Soil conservation and smallholder welfare under cassava‐based systems in Thailand

Land degradation, declining soil fertility, and erosion continue to plague agricultural production in many developing countries. In response to these farm production constraints and environmental challenges, a range of soil conservation technologies and practices have been developed and disseminated to tackle soil nutrient and fertility declines. However, evidence on the association between soil conservation, farm performance, and smallholder welfare is scarce. In this study, we examine the relationship between soil conservation, farm performance, and welfare outcomes of smallholder cassava farmers in Thailand. We use a farm household survey of 602 cassava growers and apply a doubly robust multivalued treatment effect estimator to estimate the relationship between soil conservation, farm performance, and welfare as well as the observable characteristics associated with the use of soil conservation practices. We observe a positive association between the use of soil conservation practices and cassava yields which is most likely associated with higher‐income streams. Similar insights are also observed for other welfare outcomes such as asset accumulation, including livestock which represents rural wealth to a considerable extent. The positive association between soil conservation and livestock ownership hints at some form of rural diversification. Given these insights, our analysis supports and gives credence to initiatives that promote the adoption of soil conservation as they are not only pro‐poor but also environmentally friendly with significant concerns for ecological safeguards.

also degrade soil or deplete large amounts of nutrients, leading to soil exhaustion and biodiversity loss (Delaquis et al., 2018).
In response to such farm production constraints and environmental challenges, a range of technologies and practices have been developed and disseminated to farmers in many developing nations where food insecurity and rural poverty continue to take precedence. A case in point has been the implementation of the Nippon Foundation Project (NFP) between 1993 and 2003 in Thailand and other Asian countries, which sought to reduce environmental degradation and improve soil fertility among cassava farmers. Farmers were introduced to a blanket of soil erosion and soil fertility practices that were intended to prevent the loss of soil nutrients, restore degraded and depleted soils as well as improve soil quality. For brevity, we broadly refer to these soil erosion-preventing practices and soil fertility-enhancing practices as soil conservation practices (Table 1).
This study examines the relationship between soil conservation and four welfare outcomes (yields, income, asset ownership, and livestock ownership) in cassava-based systems in Thailand. We take advantage of the implementation of the NFP between 1993 and 2003, which sought to reduce environmental degradation and improve soil fertility among cassava farmers. The NFP aimed to disseminate improved farm production practices with potentials of increasing farm performance and welfare while also preventing soil degradation by nutrient depletion and soil erosion. We revisit the project area in 2017, interviewing 602 cassava farmers and collecting extensive information on their farm management decisions. We then investigate how the adoption of soil conservation is associated with farm productivity and household welfare.
From an agronomic perspective, it has been documented that the uptake and use of soil conservation practices increase yields while reducing soil degradation (Adams et al., 2020;Adolwa et al., 2019;Nunes et al., 2018;Xiong et al., 2018). However, the impacts of the adoption of these practices may also be context-dependent. For instance, Adimassu et al. (2017) reported decreasing yields under the use of some soil and water conservation techniques. On the other hand, the use of agronomic soil and water conservation practices such as organic soil amendments, intercropping and composting were found to increase yields (Adams et al., 2020;Adolwa et al., 2019;Nunes et al., 2018;Xiong et al., 2018).
Adoption of soil conservation has also been shown to increase farm profits (Pannell et al., 2014). Beyond yield and profit increases, adoption of soil conservation could further induce welfare changes given that the benefits of adopting soil conservation are not immediate 1 (Knowler & Bradshaw, 2007;Lapar, 1999;Pannell et al., 2014). The adoption of organic soil amendments, a critical component of soil conservation, has been shown to increase food security (Tabe-Ojong, Fabinin, et al., 2022).
That said, existing studies report significant heterogeneity and contextdependency, calling for a wider evidence base to improve learning on the relationship between soil conservation and welfare.
Our study offers three main contributions to the empirical literature on the relationship between the adoption of soil conservation and smallholder welfare. In the first place, we provide evidence of the relationship between the adoption of soil conservation and welfare in the context of cassava-based systems. This is quite scarce in the adoption literature, especially as far as soil conservation is concerned.
Given that cassava cultivation can take place on marginal and degraded soils, as well as simultaneously cause environmental degradation through critical soil disturbance, understanding the welfare implications is crucial. This could avoid trade-offs while ensuring synergies in achieving some of the sustainable development goals (Barbier & Hochard, 2018;Barrett & Bevis, 2015). Second, unlike studies in the broad adoption literature, we consider a plethora of different soil fertility and erosion practices and establish how they are associated with the welfare and livelihoods of smallholder farmers.
Last but most importantly, we consider a shrubby root and tuber crop, cassava, which is mainly cultivated by resource-poor farmers who usually operate on marginal lands (Alene et al., 2018;Delaquis et al., 2018). This is, even more, the case in Thailand where cassava is a key export/cash crop 2 for smallholder farmers. Cassava is also one of the largest calorie-producing crops in the root and tubers family, making it not only an important crop in resource-poor agricultural settings but also in the development of biofuels (Zhou & Thomson, 2009). Being a tropical crop, it thrives under high temperatures and withstands seasonal droughts (El-Sharkawy, 2014). In this regard, understanding the welfare association of soil conservation under cassava-based production systems has deep implications on the ongoing debates about the hitherto considered environmentally benign crop which is increasingly argued to contribute to environmental degradation, threatening sustainable crop production (Reynolds et al., 2015).

