Determinants of pastoralists’ participation in rangeland restoration in Laisamis and Saku sub-counties, Marsabit County, Kenya

1 Introduction

Rangelands, a subsection of drylands, are defined as areas where the ratio of precipitation to potential evapotranspiration (Aridity Index) is 0.65 or less (Eldridge and Sala, 2025) and encompass a variety of ecosystems under four classes, namely dry sub-humid, semi-arid, arid and hyper-arid lands (Haddad et al., 2022). Globally, the drylands support 44 per cent of the world’s crop cultivation, are home to over 2 billion people and support around half of the world’s livestock farming (UNCCD, 2017). Drylands also contain a highly adapted variety of biodiversity, which has global economic importance and provides crucial ecosystem services that are vital for both local and global sustainability Davies et al. (2012). However, these ecosystems are increasingly threatened by anthropogenic degradation and climate change, which pose significant challenges to pastoralist communities and global environmental sustainability as enshrined in the Sustainable Development Goals (SDGs) (Schrieks et al., 2024).

Efforts to combat land and rangeland degradation have gained significant momentum through international collaboration since the 1977 United Nations Conference on Desertification (UNCOD), laying the groundwork for future global actions and policy frameworks. The legally binding United Nations Convention to Combat Desertification (UNCCD) specifically addresses land degradation in arid, semi-arid, and dry sub-humid areas, including rangelands. Rangelands have also been emphasized by the United Nations (UN) Decade on Ecosystem Restoration (2021–2030) due to threats to unique ecosystems (UNEP and FAO, 2020). These commitments are reinforced by Sustainable Development Goals (SDG) indicator 15.3.1, which targets Land Degradation Neutrality (LDN) (UN, 2015). Africa has initiatives such as the African Forest Landscape Restoration Initiative 100 (AFRI100) (FAU-AUDA-NEPAD, 2021) and the Great Green Wall Initiative, both seeking to improve resilience and livelihoods through the restoration of degraded landscapes (Goffner, 2019).

The relationship between pastoralism and rangeland degradation has long been contested in both scientific literature and policy discourse. Colonial and post-colonial narratives frequently portrayed pastoral practices, particularly livestock mobility, as inherently destructive to rangeland ecosystems (López-i-Gelats et al., 2016). This perception informed policy interventions that emphasized destocking, sedentarization, and the privatization of communal rangelands (Lankester and Davis, 2016). However, contemporary rangeland science increasingly recognizes pastoral systems as adaptive strategies that have evolved to manage spatially and temporally variable resources in non-equilibrium environments (Semplici and Campbell, 2023). Emerging evidence indicates that rangeland degradation stems not from pastoralism itself but from the disruption of traditional mobility patterns and customary management institutions resulting from changes in land tenure, population growth, and restrictive land-use policies (Boles et al., 2019; Unks et al., 2025).

Like many developing countries, Kenya faces growing concerns over environmental degradation, particularly the deterioration of its rangeland ecosystems. In response, the Kenyan government has identified rangeland restoration as a strategic priority for promoting environmental sustainability, enhancing ecosystem resilience, and improving pastoral livelihoods (Warui and Kshatriya, 2009; Ndiema, 2025; UNDP Kenya, 2024; TRI, 2024). The rangelands in Kenya occupy more than 70% of the country. The rangelands are home to the bulk of Kenya’s national parks and reserves, rich in fauna and flora (Nyongesa et al., 2023) which are key to the Kenyan tourism industry. They contribute over 12% of the 40% agricultural output. Furthermore, over 75% of cattle herds in Kenya are reared in the rangelands. The economic worth of Kenya’s pastoral sector is approximately US$1.13 billion, with the livestock sector accounting for 92% (US$1.04 billion) and the non-livestock sector accounting for 8% (US$0.0903 billion) (Nyariki and Amwata, 2019). However, the rangelands in Kenya have undergone subdivision, fencing, and conversion of rangelands for agriculture and urban development, limiting access to resources and disrupting wildlife connectivity, exacerbating degradation (Tyrrell et al., 2022). Degradation continues due to overgrazing and the spread of invasive species like Acacia reficiens and Opuntia stricta (Kimiti et al., 2020), poor resource management, and diminishing traditional ecological governance (Nyongesa et al., 2023). The persistence of rangeland degradation despite government-led restoration efforts raises concerns about the adoption of restoration practices among pastoralist communities and highlights the need to unearth the factors influencing their participation in restoration activities.

Thus, given the economic importance of Kenya’s rangelands and their continued degradation, it is imperative to among other things identify the factors that can influence the adoption of restoration practices in order to support government restoration initiatives. The success of these initiatives depends largely on meaningful local participation and the integration of indigenous knowledge into restoration planning and implementation (Kihiu, 2016). Despite increasing recognition of these considerations, significant knowledge gaps remain regarding how pastoral communities engage in rangeland restoration and the psychological, institutional, and socio-economic factors that shape their restoration behavior in Kenya.

Research on rangeland management and restoration has widely applied the Protection Motivation Theory (PMT), using its core constructs of threat appraisal (perceived vulnerability and perceived severity) and coping appraisal [self-efficacy, response efficacy, and response cost] to explain pro-environmental behavior among pastoral communities (Fenetahun et al., 2021). In this study, the PMT framework is extended by incorporating two socio-political constructs—perceived political willingness (PPW) and perceived incentives (PI)—to better capture the broader contextual factors influencing behavioral intentions and restoration actions (Walker et al., 2024).

