The dynamics of farmer's choice for rice irrigation technology in the Kilombero Valley, Tanzania: evidence from reduced farm household model

1 Introduction

Agricultural households in developing economies face complex decision-making processes that simultaneously involve production choices, consumption patterns, and resource allocation strategies. Unlike pure profit-maximizing firms, farming households must balance competing objectives including food security, income generation, and leisure allocation within resource constraints (Ditzler et al., 2019). This dual nature as both producers and consumers makes the farm household model particularly valuable for understanding technology adoption decisions in smallholder farming systems, especially in sub-Saharan Africa where adoption of improved agricultural technologies remains persistently low (Fadeyi et al., 2022; Finizola e Silva et al., 2024).

The Kilombero Valley in Tanzania presents an ideal context for examining these household-level decisions regarding irrigation technology choice. The valley, which contributes approximately 9% of Tanzania's total rice production (Musamba et al., 2011), faces escalating pressure from competing water demands. Upstream rice cultivation coexists with downstream hydropower requirements, particularly following the development of the Julius Nyerere Hydropower Project (JNHPP). Recent studies indicate that 65% of the water required for JNHPP originates from this sub-basin [Ayo, 2022; Public Procurement Regulatory Authority [PPRA], 2024]. This competition intensifies under climate variability, as the valley experiences temperatures increasing by 0.03 °C annually and rainfall fluctuating between 500 and 1,500 mm (FCFA, 2021). Pauline and Liwenga (2023) documented that small-scale farmers in the Kilombero Valley are increasingly vulnerable to these multiple stressors, facing challenges that affect their livelihoods and adaptation capacity.

To promote water use efficiency in the valley, the Ministry of Agriculture has promoted the System of Rice Intensification (SRI), which emphasizes controlled water usage through alternate wetting and drying cycles. Recent evidence demonstrates that SRI irrigation consistently produces higher yields than conventional methods; for example, reports from India showed that SRI with controlled water usage yields 6.23–6.47 tons per hectare compared to 5.36–5.59 tons per hectare under conventional flooding (Thakur et al., 2023). Despite these documented advantages in water productivity and yield potential (Kadigi et al., 2020), adoption remains limited, and over 90% of irrigable land in the valley continues under Traditional Flooding Irrigation Technology (TFIT); (Olson et al., 2015). Recent projections suggest that Africa, including Tanzania, can avoid further increases in rice imports only through substantial yield intensification, making technology adoption a critical policy priority (Yuan et al., 2024).

Previous research on irrigation technology adoption has predominantly employed discrete choice models that treat adoption as a binary outcome (Ndabila, 2018; Kaloi et al., 2021). While valuable, such approaches cannot capture the continuous nature of resource allocation decisions or the interdependencies between production, consumption, and labor supply choices that characterize farming household behavior.

This study contributes to the literature by applying a reduced farm household model (RFHM) to examine how demographic characteristics, market prices, land rental costs, and income sources jointly influence irrigation technology choice and associated resource allocation decisions. By estimating a system of equations using Seemingly Unrelated Regression (SUR), we capture correlations across decision domains while identifying factors with differential effects on SRI vs. TFIT farming systems. The SUR approach, originally introduced by Zellner (1962) and recently refined for agricultural applications (Fan et al., 2023), is particularly appropriate when error terms across equations are correlated, which is a common feature of household decision-making. The findings provide important insights for designing policies that effectively promote water-efficient rice production technologies.

2 Literature review

2.1 Empirical evidences and hypotheses development

Research on irrigation technology choice in sub-Saharan Africa has largely relied on logistic regression and probit models to map the determinants of adoption as binary responses (Ndabila, 2018; Kaloi et al., 2021). These studies consistently confirm education, extension services access, suitable land size, and off-farm income as key enablers of technology uptake. However, a bitter fact is that binary choice models are fully suitable to capturing the continuous tendency of resource allocation. The variables like the size of land, labor man-days, and capital amount devoted to next best alternative technologies (which is undeniably more policy-sound than the distinction of adoption/non-adoption itself).

An evolving scholarly works recognize that technology adoption cannot meaningfully be evaluated with no attention to household consumption function and labor supply decisions. For instance, Asante et al. (2024) unveil that understanding the drivers of climate-smart agriculture (CSA) adoption in Ghana points out the need to account for selection bias and evaluating concurrent impacts across multiple outcomes. Likewise, Fan et al. (2023) apply SUR to confirm that farming subsidies in China simultainously affect fertilizer use and cropping dynamics through intertwined household decision channels. These studies again insist toward farm household model frameworks application as more recommended analytical tools for technology adoption studies in smallholder contexts.

