1 Introduction

Achieving global food security remains a significant challenge in low- and middle-income countries (LMICs) where rural households cope with adverse food insecurity problems in the face of climate change and extreme weather shocks (Backer & Billing, 2021; UN, 2023; World Bank, 2022). Recent trends in global hunger and malnutrition rate show that the proportion of malnourished people increased from 8% of the world’s population in 2019 to 9.8% in 2021 (FAO et al., 2022; World Bank, 2022). Food insecurity is particularly severe in sub-Saharan Africa (SSA) where most rural households do not have access to sufficient food to meet their daily calorie intake requirements – a situation that existed before the COVID-19 pandemic, but that has been exacerbated by it (Amare et al., 2021; Bjornlund et al., 2022; Olabiyi, 2022). For example, in the last five years, malnutrition rates in SSA increased from 18.8% in 2017 to 23.2% in 2021, compared to Asia and Latin America where food insecurity rates increased moderately from 7.1% in 2017 to 9.1% in 2021, and from 6.4% in 2017 to 8.6% in 2021 respectively (FAO et al., 2022).

Some of the main causes of acute food insecurity and malnutrition in SSA include weak agricultural productivity due to natural resource depletion, climate change and extreme weather shocks, and political instability (Amadu et al., 2020a; Benin, 2016; Bjornlund et al., 2022; Chanza & Musakwa, 2022; Dzanku, 2019). Alternative food security pathways, such as those provided by forests, which may be more resilient to various stresses, are therefore increasingly vital for reducing acute food insecurity in LMICs like those in SSA (Ickowitz et al., 2022). There is substantial evidence that forests represent an important pathway to support livelihoods and positive social-ecological outcomes at multiple scales across SSA, Southeast Asia, and Latin America (Asprilla-Perea & Díaz-Puente, 2019; Bahar et al., 2020; Ickowitz et al., 2016; Miller et al., 2022; Pimentel, 1997). For instance, forests provide a viable source of food and medicinal products for many rural communities across diverse LMIC contexts (Bahar et al., 2020; Carignano Torres et al., 2022; Karki et al., 2018). More specifically, dietary quality is associated with tree cover in many LMICs (Hall et al., 2022; Ickowitz et al., 2016; Rasolofoson et al., 2018; Rasmussen et al., 2020; Rowland et al., 2017).

Forests can also act as a safety net providing food and other resources in times of need for forest-proximate households (i.e., households living in the proximity of forests) around the world (Andrews & Borgerhoff Mulder, 2022; Bakkegaard et al., 2017; Newton et al., 2020; Rasmussen et al., 2017; Razafindratsima et al., 2021; World Bank, 2020a; Wunder et al., 2014a, b). Forests can also enhance food security through less direct pathways like medical values and human health through ecosystem services, and poverty alleviation (Hall et al., 2022; Jagger et al., 2022; Miller & Hajjar, 2020; Mukul et al., 2016). Importantly, forests can often provide these benefits while also enhancing environmental sustainability outcomes such as carbon sequestration, water quality enhancement, and biodiversity conservation through pathways like tree planting and restocking (Cheng et al., 2019; Lepcha et al., 2019; Miller & Hajjar, 2020; Miller et al., 2017, 2022; Wolff et al., 2018).

In light of these contributions, forest are seen as a vital means for realizing several of the Sustainable Development Goals (SDGs), including on poverty (SDG 1), hunger (SDG 2), and climate action (SDG 13) (Miller et al., 2022; Timko et al., 2018). Forests can provide a sustainable pathway out of poverty, and by extension, a pathway to food security (e.g., through food availability and access) (Jagger et al., 2022; Razafindratsima et al., 2021). Forests are linked to SDGs on poverty and hunger through micro-level household experiences of food insecurity, hunger, and inequality (Carignano Torres et al., 2022; Hall et al., 2022; Jones et al., 2013). For forests to deliver on these goals requires participation by individuals, households, and communities in an array of forest sector activities including the collection and processing of timber and non-timber forest products (NTFPs) (such as fuelwood, bushmeat, and bush yam) as well as the planting of trees and shrubs. Such is the case for hundreds of millions of people around the world who live in, or near forests, and participate in different ways in the forest sector (Li et al., 2019; Newton et al., 2020).

Despite the benefits forests can bring to address hunger and other SDGs, however, analysis of the effects of forest sector participation on the achievement of food security outcomes remain limited, particularly in certain underrepresented regions like West Africa (Hajjar et al., 2021). Existing evidence from SSA has largely come from other regions, especially in Eastern and Southern Africa (e.g., Coulibaly et al., 2017; Miller et al., 2020; Rasmussen et al., 2020). Moreover, despite an increasing literature on forests and livelihoods, there is a surprising lack of micro-level evidence of the effects of the forestry sector on food security in terms of the number of months households experience food insecurity in LMICs. Much of the literature examines the relationship between forest cover and food security rather than people’s participation in forestry activities (Asprilla-Perea & Díaz-Puente, 2019; Chilongo, 2014; Hall et al., 2022).

This study seeks to address this knowledge gap by estimating the effect of forest sector participation on food security in Liberia, the most forested country in West Africa. We hypothesize that forest sector participation – namely, collecting and processing timber and non-timber forest products – can significantly reduce food insecurity (in terms of the number of months households had insufficient food) among households living in the proximity of forests in Liberia. There are several pathways through which participation in the forestry sector can contribute to food security. One such pathway is through direct consumption of harvested forest products like bushmeat and bush yam, which can reduce the incidence of food insecurity among households as noted in previous studies in Liberia (World Bank, 2020a) and elsewhere (Bista & Webb, 2006; Carignano Torres et al., 2022; Chilongo, 2014; Ickowitz et al., 2016; Yego et al., 2021).

To test our hypothesis, we used primary survey data collected from 2983 households living in the proximity of forests across all 15 counties in Liberia in 2019 (World Bank, 2020a). We used an endogenous switching poisson regression (Greene, 2009; Hasebe, 2020; Miranda, 2004; Terza, 1998) to account for endogeneity of forest sector participation and selection bias in the effects of such participation on food security (measured as a count variable: the number of months households reported having insufficient food).

The novelty of this study is its application of a rigorous resource economics approach to provide micro-level evidence of the vital contribution of forests to food security in terms of how the forestry sector can alleviate household food insecurity by reducing the number of months households living in the proximity of forests experience insufficient food. This analysis can be useful for both researchers and policymakers working in LIMCs, especially in forest-rich but underdeveloped countries like Liberia and similar countries elsewhere in SSA and beyond.