Although political willingness is frequently acknowledged as an important external driver of environmental action, it has rarely been operationalized or empirically examined, resulting in limited understanding of how government commitment, policy support, and stakeholder engagement influence restoration actions. (Bostedt et al., 2023; Chepng′etich et al., 2025; Sainz-Santamaría, 2026; Ogwu et al., 2025). Likewise, perceived incentives, defined as the economic benefits individuals expect to receive from undertaking restoration activities, may provide an additional motivational mechanism that encourages pastoralists to adopt restoration practices. Accordingly, PPW and PI were integrated into the PMT framework to examine their contribution to restoration behavior. Their inclusion provides a more comprehensive understanding of how governance support, institutional trust, stakeholder engagement, and expected economic benefits shape restoration decisions (Wainaina et al., 2021; Tedesco et al., 2023).

This study therefore advances the PMT framework by explicitly incorporating government commitment and economic incentives as determinants of restoration behavior. Integrating these socio-political dimensions is expected to enhance the explanatory and predictive capacity of PMT while providing a more contextually relevant framework for analyzing restoration behavior in pastoral systems. It further elucidates the governance structures, incentive mechanisms, and individual behavioral processes that underpin sustainable land management. By accounting for both internal cognitive processes and external institutional influences, the extended PMT offers a more nuanced explanation of how pastoralists respond to environmental threats and restoration opportunities. Consequently, the framework provides a stronger basis for designing culturally appropriate, evidence-based, and policy-relevant interventions, enabling policymakers to develop more effective risk communication strategies and restoration programs that support large-scale rangeland restoration and the achievement of Land Degradation Neutrality targets (Weyrich et al., 2020; Grelle and Hofmann, 2024; Nyanghura et al., 2024).

This paper makes three important contributions to the literature on environmental management and sustainability. First, it applies the Protection Motivation Theory (PMT) to examine factors influencing pastoralists’ rangeland restoration behavior. Specifically, the study extends the PMT framework by incorporating two additional constructs (perceived political willingness and perceived incentives) that have received limited attention in previous empirical research.

Second, while most existing studies on rangeland management have primarily focused on the influence of demographic characteristics (Tafesse, 2018; Mapiye et al., 2025), this study argues that such an approach provides only a partial understanding of restoration behavior. In addition to demographic factors, psychological factors (Kwakwa et al., 2023; Kwakwa et al., 2024), expected economic benefits (Ahmadi and Mirzaei, 2025) and government commitments (Finca et al., 2023) identified in the broader pro-environmental behavior literature as important determinants of environmentally responsible actions, should also be considered in explaining pastoralists’ restoration decisions.

Finally, within the Kenyan context, this study expands the existing body of literature on pastoralists by shifting the focus from studies that have examined factors influencing fodder purchase decisions for livestock (Mohamed et al., 2020), herd size maintenance during drought (Githu et al., 2022), adaptation strategies to changing livelihood conditions (Bostedt et al., 2023) and the determinants of livestock market channel choices (Ndiritu and Gichuki, 2025), to an investigation of the drivers of pastoralists’ pro-environment behavior in the form of rangeland restoration. In doing so, this study provides new insights into the factors that promote sustainable rangeland management practices and consequently contribute to the achievement of sustainable development among pastoral communities.

2 Materials and methods

2.1 Study area, sampling technique, and data collection

The study area, comprising Laisamis and Saku sub-counties, is located in Marsabit County, Kenya. The county is a vast and ecologically diverse savannah region covering approximately 26,103.83 km². It lies between latitudes 02°45′N and 04°27′N and longitudes 37°57′E and 39°21′E (Figure 1).

The study area was selected because Marsabit County continues to experience significant rangeland degradation despite ongoing restoration and conservation efforts by government agencies and non-governmental organizations (NGOs). These two sub-counties have soils with high erodibility, which, when combined with heavy storms and winds, result in high levels of soil erosion and gullies (Nyuma and Churu, 2022). The pastoralists practice agro-pastoral farming (Gowing et al., 2020), which is continuous cropping without adequate soil nutrient replacement or adequate fallow periods. Laisamis and Saku Sub-counties are within Kenya’s arid and semi-arid lands (ASALs) where climate variability, fragile ecosystems, pastoral livelihoods, and rapidly changing land governance intersect to create highly dynamic land systems that are increasingly recognized as global priorities for sustainable land management and restoration research. Pastoralism and livestock-related activities form significant sources of livelihood for the inhabitants of these areas. Rangelands constitute critical natural resources supporting these activities.

Within Saku is the Marsabit National Reserve (a sub-humid montane forest ecosystem). This ecological anomaly makes it a biodiversity hotspot and serves as a critical water catchment area for surrounding ASALs (Leslie et al., 2018). However, a mix of mobile and sedentary pastoralism has encouraged unsustainable agro-pastoral practices, leading to overgrazing (Kameri-Mbote et al., 2021). A recent report has also shown that the vegetation in Laisamis county has become very poor, characterised by severe to extreme vegetation deficits (National Drought Management Authority, 2026). This situation poses a serious risk to livestock production and the livelihoods of many people in the area.

Although there is evidence of county government rangeland management efforts such as the formation of Environmental Management Committees (EMCs) (Warui and Kshatriya, 2009; Ndiema, 2025), restoration projects by UNDP/ GEF ((UNDP Kenya, 2024), and The Restoration Initiative (TRI) project (TRI, 2024), restoration efforts is not yielding the desired outcome since degradation of the rangeland continue to intensify.