Nonetheless, the farm household model, as formalized by Singh et al. (1986), recognizes that smallholder households jointly maximize utility from consumption whilst optimizing farm production decisions. In non-separable household models, production and consumption decision making is jointly determined by respective household characteristics, prevailing market prices, and resource endowments, making the necessity of reduced-form estimation across multiple equations very alive. Applications of the reduced farm household model in farming technology research have extended significantly. A good example is Jogo (2010) applied the RFHM to analyze wetland resource use trade-offs in South Africa, while Sanga and Mungatana (2016) also used household decision-making into land-use externalities research in Tanzanian water catchments. More recently, RFHM-based studies have been applied by scholars to evaluate climate adaptation decision making (Asante et al., 2024) and the price responsiveness to smallholder integrated production enterprises (Ditzler et al., 2019). These applications demonstrate the framework's versatility and its appropriateness for contexts where households simultaneously face production and consumption constraints.

Within the East African setting and Tanzania in particular, SRI-specific adoption studies have reported the role of extension service access (Tusekelege et al., 2014), land availability (Kadigi et al., 2020), and labor intensity (Ndabila, 2018) as positive drivers of technology scaling. However, none of these studies directly models the interdependence of household consumption, production, and resource allocation decisions using a structured framework, and the existing study is here for addressing the gap. The current study addresses this gap.

2.2 Theoretical lens

As a unit of analysis in this study, the household often invests resources to maximize both profit and utility at the same time. After all, the household decision to use a particular irrigation technology is made after considering the risks associated with its choice. It is from these household attributes that classical profit maximization, random utility, and adoption are taken as the underpinning theories for this study.

Classical profit maximization theory mainly assumes that profit maximization is the only relevant goal that any rational producer or firm always strives to achieve (Ndabila, 2018). The theory is useful to this study especially when considering that a rational rice farmer in the Kilombero Valley, as the household, can be drawn to the irrigation technology that maximizes his desires, including profit. Random utility, on the other hand, assumes that the individual makes choices from a particular set of alternatives. This study therefore applies utility theory in analyzing farmers' choice of irrigation technology to supplement the classical profit maximization theory, which excludes the utility part. Adoption theory proposes that an economic unit needs information to make rational calculations to accept an innovation or technology with minimal chance of risk and uncertainty (Negatu and Parikh, 1999; Ngwira et al., 2014). This study borrows propositions from adoption theory since it is centered on the farming household accepting SRI irrigation or continuing with TFIT.

Against the backdrop of classical profit maximization, random utility, and adoption theories, the reduced farm household model provides a unified framework for analyzing farm production decisions alongside household consumption and labor allocation choices (Singh et al., 1986). Unlike neoclassical models that separate firm and consumer behavior, the RFHM recognizes that smallholder households simultaneously maximize utility from consumption while operating farm enterprises constrained by resource endowments, market access, and technology options. This framework has been successfully applied in recent studies examining climate-smart agriculture adoption (Asante et al., 2024) and agricultural subsidy effects (Fan et al., 2023).

Following the standard formulation, we specify household utility (Equation 1) as a function of rice consumption (XG), market-purchased goods (XM), and leisure time (LT), conditional on household characteristics (ϕ):

Max U=U(XG,XM,LT;φ)(1)

Rice production under irrigation technology i (where i = SRI or TFIT) depends on labor input (L_*i*), purchased inputs such as seeds and fertilizers (YG_i), resource endowments including land and water (δ), and technology-specific parameters (α) as seen in the Equation 2 below:

Ri=Ri(α,Li,YGi;δ)(2)

The household faces three binding constraints: a rice balance constraint ensuring consumption does not exceed production net of sales; a time constraint allocating available hours between rice production, non-rice activities, and leisure; and a budget constraint linking expenditures to income from rice sales, non-rice employment, and exogenous transfers. Solving this constrained optimization problem yields reduced-form equations expressing each decision variable as a function of exogenous prices, endowments, and household characteristics. This approach has been successfully applied in wetland resource studies and land-use externalities research that jointly incorporated household decision-making patterns (Jogo, 2010; Sanga and Mungatana, 2016).

The key insight from this framework is that irrigation technology choice cannot be analyzed in isolation; it emerges jointly with decisions about land allocation, labor deployment, rice marketing, and consumption patterns. Factors promoting SRI adoption may simultaneously affect these other margins, creating system-wide effects that discrete choice models cannot capture. By estimating reduced-form equations for multiple decision variables, we identify how explanatory factors differentially influence households practicing each technology.

