1 Introduction

One of Professor Keijiro Otsuka’s major works is investigating the adoption process and socioeconomic impacts of the rice Green Revolution. His achievements in this field are exemplified in the edited volume by David and Otsuka (1994), which summarizes the experience of the rice Green Revolution in various Asian countries. Recently, Professor Otsuka has been actively examining and advocating the possibility of applying the Asian-style Green Revolution to Africa (Otsuka and Larson 2013, 2016). In collaboration with Professor Otsuka, Yuki Mano, and myself, another edited volume will be published in 2023 on the rice Green Revolution in Sub-Saharan Africa (SSA) under a research project funded by the Japan International Cooperation Agency (JICA).

Like many other agricultural technologies, the research and development of the modern variety (MV) of rice during the Green Revolution era in Asia was a science-based innovation, and its validity has been repeatedly tested and proven before they were introduced to farmers. The Asian Green Revolution did not end with a one-shot development of miracle rice with high-yield potential; rather, it underwent continuous improvements, such as the development of rice varieties that are resistant against various pests and diseases and with better eating quality in terms of taste, softness, and aroma (Otsuka and Larson 2013). During the adoption and diffusion process, agricultural extension agents conveyed to farmers the technologies evidenced in experimental stations of scientific research institutes.

The system of rice intensification (SRI) follows a different development path from conventional plant breeding. SRI was inductively developed and assembled based on field observations and trials by Fr. Henri de Laulanié, a French missionary priest/agronomist who worked with rice farmers in Madagascar for several decades. Since the first published report in a technical journal in 1993 (de Laulanié 1993), SRI has attracted the attention of development practitioners and researchers as it has demonstrated significant yield improvements without relying on additional external inputs (e.g., MVs and chemical fertilizers). However, until recently, empirical verification of the yield potential of SRI was scarce, and many early reports on SRI were informal, anecdotal, or from gray literature rather than peer-reviewed sources with scientific rigor (Berkhout and Glover 2011). One conference report even demonstrated an extraordinarily high yield of greater than 20 tons/ha in Madagascar (Rafaralahy 2002), which invited criticism from crop scientists, such as Dobermann (2004) and Sheehy et al. (2004), who argued that the reported yield was beyond the theoretically maximum achievable output. Sinclair and Cassman (2004) derided SRI as an unconfirmed field observation (UFO).

Although SRI research in agricultural economics is still limited, empirical results have accumulated over several decades. This chapter aims to review recent literature on SRI in agricultural economics and derive a tentative but somewhat generalizable view of SRI’s yield-enhancing effects.

2 System of Rice Intensification (SRI)

2.1 Origin and Characteristics of SRI

SRI, pioneered in the mid-1980s, originated from unusual practices in the farmers’ fields that Father Henri de Laulanié observed in Madagascar (Uphoff 2006). Several farmers there transplanted single seedlings instead of clumps of three to five plants per hill, unlike prevailing worldwide practices. Moreover, they did not continuously flood their rice fields. To determine whether such low-input practices jeopardize productivity, de Laulanié attempted to grow single seedlings in unflooded soil during vegetative growth. After several attempts, he developed the practice of planting seedlings in a square grid to utilize the mechanical hand weeder perpendicularly in two directions. This, in addition to reducing the burden of weeding, helped aerate the soil. These practices, together, led rice plants, especially their roots, to grow vigorously. He then established a training center to teach young farmers the new cultivation system. From 1983 to 1984, late rains forced him and his students transplanted premature seedlings in some parts of the field. This led to the accidental discovery that transplanted young seedlings tillered better and thus, produced higher yields than transplanted older seedlings.

