Impact of Non-structural Flood Control Measures on Household Welfare in Bunyala Sub-County, Kenya

Abstract

Floods are the most destructive water-related disasters considered to have dire consequences on the livelihoods of the affected population. Structural and non-structural measures have been implemented as mitigation strategies to help cope with these disasters. Given the magnitude of the disaster from floods in Busia county, multiple agencies have come into play with alternative mitigation strategies. The strategies which directly engage the participation of the community are non-structural, which include flood forecasting and early warning systems, land-use planning with zoning, savings and credit schemes, and rainwater harvesting. Despite the fact that non-structural measures are considered sustainable, households in Bunyala sub-county are still struggling with the negative impacts of floods. It is therefore of interest to establish the welfare gains or profits households derive from using these measures. Descriptive statistics were used to analyze the variables of interest. Propensity score matching (PSM) is used to determine the impact of non-structural measures on household welfare through STATA software. The key finding is that households that participate in non-structural measures have their consumption expenditure reduced compared to non-participants. This, then concludes that the non-structural flood control measures have a positive impact on household welfare.

1. Introduction

Flood is a water-related disaster that accounts for 54% of all water-related disaster globally and 23% in Africa [1]. Flooding may be a result of torrential rainfall or manipulation of catchment areas. Activities such as deforestation and upstream land degradation may cause excess surface water run-off, hence flooding. Floods in their most immediate effects cause the destruction of property and infrastructure, loss of lives, displacement of people, and disruption of socioeconomic activities [2]. Secondary effects associated with floods include the outbreak of water and vector-borne diseases during and after floods, loss of income, disruption and setback of ongoing development programs, and disruption of normal family life [3]. Poor households suffer the most since they rely on short-term strategies that are not sustainable and cannot adapt to long-term measures [4]. In Kenya, climate variability is commonly influenced by the complex and varied topography, altitude, lake, and sea breeze. Additionally, the complex tropical climate varies significantly between regions due to regional climatic processes such as migration of inter-tropical convergence zone (ITCZ) [5].

Major flood events that have occurred in Kenya have been documented. The 1961 floods represent one of the early attempts of studies initiated to measure the extent and magnitude of the menace. Low-lying areas were extensively inundated and widespread damage to homesteads, bridges, and other facilities was experienced. Other years affected by floods were 1968, 1977/1978, 1985, and 1990. Details of these floods are however missing [6]. The 1997/1998 El-Nino floods were also one of the major events that demonstrated to Kenya the severe devastation of the floods. This flood was further associated with land degradation, increased soil erosion, and erosion of riverbanks among other damages. In April/May 2003 floods, Bunyala was also one of the hardest-hit areas. There was complete inundation of the area; data loggers, water level recorders, bridges, and river gauging structures were carried away, affecting communication and monitoring activities. Recently, flooding frequency has increased and ranges now between 2 and 5 years. Recent noteworthy events are in the years, 2007, 2011, 2015/2016, 2018, and 2020. Approximately 50,000−150,000 people are affected each year by floods in the past decade [7].

Flood disaster management is a multifaceted approach. It involves several disciplines such as public policy and planning, economics, statistics, hydrology, psychology among others. This is because of the impacts it has on the socio-economic livelihoods of people it affects directly or indirectly [8]. Most of the measures resorted to by affected communities are of a preventive or corrective nature intended to minimize damage caused by floods [9]. A comprehensive approach to disaster management should include four basic phases: preparedness, mitigation, response, and recovery. Although most countries are focused on the last two phases, the greatest potential for minimization of economic losses and reduction of disaster vulnerability, especially among low-income groups typically lies with preparedness and mitigation [10, 11].

2. Non-structural flood control measures

Measures taken to mitigate the effects of floods can be categorized into structural and non-structural measures. Structural measures include the construction of reservoirs, dykes, diversion channels, and spillways among others. The measures are meant to reduce the effects of the flood hazard by modifying the environment through construction. However, they are associated with residual risks due to the possibility of structural failures. Non-structural measures include land use planning with zoning, water harvesting, insurance schemes, flood early warning systems, and awareness campaigns [12]. These are meant to manage vulnerability hence essential to eliminate residual risks. Kundzewicz [13] reports that these measures agree better with the spirit of sustainable development in that the objectives are economically attainable, socially acceptable, and environmentally sound. Although considered complementary to structural measures, non-structural measures may also be used alone if structural measures are not considered feasible and depending on the condition of the river basin as they may be more cost-effective [14]. They also have different implementation periods: prior, during and post-flood occurrences [15].