2 Materials and methods

2.1 Study context

Liberia is the most forested country in West Africa, with 69% of its land area covered with forests or about 6.7 million hectares (World Bank, 2020a). Nearly half of Liberian households (48%) are classified as “forest proximate” household communities – they live within 2.5 km of a forest and rely on forest products like bushmeat, timber, and mushrooms for food and other livelihood needs like energy and medicine (Beevers, 2016; Hwang et al., 2020; World Bank, 2020a). Liberia is also one of the most vulnerable African countries in terms of health crises, climatic shocks, and socioeconomic problems including acute food and nutritional insecurity, which stem from micro and macroeconomic constraints like low agricultural productivity, economic stagnation, the COVID-19 pandemic, and global supply chain problems (Aggarwal et al., 2020; Reynolds et al., 2022; WFP, 2021).

The country has experienced serious food insecurity problems due to prolonged social-ecological factors, including civil conflict that led to many people being displaced from different parts of the country and exacerbated low educational achievement and environmental degradation (Agwu et al., 2013; Hilson & van Bockstael, 2012; Tsegaye et al., 2018; WFP et al., 2022). Recent World Bank reports (2020a, b) show that 46% of Liberian households experienced food insecurity in the twelve months preceding the survey on which the reports were based and that the average period of food insecurity was about three months.

The forest sector is the fourth largest sector in terms of gross domestic product (GDP) contributing about 10% of GDP (following services, agriculture and fisheries, and mining). Some 43% households reported using forest products to recover from shocks like crop failure, sickness, or a death in the family and 66% of households depend on forest products (like bushmeat) to overcome food security shocks, further demonstrating the importance of the sector (World Bank, 2020a).

Despite the potential contributions of the forestry sector towards food security and rural livelihoods in Liberia, there is a limited empirical evidence of food security outcomes attributable to forest sector participation in general, but particularly at a national scale. Such analyses are now needed to advance knowledge of the contribution forestry can make to food security and to guide development policy regarding sustainable utilization forest products and management of forests as a core natural resource to help achieve food security and other SDGs in Liberia (World Bank, 2020a).

Recognizing this need, the Government of Liberia collaborated with the World Bank to conduct a national household forest survey in 2019. The survey was conducted across the 15 Liberian counties (Fig. 1). It provides important information not only about the extent of households living in the proximity of forests in Liberia but how such households depend on forests to derive their livelihoods. For example, it shows that 70% of surveyed households in rural Liberia engaged in the collection of a wide variety of forest products (amounting to more than 40 different products like bushmeat, fuelwood, and piassava) in the twelve months preceding the survey in 2018/2019, either directly for self-consumption, or both for sale and self-consumption, with about 24% engaged in processing such products (World Bank, 2020b). Further, 98% of households reported collecting fuelwood for their energy needs and 85% used forest products to construct their dwelling/shelters (World Bank, 2020a, b).

Fig. 1
figure 1

Map of the study area in Liberia showing a counties and b household sample sites

Full size image

2.2 Data

Data for this study come from the Liberia National Household Forest Survey (World Bank, 2020b) that was conducted in 2019 by the World Bank in collaboration with Liberia’s Forestry Development Authority and Institute of Statistics and Geo-Information Services. A sample of 2983 households living in the proximity of forests in 250 Enumeration Areas (EAs)Footnote 1 across the 15 counties in Liberia was selected and interviewed for the survey (Fig. 1a). The questionnaires were adapted from the National Socioeconomic Surveys in Forestry guidebook developed by the Food and Agriculture Organization in collaboration with other institutions (FAO et al., 2016). The questionnaires were adapted to elicit information at the household and community levels in Liberia and administered by local enumerators with supervision from World Bank personnel and personnel of the respective Liberian government institutions. All enumerators were conversant with the local languages in their areas of operation.

This unique dataset implies that our study is one of the first analysis of food security to be based on data collected using a Computer-Assisted Personal Interview (CAPI) survey in Liberia (World Bank, 2020a). The households surveyed were those located in EAs with center points within 2.5 km of the nearest forest (Fig. 1b). The specific survey questions and further information about the variables we use in this study are publicly available in the Liberia Household Forestry Data (World Bank, 2020b).

2.3 Definition of main variables

Defining and measuring household food security – our outcome variable – is challenging due to varying dynamics across the study context. Moreover, food security definition has historically been a difficult subject due to many factors such as the broad transdisciplinary and multidimensional nature of what constitutes food security across context (Jones et al., 2013). The United Nations’ Food and Agriculture Organization (FAO) definition provides guidance, viz. that food security exists when there is enough food in terms of availability, access, affordability, and utilization (FAO, 1996). This definition, however, has limitations due to changing or dynamic metrics pertaining to the various domains of food security: availability, access, utilization, and stability over time and spatial context (Frongillo, 2022; Jones et al., 2013). For example, there are diverse factors that affect food security or resilience to food insecurity, which further complicates the adoption of specific broad food security definitions and measurements across context (Béné et al., 2023; d’Errico et al., 2023; Jones et al., 2013).

Some of the diverse ways previous studies have measured food security include household experience of food availability and access, coping strategies, and dietary diversity (Frongillo, 2022; Hendriks et al., 2016; Leroy et al., 2015; Msaki & Hendriks, 2014; O’Meara et al., 2022; Tambo et al., 2021). Data limitations mean that scholars have used a range of measures to indicate food security in LMIC contexts like Liberia (e.g., Cafiero et al., 2018; Coates et al., 2017; Maitra, 2017; Pereira et al., 2021).

We used an experience-based measure of food security in terms of the reported number of months a household had insufficient food during the past 12-month preceding the survey. We follow the approach in Tambo et al. (2021) whish used “the months of inadequate household food provisioning” – a subjective measure of the level of access to food. It refers to the number of months out of the previous 12 that households report having had difficulty satisfying their food needs due to depletion of own food stocks, or a lack of money to purchase food (Tambo et al., 2021, p. 104). Thus, our measure of food security is a count variable. The distribution of this variable is not linear (Fig. 2), and therefore, cannot be estimated by a simple linear regression model.