The target population of this study was pastoralists in the Laisamis and Saku sub-counties. According to the Government of Kenya Census report of 2019, the total population is approximately 114,894, with females totaling 56,265 while males are 58,615. The data were collected in January–March 2025.

However, pastoralists’ settlements are dynamic spatially and temporally, which renders conventional sampling frames inadequate. Other studies have collected data among pastoralist communities through various innovative approaches such as spatially-balanced sampling (Dumelle et al., 2023), peer-driven sampling, geospatial sampling (Wild et al., 2019), venue-based sampling (Dery et al., 2024), ethnographic mapping (Parkin et al., 2021), and compact segment sampling (Gass et al., 2020).

Given the spatially dispersed settlement patterns and mobility characteristics of pastoralist communities, a geospatially assisted and community-validated sampling procedure was adopted to facilitate comprehensive household data collection and minimize the exclusion of remote, dispersed, or administratively under-recorded homesteads (Gnyawali and Sigdel, 2025). The study covered Laisamis and Saku sub-counties, comprising five and two administrative units, respectively. To establish an up-to-date spatial sampling frame, 10-metre-resolution land-use and land-cover change (LULCC) maps derived from fused Landsat and Sentinel-2 imagery were generated using ArcGIS to identify settled areas and potential homestead locations. Differences between the mapped settlement distribution and the 2019 census data indicated that the census did not fully reflect current settlement patterns and was therefore not used as the sole sampling frame. Identified settlements and potential homesteads were subsequently validated through formative community assessments and consultations with local key informants and chiefs, considering accessibility, security, and eligibility.

Because a reliable current household denominator could not be established due to pastoralist mobility, dispersed settlements, and limitations in administrative records, a fixed allocation of 25 household representatives was selected from each of the seven administrative units, yielding a target sample of 175 respondents. This allocation ensured comparable geographical coverage across the selected administrative units rather than proportional representation based on a known household population (Case et al., 2022). Selected respondents were invited to their respective chiefs’ offices, which served as centralized venues for questionnaire administration. Participants were verified against the sampling lists and received standardized instructions concerning the study purpose, confidentiality, voluntary participation, and questionnaire completion. Respondents completed the questionnaires independently, with interpreters and research assistants providing clarification where necessary without influencing their responses. Geographic coordinates of sampled homesteads were retained as spatial reference points.

Of the 175 targeted respondents, 15 declined participation, resulting in 160 completed questionnaires and a response rate of approximately 91.4%. During data screening, 26 questionnaires were excluded because more than 50% of the required responses were missing, while a further 12 were excluded because respondents exhibited highly uniform response patterns, selecting the same Likert-scale category for more than 40% of the relevant items. Consequently, 38 questionnaires were excluded, leaving 122 valid observations for the final analysis.

Although potential selection effects arising from non-response and questionnaire exclusion cannot be completely ruled out, no evidence from the available data suggested systematic contamination of the remaining observations (Lohr, 2022; Groves et al., 2009). To assess potential attrition bias, the distributions of respondents by gender, age category, and administrative unit were compared between the completed and final analytical samples. The distributions remained broadly comparable, suggesting that the exclusions did not materially alter the sample composition or introduce substantial systematic selection bias. Accordingly, the 122 valid questionnaires were retained for the final analysis.

Although a sample of 122 may be low, the data generated on pastoralist pro environmental behaviour and the analyzed information deliver a first set of findings that are tangible, context- and site-specific. These findings, in addition to providing crucial policy insights about SDG indicator 15.3.1 and LDN tracking, could serve as an important baseline framework and can be benchmarked in other counties with the potential to be upscaled to other African studies with similar rangeland landscapes on the restoration theme.

It is worth highlighting that all research protocols were approved by the National Commission for Science, Technology and Innovation (NACOSTI), Kenya, the Director, Kenya Forest Service (Headquarters) and the Forest Conservator, Marsabit County. All methods were carried out in accordance with relevant guidelines and regulations. Informed consent was obtained from all participants. Participants’ confidentiality and the possibility to withdraw from the study at any stage were communicated effectively during the sensitization phase. Figure 2 shows the methodological flow of the study, encompassing data collection, data analysis, interpretation and discussion of the findings, and the formulation of policy implications.

2.2 Theoretical framework

Research studies on how core PMT variables interact and influence pastoralist behavior related to rangeland restoration have indicated that it is an important subject of inquiry due to the exacerbated vulnerability of dryland landscapes to natural and anthropogenic degradation (Madadizadeh et al., 2024; Cho et al., 2025; Bayranvand et al., 2025). Building on these research studies, this study adopted the PMT to identify the factors that could affect pastoralists’ willingness to engage in restoration. According to PMT, people’s decision to engage in an act is influenced by self-efficacy, response costs, response efficacy, perceived vulnerability, perceived severity, and rewards of current practices (Baah et al., 2024).

