Logistic regression analysis on factors influencing income-poverty among smallholder French bean farmers in Kirinyaga County, Kenya

The study determined factors affecting income-poverty among French bean farmers in Kirinyaga County, Kenya. Results show that 72.6% of French bean farmers were poor. Relative to the poor households, non-poor had the highest net French bean income per acre (KES. 24,133), total net crop income (KES. 150,608), total net livestock income (KES. 21,674), and off-farm income (KES. 198,070). Aversion to risks (p=0.051 and β=-2.802), household size (p=0.000 and β=-5.032), daily household expenditure per adult equivalent (p=0.001 and β=2.016), net annual household income (p=0.000 and β=7.733), access to credit (p=0.086 and β=1.187), household annual asset value (p=0.051 and β=-0.482) and age of household head (p=0.066 and β=-2.009) statistically and significantly influenced poverty status of French bean farmers. The results suggest that crop insurance, farm diversification and expansion of acreage under Global-GAP certified French bean are necessary strategies for French bean farmers to alleviate household poverty.


INTRODUCTION
Poverty indicates household well-being and according to Muyanga et al. (2006), poverty in Kenya is common in rural areas and has been increasing since the late 1980s. Nearly 80% of the total population in Kenya lives in rural areas and derive their livelihood from agriculture. A study by Kenya National Bureau of Statistics (2007) showed that poverty rate rose from 40 to 52.3% between the year 1994 and 1997. In 2007, the rate reduced to 45.9% but according to the International Fund for Agricultural Development (2013), the rate rose again to 53% in 2013.
The current national poverty rate is still high at 51.4% while rural poverty level stands at 39.9% (Oxford Poverty and Human Development Initiative, 2017). To eradicate poverty, French bean farmers in Kirinyaga County have been producing Global-GAP certified French beans with an objective of accessing lucrative export markets in Europe, in order to increase household income and thus reduce poverty. Existing studies in Kenya concur that commercial horticultural farming can improve the welfare of farmers (McCulloch and Ota, 2002;Muriithi et al., 2011;Muriithi and Matz, 2014;Chege et al., 2015). This is because most of horticultural crops are high yielding, more profitable relative to cereal crops and are thus able to reduce poverty (McCulloch and Ota, 2002;Kibet et al., 2011;Mwende, 2016) even under situations of high risks (Obare et al., 2003;Kuyiah et al., 2006).
In Central Region where Kirinyaga County is situated, poverty per adult equivalent and household stood at 30.4 and24.3% respectively in 2007 (Kenya National Bureau of Statistics, 2007). But the latest statistics indicate that poverty rate in Kirinyaga County will increase to 30.1% in future (Oxford Poverty and Human Development Initiative, 2017). However, previous studies conducted in Kirinyaga County show that compliance with the Global-GAP standards improves the households" welfare (Humphrey, 2008;Muriithi, 2008;Asfaw et al., 2009;Asfaw et al., 2010). This raises the question of whether continuous compliance with the Global-GAP standards contributes to poverty alleviation among farmers in the County or there are other factors that drive poverty situations of farmers in the County. Compliance with Global-GAP standards is associated with high costs, risks and at the same time high returns. Risk agricultural technologies in most cases are associated with high returns, which when adopted can increase household incomes and thus poverty reduction. The hypothesis on the increasing poverty in Kirinyaga County is that, since the majority of farmers have been found to be risk-averse worldwide, French bean farmers may have failed to embrace the standards by over-weighing the risks and underestimating returns associated with the standards.
For the Kenyan government to properly address the issue of poverty in rural areas where agriculture is the dominant economic activity, it needs to implement relevant, efficient and effective social protection policies. But to come up with such policies, there is a need to identify and constantly monitor all the relevant drivers of the poverty (Alemayehu et al., 2005). Several studies on the poverty at household level in Kenya have been conducted (Oyugi et al., 2000;Alemayehu et al., 2001Alemayehu et al., , 2005Nyariki et al., 2002;Republic of Kenya, 2003a;Muyanga et al., 2006;Githinji, 2011;Elhadi et al., 2012;Onyeiwu and Liu, 2013;Machio, 2015). However, none of these studies considered Global-GAP certification and risk attitudes as possible drivers of the poverty in their analysis. Studies on the poverty analysis among French bean farmers in the face of Global-GAP standards are scarce. They include: Asfaw et al. (2009), Muriithi et al. (2011) and Achieng" (2014) among others. These studies are characterized by some limitations: First, Asfaw et al. (2009) and Muriithi et al. (2011) failed to quantify poverty. In their analysis, they determined French bean farmer"s poverty status by just comparing certified and noncertified based on their welfare indicators such as consumption levels, asset values and incomes. Secondly, Asfaw et al. (2009) and Muriithi et al. (2011) failed to determine the role risk attitudes in the poverty analysis yet it is a crucial factor as reported elsewhere (Dohmen et al., 2005;Booij and De Kuilen, 2009;Kwesi et al., 2012;Ghartey et al., 2014). Achieng" (2014) quantified poverty and its determinants among French bean farmers in the face of Global-GAP but also failed to consider risk attitudes as one of the crucial drivers of the poverty.
To bridge this knowledge gap and inform further on social protection policies targeting farmers in Kenya, this study went a step ahead and quantified the poverty using the new international poverty line and considered risk attitudes and technology adoption as part of drivers of poverty in the analysis. The information generated is useful in guiding the formulation of social protection policies aimed at the poverty alleviation in Kenya by the government, Non-Governmental Organizations, private sector, and other stakeholders.