Fig. 2
figure 2

Kernel density plot of the distribution of number of months with insufficient food

Full size image

Our main policy (treatment) variable is forest sector participation, which we defined as a binary dummy variable (yes = 1, and zero otherwise) for whether a household collected or processed forest products during the 12-month period of the survey in 2019 (World Bank, 2020a). Specifically, the definition is based on whether a household engaged in forest-based activities like collection and processing of timber and NTFPs such as fuelwood, bushmeat, and bush yam (Amadu & Miller, 2024).

Other important variables include household demographic and socioeconomic factors including the age and gender of the household head, household size, the share of female labor per household, dummy variables for whether the household embarked on crop production, and if yes, the area of land cultivated, dummy variables for whether the household received any social assistance from the government, international remittances, or domestic transfers, and dummy variables for whether the household suffered any weather, market, and disease or pest-related shocks, as well as whether the households planted trees in the last 12 months, and if yes, the approximate number of trees planted. Biophysical factors include the distanceFootnote 2 from the homestead to nearest market center, the nearest main road, and the forest and a set of dummy variables (yes = 1, zero otherwise) for whether the household resides within 30 min, 31–60 min, 61–90 min, and beyond 90 min.

We controlled for such demographic factors (like the age and gender of household heads) and biophysical factors (like distance to forests and markets) as they are known to shape the extent of food insecurity in Liberia as found in relevant studies (Borlizzi et al., 2017; Coates et al., 2017; FAO et al., 2022; World Bank, 2020b, 2022). We also controlled for important community level biophysical factors, including dummy variables (yes = 1, zero otherwise) for whether there was a change in the natural forest cover in the village over the past five years (yes = 1, zero, otherwise), and the approximate area of forests (hectares) cleared in the past five years (Table 1).

2.4 Conceptual framework and analytical approach

It is well known that the problem of market failure exists in rural areas of developing countries such as Liberia. In such contexts, the production and consumption decisions of a typical household become inseparable (Roza et al., 2017; Singh et al., 1986). This implies that self-selection in forest sector participation and subsequent outcomes like food security. As such, selection bias can significantly affect our analysis of the effects of forest sector participation on food security (like the reported number of months households had insufficient food). Therefore, we must account for selection bias in our assessment of the effects of forest sector participation on food security in Liberia. To do so, we utilized a random utility framework, which assumes that an ith household in rural Liberia undertakes forest sector participate if the difference (\({P}_{i}^{*}\)) between the expected utility of participation (\({F}_{i}^{sp}\)) and non-participation (\({F}_{i}^{sn}\)) is greater than zero such that:

$${P}_{i}^{*}= {F}_{i}^{sp}- {F}_{i}^{sn}>0$$
(1)

However, we cannot simultaneously observe the two utilities (i.e., \({F}_{i}^{sp}\) and \({F}_{i}^{sn}\)) because they are subjective characteristics. Therefore, we utilize a latent variable specification structure to express both utilities as a linear combination of observable characteristics as follows:

$${P}_{i}^{*} = {\tau }_{i}{Z}_{i} + {\eta }_{i},\ \text{with}\ {P}_{i}= \left\{\begin{array}{cc}1 & if\ {P}_{i}^{*}>0, \\ 0 & otherwise\end{array}\right.$$
(2)

where \({P}_{i}^{*}\) denotes a latent variable for the probability of forest sector participation. This latent variable is determined by an observable variable \({P}_{i}\), which indicates the actual participation status observed by the researcher (Danso-Abbeam et al., 2021a; Vilar-Compte et al., 2017), \({Z}_{i}\) denotes a vectors of characterisitics affectting the participation decision and the outcome, \({\tau }_{i}\) is a vector of parameters we want to estimate, while \({\eta }_{i}\) is an error term.

2.5 Empirical strategy: Endogenous switching poisson regression model

As noted in Section 2.3, our outcome variable – the number of months households reported having insufficient food – is count data. This variable has a range of zero to twelve (12). A poisson regression model is commonly used to estimate count data (Greene, 2009). Thus, if we let \({y}_{i}\) represent the number of months a household had insufficient food for the 12 months prior to the survey in 2019, we can specify a basic poisson model for the number of event \({y}_{i}\) for an ith household as a poisson distribution with mean \({\lambda }_{i}\) and household characterisitcs denoted by \({x}_{i}\) as follows:

$${\lambda }_{i}=E\left({y}_{i}|{x}_{i}\right)= {e}^{{{x}_{i}}^{\theta }}$$
(3)

The probability of \({y}_{i}\) conditional on \({x}_{i}\) can be specified as:

$$\text{Pr}\left({y}_{i}|{x}_{i}\right) = \frac{{e}^{{-\lambda }_{i}}{\lambda }_{i}^{{y}_{i}}}{{y}_{i} !}$$
(4)

The poisson regression requires the variance of \({y}_{i}\) to be equal to the mean of \({y}_{i}\), such that:

$$\text{Var}\left[{y}_{i}|{x}_{i}\right] = {\lambda }_{i}$$
(5)

However, from Table 1, the mean of \({y}_{i}\) is 3.18, which is less than the variance of \({y}_{i}\), which is 3.61 (i.e., the square of the standard deviation, which is 1.9). This implies that our outcome variable (\({y}_{i}\)) exhibits overdispersion. Therefore, we cannot apply an ordinary poisson regression in this context (Cameron & Trivedi, 1990; Yang et al., 2007). Moreover, because we want to estimate the effects of forest sector participation (\({P}_{i}\)) on the number of months households had insufficient food in the 12 months preceding the survey in 2019 (i.e.,\({y}_{i}\)), we have a challenge of endogeneity. This is because the collection and processing of forestry products (such as timber logging, fuelwood collection, and bushmeat) by were not random. Instead, households chose to either engage in the forestry sector or not. Failure to account for the endogeneity of forest sector participation can, therefore, result in biased estimates of the effects of the forestry sector on household food security (in terms of number of months households had insufficient food).