Self-efficacy assesses whether pastoralists believe they can restore degraded rangelands. When the pastoralists feel adequately skilled in restoration knowledge, it increases their willingness to engage in activities to restore the rangeland (Baah et al., 2024). Response efficacy assesses if the pastoralists believe restoration practices can successfully reverse degradation (Haigh, 2019). When communities believe that their actions can mitigate threats, they are more likely to adopt positive environmental behaviors (Ataei et al., 2023). Response costs aims at understanding whether the pastoralists perceive restoration as too expensive or economically unfeasible (Haji et al., 2024). When entities perceive restoration activities as arduous, financially taxing, or take a long time before effects are felt, their motivation to engage declines (Liu et al., 2023). Perceived severity assesses the perception of the extent and seriousness of rangeland degradation (Madadizadeh et al., 2024). When communities feel that land degradation is severe, they are prone to undertake restoration activities. Perceived vulnerability captures the perception of one’s own risk from environmental degradation (Gebrehiwot and van der Veen, 2021). When people feel vulnerable to threats, naturally they should adopt restoration attitudes; however, maladaptive results in limited influence of perceived vulnerability in certain contexts (Schrieks et al., 2024). Rewards of current practices assesses whether individuals perceive financial benefits from maintaining their existing land use practices without SLM (Schrieks et al., 2024). When pastoralists perceive status quo practices as economically advantageous, they are more likely to oppose or not intend to engage in rangeland restoration activities (Bayranvand et al., 2025).

To contribute to the literature, we enhanced the existing PMT by including variables such as perceived political willingness and perceived incentives. These variables are underexplored, and there exist knowledge gaps on how they influence pastoralists’ restoration activities in rangelands (Bostedt et al., 2023; Chepng′etich et al., 2025). Perceived political willingness captures whether pastoralists believe the government prioritizes restoration efforts (Sharpe et al., 2021). Clear policy frameworks significantly increase community participation in rangeland restoration, leading to improved environmental outcomes (Sabogal et al., 2024). Perceived incentives indicate whether pastoralists perceive tangible economic benefits from restoration (WRI, 2022). When pastoralists are intrinsically motivated with material support, including subsidized fertilizers and seeds, access to restoration tools, or compensated for labor input, it can also substantially enhance their willingness to engage in restoration activities (El Bakali et al., 2023). Figure 3 shows the conceptual framework of the Extended Protection Motivation Theory (ePMT).

2.3 Empirical model and data analysis approach

The study adopted a 5-point Likert scale to extract information on the pastoralist’s behavior toward rangeland restoration. From a paradigm perspective, the Likert variables and assigned scales are treated as quantitative measures of latent constructs. The 5-point Likert scale ranged from “very weak” to “very strong” for the dependent variable. The assigned values were “Very weak” = 1, “Weak” = 2, “Modest” = 3, “Strong” = 4, and “Very strong” = 5.

Preliminary examination of the response distribution of the dependent variable revealed relatively few observations in the extreme categories (“Very Weak” and “Very Strong”), resulting in sparse cells. Collapsing adjacent categories increased the number of observations within each response category and improved the stability and reliability of parameter estimates in the ordered response models. Consequently, for the purpose of empirical estimation, the 5-point Likert is collapsed into three ordinal levels: weak, modest and strong under the assumption that “very weak” and “weak” are regarded as implying qualitatively similar levels of pro-environmental behavior; the same argument holds for “strong” and “very strong.” In this way, the outcome variable (i.e., pro-environmental behavior) is a 3-level ordered discrete variable where 1 = Weak, 2 = Modest, and 3 = Strong.

Accordingly, the ordered logit model (Tilahun et al., 2017; Greene, 2003) is employed to estimate the effects of PMT variables on pro-environmental behavior among pastoralists, with further focus on perceived economic incentives and perceived political willingness while controlling for relevant covariates. The choice of the ordered logit model over other methods, such as the ordered probit model, was primarily guided by considerations of interpretability and widespread application in studies involving ordinal response variables. While both models are appropriate for ordinal outcomes, they differ mainly in the distributional assumption imposed on the latent error term. The ordered logit model assumes a logistic distribution, whereas the ordered probit model assumes a standard normal distribution. In practice, the two models generally yield very similar coefficient signs, levels of statistical significance, and substantive conclusions. The ordered logit model was preferred because its coefficients can be readily interpreted in terms of odds, which is familiar to a broad readership and facilitates policy interpretation. To ensure robustness, an ordered probit model was also estimated. The results were qualitatively similar in terms of coefficient signs, statistical significance, and overall conclusions, indicating that the findings are not sensitive to the choice of link function. Therefore, the ordered logit model was retained as the main specification.

The ordered logit model is constructed on a latent (unobservable) random variable which can be expressed as shown in Equation 1:

y i ∗ = x i ′ β + ε i (1)

Where y_i* is a latent variable representing the outcome variable, pro-environmental behavior of pastoralist i; x_i is a vector of ePMT covariates comprising self-efficacy, response costs, response efficacy, perceived vulnerability, perceived severity, rewards of current practices, perceived political willingness and perceived incentives, and controls such as respondent’s gender, age, education, income level, pastoral experience, land area, for key individual demographics. β is the vector of parameters to be estimated; and ε is the random error term assumed to have a logistic distribution. Since y_i* is unobservable, the discrete responses of the observed outcome variable (y_i) are expressed in Equations 2–4 as follows:

y i = 1 if y i ∗ ≤ π 1 (2)

y i = 2 if π 1 ≤ y i ∗ ≤ π 2 (3)

y i = 3 if π 2 ≤ y i ∗ (4)

The unknown threshold parameters, π₁, and π₂ are estimated simultaneously with the coefficients. The probability that the ordered outcome variable y_i takes a different possible value is captured in Equations 5–7 as:

Pr ( y 1 = 1 / x ) = 1 − μ [ β ′ xi − π 1 ] (5)

Pr ( y 1 = 2 / x ) = μ [ β ′ xi − π 1 ] − μ [ β ′ xi − π 2 ] (6)

