equations.

(6) ${\mathrm{{\rm Y}}}_{1i}={\beta }_1{\chi }_{1i}+{\sigma }_{{\mu }_1}{\lambda }_{i1}+{\epsilon }_1$(6)

(7) ${\mathrm{{\rm Y}}}_{2i}={\beta }_2{\chi }_{2i}+{\sigma }_{{\mu }_2}{\lambda }_{2i}+{\epsilon }_{2i}$(7)

where ${\lambda }_{2i}$ and ${\lambda }_{2i}$ are the selectivity correction terms used to control for selection bias caused by unobserved attributes, and ${\epsilon }_1$ and ${\epsilon }_{2i}$ are the random error terms.

According to Lokshin and Sajaia (2004), for better identification of the ESR model, at least one identification variable other than the ones generated by the non-linearity (λ_1iand λ_2i) should be added to the selection model. The Identification variable is also known as the instrument variable. The instrument variable should influence the selection equation (2)(2) ${\mathrm{P}}_i^{\ast }=\gamma {\mathrm{Z}}_i+{\mu }_i$(2) but not the outcome equations (6(6) ${\mathrm{{\rm Y}}}_{1i}={\beta }_1{\chi }_{1i}+{\sigma }_{{\mu }_1}{\lambda }_{i1}+{\epsilon }_1$(6) and 7(7) ${\mathrm{{\rm Y}}}_{2i}={\beta }_2{\chi }_{2i}+{\sigma }_{{\mu }_2}{\lambda }_{2i}+{\epsilon }_{2i}$(7) ). From previous studies, Mojo et al. (2017) used the distance to the nearest town as an instrument variable, while Ngeno (2018) used distance to extension services.

In deriving average treatment effects on the treated ATT, we compare the incomes of farmers sold to the cooperatives with the counterfactual hypothetical case of those that did not. The expected income for households that participated in the dairy cooperatives and those that did not are expressed below (Heckman & Vytlacil, 2001).

(8) $E\left(Y_{1i}/P=1\right)={\chi }_{1i}\beta +{\sigma }_{{\mu }_1}{\lambda }_{2i}$(8)

(9) $E\left(Y_1/P=0\right)={\chi }_{2i}\beta +{\sigma }_{{\mu }_2}{\lambda }_{2i}$(9)

The ATT for participation in the cooperative is the difference between equations 8(8) $E\left(Y_{1i}/P=1\right)={\chi }_{1i}\beta +{\sigma }_{{\mu }_1}{\lambda }_{2i}$(8) and 9(9) $E\left(Y_1/P=0\right)={\chi }_{2i}\beta +{\sigma }_{{\mu }_2}{\lambda }_{2i}$(9) expressed as:

(10) $ATT=E\left(Y_{1i}/P=1\right)-E\left(Y_1/P=0\right)={\chi }_i({\beta }_{1i}-{\beta }_{2i})+{\lambda }_{i1}({\sigma }_{{\mu }_1}-{\sigma }_{{\mu }_2})$(10)

The values in a and b represented in Table 1 are the observed mean income values of participants participating in cooperative marketing and non-participants, respectively. The values on (d) provide the counterfactual expected income estimates for farmers selling to cooperatives, while (c) gives the counterfactual expected estimates for farmers selling to other milk marketing outlets.

Table 1. Treatment, heterogeneity and transitional heterogeneity effects

Sub sample	Decision stage		Treatment effects
	Participant	Non-participant
Participant	a)$E\left(Y_{1i}/P=1\right)$	c)$E\left(Y_{2i}/P=1\right)$	On the treated (ATT)
Non-participant	b)$E\left(Y_{1i}/P=0\right)$	d) $E\left(Y_{2i}/P=0\right)$	On the untreated (ATU)
Heterogeneity Effects	BH1	BH2	TH

BH1, base heterogeneity for participants; BH2, base heterogeneity for non-participants; ATT, average treatment on the treated: ATU, average treatment on the untreated; TH, transaction heterogeneity

Following Smale and Olwande (2014), the heterogeneity effects for both participants in cooperative selling and non-participant farmers can be estimated as shown in Table 1. $B{H}_1$which is the base heterogeneity for participants in cooperative and $B{H}_2$ base heterogeneity for non-participants. TH is the transitional heterogeneity determining whether participants’ income effects are small or bigger.