Following relevant literature (Greene, 2009; Hasebe, 2020; Miranda, 2004; Terza, 1998; Zakaria et al., 2020), we utilized an endogenous switching poisson regression to model \({y}_{i}\) conditional on the endogenous dummy variable \({P}_{i}\), a vector of explanatory variables \({x}_{i}\), and an error term \({\varepsilon }_{i}\) as follows:

$$Prob\left(\frac{{y}_{i}}{{\varepsilon }_{i}}\right) =\frac{\text{exp}\left\{-\text{exp}({\beta }_{i}{x}_{i}+ \Theta {P}_{i} + {\varepsilon }_{i})\right\} \{\text{exp}\left({\beta }_{i}{x}_{i}+ \Theta {P}_{i}+ {\varepsilon }_{i}\right){\}}^{{y}_{i}}}{{y}_{i}!} ,$$
(6)

where the outcome variable \({y}_{i}\) and our endogenous dummy variable \({P}_{i}\) are both independent, conditional on \({\varepsilon }_{i}\). \({\tau }_{i}\) and \(\Theta\) are parameter estimates of the explanatory variables \({Z}_{i}\) and forestry sector participation \({P}_{i}\) as noted ealier. From Eq. (2), endogenity of \({P}_{i}\) implies that \({\eta }_{i}\) and \({\varepsilon }_{i}\) are jointly correlated with a zero mean and a covariance matrix expressed as:

$$\omega = \left(\begin{array}{c}{\sigma }^{2} {\sigma }_{\rho }\\ \sigma \rho 1\end{array}\right)$$

The joint probability density function (pdf) of \({y}_{i}\) and \({P}_{i},\) conditional on \({Z}_{i}\) is expressed as:

$$f\left({y}_{i}, {P}_{i}|{Z}_{i}\right)= {\int }_{-\infty }^{\infty }\left\{\left(1-{P}_{i}\right)f(\left({y}_{i}|{P}_{i}=0, {Z}_{i},{\varepsilon }_{i}\right)\text{Pr}\left( {P}_{i}=0|{Z}_{i},{\varepsilon }_{i}\right) {Z}_{i},{\varepsilon }_{i} f\left({\varepsilon }_{i}\right)+ {P}_{i}f\left({y}_{i}|{P}_{i}=1, {Z}_{i},{\varepsilon }_{i}\right)\text{Pr}\left( {P}_{i}=1|{Z}_{i},{\varepsilon }_{i}\right)\right\} ,$$
(7)

where \(f\left({\varepsilon }_{i}\right)\) is the pdf of \({\varepsilon }_{i}\), all other variables are as defined previously. Consequently, we maximized the following log likelihood function:

$$LogL= \sum_{i=1}^{n}\text{ln}\left\{f\left({y}_{i}, {P}_{i}|{Z}_{i}\right)\right\}$$
(8)

2.5.1 Instrumental variable application for endogeneity and selection bias

Endogenous switching poisson regression does not typically require the use of instrumental variables (IVs) because the maximization process of the model uses functional forms through a Gauss-Hermite quadrature technique and a full information maximum likelihood (FIML) algorithm for identification (Miranda, 2004). However, for efficient estimation of the model, we utilize IV techniques to account for the endogeneity of forest sector participation and selectivity bias in the outcome variable, as done in relevant empirical studies (e.g., Danso-Abbeam et al., 2021b; Hasebe, 2020; Tambo & Wünscher, 2018; Zakaria et al., 2020). A valid IV should be relevant and exogenous – that is, such an IV should be correlated with our endogenous policy variable (forest sector participation), but it should not be correlated with food security (i.e., the number of months a household had insufficient food).

Finding a valid IV is often a major challenge in studies using observational data like the present one because, the variables that affected forest sector participation can also affect the number of months households had insufficient food – our outcome of interest. However, a set of variables in the household and community questionnaires were promising as IVs because they seemed exogenous and therefore, deemed to have affected household participation in the forestry sector but they did not directly affect the outcome variable in question – the number of months households had insufficient food. These include variables based on the responses to the follow questions: 1) Has there been any change in areas of natural forest cover in your village in the past five years? 2) Approximately how much forest area did the community clear in the last five years?

We used two methods to test the validity of these variables as valid IVs for their relevance and exogeneity. The first test comprised a partial correlation test (e.g., Amadu, 2022; Coulibaly et al., 2017) which we used to show that our prospective IVs were correlated with forest sector participation but they are not correlated with the food security outcome variable (i.e., the number of months households reportedly had insufficient food) (Table A1). Second, we used falsification tests, which are increasingly applied in related empirical studies to examine the exogeneity of potential IVs (e.g., Amadu et al., 2020b, 2021; Sesmero et al., 2018). The falsification tests show that both variables are valid IVs – they affected the participation decision of households in the forest sector, but they did not have statistically significant impact on food security (i.e., they can only affect food security through forest sector participation). Therefore, we conclude that they are exogenous and thereby regarded as valid instruments in this context (Table A2).

We utilized one of these IVs – change in the areas of the natural forest cover in the village/community (yes = 1, 0 otherwise) in our main regression. Note that our results are interpreted in terms of the exponential function of the coefficient estimates (i.e., exp(b)) following relevant literature (Erdogdu, 2013; Ma & Wang, 2020; Tambo & Wünscher, 2018). This is because poisson regressions like the ones we use here, do not conform to the normality assumption that underpins linear regressions.

2.5.2 Treatment effects of forest sector participation on food security

In addition to estimating the main effects of forest sector participation on household food security, we also wish to analyze the actual food security effects (i.e., treatment effects) of forest sector participation on those households that actually participated in the forestry sector. These include the average treatment effects (ATE) – i.e., the effect of forest sector participation on any random household selected from the population, average treatment on the treated (ATT) – i.e., the food security effect of forest sector participation on households living in the proximity of forests that actually engaged in the collection and/or processing of forest products, while the average treatment effect on the untreated (ATU), which is the effect for those who did not participate, had they participated. We computed the ATE, ATT, and ATU as follows:

$$ATE= \text{Exp}\left({y}_{1}-{y}_{0}|{x}_{i}^{\prime}\beta \right) + {\mu }_{i}$$
(9)
$$ATT= \text{Exp}\left({y}_{1}-{y}_{0}|{x}_{i}^{\prime}\beta , p=1\right) + {\mu }_{i}$$
(10)
$$ATU= \text{Exp}\left({y}_{1}-{y}_{0}|{x}_{i}^{\prime}\beta , p=0\right) + {\mu }_{i}$$
(11)

where y1 and y0 denote food security effects of forest sector participation and non-participation, respectively, \(p=1\) and \(p=0\) denote forest sector participants and not-participants respectively, \({x}_{i}\) constitute household characteristics, the \(\beta\) terms are parameters to be estimated, while \({\mu }_{i}\) are errors terms. We estimated these equations using the “teescount” user-written STATA routines proposed by Hasebe (2020).Footnote 3 We used these algorithms in STATA 17.