Pr ( y 1 = 3 / x ) = μ [ β ′ xi − π 2 ] (7)

Where μ denotes a logistic distribution and cut-points, π₁ and π₂, divide the categories of the outcome variable. The maximum likelihood method was used to estimate the parameters of the ordered logit model, which gives the coefficients and the direction of the effect of the explanatory variables. Hence, to estimate the magnitudes of the effects of the explanatory variables, the marginal effects are also estimated and expressed as follows (Equations 8–10):

dPr ( y 1 = 1 / x ) / dx = μ [ β ′ xi − π 1 ] β (8)

dPr ( y 1 = 1 / x ) / dx = μ [ β ′ xi − π 1 ] − μ [ β ′ xi − π 2 ] β (9)

dPr ( y 1 = 1 / x ) / dx = μ [ β ′ xi − π 2 ] β (10)

Thus, the marginal effects determine how the distribution of the outcome variable, that is, all outcome probabilities, varies as x i changes. The measurement of the variables is shown in Table 1.

Table 1. Description of variables and their measurement.

^aIn general, Cronbach’s alpha exceeded the minimum threshold of 0.60, which is often considered acceptable for exploratory research and newly or adapted measurements and many surveys, as it is in our case (Taber, 2018; Hair et al., 2021; Hair et al., 2019).

STATA version 10 software was used to analyze the data from the questionnaires. As indicated earlier, of the administered questionnaires, 38 were excluded due to some significant missing data and errors. This could lead to an overrepresentation of respondents who are more accessible and/or more educated, potentially skewing estimates of environmental perceptions, intentions and behaviors. Due to limited information on excluded respondents, the extent and nature of the bias remain undetermined. Therefore, the final findings should be viewed as representative of the analyzed sample, acknowledging limited generalizability to the broader pastoral population.

3 Results

3.1 Descriptive statistics

The socio-demographic profile of respondents shows a gender disparity, with men constituting (81) 66.4% and women (41) 33.6%, suggesting a male-dominated nature of pastoral activities in Laisamis and Saku rangelands. Yurco (2023) also found that men largely accomplish pastoralism in northern Kenya due to culturally gendered land administrative processes. However, the significant number of participating women (over one-third) is a clear indication of a shift toward gender equity in pastoral labor, likely driven by male outmigration and livelihood diversification pressures (Galwab et al., 2024). The age distribution shows a concentration in the under 27 (22.9%) and 27–37 years (36.1%) categories, implying a relatively young pastoral population. Education is markedly low, as 69.7% of the total population lacks formal education and only 5.7% has attained secondary education or vocational training. This low literacy level has impacts on participation in formal restoration schemes and comprehension of policy instruments such as the Land Degradation Neutrality (LDN) framework or Payment for Ecosystem Services (PES) (Gichenje et al., 2019). Economic indicators show high vulnerability, with only 1.6% exceeding KES 100,000, while 72.1% earn below Kenyan Shilling (KES or currency code KSH) 5,000 monthly. Low incomes in pastoral setups impact investment in restoration activities (e.g., reseeding, fencing, or erosion control). Mutio et al. (2025) found that income constraints are the strongest predictor of non-adoption of restoration practices in arid Kenya. Overall, our findings reveal a community characterized by low education, low income, but high experiential capital and youthful demographics, a combination that demands participatory, incentive-driven restoration policies.

To make sense of the ePMT variables, we constructed their respective averages based on pastoralists’ responses to the various related questions. Thus, in Table 2, we report these averages for the ePMT variables utilized in this study, which are then deployed in the regression analyses. This table provides strong empirical support for Protection Motivation Theory in rangeland restoration behavior. The dependent variable, pro-environmental behavior, shows a strongly skewed distribution toward the high category. Out of the 122 respondents, 76 individuals (62.81%) exhibited high pro-environmental behavior, while 32 (26.45%) fell into the moderate category and only 13 (10.74%) were in the weak category. This pattern indicates that most pastoralists in the study area are already positively inclined toward environmental protection and restoration practices. There is a strong readiness for scaling up restoration and conservation programs in the rangelands. The distribution of the dependent variable shows that pro-environmental behavior is generally strong in the study population, but not uniform.

Table 2. Descriptive statistics.

SD, standard deviation.

3.2 ePMT and the decision to participate in restoration activities (ordered logit results)

Tables 3–6 report the results of the ordered logit estimation investigating how ePMT factors influence pastoralists’ decisions to participate in weak, modest or high pro-environmental behavior (restoration activities) while controlling for relevant covariates. The last column reports the coefficients. Prior to estimating the ordered logit model, the proportional odds (parallel lines) assumption was tested using the Brant test. The global test statistic was not statistically significant (χ² = 12.45, p = 0.132|χ² = 10.71, p = 0.161|χ² = 11.02, p = 0.144|χ² = 13.58, p = 0.185|for the models in Tables 3–6 respectively), indicating that the proportional odds assumption was satisfied in all instances. Therefore, the ordered logit model was considered appropriate for the analyses.

Table 3. Baseline marginal effects.

Robust standard errors in parentheses|***p < 0.01, **p < 0.05, *p < 0.1. The loss of observations to 116 is due to 6 missing observations for Response cost, as STATA automatically uses non-missing data points.

Table 4. Marginal effects of perceived incentives on pro-environmental behavior.

Robust standard errors in parentheses|***p < 0.01, **p < 0.05, *p < 0.1. The loss of observations to 116 is due to 6 missing observations for response cost, as STATA automatically uses non-missing data points.