2.1 Land use planning with zoning

Subsistence farming is the main land activity for households in the area with farms averaging 2 acres. The main food crops grown in this area include maize, sorghum, and beans. Livestock farming mainly involves the rearing of free-range indigenous chickens at a small scale ranging from 5 to 20 birds per household. Households also engage in sand harvesting as a source of livelihood (Busia [16]).

Land use planning involves regulating how land is used and zoning endangered resources in order to ensure efficient resource use and desirable environmental outcomes.

2.2 Water harvesting

Water harvesting is among the measures earmarked for support by KIWASH through WRUAs in Busia County through the construction of water intakes, storage tanks, distribution pipelines, and pumping systems. Additionally, the program was envisaged to support capacity building among stakeholders to ensure that households benefit through access to water and awareness creation on the protection of water sources [17]. Households in Busia county have embraced small-scale water harvesting, which involves redirecting, storing, capturing rainfall, runoff, and groundwater [18].

2.3 Savings and credit schemes

Owing to the risks of floods for people residing in Busia and the possibility of destruction of crops, livestock, and homesteads, there is a need for some form of insurance. This enables households to be less vulnerable to future poverty as they can smooth their consumption in the presence of shocks and bounce back after the disaster through individual claims in case of floods. Insurance policies set in Busia county is known by very few and additionally, they would not be able to afford it. Households are therefore largely engaged in savings and credit schemes as a way of insurance from floods through engagement in social groups and cooperatives [19].

2.4 Flood forecasting and early warning systems (FEWS)

Flood forecasting and early warning systems (FEWS) is a tool that cover flood levels, likely impacts of a flood, disseminating warning messages as well as reviewing the effectiveness of the system following an event. There are significant numbers of institutional initiatives currently active in the African continent. Information regarding many of these initiatives is not publicly accessible, which results in underestimation by the wider scientific community of the amount of flood forecasting activity undertaken in the continent [20]. In 2008, the Western Kenya Community-Driven Development and Flood Mitigation (WKCDD & FM) Project was initiated by the Kenyan government with support from the World Bank to address flooding problems. This project collaborated with the Kenya Meteorological Service (KMS) which took to establishing the flood early warning systems [21].

The FEWS developed was motivated by experiences from the lake Victoria basin which recommended the integrated flood management approach. It was reported that in the 2011 flood events, there were no casualties in Bunyala sub-county as communities were warned early enough. Additionally, 1 million USD is saved annually as a result of the flood mitigation policies that were implemented. However, a picture of policy influence at the household level is lacking.

This system consists of four components that complement each other in ensuring that the system is complete and effective. These components are explained as follows:

Risk knowledge: this involves the systematic assessment of risks and hazards; mapping their trends and patterns. Upon understanding these risks, weather and river gauging stations are established considering the basin’s topography, geology, and soils [12]. There are three synoptic stations in the Nzoia basin: 16 automatic hydrometric systems, three rainfall stations, and three radar water level stations: Rwambwa, Semogere, and Webuye [22].

Figure 1 shows the distribution of synoptic stations in the Nzoia Basin. These resources are used for data acquisition, which helps in the flood diagnostics and forecasting center. 1EF01 is the main pour point, while 1 BC01 and 1DA02 are sub-basin discharge stations.

Figure 1.

Location of hydrometric and rainfall radars in the Nzoia Basin.

Monitoring and warning system: This involves establishing sensors measuring water levels at relevant sites in local waterways and linking them to the local database. This system consists of data collection, observation, rapid communication system, processing, analysis, and database management systems. This center should be operational 24 hours a day, 7 days a week, all year round and have skilled personnel in hydrology and hydrometeorology and advanced meteorological telecommunication facilities to ensure access to data [23].

Dissemination and communication: Information is then disseminated to the dissemination center from where it is distributed to the target population and stakeholders. In Kenya, the flood warning and dissemination centre (FWDC) is responsible for this function. In the Nzoia basin, some users of this information include the public, Bunyala community, and emergency management like the REDCROSS, government sector, private sector, and civil societies among others. Some methods used to convey information include emails, internet websites, weather radio, and mobile short messages among others [24].