Study area
The study was conducted in Kirinyaga County within Kirinyaga Central, Kirinyaga East, Kirinyaga West, Mwea East and Mwea West Sub-counties ( Figure 1) because of the increasing importance of French bean farming and the implementation of Global-GAP standards in the County. The County is located 120 Km North West of Nairobi and has a total population of 153,095 (Economic Survey, 2009). Apart from French beans, rice, maize, and horticulture (onions, tomato, snow peas, avocado, mango and pawpaw) are also commonly grown in the County. French beans are mainly produced under irrigation and rain-fed.

Sample size
Lists containing French beans farmer"s certification details were obtained from Kirinyaga County Agricultural Office, farmer groups and exporters of French beans contracting farmers. The lists were then used to generate a sampling frame of 1,943 certified and noncertified farmers. The formula by Krejcie and Morgan (1970) was further used to determine the sample size. Mathematically, the formula is given as: where s is the required sample size, χ 2 is the table value of Chisquare for 1 degree of freedom at the desired confidence level (1.96 × 1.96 = 3.84), N is the population size, Pp is the proportion of the sample size to population size, d is the degree of the accuracy expressed as a proportion (0.05). Using the formula, the sample size corresponding to N = 1,943 is 322. However, due to the availability of funds, and the need to increase the accuracy of the results, the sample size was increased to 492.

Sampling procedure
The study adopted a multistage sampling procedure in the selection of respondents. Within Kirinyaga County, Kirinyaga Central, Kirinyaga East, Kirinyaga West, Mwea East, and Mwea West Sub-Counties were purposively selected because this is where French beans are mainly produced. Mutige in Kirinyaga Central, Gitaku, and Kathare in Kirinyaga West and Kamunyange and Mwea in Mwea East were randomly selected. French beans farmers in each area were then stratified into two groups: Global-GAP certified and non-certified. Sampling frames for the certified and non-certified French beans farmers were generated with the help of village elders, French beans farmer"s group leaders, and County Agricultural Extension officers. The study applied a systematic random sampling procedure to select the certified and non-certified French beans farmers from the sampling frames. Both certified and non-certified French beans farmers were selected proportionately depending on the respective generated sampling frames to give a total sample size of 492 respondents. All five sub-counties were sampled proportionately based on the certification status of the French beans farmers.

Data and data analysis
Single cross-sectional data was used in the study. The data was obtained from the 492 randomly selected respondents using structured and unstructured questionnaires covering 2014 cropping season. Data collected include general household socioeconomic, institutional and psychological characteristics. Psychological characteristics (risk attitudes) were captured using a 5-point Likert scale. Data on total annual household income was captured as the sum of net crop income, net livestock income, and total off-farm income. Total household expenditure comprised of food and nonfood items. Total asset values were computed based on respective market prices existed in the year 2014. Non-normal data were log transformed to approach normality. Means with 95% confidence intervals were presented. Differences between certified and noncertified in terms of their socioeconomic, institutional and psychological characteristics were analyzed using a t-test for continuous data and Chi-square for categorical data. Statistical analyses were performed using SPSS, STATA and Ms. Excel computer programs.

Theoretical framework
The study was based on the individualistic theory of poverty outlined in Asen (2002). The theory further states that individual characteristics such as low intelligence levels and market factors such as externalities, moral hazard and uncertainty are likely to drive one to poverty. Incomplete information in the presence of shocks/risks makes one prone to poverty.