Drawing from the literature evidence reviewed above, the hypotheses tested by this study include, household head education, male gender, remittances are positively related to land allocated to SRI and external income, and negatively related with rice consumption. Also, escalating rice price reduce rice consumption more sharply among TFIT households, largely farming rice subsistence, than among SRI households who are more market-focused and with diversified dietary consumption.

2.3 Conceptual framework

As derived from utility and profit theories in totality, rice-producing households strive to maximize both utility and profit at the same time. In this regard, the framework is made up of five interlinked components, namely the farming household characteristics, household consumption function, household production function, technology economic feasibility and eventually the irrigation technology choice. In addition, understanding that the interlinkages are not taking place in the vacuum, there also so many factors as external influence component can also influence the outcome variables. The structure of this relationship is explained hereafter. The farm household has different characteristics demographically and so forth, which literally determine what irrigation technology to use for rice farming, but that relationship is not that explicit since the household thinks about many things before arriving to that choice. The household produces not necessarily rice alone, but with other crops and engaging in non-farming activities to make a living. Further, the household consumes not necessarily rice only, but again other market goods for attaining highest possible satisfaction. How much to produce and consume heavily rely on the composition of household itself in terms of household size and resources as examples. Given the sharing of household resources, household choice for irrigation depends on household production and consumption functions as clearly depicted in Figure 1. Equally important, such a household choice making also considers the economic feasibility of irrigation technology itself in terms of anticipated costs and benefits.

Moreover, the farming household passes through a process with many things to consider before deciding which irrigation technology to use for rice farming. In this regard, the farming household considers more than profitability to adopt an irrigation technology, as pointed out earlier in the theories. Before adopting an irrigation technology in a particular cropping season, the rice-producing household considers things such as the resources to allocate in farming, the quantity of rice to produce, the quantity of rice to consume, the quantity to demand of other market goods, and the external income to generate, because these have implications for the irrigation choice made. For instance, irrigation technology is costly, but a farming household may still adopt it given that the household has external income to cover additional costs. Another example is that if a household consumes a lot of what it does not produce, the household will look for an irrigation technology (SRI or TFIT) with the possibility to produce more surplus rice that it will eventually sell to purchase other market goods to meet its consumption. This is the rationale for including variables like these in the model for the evaluation of choice determinants. These are distinctive considerations that the rice-producing household weighs before adopting a technology, and they are discussed in detail in the methodology that follows.

3 Materials and methods

3.1 Study area

This study was carried out in Kilombero District, the district on which the largest portion of the Kilombero Valley relies. The Kilombero Valley, situated between longitudes 34.563° and 37.797°E and latitudes 7.654° and 10.023°S, constitutes Tanzania's most significant wetland rice production zone. The valley lies within the larger Rufiji River Basin and provides critical ecosystem services including water storage, flood attenuation, and agricultural production. Recent land-use studies document substantial agricultural expansion—agricultural land increased by 11.3% between 1990 and 2016 while floodplain wetland area decreased from 4.6 to 0.9% (Khatun et al., 2019). Commercial and smallholder agriculture has expanded substantially—sugarcane cultivation grew from 3,500 hectares in the 1990s to 15,000 hectares by 2014 (Wilson et al., 2017), while rice farming has intensified under the Southern Agricultural Growth Corridor of Tanzania (SAGCOT) initiative.

Kilombero District has a population of 582,960 (URT, 2022) and approximately 400,000 hectares of arable land, and it represents the core of the valley's rice production. Rice serves as both the primary cash crop and staple food, accounting for over 90% of local food consumption. Projects such as IUCN's SUSTAIN have been working with farmers in this area to promote resilient agricultural practices, with farmers in Mkula ward reporting rice harvest declines from 30 to 10 tons per season due to climate impacts (IUCN, 2024). Four villages containing irrigation schemes where both SRI and TFIT are practiced were purposively selected: Sululu (Signal ward), Mkula (Mkula ward), Sanje, and Msolwa Ujamaa (both in Sanje ward; Figure 2).

3.2 Sampling and data collection

We employed stratified random sampling to ensure balanced representation of both technologies. From households within irrigation schemes, 50 SRI-practicing and 50 TFIT-practicing households were randomly selected, yielding 100 total observations. Village-level distribution reflected relative scheme sizes: Sululu (24), Mkula (25), Msolwa Ujamaa (22), and Sanje (29) households. While the sample size is modest given the complexity of the Seemingly Unrelated Regression (SUR) system, it reflects the population of households formally enrolled in these specific irrigation schemes and provides sufficient degrees of freedom for the six-equation system estimated.

Primary data were collected through structured household questionnaires administered via face-to-face interviews. The questionnaire captured detailed information on household demographics, land ownership and rental arrangements, labor allocation across activities, input purchases and prices paid, rice production and marketing, consumption patterns, and income sources.