Building on these observations and further trial and error (e.g., shifting from the use of chemical fertilizer to compost after the government removed the fertilizer subsidy in the late 1980s), de Laulanié and his successors gradually synthesized the following set of high-yielding practices: (1) Early transplanting of young seedlings (less than 15 days old, preferably 8–12 days), contrary to the conventional practice of transplanting 20–60 day-old seedlings; (2) transplanting one or two seedlings per hill, in contrast to a bundle (4–5) of seedlings per hill; (3) wide spacing (more than 20 × 20 cm [cm]), in contrast to narrow (10–15 cm) or random spacing between hills; (4) alternate wetting and drying (AWD) to maintain moist, aerobic soil conditions, in contrast to continuous flooding from transplanting to maturity. Moreover, the proponents of SRI often advocate using compost or manure instead of chemical fertilizers to enrich soils with organic matter (Raharalahy 2002; Uphoff 2006), even though de Laulanié did not initially emphasize any fertilization method. Early and regular weeding, preferably with a mechanical rotary weeder, is also highly recommended as a rotary weeder helps remove weeds that can defeat young seedlings and churn and aerate the soil.

Unlike the Asian Green Revolution technologies, SRI does not require additional external inputs, such as fertilizer-responsive MVs, which genetically improve yield potential, chemical fertilizers, and protective agrochemicals. Instead, SRI has been claimed to elicit more productive phenotypes from existing rice genotypes by changing the management of the plants, soil, water, and nutrients (Uphoff 2003; Sato and Uphoff 2007). The reduced costs of external inputs (e.g., seeds and chemicals) were important considerations as many Malagasy farmers were poor and credit constrained. Although SRI generally requires more labor inputs for careful crop, soil, and water management, labor is relatively abundant for the poor. Therefore, SRI is often considered a pro-poor management practice despite its complex, knowledge-intensive nature, and has indeed been disseminated among low- and medium-income farmers in more than 50 developing countries.

The four to six core SRI principles described above are typically recognized as a package, as they are believed to have synergistic effects (Stoop et al. 2002; Uphoff et al. 2008). However, actual practices can vary among farmers across places as SRI can be adapted to each specific locality and has been continuously evolving based on participatory on-farm trials. The age and space of transplanted seedlings are flexibly adjusted. SRI variants can even include direct seedlings under rainfed conditions, wherein careful crop and water management may be difficult. Uphoff (2003) claims that SRI is a ‘philosophy’ or even ‘paradigm shift’ rather than a prescribed set of technologies that farmers should firmly follow, as is common in many other agricultural technologies (Berkhout and Glover 2011).

2.2 Controversy Surrounding SRI Yield Potential

Since its first introduction by de Laulanié in 1993, various on-farm trials have demonstrated significant SRI yield gains over conventional management practices (Uphoff 1999, 2003; Stoop et al. 2002; Sato and Uphoff 2007). However, partly because of its nontraditional development path, SRI has proven extremely controversial.

This debate escalated after Rafaralahy (2002) reported that yields per hectare exceeded 20 tons in Madagascar. For example, Sheehy et al. (2004) conducted comparisons between SRI and conventional management practices at three well-controlled experimental stations in China and found no significant yield differences among them. Sheehy et al. (2004) also presented a theoretical model of rice yield and emphasized that the reported ‘fantastic’ SRI yields by Rafaralahy (2002) would exceed the photosynthetic efficiency of rice and are likely the result of measurement errors. Similarly, Dobermann (2004) argued that SRI methods do not have inherent advantages over conventional best management practices, and their performance may depend on soils and other environments. He hypothesized that SRI may be particularly suitable for poor cultivation environments with acidic iron-rich soils and limited water availability, as in Madagascar, but its benefits would be small in more favorable environments. McDonald et al. (2006) reviewed 40 published journal articles and concluded that outside of Madagascar, where soil conditions are especially suitable for SRI practices, SRI has negligible or even negative impacts on rice yields relative to best management practices. Sinclair and Cassman (2004) echoed their claims and labeled SRI an agronomic UFO.