Table 5. Marginal effects of perceived political willingness on pro-environmental behavior.

Robust standard errors in parentheses|***p < 0.01, **p < 0.05, *p < 0.1. The loss of observations to 116 is due to 6 missing observations for response cost, as STATA automatically uses non-missing data points.

Table 6. Marginal effects of all ePMT constructs on pro-environmental behavior.

Robust standard errors in parentheses|***p < 0.01, **p < 0.05, *p < 0.1. The loss of observations to 116 is due to 6 missing observations for response cost, as STATA automatically uses non-missing data points.

In the first ordered logit model, that is, the baseline model (Table 3), the results indicate that higher self-efficacy and response efficacy significantly shift people away from low pro-environmental behavior toward high pro-environmental behavior. In contrast, the opposite is observed for response cost. A unit increase in self-efficacy and response efficacy reduce the probability of exhibiting weak pro-environmental behavior by 0.077 and 0.109 percentage points, respectively, and increases the probability of formulating strong pro-environmental behavior by 0.178 and 0.252 percentage points, respectively. In contrast, a unit increase in response cost increases the probability of weak pro-environmental behavior by 0.061 percentage points. It decreases the probability of strong pro-environmental behavior by 0.141 percentage points.

In the second and third ordered logit models (Tables 4, 5), we found perceived incentives and perceived political willingness to trigger the drive toward higher pro-environmental behavior. Specifically, a unit increase in perceived economic incentives and perceived political willingness reduces the probability of weak pro-environmental behavior by 0.0464 and 0.037 percentage points, respectively, and increases the probability of strong pro-environmental behavior by 0.121 and 0.848 percentage points, respectively.

In Table 6, regression results for the model that contains perceived incentives and perceived political willingness, traditional PMT constructs alongside the control variables show an outcome that is not much different from what is reported in Tables 3–5. The discussion of results will, however, refer to the output in Tables 3–5.

In all specifications, females are less likely to engage in pro-environmental behavior than males. Also, some level of education and income improve restoration behavior while pastoral experience and land area have no effect. Similarly, perceived severity and rewards of current practices are not statistically related to the respondents’ pro-environmental behavior.

4 Discussion

The findings from the models indicate that response efficacy, self-efficacy, perceived political willingness, and perceived incentives collectively increase restoration behavior. Conversely, response costs significantly reduce restoration behavior. Among these, response efficacy emerged as the strongest predictor in the model (β = 1.629), followed by perceived vulnerability (β = 1.203), with self-efficacy (β = 1.150) and perceived incentives (β = 0.811) occupying the third and fourth positions, respectively, and political willingness was fifth (β = 0.571). Perceived severity did not have a significant influence on pastoralist behavior toward restoration. This implies pastoralists are not responding solely to ecological concerns but responding to perceptions of government commitment, policy support, institutional backing and implementation capacity. We can therefore say that our findings demonstrate that restoration outcomes may not be driven by awareness of environmental threats (perceived severity) but by the individual’s efficacy beliefs, perceived political willingness (visible restoration activities) and perceived incentives from improved pasture and improved livestock productivity accrued directly from restoration activities.

4.1 Effect of self efficacy

Self-efficacy has a positive and significant effect on pastoralists’ decision to restore degraded rangelands. This implies that when the pastoralists feel adequately skilled in restoration knowledge, it increases their willingness to engage in activities to restore rangeland (Sircely et al., 2022). This means that respondents with strong self-efficacy are likely to engage in reforestation, be involved in re-shrub or re-grass and to practice rotational grazing for rangeland restoration, while those with weak self-efficacy are less likely to engage in rangeland restoration. Knowledge-based behavior has been found to support rangeland management, thereby directly contributing to halting degradation (Mahdavi et al., 2023). Studies indicate that self-efficacy can be bolstered by successful past restoration experiences (Bruyere et al., 2022) and extensive indigenous knowledge in rangeland management (Mahdavi et al., 2023). Further, individuals who have undergone practical training programs with strong peer networks exhibit strong self-efficacy, which leads to uptake in restoration activities. This finding is highly relevant for Kenya, where restoration success is determined by local land users’ belief in their own capacity in terms of knowledge, skills, labor, and resources necessary to implement restoration activities (Agholor, 2019). In Laisamis and Saku, high self-efficacy may be attributed to consistent NGO interventions, functioning community institutions, and perceived tenure security (Leslie et al., 2018). Since 2009, the county government, together with NGOs, supported the formation of Environmental Management Committees (EMCs), which were responsible for sustainable woody-vegetation harvesting (Warui and Kshatriya, 2009; Ndiema, 2025). The findings in this study align with studies by Kothe et al. (2019); Baah et al., 2024, who identified self-efficacy as a critical determinant of behavioral action.

4.2 Effect of response efficacy

Response efficacy demonstrated a strong positive and significant effect on pastoralists’ decision to restore degraded rangelands. This implies that when pastoralists view restoration practice as successful in reversing degradation and improving pasture, it increases their willingness to engage in activities to restore the rangeland. Since response efficacy has been shown to be a strong predictor for adoption toward restoration activities, it is suggestive that it will contribute toward vegetation recovery and productivity and therefore produce measurable LDN outcomes (Hedayati et al., 2023). The positive effect of response efficacy in the county may be attributed to several interrelated factors, including adaptive management practices like enclosures and rotational grazing, economic diversification through tourism and agriculture, and participation in community-based restoration approaches (Young et al., 2018). This is supported by Ataei et al. (2023), who stated that if communities believe that their actions can mitigate threats, they are more likely to adopt positive environmental behaviors. Other findings, such as (Schutte and Bhullar, 2017) assert that the lack of response efficacy can constrain intentions to rehabilitate rangelands. Other authors found that when pastoralists perceive some interventions as interfering with pasture productivity, skepticism might arise (Briske et al., 2024).