2.5.3 Heterogeneity and pathway analyses

We estimated the pathways through which forest sector participation achieves food security. To do so, we re-estimated our main treatment effect models in Eqs. (9) through (11) conditional on the two main channels of forestry sector participation: i). timber and other wood products and ii). NTFPs. We present these as Eqs. (12) – (14) for timber and other wood products, and as Eqs. (15) – (17) for NTFPs as follows:

For timber and wood products:

$${ATE}_{Timber+}= \text{Exp}\left({y}_{1}-{y}_{0}|{x_i}^{\prime}\beta \right) + {\mu }_{i}$$
(12)
$${ATE}_{Timber+}= \text{Exp}\left({y}_{1}-{y}_{0}|{x}_{i}^{\prime}\beta , p=1\right) + {\mu }_{i}$$
(13)
$${ATE}_{Timber+}= \text{Exp}\left({y}_{1}-{y}_{0}|{x}_{i}^{\prime}\beta , p=0\right) + {\mu }_{i}$$
(14)

For NTFPs:

$${ATE}_{NTFPs}= \text{Exp}\left({y}_{1}-{y}_{0}|{x}_{i}{\prime}\beta \right) + {\mu }_{i}$$
(15)
$${ATE}_{NTFPs}= \text{Exp}\left({y}_{1}-{y}_{0}|{x}_{i}^{\prime}\beta , p=1\right)+ {\mu }_{i}$$
(16)
$${ATE}_{NTFPs}= \text{Exp}\left({y}_{1}-{y}_{0}|{x}_{i}^{\prime}\beta , p=0\right)+ {\mu }_{i}$$
(17)

where \({ATE}_{Timber+}\), \({ATT}_{Timber+}\), and \({ATU}_{Timber+}\) denote the ATE, ATT, and ATU estimates conditional on timber and other wood products, while \({ATE}_{NTFPs}\), \({ATT}_{NTFPs}\), and \({ATU}_{NTFPs}\) denote the ATE, ATT, and ATU attributable to NTFPs.

2.5.4 Multiple imputation

Furthermore, we used multiple imputation with several estimation techniques including mi-poisson and negative binomial regression to account for missing data in our dependent variable – number of months with insufficient food. As such, our analysis of the imputed data followed several relevant studies (Johnson et al., 2011; Lee et al., 2011; Royston, 2004).

3 Results

3.1 Summary statistics

Table 1 presents descriptive and summary statistics of the main variables. The number of observations is 2983 for all variables except for the number of months with insufficient food, which was 1408, with a significant proportion of missing values (about 1576). The average rate of forest sector participation, our main independent variable, was 68%. Of those participating, the collection of timber forest products and other wood products was 20%, while the rate of non-timber forest products was 18%. Non-enterprise forest income was $236 on average while the value of all forest-related incomes (including both enterprise and non-enterprise forest activities) was $245 per annum. These suggest that the forestry sector makes some significant contribution to household incomes and livelihoods in Liberia (see Amadu & Miller, 2024).

The average age of the household head was 44 years, while 23% of households were female-headed. Moreover, the share of female labor per household – i.e., the proportion of household female members of a working age between 15 and 60 years was 48% compared to 46% for males. Also, 48% of surveyed households engaged in crop production, with a reported land size 0.93 hectares, meaning that the average household in rural Liberia is a smallholder farmer.

Table 1 Variable description and summary statistics
Full size table

There were significant county-level variations in food security in terms of the number of months with insufficient food (Table 2). Specifically, River Cess, Grand Bassa, Margibi, and Sinoe counties were the four most food insecure counties, respectively having 4.01 months, 3.67 months, 3.51 months, and 3.39 months with insufficient food on average. On the other hand, the counties that had the lowest average number of months with insufficient food include River Gee, Grand Kru, Maryland, and Grand Cape Mount with 2.85, 2.53, 2.51, and 2.25 months, respectively (Table 2). Likewise, there was significant variation in the rate of forest sector participation across counties with the top four counties being Maryland, River Gee, Grand Gedeh, and Nimba county, which had on average, participation rates of 92%, 90%, and 83% respectively, while Montserrado, Bomi, and Gbarpolu are the counties with the lowest participation rates at 43%, 58%, and 63% respectively (Fig. 3 and Table 2).

Fig. 3
figure 3

Forest sector participation rate per county in Liberia

Full size image

These statistics suggest that county-level heterogeneity analysis of the effects of forest sector participation on household food security might be important in understanding the impacts of the forestry sector on food security in Liberia.

Table 2 Distribution of main outcome and treatment variables by county
Full size table

3.2 Summary statistics by forest sector participation

Table 3 presents conditional summary statistics by households categorized as forest sector participants and non-participants in rural Liberia. There was a statistically significant difference between forest sector participants and non-participants in terms of our food security proxy variable (i.e., the number of months households reportedly had insufficient food), with a mean difference of 64% between non-participants and participants. The mean number of months with insufficient food is 3.02 among participants and 3.67 among no-participants. This value implies that food insecurity is a serious challenge in the country.

It is important to note that although non-participants reported having collected or processed some forest products, there was zero forest-related income among non-participants compared to participants. This implies that the livelihoods of non-participants were not dependent on the forestry sector. Significant differences also existed for many demographic and socioeconomic variables, such as the age and gender of the of the household head, household size, crop production and the area of land cultivated, and variables related to shocks like weather, markets, and disease-related problems experienced by households.

Moreover, there were significant differences in biophysical factors like the distance to the nearest market centers, change in the natural forest cover, and the community forest areas cleared (Table 3). For example, 41% of participants reported perceiving a change in the area of natural forest cover in and near their villages, whereas only 20% of the other households reported such a change. Likewise, households that participated in the forest sector reported noticing approximately 17.6 ha of forest cleared from their communities compared to only 7.25 ha of households cleared in other villages.

There were also significant differences at the regional cluster and county levels between the proportion of households participating in the forest sector and other households. For example, while the difference between participants and non-participants was not significant in Central cluster, there were significant differences between participants and non-participants in the Eastern and Western clusters at about 14% each. These differences may have shaped the rate of forest sector participation and the subsequent effects on food security in the country. Our analytical approach has therefore used rigorous techniques to account for such differences and avoid biased estimates.

Table 3 Summary statistics conditional on forest sector participation
Full size table

3.3 Econometric results

3.3.1 Determinants of forest sector participation

Here we present (Table 4) the main result of our endogenous switching poisson regression. The estimation technique provided two sets of outputs – one conforming to the selection equation (i.e., the determinants of forest sector participation) while the other equation represents the food security outcome equation – i.e., the factors affecting food security (in terms of the number of months households had insufficient food). There are many statistically significant determinants of forest sector participation.