Response capability: Response plans by the communities and forecast review and development are also included in the components of EWS. It is important to have training and communication centers to create awareness and enhance preparedness. Communities are expected to also keep watch of the rainfall and water levels, provide indigenous information that can be integrated into EWS, provide flood information, provide information on impacts of floods, provide security for flood monitoring equipment and disseminate information in the local language [21].

Given the magnitude of flood disaster in Bunyala sub-county, various agencies and individuals have directly been involved in the implementation of the non-structural measures. Despite these efforts, households are still struggling with the negative flood effects. The purpose of this study is to there determine the benefits or losses households derive from these measures. The measure of welfare used in this study is household consumption expenditure.

3. Welfare and theories

The concept of household welfare explains the most commonly used proxies of household welfare, their advantages and disadvantages, and the reason behind choosing household consumption expenditure as the proxy in this study.

Theories are used to explain household preparedness and actions upon receiving warning information on impending risks. They capture cognitive skills, household needs, cultural and economic conditions that influence households’ decisions to use or fail to use non-structural flood control measures.

3.1 Concept of household welfare

Welfare measures allow for the estimation of patterns in standards of living across populations over time. Consumption expenditure, asset accumulation, and income are commonly used as proxies for household welfare. Income and consumption have been debated intensely by researchers with a clear consensus on favoring consumption over income [25]. First, consumption seems to be better in capturing the concept of standard of living since individuals derive material well-being from the actual consumption of goods and services rather than from income. Consumption better reflects long-term income as it is not closely tied to short-term fluctuations; it smoothens over seasons and is less variable than income [26].

Income is more likely to be affected by seasonal patterns resulting in either an underestimation or overestimation of real income. Although collecting data on consumption is usually very time-consuming, the concept of consumption is usually clearer than the concept of income. Furthermore, it is extremely difficult to accurately measure household income, especially for self-employed households and those working in informal sectors. Finally, income is likely to be a more sensitive issue for respondents than consumption. Those who are well-off are less likely to participate in the survey or respond leading to an underestimation of income inequality among the population [27].

Assets indices are also an alternative measure of welfare. In recent years, the use of asset-based wealth indices as an alternative metric has become increasingly prominent. It has been considered superior to consumption and income as wealth better reflects long-term welfare as it is less volatile than income and consumption [28]. It is suitable for analyzing multidimensional poverty and less data-intensive hence easier to calculate. These features, however, make wealth index a specific indicator such that it cannot be comparable to conventional measures of economic status. Different studies report that the asset index is a poor proxy for current household income or expenditure even though it may reflect permanent income [29]. Some reasons limiting the use of asset bases indices are: first, this index measures household wealth relative to other households in the sample but does not quantify the households’ current levels of welfare or poverty. Secondly, it has been found to have an urban bias and limited discriminatory power at the lower end of the wealth distribution. Thirdly, differences in price levels, as well as asset quality across regions, are not taken into account in the asset-based approach [29]. Wealth index, therefore, cannot be used as a perfect substitute for income or consumption, which among other considerations remain the most common and accepted measures of welfare.

3.2 Theoretical framework

A combination of theories is better in practice when explaining human behavior toward risk, especially in the natural environment [30].

The cognitive theory emphasizes the role of thinking, imagination, emotions, and values in human action. Human behavior is formed by how the person processes information perceived by the environment. According to this theory, people who receive risk information go through a sequential process that shapes their perception and behavior. Perception, in this case, refers to what people understand and believe, while the response is what people decide about alternatives in preparing and mitigating actions. This theory presumes that preparedness and mitigation behavior is a consequence of perceived risk [31].

The need theory stresses the importance that the individual places on the meaning of events and experiences. According to Maslow’s theory of human behavior based on humanistic need principles, security and safety needs are important, but not more than physiological needs. The hierarchical system of needs vary between households and communities. Unless basic needs are fulfilled, safety needs such as those against natural disasters may be considered unimportant hence unobserved mitigation behavior [32].

Another theory is the cultural theory. Culture is seen as a powerful influence as it organizes a societal social structure which in turn govern people’s behavior. Renn [33] has demonstrated that responses to risks are a function of cultural belief systems. A study by [30] reveals that some cultures may consider natural disasters as acts of God, inevitable and beyond human control even to mitigate the consequence; hence, very limited mitigation behavior and acceptance of recommendations of mitigation measures. On the other extreme, other cultures adopt a positive outlook, believing that technology and government action can mitigate their worst impact.