2.3.1. Model for estimating potential endogeneity

There is a potential for an endogeneity problem arising while estimating equation 2(2) ${\mathrm{P}}_i^{\ast }=\gamma {\mathrm{Z}}_i+{\mu }_i$(2) from variables like access to off-farm income. Cooperatives can encourage farmers to participate in off-farm work to supplement household income. Participating in off-farm activities can also influence farm production and cooperative membership. This study used an approach by Rivers and Vuong (1988) where the potential endogenous variables are expressed as a function of the explanatory variables used to determine the choice of cooperative alongside the instrument variable, as shown in equation 11(11) $G_i=Z_i\beta +Si\omega +e$(11) .

(11) $G_i=Z_i\beta +Si\omega +e$(11)

Where is G_i a vector of observed potential endogenous variables such as access to off-farm income, ${\mathrm{Z}}_i$ is a vector of exogenous variables influencing the choice of cooperatives, Si is a vector of instruments and ${\mu }_i$ is the random error term. The instrumented variable used for access to off-farm income was the main occupation. The predicted residuals for the endogenous variables (off-farm income) were generated after regression. Lastly, the predicted and observed factors are used in the choice of cooperatives. The endogenous variables, therefore, become exogenous in the second stage of the ESR following (Wooldridge, 2015). Moreover, the Hausman test for endogeneity becomes robust. The collected data underwent analysis through the use of Stata.

3. Results and discussions

3.1. Descriptive statistics

Table 2 describes the variables used in the ESR model with their respective descriptive characteristics. The sampled households were comprised of 42% participants in cooperative marketing. The household heads comprised 76% males with an average age of 57. The number of years of schooling was 10, indicating that the household heads sampled had all completed primary school education. More than a third of the households had adopted the practice regarding improved fodder growth.

Table 2. Covariates for ESR and descriptive statistics

Variable	Description	Mean (Std. Dev)
Treatment variable	Participants 1= cooperative 0 otherwise	0.42(0.49)
Outcome variable	Annual income in KES	368,812.60(25508.37)
Independent variable
Age	Age of household head (years)	57.04 (13.80)
Gender	Gender of household head (male dummy)	0.76 (0.43)
LevEdu	Years of schooling	10.80 (4.56)
Extension acc	Extension services access (yes = 1, no = 0)	0.58 (0.5)
CrdAcc	Access to credit	0.37 (0.48)
Farm size	Farm size in acres	4.34 (0.25)
Lactating cows	Number of lactating cows	1.5(0.07)
Offarmincome	Access to off-farm income (1=yes,0 No)	0.64 (0.48)
H/size	Household size	5.3 (2.2)
Paymthd	Payment duration (daily = 1, 0 otherwise)	0.52 (0.5)
DistMrk	Distance to the nearest market centre	1.25 (1.58)
Fodder growth	1=yes 0 = no	0.80 (0.02)

Figures in parentheses are standard deviations associated with the means of the variables indicated.

Table 3 describes farmer and farm characteristics with t-test statistics indicating the differences in the mean between participants and non-participants of the cooperatives. Statistical differences were observed in schooling, farm size, access to extension, household size, fodder growth, and annual incomes. Participants of cooperatives had more years of schooling, implying that education could be an essential factor influencing participation. The results indicated that 67% of participants and 50% of non-participants received extension services.

Table 3. Farm and farmer descriptive characteristics

	Participant Mean Std. Dev		Non-participants Mean Std. Dev		difference t-stat
Variables
Age (years)	56.93	13.38	58.43	14.10	0.8818
Lactating cows (count)	2.0	0.12	1.0	0.07	−6.342***
Household size (count)	4.9	1.92	5.5	2.32	2.439**
Gender (male dummy)	75.86	0.43	75.78	0.43	−0.016
Education (yrs of schooling)	12.08	4.33	9.84	4.57	−1.175***
Payment(1=daily,0 =otherwise)	0.009	0.29	0.83	0.37	17.63
Collection point distance (Km)	1.42	1.63	1.13	1.53	−1.485
Credit access (1=yes,0=no)	0.35	0.48	0.39	0.49	0.536
Extension access (1=yes,0=no)	0.67	0.47	0.51	0.50	−0.045***
Fodder growth (1=yes,0=no)	0.76	0.43	0.27	0.44	−9.214***
Farm size (count)	5.28	0.46	3.65	0.28	−3.250**
Off-farm income (1=yes,0=no)	0.61	0.47	0.66	0.49	0.167
Annual income (KES)	509806	453806.61	267190	371585.9	−4.883***

Abbreviation: KES, Kenyan Shillings.