Poverty estimation
Poverty estimation involves the determination of income per adult equivalent then compared with the predetermined poverty. There are many ways of determining Adult Equivalent Values (AEV), viz; World Health Organization (WHO) adult equivalent conversion factors (Table 1), Organization for Economic Co-operation and Development (OECD) method, and Harold Watts"s approach. WHO approach considers the effect of household gender, age, and size in the determination of adult equivalent values per household. Men are given higher weights than women while children and the old are given lower weights (OECD, 2008(OECD, , 2011. Relative to Watts" and OECD approaches, WHO approach is multidimensional in the determination of AEVs. It considers not only the economies of scale (household size) but also gender and age issues. In this study, AEVs were determined using the WHO approach. Following Kirimi et al. (2013), household income, household expenditure, and asset values of French beans farmers were computed per annum, monthly and daily. Daily household income and expenditure of each household were then divided by the AEVs from the WHO approach to get daily income and expenditure per adult equivalent. Asset value per adult equivalent was determined by dividing total annual asset value by adult equivalent values generated from the WHO approach.

Determination of the poverty line
The rural income-poverty line of KES 1,562 per month per adult equivalent generated by the Kenya National Bureau of Statistics was last updated in 2006 (Kenya National Bureau of Statistics, 2007). The poverty line has been widely applied in many studies, including Kirimi et al. (2013). The latest survey launched by Kenya National Bureau of Statistics (KNBS) to update the poverty line was launched in 2015, and that was after the collection of household survey data for this study. The first international poverty line was developed by Ravallion et al. (1991) in 1985 and was set at $1.01. The poverty line was later updated to $1.08 per adult equivalent per day by Chien and Ravallion (2001). Ravallion and Shaohua (2009) further updated to $1.25 per adult equivalent per day. Because of the increasing cost of living in developing nations due to inflation, it makes the poverty lines of KES 1,562 and $1.25 less effective. World Bank has been using the poverty line of $1.25 from 2009 to 2014 when Narayan et al. (2015) updated it again to $1.90. This new poverty line is highly recommended for poverty estimations, especially in developing countries. This is because it was partly generated from data collected in African nations (Narayan et al., 2015). Given that Kenya is a developing country and the fact that few studies have applied the new poverty line, it was imperative to embrace it in this study. Certified and non-certified French beans farmers were categorized as poor if their daily income per adult equivalent fall below the poverty line of KES 193.56 (that is, $1.90 at the exchange rate of KES 101.87 per dollar during data collection period) and non-poor if equal or fall above the poverty line.

Analytical technique
French beans farmers" poverty status (PV i ) was captured as binary such that PV i = 1 indicates not poor while PV i = 0 indicates otherwise. The factors influencing binary dependent variable can be estimated using binary Logit or Probit model. Binary Logistic model, as outlined in Nyota (2011), was used to determine factors affecting poverty among French beans farmers. The study assumed that the probability of French beans farmer i being either poor or non-poor (PV i ) is subject to his/her socioeconomic, institutional, and psychological characteristics (X i ) as indicated in Equation 2.
An underlying unobserved or latent variable (PV i *) can be defined to denote the level of poverty and the unobservable variable is related to the characteristics X i of the farmer. That is assuming there are no ties, then Where Β i is parameters estimated while ε i is the error term that captured unobserved variations in French beans farmers" poverty status. Functionally, this is given as: If the residuals are independent and identically distributed with a , an analytical solution exists, and the probability of a given choice alternative for the i th French bean is given as: where Prob(Not poor = 1) denotes the probability of French beans farmer i being poor, X i is a vector of the farmer characteristics while β j are the parameters of the exogenous variables estimated. The parameters were estimated using maximum likelihood (ML) method. Binary logistic regression can yield either the odds ratio or marginal coefficients. Odds ratios mean a unit change in an exogenous variable leads to changes in the probability of French beans farmers not being poor (Prob(Not poor = 1)) by a factor of exp β. On the other hand, marginal coefficients indicate the effect of each exogenous variable on the probability of French beans farmers being poor, ceteris paribus, are interpreted as typical beta coefficients in a linear regression model (Nyota, 2011). According to Laduber et al. (2016), the slope of a logistic regression function tells how the log odds ratio in favor of not being poor changes as explanatory variables change. For instance, given that Prob(Not poor = 1) is the probability of not being poor then, (1-Prob(Not poor=1)) represents the probability of being poor. Mathematically this is given as: Given the equations above, the odds ratio equation is given as: such that Prob ( Not poor = 1) 1− Prob (Not poor = 1 ) ¿¿ is the odds ratio in favor of Global-GAP compliance. That is the ratio of the probability that the farmer would comply with Global-GAP standards to the probability that the farmer will not comply with the standards. Factors estimated on the observed poverty are described in Table 2. Empirically, the model was determined as given in Equation 8: Proper estimation of binary logistic regression depends on certain assumptions. First is the assumption that the dependent variable should be ordinal. Secondly is the linearity assumption, which states that independent variables should have a linear relationship with the dependent variable. Thirdly is the assumption of independent errors, which states that errors should not be correlated. The fourth and last assumption is that there should not be multicollinearity 1 . Table 3 indicate that the majority of French bean farmers, irrespective of Global-GAP certification status, were poor (72.6%). The possible reason is that income from French beans was not sufficient to move the households out of poverty brackets due to low acreage cultivated. Poor farmers cultivated an average of 0.5 acres of French beans while non-poor ones cultivated an average of 0.5 acres as indicated in Table 4.