Skepticism of SRI has been addressed by Uphoff (2004), Stoop and Kassam (2005), and Uphoff et al. (2008), among others. Uphoff (2004) challenged the claims of limited applicability and gains of SRI by highlighting that SRI has been practiced in many regions beyond Madagascar, which is evidence of its intrinsic value. Stoop and Kassam (2005) criticized Sheehy et al. (2004), arguing that a single standardized field experiment cannot reveal SRI’s full potential. Uphoff et al. (2008) emphasize that the methodology used by McDonald et al. (2006) is flawed because selected observations do not represent the universe of SRI experiences and because many of their observations only partially adopted the core principles of SRI, hindering full synergistic advantages among practices.

Some controversy seems to arise as SRI is not clearly defined, even by its proponents. As explained earlier, Uphoff, a champion of SRI proponents, tended to emphasize SRI as an adaptable suite of principles for rice cultivation. However, when he criticizes others, he highlights the importance of adopting all principles as a package. As Bouman (2012) precisely emphasizes, “[p]roponents of SRI highlight results of superior SRI performance with practices that only partially follow the original or that are heavily modified, while they reject results of inferior SRI performance based on the fact that practices were incompletely or wrongly implemented (p. 2)”. According to this view, rather than SRI triggering positive impacts on yield, the package of rice management practices providing better yields should be called SRI. The lack of a clear working definition of SRI makes this controversy fruitless.

This chapter does not endorse or criticize SRI’s potential by revisiting the aforementioned physiological controversy. Rather, the recent empirical literature on agricultural economics in various contexts are reviewed, keeping internal validity (consistency and unbiasedness) and external validity (generalizability) in mind. I will attempt to explain the definition of SRI and the identification strategy employed in each paper as clearly as possible so readers can evaluate the credibility of the results on their own.

3 Assessment from the View of Agricultural Economics

3.1 Conceptual Framework

Before going into the details of each paper, let me explain the basic concept of how to compare adopters and non-adopters of SRI. All agricultural economics papers cited in this chapter explicitly or implicitly follow Rubin’s causal model in the following form:

$${y}_{i}={D}_{i}{y}_{i1}+\left(1-{D}_{i}\right){y}_{i0},$$
(7.1)

where \({D}_{i}\) represents an adoption indicator with \({D}_{i}=1\) if farmer/plot i adopts SRI, and \({D}_{i}=0\) otherwise. The ith unit has two potential outcomes: \({y}_{i1}\) (outcome of adopting SRI) and \({y}_{i0}\) (otherwise). The causal impact of SRI adoption is expressed as

$${y}_{i1}-{y}_{i0}.$$
(7.2)

Equation (7.2) can create an ideal ceteris paribus condition wherein any observable and unobservable factors other than SRI adoption are ‘identical.’ However, because two potential outcomes cannot be observed simultaneously for any particular individual unit, most economics studies estimate the average treatment effects (ATE), \({E[y}_{i1}-{y}_{i0}]\) (where \(E[]\) is an expectation operator), or the average treatment effects on the treated (ATT), \({E[y}_{i1}-{y}_{i0}|{D}_{i}=1]\), by creating appropriate counterfactual groups based on a randomized controlled trial (RCT) or other statistical methods.

Note that as the two statuses above are distinguished only between with and without SRI, any associated changes cannot be controlled for. For example, if SRI requires more labor inputs but requires less water, seed, and fertilizer expenses, crop scientists may note that the resultant yield differences cannot be sorely attributed to the difference in technology parameters with and without SRI as input uses are different between the two. Although economics has developed a technique for estimating the total factor productivity (TFP), few studies have adopted it in SRI studies (Berkhout et al. 2015). Thus, this chapter considers the overall impact of SRI adoption (incorporating mediating impacts through changing input uses) when referring to yields. While past controversy has centered on yield potential, I will also discuss income and profitability as much as possible, allowing us to properly assess SRI’s true potential, taking costs and other associated changes in adoption into consideration.