*** and **represents statistical significance at 1%, and 5% levels, respectively.

Figures in parentheses are standard deviations associated with the means of the variables indicated.

The household sizes were 5 and 6 people for participants and non-participants, respectively. The household size determined the amount of household labour available from family members. Participants of the cooperatives had an average of 2 lactating cows, while non-participants had 1. More than three-thirds of cooperative participants grew improved fodder and had larger farm sizes. Furthermore, the study showed that participants had a higher average annual income despite receiving lower prices per litre of milk. The results showed that participants received a mean income of Ksh 509,857 (USD 3333) (as compared to the non-participants, who had Ksh 267,190 (USD 1747).

3.2. Empirical results

3.2.1. Determinants of choice of cooperative

Table 4 presents the Full Information Maximum Likelihood (FIML) endogenous switching regression model results, wherein the coefficients are interpretable as normal Probit. The wald chi -square statistic significantly indicated the goodness of fit of the ESR model at a 1% significance level. The wald chi-square statistic results also showed an endogeneity problem controlled for using the ESR model. The likelihood ratio test of independent equations was highly significant, indicating that the null hypothesis (no correlation) between participants in cooperatives and household annual income was rejected.

Table 4. ESR results for impact of choice of market outlet

	Selection model		Household income (Participants)		Household Income (Non-participants)
Variables	Coef	Std. Err	Coef	Std. Err	Coef	Std. Err
Education (years)	0.069	0.054	0.067***	0.016	0.041**	0.018
Age (year)	0.003	0.010	0.005	0.005	0.001	0.006
Gender (male dummy)	−0.327	0.335	−0.132	0.142	−0.034	0.195
Off-farm income (0= no, 1=yes)	−0.830	1.280	0.926***	0.134	0.835***	0.164
Farm size (acres)	−0.051*	0.026	0.041***	0.012	0.028	0.023
Household size	0.036	0.079	−0.018	0.032	−0.008	0.035
Credit access (0= no, 1=yes)	0.130	0.275	−0.032	0.129	0.114	0.158
Extension access (0= no, 1=yes)	−0.002	0.293	0.273**	0.136	−0.031	0.158
Lactating cows	0.280**	0.122	0.103**	0.048	0.100	0.088
Homa bay (0= no, 1=yes)	−2.553***	0.373	0.531	0.362	0.101	0.232
Market Distance (Km)	0.193**	0.095	0.103***	0.038	−0.079	0.052
Fodder growth (0= no, 1=yes)	−0.231	0.479	−0.212	0.207	0.153	0.181
Payment period (1=daily 0=otherwise)	−2.020***	0.296
Off-farm res	0.7249	1.2972
_cons	0.733	0.931	10.760***	0.373	10.494	0.499
rho_1	−0.327	0.251
rho_2	−0.629**	0.180
Log-likelihood - −377.98, LR test of independent equations χ2(1) 7.02 ***

Abbreviations: ESR, endogenous switching Regression

Note: ***, **, * denote statistical significance at 1%, 5% and 10% levels, respectively. A log of income was used.

The covariance terms for both participants and non-participants (rho_1 and rho_2) are negative and statistically significant, indicating the presence of selection bias related to unobserved factors, thus justifying the use of the ESR model. The instruments used in this study are statistically significant in influencing the choice of cooperatives but did not affect incomes. We posit that payment duration, whether in terms of daily cash or otherwise, influenced the choice of marketing outlet but did not directly impact the annual income received by a farmer. We confirmed the validity of the instrumental variable through the falsification process.

The payment period negatively and significantly affected the selection of cooperatives. When cooperatives delay payments to farmers for milk delivered for 30 days or more, farmers are more likely to opt for informal markets that offer daily or weekly payments. A plausible reason for this behaviour is that farmers may need cash to cover their daily expenses. Wangu et al. (2021) found that farmers prefer brokers who provide daily cash payments over processors and dairy cooperatives. Additionally, farmers in Homa Bay County are less inclined to sell to cooperatives than those in Nyamira County.