Lastly, the economic theory argues that human response to environmental risks is influenced by economic resources, hence, safety is a function of income and wealth [32]. Poorer people are more vulnerable to environmental risks because they live in houses with fewer safety measures or locations prone to various disasters [34]. Low income, therefore, prevents voluntary mitigation actions against risk. The attributes in these theories intertwine to determine the preparedness and mitigation decisions made by households. A household, for example, may be aware of the impending disaster and the mitigation measures recommended but are constrained financially or by cultural beliefs and so on as scenarios are different for individual households.

4. The study area

Bunyala sub-county is one of the seven sub-counties located in Busia County in western Kenya. The sub-county borders Samia to the north, Siaya to the east, Bondo to the south, and lake Victoria to the west (Figure 2).

Figure 2.

Map of the study area.

The area experiences an average rainfall of about 750 mm and 1015 mm and has alluvial soils which support small-scale agriculture of both crops and livestock. Other activities include fishing and non-farm activities, such as petty trade. The long rains are at a peak between the months of March and May, while the short rains fall between August and October. The dry season with scattered rains falls between December and February [18].

River Nzoia drains into Lake Victoria through the Bunyala plains. Bunyala is a low-lying area with a generally flat landscape that predisposes the area to recurrent floods that occur after every 2 years on average. This can be attributed to the overflowing of River Nzoia due to the bursting of the riverbanks. This situation is aggravated by the backflow of Lake Victoria due to siltation in river Nzoia causing inundation for long periods (Busia [16]). Floods affect all the locations in the sub-county, however, those closest to river Nzoia and Lake Victoria are the hardest hit. Three sub-locations closest to the lake were therefore randomly selected for study, namely, Bukoma, Rukala, and Obaro.

5. Methodology

This section involves sample size determination and distribution among selected sub-locations. Model specification gives step-by-step procedure that will be used to determine difference in household consumption between participants and non-participants in non-structural flood control measures.

5.1 Sample size determination

The population from which the sample will be drawn consists of households in Bunyala sub-county. The sample size will be determined using Cochran [35] formula.

$$n=\frac{\mathit{pq}{Z}^2}{e^2}$$

$$n=\frac{{1.96}^2\times 0.5\times \left(1-0.5\right)}{{0.05}^2}=384$$

where n - sample size.

p - Sample proportion (use 0.5 if p is not known).

q - 1-p.

d - Confidence coefficient (precision).

Z - Standard deviate at 95% significance level.

5.2 Sampling procedure

This study used a multistage sampling procedure. Bunyala sub-county was purposively selected as it is one of the most flood-prone areas in the country located in the lower reaches of River Nzoia.

Bukoma, Bukala and Obaro sub-locations were purposively selected due to their proximity to water bodies.

Sub-county staff, assistant chiefs and village elders were consulted to help in generating a list of households that have been in the area for at least 5 years as they have experienced at least two flood events. Respondents were drawn from the sub-location using simple random procedure. The sample size was divided in proportion to the size of the household population of the three sub-locations. According to the 2019 census, the estimated household population for Bukoma, Bukala and Obaro was 1558, 896, and 576, respectively (Table 1).

Sub-locations	Population	Sample Size Proportion	Participants proportion	Non-participants proportion
Bukoma	1558	197	131	66
Bukala	896	114	76	38
Obaro	576	73	49	24
Total	3030	384	256	128

Table 1.

Sample size distribution.

5.3 Model specification

Propensity score matching (PSM) was used to determine the impact of non-structural flood control measures on household welfare. In this study, households that participate in non-structural measures are used as the treated, hence participants while those who do not are used as the control group hence non-participants. The purpose was to compare the consumption expenditure for those who took up these measures and those who did not.

Upon collecting data, there was fear that selection bias may exist, hence choosing PSM for this objective. To eliminate the selection bias, PSM uses the probability of employing the treatment propensity score (PS) to match individuals in the treatment and control group. Propensity Scores remove dimensionality issues and compress relevant information into a single value, hence making it easy to match individuals. In the estimation of the predicted values of the probability of participation, a probit model will be used as shown:

$$\mathrm{P}\left({\mathrm{X}}_i\right)=Pr\left(D=1|{\mathrm{X}}_i\right)$$E1

where $\mathrm{P}\left({\mathrm{X}}_i\right)$ is the probability of participation in non-structural control measures, D = 1 for participants and D = 0 for non-participants. The regression function is as shown:

$${\mathrm{P}}_i=\phi \left({\theta }_0+{\theta }^{\prime }{\mathrm{X}}_i+{\epsilon }_i\right)$$E2

where $\phi$ is the standard normal distribution, ${\theta }^{\prime }$ is the vector of coefficients, ${\mathrm{X}}_i$ is the vector of explanatory variables; containing confounding variables that are both related to participation and outcome and ${\epsilon }_i$ is the error term which is assumed to be normally distributed.