Variables that positively affected forest sector participation include the average years of education of household members in the labor force (age 15–64 years), which had a 10% effect on forest sector participation, crop production with an effect size of 74%, which suggests that households which undertook crop production were 74% more likely to engage in the forestry sector compared to other households. This result implies that agriculture and the forestry sector are highly correlated in rural Liberia. However, the area of land cultivated did not significantly affect forest sector participation, which implies that the size of farms that households living in the proximity of forests do cultivate does not have the same effect on their decision to engage in forest-based activities as does their decision to cultivate crops. Moreover, the share of female labor in the household significantly enhanced the probability of household participation in the forest sector by 635%. Forest income was another important determinant of households’ forest sector participation, which was at 106%. Likewise, the proportion of forest income to total household income significantly affected forest sector participation by 37%.

These results suggest that household income from forests remain a crucial factor in households decision to undertake forest-based activities, and thus imply that forests contribute significantly to household income in Liberia – as noted in relevant studies from this context (Amadu & Miller, 2024; World Bank, 2020a, b). Other positive determinants include variables that are pertinent to social capital and safety sets such as social assistance coverage (15%), international remittances (0.1%), private domestic transfers (0.3%); and shocks such as weather (8%), market (14.2%), and diseases/pest outbreaks (21%). This suggests that forest sector participation is crucial for helping households navigate difficult challenges like shocks.

Variables that negatively affected forest sector participation include gender and external shocks including weather and market-related shocks, which reduced the participation of households in the forestry sector by 29% and 30% respectively. For example, female-headed households are 42% less likely to participate in forest-product collection and processing compared to male-headed households.

3.3.2 Main result: The effects of forest sector participation on food security

The fifth column of Table 4 presents the effects of forest sector participation and covariates on food security in terms of the reported number of months households had insufficient food. The seventh column presented the exponential values of these estimates (as noted in Section 2.5.2). We found a negative and statistically significant effect of forest sector participation on food security: the number of months households had insufficient food decreased by 84% (i.e., exp(-0.176) *100), which amounts to a reduction in the number of months with insufficient food by about 2.7 months per year. This result implies that participation in the forest sector helped to reduce the number of months households had insufficient food by about three months per year – an important contribution to food security in the country.

Our result supports our hypothesis that forest sector participation can reduce food insecurity in Liberia in terms of the number of months households had insufficient food. Our result is consistent with relevant studies in Liberia, elsewhere in Africa, and beyond (e.g., Agwu et al., 2011; Ahn et al., 2020; Beevers, 2015; Nicholson et al., 2021; World Bank, 2020a). This finding is important because it serves as a proxy measure not only for access to food but also the potential sustainability of food availability among vulnerable households in Liberia, which is especially crucial in the face of climate-induced food insecurity.

Other important determinant of food security in Liberia included households-level characteristics such as the area of land cultivated, which reduced food insecurity by 103%, the share of male and female labor in the households, both of which reduced food insecurity by 76% and 68% respectively. These results suggest that female labor share in rural households are more crucial for eradicating food insecurity constraints in forest-proximate household communities across Liberia. The results imply that understanding gender roles in forest sector participation can significantly help to alleviate food insecurity among households living in the proximity of forests in Liberia.

Our result is plausible because women are more likely to engage in the collection and processing of forest products like bushmeat, bush yam, and fishing, which can enhance their benefits of forest sector participation, as found in other relevant studies (Blesh et al., 2019; Brucker et al., 2022; Harris-Fry et al., 2015). Likewise, social assistance coverage reduced household food insecurity by 83%, which implies that social capital and safety nets are crucial in eradicating food insecurity constraints in poor households such as those in rural communities not just in Liberia but elsewhere (Singleton, 2022). Moreover, these results support the notion that demographic and socioeconomic factors affect food insecurity in rural Liberia and are consistent with similar studies in Liberia and elsewhere (Bahar et al., 2020; Bakhtsiyarava et al., 2021; Blesh et al., 2019; Fujimori et al., 2022; World Bank, 2020a).

Diagnostic checks at the bottom of Table 4 indicate a good model fit as shown by the statistically significant Log likelihood and Wald Chi-square values. Likewise, the statistically significant value of rho value – the correlation between unobservable factors that affected forest sector participation and food security impacts – suggest that unobservable factors and endogeneity issues affected household participation in the forestry sector across Liberia, and that such factors could have affected our estimate of food security effects of forest sector participation had we not accounted for endogeneity in our analytical approach.

Table 4 Endogenous switching poisson regression results of the impact of forest sector participation on the number of months with insufficient food
Full size table

3.3.3 Treatment effects of forest sector participation on food security

For policy analysis, we extended our espoisson regression to a treatment effects estimation using the teescount user-written STATA routines developed by Hasebe (2020) as discussed in Section 2.5.3. Using this approach, we estimated the ATE, ATT, and ATU. Table 5 shows the results of these estimation techniques for the main regression (i.e., the effects of forest sector participation of food security estimates on the full sample of 1408 households).

Our results show statistically significant effects of forest sector participation in Liberia. Specifically, we found statistically significant estimates of the ATE at 78%, the ATT at 77%, and the ATU at 80%. These results imply that forest sector participation significantly reduced food insecurity in Liberia. Our results support our hypothesis that forest sector participation significantly affects food security in Liberia by reducing the number of months forest-proximate households had insufficient food.

These findings are consistent with previous studies, which suggest that forests are crucial for food security not only in Liberia, but similar contexts elsewhere in SSA, and beyond (Agwu et al., 2013; Ahn et al., 2020; Asprilla-Perea & Díaz-Puente, 2019; Bakhtsiyarava et al., 2021; World Bank, 2020a). An interesting finding is that the statistically significant and larger ATU estimate suggests that non-participant households would have experienced higher reduction in the number of months they had insufficient food, had they participated in the forestry sector. Further research would be needed to understand why such households, however, failed to participate in the forestry sector.