Descriptive characteristics of the respondents
Socioeconomic and institutional factors were considered in comparing the poor and non-poor groups of farmers. Given the poverty line, both poor and non-poor French beans did not statistically and significantly differ in terms of household size, age, household adult equivalent, total land size owned, distance to the market, years of experience in farming, total annual asset value, annual expenditure on non-food items, number of training attended and the acreage under French beans. On the other hand, the poor and non-poor farmers differed based on costs incurred per acre of French beans, net income per acre of French beans, annual net crop income, annual net livestock income, annual off-farm income, annual total income, annual expenditure on food items, and annual total expenditure per adult equivalent (Table 4).
Relative to the poor, non-poor had the highest net annual French beans income ( Poverty status of the respondents statistically and significantly varied according to the location (p = 0.020), an education level (p = 0.000) and risk preferences (p = 0.026) of French beans farmers. On the other hand, poverty status did not vary according to Global-GAP certification status, gender, and marital status. The majority of poor farmers are located in Kirinyaga West Sub-County (83.2%) followed by Kirinyaga East Sub-County at 75%. Mwea West Sub-County had the lowest poverty rate at 60%. The majority of those who were poor (88.9%) had no education at all while the majority of those who were non-poor had certificate and diploma (Table 5). Education enables French beans farmer"s access to formal employment, which later increases their income and thus reduces poverty. Farmers can also use the knowledge acquired from the education process to improve farming activities such as identification and uptake of high yielding agricultural technologies, which eventually increases farm yields and income and hence poverty reduction.
Finally, among the poor French beans farmers, the majority never like taking risks (87%) and in most cases do not like taking risks (77.6%) while among the nonpoor, the majority always liked taking risks (40%). The results show that aversion to risks positively relates to poverty. The findings concur with those of Hulme and Shepherd (2003) who found that risk-averse persons in the face of risks are more likely to be poor because they prefer low risky ventures which are associated with low returns. Mosley and Verschoor (2005) found a very weak link between farmers" risk attitudes and poverty.