According to Brookhart et al. [36], one should include variables that are thought to be related to the outcome regardless of whether they are related to the exposure. This is because even if a covariate is theoretically unassociated with participation, there can be some slight chance of relation for any given realization of a data set. Including such a covariate in a PS model corrects for small amounts of chance bias, hence improving the precision of the estimator.

Baseline confounders include age, gender, education level, household size and various household characteristic. The intuition behind the inclusions of outcome variables in this study are based on the elements of the Crichton risk triangle, which capture hazard, vulnerability and exposure as the three risk components (Table 2) [37].

Once the model is estimated, the balancing assumption will be tested using t-tests. The sample will be stratified by PS and tested for lack of difference between the control and treatment of each stratum. After balancing, the matching process will be done.

There are different matching techniques. First, the nearest-neighbor matching which involves matching the person with the closest PS in the control group. This type of matching can be done with or without replacement. Matching with replacement is whereby a person in the control can match more than one treated while without replacement is when once a control has been matched, it cannot be used to match another treated. Second, the caliper or radius matching involves matching individuals in the control and treatment group that lie within a bandwidth around the interested PS. Bad matches are reduced due to the bandwidth, however, if no agent is located inside the radius, then there is no match for them. Third, the stratification matching where the area around the PS overlap is partitioned into strata. Each stratum is defined over a specific range of the PS and within each stratum, there is no statistically significant differences between the treatments and control groups. Lastly, the Kernel technique where the weighted average of all observations in the control group is used to create matches for the members of the treatment group. The greater the distance between the PSs, the lower the weight. In this model, all members of the control group are used to create a counterfactual for the treatment hence, bad matches will be included. The weighting process however reduces the influence of bad matches. Bandwidth is very important here as it determines the degree of smoothing, however, it is unclear what the correct bandwidth is hence its selection is treated as a tradeoff between bias and variance.

The nearest-neighbor method will be used to match the control and the treatment group. Each treatment is matched to the suitable control with the closest PS. However, it may be that the nearest neighbor is very far in terms of the PS. Matching with replacement will be used to address this issue hence ensuring the reduction of bias.

After matching, the average treatment effect (ATE) is estimated. The average outcome of the treatment is compared to the average outcome of the control group. The difference between the outcomes is the impact of non-structural measures on households as shown:

$${\delta }_i=y_{1i}-y_{0i}$$E3

hence,

$$\mathit{ATE}=E\left({\delta }_i\right)=E\left(y_1-y_0\right)$$E4

where $y_1$ is the outcome of the participants and $y_0$ is the outcome of non- participants.

Variable	Description and Measure	Expected sign
Dependent variables
monthlyexp	Household consumption expenditure; Ksh	+
Independent variables
hhAge	Age of household head age; Years	±
gender	Gender of the household head; 1 = male 2 = female	±
hhsize	Household size; Number	+
landownsp	Household land ownership; 1 = yes 0 = no	+
monINC	Household monthly income; Ksh	+
educ	Education level of the household head; 1 = primary2 = secondary3 = tertiary; 4 = none	+
hhoccup	Household head occupation; 1 = farmer2 = government sector3 = private sector4 = self-employed5 = unemployed6 = other(specify)	±
accesscredit	Access to credit; 1 = access 0 = no access	+
EWduration	Time in days	+
floodfreq	1 = always 2 = never 3 = often 4 = sometimes	—
Savings	1 = no 2-yes	+
Consecutivefloods	1 = no 2 = yes	—

Table 2.

Covariates for propensity score matching and their measurements.

6. Results and findings

6.1 Region of common support

After the estimation of propensity scores, an optimal number of blocks is identified, in this case five. This is done so as to ensure the mean of propensity score is not different between the treated and control groups in each block.

When the balancing property was satisfied, the region of common support was selected. These scores lie between 0.189 and 0.996 as indicated in the table below. Observations for participants and non-participants within the region of common support were compared. Matching takes place only in the region of common support [38]. Therefore, observations showing propensity scores below 0.189 and above 0.996 were discarded.