| CASSAVA PRODUCTION AND SOIL CONSERVATION
Cassava production, both in terms of area under cultivation and yields has increased over the last 50 years and is mostly cultivated in the T A B L E 1 Soil conservation practices promoted by the Nippon Foundation tropics (Delaquis et al., 2018;Molua et al., 2020 South-East Asian countries is cultivated and regarded as a cash crop with a more industrial orientation. However, production is in the hands of small scale and usually resource-poor farmers. In these countries, cassava is mainly processed into dry chips and pellets and exported to be used as starch, and animal feed (Alene et al., 2018).
The growing area and importance of cassava throughout the global tropics are due to its attractiveness to small-scale and resource-poor farmers who constitute the majority of households in these tropical and subtropical regions. Given this, cassava cultivation has the potential to achieve targets such as reducing hunger and food insecurity, and increasing smallholder incomes and welfare while stirring development in rural areas. Beyond serving the purposes of food and improving welfare, cassava is also used as a feedstock in the production of biomass energy (Zhou & Thomson, 2009 Because cassava is cultivated and thrives well on degraded and marginal lands, there is usually little addition of chemical fertilizers or organic soil amendments to improve the fertility of the soil (Delaquis et al., 2018). This could explain the global low yield levels and yield gaps observed in many cassava-producing regions (Fermont et al., 2009). Closing these yield gaps and achieving yield increases is possible under the use of effective weed management strategies, chemical fertilizers, organic soil amendments and the use of various soil fertility and erosion practices that improve soil fertility, and nutrient retention and prevent further soil erosion, exhaustion, and degradation (Fermont et al., 2009).
Cassava cultivation could also lead to soil erosion and degradation (Reynolds et al., 2015;Valentin et al., 2008). This is due in part to its mode of harvesting which involves critical soil disturbance, poor canopy cover which is less resistant to heavy rains, and its potential to thrive on degraded soils (Valentin et al., 2008). Despite this, cassava cultivation is also prone to emerging infectious diseases arising from arthropod pests, pathogens, and other soil microorganisms when the soil biological function declines and the plant builds a low immune response due to inadequate management of the crop (Graziosi et al., 2016;Vurro et al., 2010). All these reduce yields and make it seemingly hard to close the yield gap. However, the use of various soil fertility-enhancing techniques and soil erosion and exhaustion preventive practices may make it possible to fully explore the benefits inherent in the cultivation of cassava while tackling some of its associated environmental implications.  Dalton et al., 2011). Increase in adoption and use of these management innovations may have led to a substantial increase in cassava productivity with ensuing implications on other welfare outcomes. Some of the practices promoted by the NFP are shown in Table 1.

| FARM HOUSEHOLD SURVEY
An ideal data set to use to understand the relationship between soil conservation, farm performance, and welfare outcomes would either were shared with the wider community.
We used a purposive and stratified sampling strategy to first select villages in the main cassava-producing provinces, and secondly to randomly select households within these villages. From each province, we included all the project villages (eight villages). We then