Determinants of observed poverty
A binary logistic regression analysis was conducted to determine the factors affecting observed poverty among French beans producers in Kirinyaga County, Kenya ( Table 6). The dependent variable (poverty status) was captured as poor = 0 and Not poor = 1 based on the poverty line of KES 193.56 ($1.90 per day per adult equivalent. Hosmer and Lemeshow Test are statistically insignificant (p = 0.281) indicating that the model fits the data well. The model is statistically significant, indicating that the explanatory variables estimated reliably distinguished between the poor and non-poor (Chisquare ( χ 2 ) = 153.314, p = 0.000). Nagelkerke R-square value is 0.626 indicating that 62.6% of the variation to be observed in the poverty situations of the French beans Poverty status s was generated using the international poverty line of $1.90 per day per adult equivalent or KES 193.56 per day per adult equivalent. ***p < 0.01, ** p < 0.05, *p < 0.1. farmers were explained by the combined effects of all the independent variables in the model specified. Binary logistic regression is based on four crucial assumptions that need to be addressed. First, the dependent variable should be ordinal. In this study, the dependent variable was captured as a binary variable. Secondly is the linearity assumption. Linearity in the binary logistic model assumes that independent variables have a linear relationship with the dependent variable. This assumption can be verified by checking the model fit statistics and pseudo-R-Squared 2 .
In this study, the model fit statistics and pseudo R 2 in Table 6 indicate that the model well fits data. Also, binary logistic regression does not need a linear relationship between the dependent and independent variables because it applies a non-linear log transformation to the predicted odds ratio 3 . Thirdly is the assumption of independent errors. The assumption of independent errors states that errors should not be correlated for two observations. That is, data should be drawn from independent samples and not dependent samples such as before and after or matched pairings. In this study, 2 See www.restore.ac.uk/srme/www/fac/soc/wie/researchnew/srme/modules/mod4/9/ index.html. 3 See www.statisticssolutions.com/wp-content/uploads/wp-post-to-pdfenhanced-cache/1/assumptions-of-logistic-regression. single cross-section data was drawn from an independent sample. The fourth assumption is that there should not be multicollinearity. Binary logistic regression requires that independent variables should not be highly correlated with each other, but to some degree. Correlation analysis is one of the tests that can be used to check the existence of multicollinearity. Pearson"s value (r) that is equal to 0.8 or above indicates a serious problem of multicollinearity 4 . In the study, correlation analysis was conducted, and results presented in Appendix A1. The results show that all the Pearson"s values (r) are below r =0.8, and hence, multicollinearity was not a problem.
Variables indicating certification status and net income from French beans are statistically insignificant, with a positive relationship with poverty reduction. The results suggest that income earned by the farmers from participating in the production of Global-GAP certified French beans was not sufficient to move households out of poverty brackets. The reason may be due to the low acreage of French beans cultivated. Risk preference variable was coded using a 5-point Likert scale starting from "I never like taking risks" to "always I like taking risks." A response indicating, "I never like taking risks" is statistically significant (p = 0.051) with a negative Poverty status means poverty status of farmers was generated using the international poverty line of $1.90 per day per adult equivalent or KES 193.56 per day per adult equivalent. ***p < 0.01, ** p < 0.05, *p < 0.1. coefficient (β= -2.802). The results suggest that French beans risk-averse farmers are likely to be poor and vice versa. The odds ratio = 0.061 means that, ceteris paribus, a farmer who is risk-averse increases his/her log odds of becoming poor by 0.061 times. The findings are in line with those of Moscardi and De Janvry (1977), Dillon andScandizzo (1978), Binswanger (1980), Antle (1987), Hulme andShepherd 2003, Dohmen et al. (2005), Booij andDe Kuilen (2009), Kwesi et al. (2012), and Ghartey et al. (2014) who found that risk-averse farmers are more likely to be poor.
Household size variable is highly significant (p = 0.000) with a negative coefficient (β = -5.032). The results revealed that an increase in household size increases farmer"s chances of becoming poor. The odds ratio value of 0.007 indicates that ceteris paribus, an increase in household size by one adult equivalent increases household log odds of becoming poor by 0.007 times. An increase in household size constraints existing incomes. Reduction in income reduces household consumption expenditure and hence an increase in household poverty. Previous studies in Kenya and elsewhere report similar findings. For example, Oyugi et al. (2000), Nyariki et al. (2002), Alemayehu et al. (2005), Geda et al. (2005), Muyanga et al. (2006), Muriithi (2008), Elhadi et al. (2012), Onyeiwu andLiu (2013), Achieng" (2014) and Macho (2015) found that an increase in household size, directly and indirectly, increases household poverty through reduction in income per adult equivalent which eventually impairs standard of living. The findings also concur with those of Swanepoel (2005) and Igbalajobi et al. (2013) who found that large family size with more dependants increase the severity of poverty because it decreases per-capita expenditure. A study by Megersa (2015), however, reported contrary findings. Variable indicating daily household annual expenditure per adult equivalent is statistically significant (p = 0.001) with a positive coefficient (β = 2.016). The results suggest that households that spend more are less likely to be poor. The odds ratio = 7.509 means that ceteris paribus, an increase in annual expenditure per adult equivalent by KES 1 decreases household log odds of becoming poor by 7.509 times and vice versa. Expenditure is a welfare indicator, such that households, which spend equal to or above the poverty line of KES 2,900 per month per adult equivalent, are regarded as non-poor and vice versa. It is clear in Table 4 that non-poor and poor households statistically and significantly deferred in terms of household expenditure per annum. Poor households spent KES 89,141 per annum relative to non-poor counterparts who spent KES 141,683 within the same period. Similar studies indicate that household"s expenditure on aspects such as education increases their chances of access to well-paying jobs, which in turn increases household income and hence reduction in poverty. Expenditure on health and household food increases household member"s productivity, which in turn translates to high income and consequently, reduction in household poverty (Kiiru, 2010;Edoumiekumo et al., 2014).