3.2 Empirical Evidence of the Impacts of SRI

Barrett et al. (2004) presented the first seminal work on the yield potential of SRI. They collected original data from 111 randomly selected farmers from four sites in Madagascar who practiced both SRI and traditional cultivation methods (referred to as SRT). Although no explicit delineation between SRI and SRT was provided, footnote 6 (Barrett et al. 2004, p. 872) explains that farmers closely adhere to all the recommended core practices of SRI, which is plausible, considering that Madagascar is a mecca of SRI. SRI had an unconditional average yield of 6.3 tons/ha, compared to SRT’s 3.4 tons/ha. As this yield difference includes selection effects, the authors employed a variant of the random coefficient switching regression model, referred to as differential yield function estimation. This allowed them to decompose the total yield-enhancing effects into SRI itself and other plot and household characteristics. Their findings indicate that SRI increases yield by 84% relative to SRT, of which half is attributable to the adoption of SRI itself, including accompanying changes in input uses (and the rest to differential household and plot characteristics).

Approximately a decade since Barrett et al. (2004) published their article in the American Journal of Agricultural Economics, two papers from Southeast Asia were published: one by Noltze et al. (2013) with data from Timor Leste, and the other by Takahashi and Barrett (2014) with Indonesian data.

Noltze et al. (2013) applied the endogenous switching regression model (ESR) to examine SRI impact. Their sample comprised 475 plots from 397 households, of which approximately 35% used SRI. SRI plots were defined as plots adopting the four core principles (early transplanting, single seedling, wide spacing, and AWD). The unconditional average yield of SRI was 2.9 tons/ha, whereas that of non-SRI was 3.2 tons/ha. While ESR can be an unbiased estimate even when all covariates in the first-stage estimation of SRI adoption overlap with the second-stage estimation of the yield function, it provides more credible estimates with at least one instrumental variable (IV). The authors used the percentage of SRI training participants in the farmer’s village as IV. ATT estimated from ESR revealed that, on average, SRI will have 46% yield gains relative to conventional practices. However, its yield gains are not transformed into sufficient household income gains: SRI farmers increase their household incomes by only 2.3%.

My work in Takahashi and Barrett (2014) reached similar conclusions. In addition to substantial yield gains with negligible income gains, we identified the underlying mechanism. We used data from 1,202 plots among 864 sample farmers in South Sulawesi, Indonesia, wherein about 14% adopted SRI. An SRI plot is defined as one adopting at least one of the four core principles of SRI. The unconditional average yield of SRI was 5.5 tons/ha, whereas that of non-SRI was 3.0 tons/ha. Because of the lack of plausible IVs, we employed propensity score matching (PSM) for the estimation. As PSM can control only selection-on-observables, we also applied sensitivity tests to check whether unobservable factors contaminated our estimation results. ATT from PSM revealed 64% yield gains from adopting SRI. However, household income was not significantly different between farmers who adopted SRI practices and those who did not. This is because SRI increases rice income at the expense of off-farm incomes, owing to larger requirements of family labor for careful water and crop management, which induces labor reallocation into rice cultivation away from off-farm activities. Sensitivity tests revealed that unobservables may not significantly change the statistical inference. Because it may be questionable to call a plot applying only one principle as an SRI plot, we further implemented a robustness check with a stricter definition of SRI, wherein an SRI plot is defined as one that applied all four core principles. The results remained the same qualitatively.