4.3 Effect of response costs

Our findings in the baseline model indicate that response costs exert a significant negative influence on restoration adoption intentions by pastoralists. This means that respondents who feel response cost is too high due to expensive labor inputs or long periods before returns are achievable are not likely to engage in rangeland restoration. This implies that response cost is the choke point in the restoration uptake. Case studies show that households in Marsabit experience fluctuating incomes driven by changes in climate, droughts, loss of grazing opportunities, lack of access to credit to cover seeds, fencing or labor due to lack of asset guarantee (Kimiti et al., 2020). Coordination failure among pastoral communities and inadequate customary law enforcement amplify perceived response costs (Osewe et al., 2023), which can affect restoration efforts in the study area. The findings from this study align with authors such as (Hirche et al., 2011; Liu et al., 2023) who found that higher response costs remain a central deterrent in rangeland restoration decisions and are associated with lower adoption rates of restoration practices, emphasizing the importance of economic considerations in policy design.

4.4 Effect of perceived vulnerability

The findings reveal that the more vulnerable pastoralists perceive themselves to be to the adverse effects of environmental degradation, the more likely they are to engage in rangeland restoration activities. This implies that an increased awareness of the ecological and livelihood risks associated with rangeland degradation, including declining pasture quality, water scarcity, and increased vulnerability to climate variability, motivates pastoralists to adopt restoration practices as a means of safeguarding their livelihoods. The result supports the threat appraisal component of Protection Motivation Theory, which posits that individuals are more inclined to adopt protective behaviors when they recognize that they are personally susceptible to a significant threat. Consequently, interventions aimed at promoting rangeland restoration should strengthen pastoralists’ understanding of the long-term consequences of land degradation through targeted environmental education, risk communication, and community-based awareness programs while simultaneously providing practical restoration options to translate heightened risk perception into sustained restoration action.

4.5 Effect of perceived political willingness

The findings showed PPW has positive and significant effect pastoralists’ restoration efforts. This means that respondents are aware that “there are sufficient county government initiatives aimed at preventing rangeland degradation.” Such respondents are more likely to engage in rangeland restoration activities, while those who perceive political support as inadequate are less likely to engage in rangeland restoration. The reason that perceived political willingness is positive may be due to ongoing projects by UNDP/GEF, which have carried out in situ restoration, provided training and in-kind restoration inputs to the pastoralists in the Nature and People as One (NaPO’s) RAMAT model (UNDP Kenya, 2024). Other Activities include “The Restoration Initiative” (TRI) project (2018–2025), which provided community capacity building, water-management works, and seeds (TRI, 2024). This implies that visible government restoration activities are a key factor influencing positive outcomes and result in the pastoralists’ participation in restoration activities.

4.6 Effect of perceived incentives

Perceived incentives exerted a positive and statistically significant effect on the adoption of rangeland restoration practices. This finding indicates that pastoralists who believe that restoring degraded rangelands will improve pasture quality, enhance forage availability, and increase livestock productivity are more likely to participate in restoration activities. Conversely, those who perceive limited or no economic and livelihood benefits from restoration are less inclined to invest their time and resources in such practices. This underscores the importance of expected economic returns as a key motivational factor influencing restoration behavior.

A plausible explanation for this finding is that many pastoralists may have observed tangible improvements in pasture conditions and livestock performance resulting from ongoing restoration initiatives, such as The Restoration Initiative (TRI) project (2018–2025). First-hand experience or observations of successful restoration outcomes within their communities or elsewhere may have strengthened their confidence in the benefits of restoration, thereby reinforcing their willingness to adopt similar practices. These findings suggest that demonstrating the economic and livelihood benefits of restoration through pilot projects, demonstration sites, and farmer-to-farmer learning can serve as powerful incentives for increasing participation in rangeland restoration programs.

5 Conclusion

This study employed an extended Protection Motivation Theory (ePMT) framework to examine the drivers of pro-environmental behavior among pastoralists. Reliability analysis confirmed that the ePMT constructs adequately captured the psychological, institutional, and economic factors influencing restoration intentions. Ordinal logistic regression was used to evaluate the effects of PMT constructs and socio-economic characteristics on the adoption of restoration practices. The results indicate that response efficacy, perceived political willingness, perceived vulnerability, self-efficacy, and perceived incentives significantly and positively influence pastoralists’ adoption of restoration practices. Among these predictors, response efficacy emerged as the strongest determinant, followed by perceived vulnerability, self-efficacy, and perceived incentives.

The findings demonstrate that both threat appraisal and coping appraisal significantly influence restoration behavior; however, coping appraisal exerts the greater influence, as evidenced by the larger coefficient and marginal effects associated with response efficacy. This suggests that pastoralists’ restoration decisions are shaped not only by their awareness of the risks and severity of rangeland degradation but also by their confidence in the effectiveness of restoration measures and their ability to implement them successfully. Moreover, the significance of perceived political willingness indicates that restoration behavior is also conditioned by confidence in government commitment, policy support, and institutional backing. Consequently, raising awareness of environmental degradation, although necessary, may be insufficient to stimulate sustained restoration behavior. Behavioral change is more likely when awareness is complemented by interventions that strengthen self-efficacy and response efficacy while reducing the financial, labor, and opportunity costs associated with restoration practices.