| Measurement of soil conservation
As earlier introduced, soil conservation refers to a suite of practices either used to improve the quality and fertility of the soil or to prevent further soil loss, erosion, and degradation. For this study, we refer to soil conservation practices as either soil fertility practices or soil erosion practices. While soil fertility practices include crop rotation, green manuring, intercropping, use of chemical fertilizers, organic fertilizers, use of crop residues, foliar fertilization, and the practice of subsoiling and the use of inoculants, soil erosion practices include the use of hedgerows, contour ridges, water harvesting, terracing, mulching, soil and stone bunds, minimum tillage, and tree planting (full list of practice is shown in Table 1). The summary statistics of the use of the various practices are shown in Tables S1 and S2 in the appendix.
Information on practices was asked at the level of each plot. In this case, only those corresponding to the cultivation of cassava were used, and the same farmer could use all the practices or not use any, as well as use it on multiple plots or only on one plot. As can be seen from these tables, the adoption of some of these practices is very low (<1% of farmers are using them). We thus define adoption based on whether every representative farmer is using either one of the soil fertility enhancing practices or soil erosion preventive practices. Under the use of soil fertility enhancing practices, about 80% of farmers are applying chemical fertilizers on their farms. Given that the use of chemical fertilizers is conceptually and technically different from the use of other soil fertility-enhancing techniques which can be mainly considered as the use of organic soil complements, we created a separate adoption category for the use of chemical fertilizers. For instance, it can be argued that the marginal yield response to fertilizers may be different based on either the use of organic fertilizers or chemical fertilizers. Yield increases due to chemical fertilizers may be lower on degraded soils and soils with low soil organic matter as compared to fertile soils (Barrett & Bevis, 2015;Rowe et al., 2006). Thus, we have three categories of adoption as shown in Table 2: use of chemical fertilizers, use of soil fertility practices, and use of soil erosion preventive practices.

| Measurement of outcome variables
We test the association between the adoption of soil conservation practices and several welfare indicators. To begin with, yield was measured using the total quantity of cassava produced and divided by the area under production. This information was gotten at the plot level, for which the total amount of production and the corresponding areas for each plot were added and later collapsed at the farmer level. Given that yield increases could translate to other welfare outcomes, we also used proxies for income and asset ownership. Under income, we used the share of farm income obtained from the sales of cassava. Given the increasing relevance of asset-based outcomes as better proxies of rural poverty, as opposed to income and consumption outcomes which are liable to short-term fluctuations, we use the total value of all assets owned by a household (Brockington,

| MATERIALS AND METHODS
A preferred empirical strategy to understand the association between soil conservation, farm performance, and welfare outcomes would be to leverage an estimation strategy like the difference in difference technique using a multi-wave data set or better still panel data methods that can control for unobserved time-invariant heterogeneity and establish some dynamics. However, it is impossible to reconstruct a baseline for this study. This makes it hard to observe the impacts of soil conservation on welfare. That notwithstanding, we make use of the cross-sectional data to understand the association between soil conservation and welfare. In doing so, we control for observed confounding factors and employ some rigorous estimation techniques in establishing this association.
We use the doubly robust multivalued treatment effect estimator (Cattaneo, 2010;Cattaneo et al., 2013; Tabe Consider the following regression model depicting the potential outcome framework (Rubin, 1974): Where: Y i refers to the outcome variables (farm performance and welfare indicators), C i is the multivalued treatment variable which take on integer values between 0 and 3 where 1 represents the use of soil erosion practices, 2 is soil fertility practices, 3 is the use of chemical fertilizers and 0 is non adoption. Y ic represents the potential outcome for every household i for which C i ¼ c. For each household, only one of the potential outcomes is observed, depending on the adoption of the soil conservation practice. D it C i ð ) is an indicator for adopting any soil conservation practice, c by a representative household i and is represented as More formally, the treatment model and the outcome models can be represented as An important assumption of doubly robust estimators is that conditional on the specified control variables, each observation has a non-negative probability of being an adopter. This assumption, also known as the overlap assumption, creates a counterfactual adoption observation for each adopting observation. The violation of this assumption may lead to less precise estimates. The tolerance level for the probability of being an adopter ranges between 0 and 1. We observe sufficient overlap in the probability of adopting soil conservation. We also perform some balance checks to assess pre-estimation balancing between treated and control units. As shown in the supplementary material, few of the weighting variables have a standardized difference greater than 0.25, which is within the conventional thresholds without reducing our sample size (Imbens & Wooldridge, 2009).
The variance ratios are also close to one. We also perform some overidentification tests where we examine the null hypothesis that covariates are balanced. The test statistics are χ 2 (15) = 1.049 with p > χ 2 = 1.00. This suggests that the weighted samples are well balanced. All these checks provide sufficient credence to our estimated coefficients.