The distance to the milk market centre positively and significantly affected the choice of cooperatives. An increase in distance by one km increased the likelihood of selling to the cooperatives by 17%. This effect may be attributed to cooperatives transporting milk from farmers to the market at a lower cost, thus reducing transaction expenses. Jitmun and Kuwornu (2019) found out that farmers far from the marketing centres preferred selling collectively to reduce the cost of transportation. Furthermore, Ahmed and Mesfin (2017) argued that farmers closer to market centres have better access to market information and are less reliant on cooperative marketing.

An increase in the number of lactating cows positively and significantly increased the likelihood of farmers selling to the cooperatives by 28%. Earlier studies by (D’ Antoni et al., 2013; Mutura et al., 2015) argued that an increase in the number of lactating cows resulted in a higher marketable surplus and the ability to sell to multiple marketing outlets. Furthermore, Aderinkola et al. (2022) found that an increase in the number of lactating cows led to a higher volume of milk supplied to milk collection centres in Nigeria. Similarly, Ayyano et al. (2020) found that private milk traders and cooperatives prefer farmers with higher milk production due to lower transaction costs, and farmers benefit from larger volumes of milk they can sell to these outlets.

3.2.2. Determinants of household income

The coefficients in the fourth and sixth columns of Table 4 present the determinants of farmers’ income for participants and non-participants of the cooperative marketing, respectively. The results showed that the level of education positively and significantly affected household incomes for both the participants and non-participants at least at the 5% significance level. The result showed that more educated farmers had more income than the less educated. This could be due to the ability of learned farmers to adopt reliable milk marketing outlets and better production practices. This finding echoed studies by (Abdulai & Huffman, 2014; Ngeno, 2018; Sharma, 2015), who revealed that farmers with more years of education were more likely to adopt strategies that increase their incomes due to their knowledge and skills.

As expected, access to sources of income other than from dairy and crop farming increased farmers’ income at a 1% significance level for both participants and non-participants of cooperatives. This could have resulted from farmers receiving income from other sources to cushion them from crop failures and agricultural commodity price variabilities. Income from non-farm activities could also be used to finance dairy farming for resource-poor households who cannot obtain loans from financial institutions. Zheng et al. (2023) confirmed that off-farm activities increased access to farm inputs and labour thus increasing productivity. This result is confirmed by Twumasi et al. (2021) who found that off-farm jobs increased household income for members of cooperative societies. Nonetheless, Dan et al. (2021) found a negative and significant effect between off-farm income access and agricultural incomes of cooperative members.

The number of lactating cows owned by the farmers positively impacted the participants’ income at 5% significance. A plausible reason for this is that the quantity of milk produced is directly correlated with the number of lactating cows. According to African tradition, the number of animals a farmer owns can be used as a proxy for wealth. Ngeno (2018) ascertained that an increase in the tropical livestock unit owned by participants of the dairy hub increased their net returns. The author reasoned that farmers with many livestock are considered wealthy, can finance dairy production, and bear the production risks.

The size of the farm owned by households positively influenced participants’ income, with statistical significance observed at the 1% significance level. This result indicated that farmers managing larger agricultural holdings tend to generate higher annual incomes when compared to those with smaller plots of land. This finding aligns with previous studies by Li et al. (2020) and Ma and Abdulai (2016), who found a positive and significant correlation between farm size and the preference for cooperatives as a marketing outlet. However, it is essential to note that Ngeno (2018) reported a contrasting result, revealing a negative and significant relationship between farm size and household income. Ngeno attributed this adverse impact to the presence of diseconomies of scale. Moreover, Bidzakin et al. (2019) linked the negative association between farm size and gross margin to inefficient production practices and suboptimal land management.

Distance to the market centre positively and significantly affected participants’ income. Participants of the cooperatives lived far from the market centers and relied on selling through the cooperatives to minimize costs and get better prices. A plausible reason for this is that the cooperatives own vehicles that transport milk from the farmer to the cooling facility at a fixed low cost per litre. However (Ngeno, 2018), found an inverse relationship between distance and income of dairy farmers. Musyoki et al. (2022) found no relationship between distance to market and income levels between participants and non-participants of milk processing plants in Kenya.

Access to extension services impacted income positively and significantly for participants of the cooperatives. The result implied that farmers with frequent access to extension services had increased incomes. Extension access results in better access and adoption of new technologies, which improve dairy performance. Mutonyi (2019) vindicated that the knowledge obtained from collective action had a spill-over effect on the production of other crops, thus resulting in increased income. This result concurs with earlier studies by (Abdul-Rahaman & Abdulai, 2020; Musyoki et al., 2022).