Table 5 Teescount estimation of treatment of effects of forest-sector participation on food security
Full size table

3.3.4 Pathways of the effects of forest sector participation on food security

Table 6 presents the results of the pathways for the effects of forest sector participation on food security. As noted in Section 2.5.4, we wish to determine the most viable pathways through which forest sector participation affected food security in Liberia. To do so, we re-estimated the treatment effect models for the ATE, ATT, and ATU. The results show statistically significant effects of both timber and other wood products as well as NTFPs. The ATE, ATT, and ATU estimates for timber and other wood products were 66.8%, 66.2%, and 68.9% respectively. Likewise, the ATE, ATT, and ATU for NTFPs were 80%, 79.5%, and 81.7% respectively.

Taken together, these results show that the collection or processing of forest products (in terms of both timber and wood products and NTFPs) significantly reduced the rate of food insecurity in Liberia. The higher effects of NTFPs suggests that NTFPs are indeed crucial pathways for food security benefits of forest sector participation among forest proximate households in rural Liberia – a finding that is consistent with many previous studies which identified NTFPs as a key pathway for food security impacts of forests in LMICs (Bakkegaard et al., 2017; Barany et al., 2004; Gurung et al., 2021; Ickowitz et al., 2016; Liu & Xu, 2019; Rasmussen et al., 2017). The extent to which these effects persist in Liberia and implications for long-term sustainability of forestry in line with environmental management and sustainability concerns in the face of climate change and global food insecurity merits further research.

Table 6 Impact pathways: Teescount estimation of the effects of forest-sector participation on household food security
Full size table

3.4 Robustness checks

We ran several robustness checks to ascertain the robustness of our main estimates. First, we re-estimated the main endogenous switching poisson regression model with a different specification: endogenous treatment poisson regression (ETPR). We then ran a series of pos-hoc estimations to determine the food security effect of forest sector participation, both in terms of distance from a forest, and by regional clusters (to account of regional and county-level variations in food security). Moreover, we ran a multiple imputation poisson and negative binomial regressions to determine if the results would be significantly different from our main estimates.

3.4.1 Endogenous treatment poisson regression

First, we ran an ETPR, which, unlike the espoisson model, typically requires an IV for identification (Ma & Wang, 2020). Our use of the ETPR can demonstrate the validity of, and sensitivity of our main estimate. Using the same analytical approach pertaining to the FIML algorithm and the exact covariates used in the main espoisson regression, we found that forest sector participation reduced the number of months households had insufficient food by 78% (Table A3). This result supports our main result in Table 4. Note that like the espoisson regression, ETPR also has a first stage regression result, which shows the determinants of forest sector participation as well as diagnostics tests such as the rho and Wald Chi-square test of independent equations, which show whether our model accounts for endogeneity. These estimates, like the espoisson model, show a good model fit (based on the Log Likelihood ratio and its significant Wald Chi-square value). They also show that unobservable factors, which affected forest sector participation also affected the food security effects of participation. These results suggest that our main estimates are robust.

3.4.2 Post-hoc estimation as additional robustness checks

In addition to the ETPR as a main robustness check for the main estimate, we ran a series of post-hoc estimations by sub-dividing our data into several sub-categories based on: i). distance from the household to the nearest forest and ii). by regional clusters to determine if the estimates are consistent with our main results. The post-hoc estimation results (Tables A4 – A6) support our main estimates. Likewise, multiple imputation results (Table A7) supported our main estimates. The result also implies that missing values and omitted variables did not adversely affect our analysis – consistent with relevant prior studies elsewhere (Jamshidian & Jalal, 2010; Royston, 2004).

Taken together, the results of these additional specificationsFootnote 4 (Tables A3 – A7) supported our estimates and thus support our hypothesis that forest sector participation reduces food insecurity in Liberia.

4 Discussion

Food insecurity through hunger and poverty remains a vexing problem in LMICs like Liberia (World Bank, 2022). Forests can contribute addressing this challenge and achieving the first two SDGs: on poverty and hunger, respectively (Miller et al., 2022; Timko et al., 2018). For example, a recent study from Liberia (Amadu & Miller, 2024) showed that households living near forests who participated in the forest sector through the collection and processing of forest products had per capita annual income gains of 139% compared to other households. This and other results imply that forests can provide a pathway out of poverty and food insecurity. Yet, a rigorous analysis of food security effects of this pathway has been lacking.

We have addressed this gap by estimating the effects of forest sector participation on household food security in terms of the number of months households living in the proximity of forests in Liberia reported having insufficient food. Our topline finding is that forest sector participation increased food security in Liberia: participants saw their food insecurity rate decrease by 84% in 2019. We also found that both timber and other wood products as well as NTFPs increased food security among Liberian forest-proximate households. The average treatment effect of the collection or processing of these products reduced the number of months households reported having insufficient food by 66% for timber and other wood products, and 80% for NTFPs.

Our study makes at least three contributions. First, we contribute to the literature at the intersection of forests and food security by building knowledge on the role of household participation in the forest sector – rather than forest cover per se as commonly done (Asprilla-Perea & Díaz-Puente, 2019; Chilongo, 2014; Hall et al., 2022) – and food security outcomes in a country, Liberia, where forests, forest reliance, and food insecurity are high and overlapping.

Second, our use of count data models that carefully account for endogeneity and selection bias remains rare in studies of the relationship between forests and food security impacts. Despite a growing literature on the connection between forests, trees, and food security (e.g., Gurung et al., 2021; Hickey et al., 2016; Ickowitz et al., 2016), there remains a lack of micro-level evidence of the effects of the forestry sector on food security in terms of the number of months households experience food insecurity in LMICs. Our empirical approach addresses potential issues relating to endogeneity of forest sector participation and selection bias in the estimates of food security effects of participation in the forestry sector. In so doing, it helps to enhance analysis of food insecurity and malnutrition outcomes in forest-rich regions in the face of climate change and extreme weather shocks (Acevedo et al., 2018; Bahar et al., 2020; Fujimori et al., 2022; Hickey et al., 2016; Kreidenweis et al., 2016).

Finally, our study makes an empirical and policy-relevant contribution by using a newly-available and still uncommon nationally-representative dataset with forest-related information (World Bank, 2020b). The national level is often especially important for forest policymaking (McDermott et al., 2010). By using household-level information that is representative at a national-scale, our results therefore have special relevance to forest and land use policy as well as food security and poverty reduction in Liberia with potential insights for similar contexts elsewhere. This analysis can help shed some light on the importance of forests as a viable pathway to sustainable food security and related SDGs like zero hunger and climate action (UN, 2023). Our analytical approach can be applied to analyses in other countries where similar data are available and of other social-ecological impacts of household participation in activities related to forests in Liberia and elsewhere in SSA (e.g., Agwu et al., 2011; Kamugisha et al., 2022; World Bank, 2020a).