The next three papers used samples from the same project in Tanzania. The project site is in the Kilombero District, Morogoro Region, wherein modified SRI was promoted by a large-scale rice farming company, Kilombero Plantation Limited. Unlike the four core principles of SRI typically promoted, the following six technologies were considered SRI practices at this site: (1) sorting of rice seeds, (2) direct planting or transplanting of one or two seeds per hole/hill, (3) wide-spaced seeds/seedlings on a 25 × 25 cm square grid pattern, (4) mechanical weeding using simple handheld weeders, (5) use of chemical fertilizer, and (6) use of an improved seed variety known as SARO 5. Alem et al. (2015) defined SRI plots as those in which four of the six technologies were practiced. According to this definition, 194 plots in their sample were SRI plots, and 140 were non-SRI plots. The unconditional average yield of SRI was 2.7 tons/acre (equivalent to 6.6 tons/hectare), whereas that of non-SRI was 1.06 tons/acre (equivalent to 2.6 tons/ha). The estimated results via ESR, using the number of years farmers have lived in the village and social networks measured by the number of group memberships as IVs, indicated that SRI increases yield by 58%. An obvious concern is that such yield gains stem from the use of MVs and chemical fertilizers, as only 12% and 5% of non-SRI plots (vs. 97% and 86% of SRI plots) apply them, respectively. To address this valid concern, the authors redefined SRI as plots that apply practices (1), (3), and (4) previously mentioned. This robustness check provided a qualitatively similar result, with 60% yield gains from SRI over conventional practices. As for household income, the results are mixed, depending on the price of rice used for calculation. The reported farm gate price is approximately 46% lower for MVs than for traditional varieties. Thus, if the price differences are genuine and SRI farmers tend to face lower prices, their household income becomes lower than non-SRI farmers. Conversely, if farmers are assumed to face the same rice price, SRI farmers generate approximately 40% higher incomes than non-SRI farmers. While income effects are thus ambiguous, robust yield-enhancing effects have also been reported by Sarr et al. (2021), who used the same dataset as Alem et al. (2015). Applying an ESR variant, Sarr et al. (2021) showed that SRI increased yield by 43% on average.

Nakano et al. (2018), in which Professor Keijiro Otsuka was a coauthor, used their original dataset from Kilombero District, Tanzania. This sample comprised 398 plots from 281 households with 110 SRI and 79 non-SRI plots owned by SRI adopters; the rest were owned by non-SRI adopters. The distinction between SRI and non-SRI plots was based on farmers’ self-reports. The unconditional average yield of SRI was 4.7 tons/ha, and that of non-SRI was 2.6–2.9 tons/ha. The difference-in-differences and PSM methods were separately used to address potential endogeneity. Their estimation results on ATT via PSM showed that SRI yield increased by 62%, and rice profits, defined as the gross output values minus paid-out costs and imputed costs of family-owned resources (e.g., labor and capital), increased by 285%, demonstrating a substantial profitability potential. While their study does not investigate the impact on total household income, any offsetting loss of earnings from other income sources is properly considered as long as the computation of imputed values of family-owned resources is correct.

Taken together, these observational studies indicate robust findings of the yield-enhancing effects of SRI over conventional practices. However, it must be noted that each study may have several caveats. For example, while Barrett et al. (2004) provided one of the most reliable estimates, their method was somewhat complex, and Chen and Yen (2006) highlighted several analytical flaws challenging its internal validity. Other studies using ESR may encounter a problem in that the selected IVs do not satisfy exclusion restrictions, whereas the studies that use PSM, including my own, may be criticized because it can control only selection-on-observables, no matter what sensitivity tests to unobserved confounders are executed.

To date, the most credible estimate of SRI impacts has been conducted by Barrett et al. (2022) based on an RCT, which is considered the gold standard for policy evaluation. They collaborated with the BRAC in rural Bangladesh and randomized SRI training intensities to farmers, where some villages received two-year training, some villages received one-year training, and the rest received none. Moreover, eligibility for participation in training within a village was randomized. So, their sample comprised 1,464 farmers in control villages (C or no one received training); 806 eligible and 507 non-eligible households in one-year training villages (T1 and U1, respectively); and 892 eligible and 462 non-eligible households in two-year training villages (T2 and U2, respectively). While BRAC promoted the four core principles of SRI, plus the use of organic matter amendments (e.g., compost and manure) and mechanical weeders, the adoption rates of these additional practices and AWD were low. Accordingly, Barrett et al. (2002) defined SRI plots as following the three basic principles of plant management: (1) early transplanting, (2) one to two seedlings per hill, and (3) wide spacing. SRI’s adoption rate, thus defined, was 0% among C, 9% among U1, 38% among T1, 12% among U2, and 53% among T2 groups two years after the first training intervention, with the unconditional average yield of 5.3 tons/ha, 6.1 tons/ha, 6.2 tons/ha, 6.2 tons/ha, and 6.3 tons/ha, respectively. The local average treatment effect estimator, which used the random assignment of training as IV, showed statistically significant impacts of SRI on yield and rice profits with magnitudes of 25% and 44% growth, respectively.