| Regression results
Here, we present the results of the multivalued treatment effect model for the various welfare outcomes: yields, income, asset value, and index as well as livestock value and index.

| Soil conservation, yields, and income
Estimating the relationship between soil conservation and on-farm performance outcomes like yields and farm income, we observe intuitive and expected results as shown in Table 3. Adoption of soil erosion practices is associated with yield increases of about 21%. This is also very similar to the adoption and use of soil fertility enhancing practices and chemical fertilizers which are associated with yield increase of approximately 24% and 23% respectively. Yield increase from the use of these soil conservation practices is expected given that these measures are not only meant to retain and maintain soil nutrients but also to add soil nutrients in the case of soil enhancing measures like the application of organic and chemical fertilizers. Previous empirical analysis (Adams et al., 2020;Adolwa et al., 2019;Nunes et al., 2018;Xiong et al., 2018) found similar results on the positive association between the adoption of soil conservation and yields. In some cases, the adoption of soil conservation practices such as the use of organic soil amendments has been shown to be associated with food security (Tabe-Ojong, Fabinin, et al., 2022). For the more specific case of Thailand, earlier analysis from this same project area found substantial yield increases because of the farmer participatory research and extension approach (Dalton et al., 2011). Observations 600 600 Note: Additional controls include the gender of the household head, age and educational level of the household head, educational level of a spouse, household size, number of working-age adults, access to credit facilities, size of landholding, number of crops cultivated, the incidence of pest and diseases, drought shock, flood shock, training and number of known users of soil conservation practice. Robust standard errors in parenthesis *p < 0.1 **p < 0.05 ***p < 0.01

| Correlates of the adoption of soil conservation
Now that we have established the positive association between soil conservation, farm performance, and welfare outcomes, we are interested to know and understand the correlates of the use of these various farm practices that not only conserve the soil but have implications for nutrient increases and soil fertility. Our use of the doubly robust multivalued treatment estimator makes it possible to easily obtain correlation as shown in Table 5

| Limitations of the analysis
Our analysis has some limitations that could be taken up in future research endeavors. In the first place, we refer to the estimated Note: Robust standard errors in parenthesis *p < 0.1 **p < 0.05 ***p < 0.01

| CONCLUSIONS
In this study, we examine the relationship between soil conservation, farm performance, and welfare outcomes in cassava-based production systems in Thailand. Adoption of various soil conservation practices such as soil erosion reducing practices, and soil fertility enhancing practices have been promoted and disseminated to farmers as part of the Nippon Foundation Project in Vietnam, Thailand, China, andIndonesia between 1993 and2003. We return to these sites in Thailand approximately close to two decades later to understand the adoption of these soil conservation practices and their association with on-farm performance outcomes such as yields and farm income, as well as rural diversification outcomes such as off-farm income and livestock ownership and longer-term measures of poverty such as asset accumulation.
Leveraging a farm household survey of 602 smallholder farmers in Thailand and employing a doubly robust augmented inverse probability weighting multivalued estimator, we report the positive correlation between soil conservation and yields. Such yield increases could explain income increases which could also be translated to investments in livestock accumulations. Investments in livestock represent rural diversification to a considerable extent and offer some insights into the complementary gains inherent in the adoption of soil conservation. We also report a positive correlation between the adoption of soil conservation and asset ownership which not only represent rural wealth in most farming settings but can also be considered to represent longer-term measures of poverty more fully.
Based on the above results which underscore the positive short and long-term welfare implications of the adoption of soil conservation. Our analysis gives credence to the promotion, distribution, and diffusion of both soil erosion preventing practices and soil fertility enhancing practices as they can be both pro-poor and environmentally friendly especially in the context of land degradation and soil erosion. In such difficult zones, the cultivation of a shrubby tuber crop like cassava is important as it not only survives in such difficult soil conditions and marginal soils but also is less costly in terms of the required inputs to increase productivity and close existing yield gaps.
Hence, cassava can be regarded as a crop that is attractive to resource-poor farmers. Despite this, some opponents of the crop have it that the cultivation of cassava could also further degrade the soil and lead to soil erosion through its means of harvesting which critically disturbs the soil. While this may be true, the adoption of soil conservation practices in cassava-based production systems could address these concerns and offset these environmental losses.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are openly available in  (Beaman & Dillon, 2018).