4.1 Limitations and future research implications

Although the analytical approach we utilized has proven quite robust, there are some limitations that we wish to highlight. First, our use of cross-sectional data without an explicit baseline limits us from explicitly addressing all potential issues of endogeneity in this study. The analytical approach we employed is as rigorous as possible given available data, but future research could address this potential shortcoming by collecting and using panel data, with the data we used as a baseline.

A second limitation is that the dataset we used focuses mainly on households living in the proximity of forests in the rural communities across Liberia, but does not capture the entire population of Liberian households in all rural or urban areas. However, our sample is representative of forest proximate households in all 15 counties in Liberia. Further, we applied post-hoc estimations at various distance categories to analyze the extent to which distance from a forest affect food security and found our results to be quite robust. Likewise, we estimated regional cluster-level post-hoc analysis to examine the differences in food security of forest sector participants at the regional cluster levels. Future research can complement our results by also looking at these larger populations to gain a sense of whether and how distance to a forest matters to food security outcomes.

Additionally, data limitations prevented analysis of detailed information on which foods were scarce during the 3.2. average month-long hungry period or on which foods were cultivated on an average of the 0.9 hectares per household. Likewise, we do not have information on the nutritional status of women and children at the household level (such as information on stunting and wasting in children, or iron deficiency in women and teenage girls). However, the dataset did provide relevant information on constraints like shocks such as weather, market, and diseases and pests, which are important factors affecting household-level food security in LMICs like Liberia (Niles & Brown, 2017; Waage et al., 2022).

Furthermore, we do not have data on the exact months of the calendar year that households reportedly had insufficient food. For example, the predominantly hungry months include July, August, and September (from personal conversation). However, the data we utilize did not include this crucial piece of information. Therefore, future research using data that captures such nuanced details can help provide rigorous food security estimates of forest sector participation in Liberia.

An additional limitation is the number of missing values in our sample. Although we analyzed data from 2983 households, there were only 1408 cases that reported the number of months they had insufficient food. About 53% of households had missing data on this important variable. Thus, our estimation methods were based on these 1408 complete cases, which means that we might have lost some important information in the missing data. We were not able to identify a consistent reason why these values might be missing and if the missing values found have substantive value, or if they are an artifact of other unobserved factors. We have sought to address this issue using a full information maximum likelihood analytical technique that accounts for missing data, but it should be noted as a potential limitation.

Our definitions of food security and forest sector participation may also be limited because we do not directly observe the households and the precise dates when the households undertook specific forest sector-related and other activities are not specified. However, such more precise data remain very rare (Hajjar et al., 2021) even as their availability could form the basis for even more reliable results. Additionally, our definition and measurement of food security mainly concerns with households reported experience of food availability and does not capture other important aspects of food security such as economic access to food, food utilization – in terms of the quality of food and how different members of the households consumed such available food, and the stability of food security over time (Jones et al., 2013). Future analyses of the food security effect of forest the forestry sector should employ such metrics.

It is worth nothing that because different organizations often utilize different measures of food security, it is increasingly important for any metrics used in the analysis of food security to be validated against other existing metrics across context (Béné et al., 2023; Frongillo, 2022; Jones et al., 2013). Data limitations prevented us from comparing our food security metric to other metrics like the Household Food Insecurity Access Scale, Household Dietary Diversity Score, and Months of Inadequate Household Food Provision. For example, a recent study (Charamba et al., 2023) proposed the application of item response theory modelling for measuring an aggregate food security access score comprising such popular metrics. Future empirical research from Liberia or similar context can test our approach against these metrics to ensure a more robust impact assessment of food security attributable to the forestry sector.

Relatedly, because the survey data we used focused on forest-based activity consumption outcomes (e.g., the food security proxy variable of the number of months households reported having insufficient food in the 12 months immediately before the survey), questions on other sources of food security are more aggregated and receive less emphasis in this analysis. As such, our results may be affected due to non-forest-based food consumption reporting, compared to, for example, a detailed household and farm-level agricultural survey capturing more detail on the food security status of study households.

Furthermore, a set of regression analyses of income and expenditure on food can be quite useful in showcasing the food security effects of the forestry sector in rural Liberia and in similar contexts elsewhere in West Africa. Although that line of analysis was beyond the scope of this paper, we recommend that future research should consider such an important analysis to ensure a more robust and policy-relevant estimate of food security effects of the forestry sector.

Like other relevant studies that have applied falsification tests to ascertain the validity of their exclusion restriction in impact assessment research (e.g., Amadu & Miller, 2024; Coulibaly et al., 2017; Danso-Abbeam et al., 2021a, b), our exclusion restriction and the result of our analyses should be interpreted with caution due to potential limitations with instrumental variable techniques compared to the use of other rigorous techniques like randomized control trials (RCTs). Future analyses can use RCTs to provide more robust estimates of the effects of forest sector participation on food security in Liberia and similar contexts elsewhere.

A final important research direction is to examine environmental sustainability of our food security findings. Collecting forest products, a fundamental aspect of forest sector participation, may lead to deforestation and other negative environmental outcomes if not properly coordinated at the community level or incentivized in policy. Further research is needed to examine multiple dimensions of sustainability simultaneously in our study area as more generally (Agrawal & Chhatre, 2011).

5 Conclusion

This study has estimated the effects of forest sector participation on household food security among households living in the proximity of forests in rural Liberia. Forests are increasingly viewed as a crucial pathway to sustainable food security in low- and middle-income countries like Liberia in the face of climate change and extreme weather shocks. However, prior studies have not provided rigorous empirical evidence about the food security effects of household participation in the forestry sector in such contexts. We have responded to this gap by estimating the effects of forest sector participation on household food security in terms of the number of months households reported having insufficient food per year in Liberia.

We conclude that forest sector participation has had a statistically significant and substantively meaningful effect on food security in Liberia. We also conclude that both non-timber forest products (NTFPs) and timber and wood products are significant pathways for household food security in Liberia. Of these two pathways, study households relied more on NTFPs for their food security needs than timber and wood products. This finding suggests the need for policies and programs to maximize and sustain the returns of NTFPs to enhance sustainable food security in rural Liberia.

Taken together, the results of this study lead us to conclude that forests can contribute to food security and enhance the achievement of the Sustainable Development Goals on hunger in Liberia and similar contexts elsewhere.