Overall, while some ambiguities may remain due to the lack of a firm definition of what constitutes SRI, all these cited studies identify the positive impact of SRI on yield in different contexts. Each study’s results will not be necessarily generalizable as they are contingent on and specific to the area and period studied. However, repeated observations of yield-enhancing effects of SRI and its variants from one place to another may collectively indicate that the results might be externally valid (i.e., generalizable). Therefore, the key takeaway message of this chapter is that the yield-enhancing effects of SRI are becoming less controversial compared to conventional practices (if not to alternative best management practices) and seem to no longer be a UFO in the field of agricultural economics. Meanwhile, there should be some reservations about whether SRI is a truly promising technology in terms of profit maximization given the increased labor requirement, at least in the early phase of adoption (Barrett et al. 2004; Takahashi and Barrett 2014), even though accumulating evidence demonstrates positive returns (Nakano et al. 2018; Barrett et al. 2022).

4 Discussion and Conclusion

This chapter reviewed recent literature in agricultural economics on the yield-enhancing effects of SRI and found positive impacts under various circumstances. Although I am not a crop scientist, I conjecture that careful management practices from nursery to harvest would reduce the efficiency loss to reach the production possibility frontier, if not advance the frontier itself. Indeed, while the Asian rice Green Revolution has often been linked to intensive use of MVs and agricultural chemicals only, Otsuka et al. (2022) argued that it was management-intensive in nature, requiring the selection of good quality seeds, the implementation of proper land preparation (e.g., leveling and bunding), and transplanting in rows, with careful weeding and water management. Genetic improvement through the development of new fertilizer-responsive MVs and application of chemical fertilizers may enhance intrinsic yield potential, while improved management practices might be necessary for exploiting its full potential. In my (and in all likelihood, Professor Otsuka’s) view, these combinations of science-based innovations and farmers’ adaptation were key to the success of the Asian rice Green Revolution. The lack of understanding of the importance of disseminating appropriate management practices through agricultural training and farmer-to-farmer extension is a key bottleneck in applying Green Revolution technologies to SSA (Otsuka et al. 2022). Hence, it appears that a part of SRI’s system and even philosophy, whether SRI proponents like it or not, is not very different from that of the Green Revolution. While exceptional yield reports, such as those exceeding 20 tons/ha, once sparked ‘rice wars,’ the solid positive findings on average yields in different contexts indicate that the ‘rice wars’ over SRI are ending, at least among agricultural economists, and attention should now shift to other unresolved issues, including its profitability.

Recollections of Professor Keijiro Otsuka

I owe what I am today to Professor Otsuka. When I was still in the master’s program at GRIPS, I told him that I wanted to pursue a doctoral degree. He said, “If you go to universities in the US, the risk for failure is high, but returns on success will also be high. If you stay in a Japanese university, the risk is low, but the return may not be so high.” Hearing that, I chose the latter because of less risk. But the doctoral course at GRIPS was much harder than I thought, and I couldn’t complete it within the standard time frame. However, thanks to Professor Otsuka’s persistent guidance, I eventually completed it in six and a half years. I am very deeply moved to be able to currently work as a professor at GRIPS and work in the office that Professor Otsuka used to use. I was almost tricked by Professor Otsuka. Studying under his supervision in Japan has provided me with very high and long-lasting returns!