Table 3 reports the distribution of observations in the region of common support. From 128 non-participants (control group coded 0), 57 were dropped from the analysis; these are observations whose propensity scores were below and above the minimum and maximum scores [39] while 95 were retained. For the participants (treated group coded 1), 31 observations were dropped while 225 observations were retained.

Inferior of block of p-scores	Participation		Total
	0	1
0.1896932	13	3	16
.2	13	9	22
.4	18	18	36
.6	14	43	57
.8	10	25	35
.9	3	28	31
.95	0	99	99
Total	71	225	296

Table 3.

Participant and non-participant support distribution.

6.2 Average treatment effects on household consumption expenditure

To determine the impact of participating in non-structural flood control measures, the average monthly household consumption expenditure for participants and non-participants was compared. Matching between these two groups was attained through the nearest neighbor matching method with replacement. This is because this method is considered simple and, in case the nearest neighbor is far in terms of PS, a replacement can be found.

The household expenditure was aggregated from the following expenses: food items, toiletries, housing and energy, healthcare, education, transport and communication, insurance, gifts, donations and church offerings.

Table 4 results show there is a significant difference (z > 1.96) in the household expenditures of participants and non-participants at 5% significance level. Participants are seen to have their consumption expenditure reduce by Ksh 3860.703 compared to non-participants. It can be therefore concluded that non-structural flood control measures have a positive impact on welfare.

Monthly expenditure	Coef.	AI Robust Std. Err.	z	P > |z|	[95% Conf. Interval]
ATE; Participation; (1 vs. 0)	−3860.703	794.3921	−4.86	0.000	−5417.683; −2303.723

Table 4.

ATE effect on household expenditure.

6.3 Testing for covariate balance

There are differences that exist in observed covariates of the matched groups; hence the propensity score is used to reduce bias and balance between the treated and control group [40]. It is therefore important to access the balance of the measured covariates between the two groups. Balance means similarity in covariate distribution [41].

Covariate balance was checked after estimating the average treatment effect; to check if observations have the same distribution of estimated propensity scores. Figure 3 reports the covariate balance using propensity scores. The factors used are as follows: household occupation, household size, education level, household age, savings, and access to credit, gender, monthly income, flood frequency, early warning duration, consecutive floods and land ownership. A perfectly balanced covariate has a standardized difference of zero and variance ratio of one [42].

Figure 3.

Covariate balance on propensity scores.

The density plot for the matched sample is nearly the same, implying that matching on the estimated propensity score balanced the covariates. The conclusion is that the covariates were well balanced and distributed in matching the participants and non-participants.

7. Conclusion and recommendations

The objective of this study was to determine the impact of non-structural flood control measures on household welfare, which was measured using consumption expenditure. Primary data were collected from 384 randomly selected respondents. Among the respondents, 256 were participants (treatment group) while 128 were non-participants (control group). To estimate the impact of these measures, household consumption expenditure was compared between the two groups, Average treatment effect (ATT) was calculated using PSM.

Results showed that the participants had their expenditure reduced by USD 38.6. Households have however reported struggles with these measures. Environmental degradation, especially land through deforestation, sand harvesting, farming along the banks, and pollution of watershed catchment areas, are features that make communities in Bunyala more vulnerable to floods [43]. I recommend that households should be given awareness and support on how to protect land and water resources so as to reduce the vulnerability of communities.

Water harvesting has also been mainly small scale and households rely on river or flooded water for consumption due to lack of resources to explore better water harvesting options. I recommend that the government in partnership with community-based organizations (CBOs) and non-profit organizations (NGO) should combine efforts in helping households install better water storage facilities.

Savings and credit schemes strategy, due to the low incomes, households tend to save less or nothing since all their income is used for consumption. I would recommend that the government with insurance companies should help households come up with a sustainable saving plan that will encourage households in the practice of saving,

For FEWS, due to unreliable information in the past, they are unable to trust this information, hence not using it. I would recommend that the KMD improve on consistency and timeliness that will relay information more accurately and hence can be trusted.

Despite all the challenges, these measures have proven essential in improving welfare.

Acknowledgments

I would like to recognize and thank my supervisors, Professor Hillary Bett and Job Lagat for their support, guidance, and contribution to this paper. I would also like to that my colleagues for their ideas and the encouragement they have accorded me, this has given me much confidence in my work.

I would also like to appreciate AERC and the support team for the research grant that has assisted a great deal in data collection. The AERC support staff for their support and encouragement.