Variable indicating access to credit is statistically significant (p = 0.086) with a positive coefficient (β = 1.187). The results revealed that access to credit reduces household poverty among French beans farmers. The odds ratio = 3.277 means that ceteris paribus, access to credit reduces household log odds of becoming poor by 3.277 times and vice versa. Credit access provides capital to purchase key inputs of production, which increase yields, income, and savings, and eventually a reduction in household poverty (Igbalajobi et al., 2013). Access to credit also provides income for spending on necessities such as medical, school fees, food, social emergencies, and farm inputs (Muyanga et al., 2006;Owuor et al., 2007). Furthermore, access to credit enables farmers to acquire modern farming techniques and good farm management principles which can improve farm productivity and thus poverty reduction (De Janvry and Sadoulet, 2000;Apata et al., 2010). A study by Machio (2015) found no effect.
Variable denoting household asset value is statistically significant (p = 0.051) with a negative coefficient (β = -0.482). The results suggest that as households accumulate more assets, their chances of becoming poor increases. The odds ratio = 0.618 indicates that, ceteris paribus, an increase in household asset accumulation by KES 1 increases household log odds of becoming poor by 0.618 times and vice versa. Two types of assets exist: assets that generate income for the households and those that do not. Accumulations of assets that do not generate income do leave households with little or no cash to transact daily household needs, and this may lead to increase in household poverty. In this case, French bean farmers seem to be accumulating assets that do not generate income and hence their high chance of becoming poor. Similar findings are reported in Achieng" (2014) who found that, ceteris paribus, an additional high valued asset positively influences severity of poverty by 0.280 times among French bean farmers in Kirinyaga County while contrary findings are reported in studies conducted by Mariara (2002), Muyanga et al. (2006) and Mbakahya and Ndiema (2015) in Kenya. For instance, a study by Mariara (2002) found that asset accumulation is critical in poverty alleviation among pastoralists in Kajiado County, Kenya.
Variable indicating total household annual income is statistically significant (p = 0.000) with a positive coefficient (β = 7.733). The results demonstrate that an increase in household income decreases its chances of becoming poor. The odds ratio = 2282.8 means that the odds ratio in favor of not being poor decreased by a factor of 2282.8 per unit increase in annual household income. The findings concur with those found in Elhadi et al. (2012). The study revealed that income diversification significantly reduces household poverty. Oyugi et al. (2000), Alemayehu et al. (2005), Burke et al. (2007) and Machio (2015) demonstrated that livestock provides milk, meat, and other products which increases household income and subsequently reduces household poverty. Similarly, Geda et al. (2005), Muyanga et al. (2006), Achieng" (2014, and Mwende (2016) found that an increase in income from crop diversification activities significantly reduces household poverty. However, contrary findings are reported in Mwabu et al. (2000) and Machio (2015) who found that dependence on agriculture and cash crops respectively increases the probability of farmers being poor.
Variable indicating the age of household head is statistically significant (p = 0.066) with a negative coefficient (β = -2.009). The results means that an increase in the age of household head increases his/her chance of becoming poor and vice versa. The odds ratio = 0.134 shows that ceteris paribus, an increase in the age of household head one year, increases log odds of a household becoming poor by 0.134 times and vice versa. As the age increases, the productivity of household head decreases due to poor health associated with old age. The findings concur with those in Barrientos (2007), Mwanyangala et al. (2010), Harvison et al. (2011), Muleta andDeressa (2014) and Khamaldin et al. (2015). The studies revealed that the aging of the household head tends to increase the household probability of falling into poverty. Contrary results are reported in Akona (2014), who found that an increase in the age of household head significantly reduces household observed poverty. The study argued that as the household head grows older, he/she should accumulate more income that is sufficient to move their households out of poverty. Deressa (2013) concur with Akona (2014) that as the age of the household head increases his/her skills, experience, and assets and thus, the low probability of falling into poverty.

Policy implications
Since the positive relationship exists between risk taking and poverty reduction in the face of Global-GAP certification, French bean farmers are advised to continue producing and expanding acreage under Global-GAP certified French beans. Risk and loss averse farmers are encouraged to venture into production of Global-GAP certified French beans because it is a profitable venture that can eventually reduce household poverty. French bean farmers should also diversify household income sources in order to increase household income and hence poverty reduction. Both national and County Governments in collaboration with financial institutions (insurance companies and banks) should develop insurance and credit products relevant to farmers producing vegetables for export. This will mitigate aversion to risks and a lack of capital respectively among vegetable farmers.