3.2.3. The treatment effects on smallholder dairy farmers’ income

Farmers participating in cooperative society marketing formed the treatment group, while those participating in other milk marketing formed the control group. Table 5 summarizes the mean treatment effects and their counterfactual effect on dairy farmers’ incomes. The values in (a) and (d) show the actual log of incomes for participants and non-participants, respectively. (c) and (b) are the counterfactual income.

Table 5. Mean treatment effect on farmers’ income

	Sub-sample	Decision
Outcome		Participants	Non-participants	ATE
Total income	Participants	(a) 12.70	(c) 11.74	ATT = 0.96***
	Non-participants	(d) 12.87	(b) 11.92	ATU = 0.95 ***
Heterogeneity Effect		BH1= −0.17	BH2= −0.18	TH = 0.1

Abbreviations: BH₁ base heterogeneity for participants; BH₂ base heterogeneity for non-participants; ATT average treatment effects on the treated; ATU, average treatment effect on the untreated; TH transaction heterogeneity, ATE Average treatment effect

***represents significance at 1%. Note the values are logs of income.

ATT and ATU demonstrated a positive and significant effect on household incomes at a 1% significance level. The mean household income of participants would have reduced by 8.18% had they not participated (from 12.70 to 11.74). Similarly, non-participants would have increased their incomes by 7.97% (from 12.87 to 11.92) had they participated. These findings are consistent with (Kiprotich et al., 2017; Ngeno, 2018; Rao et al., 2016), who revealed that formal milk market participation positively impacted farmers’ incomes.

The negative base heterogeneity BH₁ showed that non-participants would earn an increase of 0.17 in their log of income had they participated. Likewise, the negative BH₂ suggested that participants would have reduced their log of income by 0.18 had they not participated. The positive transitional heterogeneity (TH) implies that income effects are more remarkable for the participants than for non-participants.

3.3. Robustness check

A robustness check was done using propensity score matching (PSM) and inverse probability weighted regression adjustment (IPWRA) to determine the consistency of the ESR. The average treatment effects of cooperative participation were positive and significant, as shown in Table 6. The ATT results from PSM and IPWRA estimations of household income are 0.716 and 0.493, respectively.

Table 6. ATT results using PSM

Outcome	ATT(PSM)	ATT(IPWRA)
Household income	0.891 (0.315) ***	0.461(0.217) ***

Abbreviations: ATT, average treatment effects on the treated; IPWRA, inverse probability weighted regression adjustment; PSM, propensity score matching.

***significance at 1%. Note the values are logs of income.

4. Conclusion and recommendations

This study used the ESR model to determine the impact of cooperatives on smallholder dairy farmers’ income. The ESR model accounted for the observed and non-observed factors due to selection bias. The ESR model’s first stage selection equation presented the factors influencing cooperatives’ participation. The Probit results showed that the number of lactating cows, farmer location, farm size, distance to the market centre and payment period significantly determined the selection of the cooperatives. Furthermore, the second stage of the ESR model showed that participating in cooperatives improved farmers’ income in Kenya. Education, access to off-farm income, number of lactating cows, farm size, extension access and distance to market influenced farmers’ incomes.

This study recommends policies that enhance cooperative market participation. Initially, it is essential to raise awareness among farmers in Homa Bay and Nyamira about the potential benefits of participating in cooperatives. The farmers should also be encouraged to increase the number of milk-producing cows relative to the bulls. In addition, farmers should embrace the use of better breeding mechanisms like Artificial insemination and proper routine management to ensure more milk production.

The cooperatives should also consider paying farmers at the agreed time to prevent side selling. Furthermore, the cooperatives should seek partnerships with organizations that promote the development of small businesses so that their members can consider venturing into other sources of income. Lastly, the government should review milk prices so that farmers do not suffer from meager prices offered by the cooperatives.

Acknowledgement

We thank the Center of Excellence in Sustainable Agriculture and Agribusiness Management (CESAAM), Egerton University and the International Livestock Research Institute (ILRI) Kenya for their support. I also thank the farmers for participating in the survey.

Disclosure statement

No potential conflict of interest was reported by the author(s).