3.3 Model specification

From the theoretical model's first-order conditions, we derived six reduced-form equations (Equations 3–8) capturing key household decisions under each technology system. Each decision variable is expressed as a function of household characteristics, resource endowments, and market prices. The decision variables in the six equations evaluated in this study include the quantity of rice harvested by household under the technology applied (Rs_*i*), quantity of rice demanded by household (Rd_*i*), household demand. For other non-market goods (Od_*i*) Household's external income (EY_*i*), household's land allocated to rice production (LR_*i*) and household's labor demand for rice production (Md_*i*)

Rsi=Rst(L,E, φ,Ps,Wo,β,α,δ,ε)(3)

Rdi=Rdt(L,E, φ,Ps,Wo,β,α,δ,ε)(4)

Odi=Odt(L,E, φ,Ps,Wo,β,α,δ,ε)(5)

EYi=EYt(L, φ,Ps,Wo,β,α,δ,ε)(6)

LRi=Rdt(Rs,E, φ,Ps,Wo,β,α,δ,ε)(7)

Mdi=Mdt(Rs,E, φ,Ps,Wo,β,α,δ,ε)(8)

where ε represents the error term and Ps denotes prevailing market prices for rice grain, inputs, and market goods. These reduced-form equations are estimated empirically using a linear (levels) specification. For more details regarding definition of all variables used in the model.

3.4 Estimation strategy

Given that these six equations share common explanatory variables and likely correlated error terms, separate OLS estimation would yield inefficient estimates (As described in Table 1). We therefore employ Seemingly Unrelated Regression (SUR), which accounts for cross-equation error correlation while estimating the system jointly (Zellner, 1962). The SUR approach has recently been applied in agricultural contexts including the analysis of China's agricultural subsidies on fertilizer use (Fan et al., 2023) and crop yield modeling under climate change (Hu et al., 2024). The Breusch-Pagan test confirmed significant error correlation (χ² = 35.6, p < 0.01), validating the SUR approach over equation-by-equation estimation.

Table 1. Comprehensive variable description table.

+ means Positive expected relationship.

− means Negative expected relationship.

+/− means ambiguous/uncertain relationship.

The study acknowledges that farmers self-select into SRI or TFIT based on observable and potentially unobservable characteristics, which sometimes may bring selection bias into the coefficient estimates. For the sake of partially address this issue, the study controls for key observable determinants of technology choice like education, land size, gender, and extension access in each equation. The study nonetheless acknowledges that approaches like full endogenous switching regression and Three-Stage Least Squares (3SLS) would also provide additional strength but often require more robust instrumental variables than which are available in the existing dataset; these are marked as an important target for future research. Looking forward to another econometric diagnostic test for multicollinearity, this was assessed using Variance Inflation Factors (VIF) and all VIF were < 5 implying no issue, and heteroskedasticity-robust standard errors were generated for all equations.

4 Results

4.1 Descriptive statistics

The descriptive statistics reveal critical human capital disparities between SRI and TFIT adopters (Table 2). SRI farmers possess significantly higher education levels (10.38 vs. 7.37 years, p < 0.001), representing a 41% education premium. Interestingly, TFIT farmers demonstrate 60% more rice farming experience (7.73 vs. 4.82 years, p < 0.001) and cultivate 82% larger landholdings (2.90 vs. 1.59 acres, p < 0.001).

Table 2. Descriptive statistics of continuous variables by irrigation technology.

Standard deviations in parentheses. ^***p < 0.01, ^**p < 0.05, ^*p < 0.1 (two-tailed t-test). SRI, System of Rice Intensification; TFIT, Traditional Flooding Irrigation Technology; HH, household.

Further, SRI farmers receive 52% more extension visits (4.04 vs. 2.65 visits, p < 0.05), indicating that institutional support and knowledge transfer mechanisms remain critical for SRI adoption in Tanzania. While SRI farmers achieved a higher average rice harvest (1,975 kg vs. 1,625 kg), this difference was not statistically significant (p > 0.10). However, the pattern suggests SRI's yield potential may be constrained by factors beyond the technology itself. More revealing is the marketing behavior, SRI farmers retain less rice for home consumption (464 kg vs. 407 kg) and sell slightly more (1,489 kg vs. 1,218 kg), indicating greater market orientation, a pattern consistent with their higher education and commercial farming outlook.