The findings further reveal that restoration behavior is embedded within a broader institutional and policy context. Perceived political willingness and tangible incentives, such as improved pasture quality and enhanced livestock productivity, emerged as important enabling conditions that reinforce individual motivation and reduce uncertainty regarding restoration investments. These results underscore the importance of integrating institutional support with behavioral interventions. Governments and development partners should therefore move beyond conventional awareness campaigns by providing consistent policy commitment, secure institutional arrangements, effective extension services, demonstration of successful restoration outcomes, and incentive-based mechanisms that reduce barriers to adoption. Such measures can strengthen pastoralists’ confidence in restoration interventions, lower the perceived risks of investment, and promote sustained participation in rangeland restoration initiatives.

6 Recommendations, limitations, and suggestions for further studies

Based on the findings of this study, the following recommendations are proposed for policymakers, government agencies, non-governmental organizations (NGOs), county governments, and development partners to guide rangeland restoration efforts in Laisamis and Saku Sub-counties and other areas with similar ecological, socioeconomic, and climatic characteristics.

First, it is important to strengthen demonstration and awareness programs on restoration effectiveness. Since response efficacy emerged as the strongest predictor of restoration behavior, government agencies, non-governmental organizations, and development partners should establish demonstration sites and disseminate evidence of successful restoration interventions. Showcasing tangible improvements in pasture quality, vegetation cover, and livestock productivity will increase pastoralists’ confidence in the effectiveness of restoration practices and encourage wider adoption.

Second, pastoralists’ technical capacity and self-efficacy should be improved. Since self-efficacy emerged among the top three significant factors of restoration behavior, extension services should provide regular training, technical support, and practical skills in sustainable grazing management, reseeding, soil and water conservation, and other restoration techniques. Strengthening local knowledge and skills will improve pastoralists’ confidence in implementing restoration practices successfully.

Third, there should be incentive-based restoration programs. Given the positive influence of perceived incentives, governments and development partners should introduce financial and non-financial incentive mechanisms to encourage restoration adoption. These may include input subsidies, payments for ecosystem services, access to improved grazing resources, livestock productivity enhancement programs, credit facilities, and market support for pastoral households undertaking restoration activities.

Next, the significant effect of perceived political willingness highlights the importance of visible government support. Policymakers should strengthen institutional frameworks, ensure consistent implementation of restoration policies, improve coordination among relevant agencies, and actively involve pastoral communities in planning and decision-making processes. Such measures will enhance trust, increase stakeholder participation, and promote long-term restoration efforts.

In all, policymakers should adopt integrated restoration policies that address behavioral, institutional, and economic factors simultaneously. Since restoration behavior is influenced by psychological, institutional, and economic drivers, restoration programs should move beyond conventional awareness campaigns and incorporate behavioral change strategies, institutional support, and appropriate incentive mechanisms. Such an integrated approach is more likely to achieve sustained participation in restoration activities and contribute to the attainment of Land Degradation Neutrality and Sustainable Development Goal 15.3.1.

This study has several limitations. First, the analysis is based on cross-sectional data and therefore identifies associations rather than definitive causal relationships between explanatory factors and participation in rangeland restoration. Second, although the study extended the Protection Motivation Theory (PMT) to assess the effects of perceived incentives and political willingness as well as the traditional PMT constructs on restoration behavior, some potentially relevant factors, such as social networks and risk preferences, were not directly observed.

Another limitation relates to sample attrition during data cleaning. Comparisons between retained and excluded respondents revealed no significant differences in observable characteristics such as gender, age group, and administrative unit, suggesting that substantial selection bias is unlikely. However, the possibility that excluded respondents differed in unobserved characteristics cannot be eliminated. Consequently, the findings should be interpreted with this consideration in mind, and caution should be exercised when generalizing the results beyond the study population. Finally, the study was conducted within a specific pastoral context in Kenya, and caution should therefore be exercised when generalizing the findings to other rangeland systems with different ecological and institutional characteristics.

In future studies, we propose:

• The adoption of Bayesian approaches which provide a framework for incorporating prior information and handling complex data structures, such as clustered or longitudinal data. These models can be particularly useful for estimating parameters in the presence of missing data or when modeling agreement and evaluating diagnostic tests (Agresti and Natarajan, 2001) Moreover, Bayesian cumulative probit models have been shown to outperform linear models.
• Move from Ordinal Logit to Governance-Extended Structural Equation Modeling (SEM), which would enable the testing of Threat Appraisal →Coping Appraisal →Restoration practices while simultaneously examining Political Willingness →Perceived Incentives →Restoration practices.
• The use of longitudinal data to strengthen causal inference.
• There is a need to examine additional constructs like social networks, risk preferences, community leadership, customary institutions and assess how they influence collective restoration actions.

Generative AI statement

The author(s) declared that Generative AI was used in the creation of this manuscript. ChatGPT was used to edit the manuscript to improve its readability, clarity, grammar, and overall language quality. The authors reviewed and verified all revisions and retained full responsibility for the content of the manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Tong Li, The University of Queensland, Australia

Reviewed by: Harineck Mayamiko Tholo, Malawi University of Science and Technology, Malawi

Marcela Pimid, Universiti Malaysia Sabah, Malaysia