Table 3 reveals that male-headed households comprise 65.3% of SRI adopters vs. only 43.1% of TFIT users (p < 0.05). This gender gap in adoption reflects broader patterns of differential access to resources, extension services, and decision-making authority in agricultural technology adoption. Equally noteworthy is the education-quality perception link: 24.5% of SRI farmers attained tertiary education vs. only 2.0% of TFIT farmers (p < 0.001). Correspondingly, 71.4% of SRI farmers rate their rice grain quality as good compared to only 9.8% of TFIT farmers (p < 0.001).

Table 3. Categorical variables by irrigation technology.

Percentages are shown for each irrigation group. ^***p < 0.01, ^**p < 0.05, ^*p < 0.1 (chi-square test).

Further, group membership is significantly higher among SRI farmers (69.4% vs. 49.0%, p < 0.10), indicating that participatory approaches through farmer groups encourage uptake of complete technology packages and lead to significantly higher yields. Although the majority of households in both groups are self-sufficient in terms of food security, the SRI group outperformed its counterpart by reporting having enough food, and the food sufficiency differences were statistically significant at the 0.05 probability level.

4.2 Factors influencing resource allocation and production decisions

Table 4 presents SUR estimates for the six reduced-form equations, separately for SRI and TFIT farming systems. Several findings merit detailed discussion.

Table 4. Factors influencing the household decision on choice of rice production technology to use and resource allocation.

Values in parentheses are z-statistics. ^** and ^* denote variables significant at the 5% and 10% significance levels, respectively. All models are estimated in levels. Coefficients on price variables therefore represent marginal effects in natural units (e.g., kg of rice, hours of labor)

Table 4 presents results from the household reduced model showing factors influencing the household decision on allocating resources to rice production and on the type of technology to use. As indicated in the conceptual framework, a household is considered as a decision unit in choosing the production technology to use and allocating resources at its disposal to various production ends and technologies. The results show that the size of land owned by the household significantly determined the amount of labor demanded for rice production. An acre increment in the land owned by a household would cause an increase of about 11 and 6 man-days per season for rice production under SRI and TFIT, respectively.

Also, land owned for agricultural purposes has shown a positive and significant impact on rice supplied by a particular household. Each acre increment in land size owned by a household for rice production caused an average increase in rice production of about 1,180 and 262 kg of rice under SRI and TFIT, respectively.

Results have also shown that household size positively and significantly influences the decision on how much land to allocate to each rice production technology for a household practicing both technologies. The results describe that each increment of one household member leads to an expansion of 0.6 and 0.8 acre for SRI and TFIT, respectively.

As expected, household size also has a very strong influence on the household's demand for rice as food and for other market goods. Table 4 shows that as one member is added to the household, demand for rice as food would increase by 2.21 kg (statistically significant for households practicing TFIT) and demand for other market goods would increase by 4.69 kg (statistically significant for households practicing SRI).

Moreover, with respect to household head gender, results indicate that households headed by males allocate more land for rice production than those headed by females in both technologies. This can be attributed to the fact that in the study area land ownership is gender biased; males are the owners of land and hence are the ones who make the final decision on how to allocate land for rice production.

Equally important, in both technologies, female-headed households on average outperformed male-headed households by supplying more rice. A possible explanation is that females are relatively more engaged in farming activities for the purpose of ensuring food security for their families. Nonetheless, the gender effect on the quantity of rice supplied was not significant for households practicing TFIT, probably because some female farmers were not in position to follow proper agronomic practices. This implies that extension services in the area are gender biased.

Education was found to be statistically significant in influencing household rice farming decisions. Results from Table 4 show that, each increment in years spent in education by the household head leads to an average increase in external income of TZS 131,089 and 35,380 for the SRI and TFIT groups, respectively.

Furthermore, demand for rice as food was also found to be indirectly influenced by education through external income, as elaborated above. It was found that as the household head gains more education, demand for rice as food falls among households by 2.7 kg (SRI) and 1.5 kg (TFIT). Education probably determines the quantity of rice demanded through the issue of external income. As highlighted above, increasing years of education means more external income for the farming household. In this sense, the ability of the household to afford non-rice foods may increase because of the increased income.

Education was also found to predict the amount of rice a farming household supplies to the market per season. Each increment in years spent in education caused a positive effect of about 72.76 kg of rice for SRI and a negative effect of about 41.23 kg of rice for TFIT. This is because SRI irrigation requires more knowledge, and educated households are more likely to acquire it quickly than less educated ones.

Equally important, education has shown a positive contribution to household demand for other market goods. An increment in education years would cause an increase of 3 and 2 kg of other food demanded by households in SRI and TFIT, respectively. The observed differences are because households practicing SRI were more educated than those practicing TFIT hence external income favorites as explained earlier.

The experience of a household head in farming was also found to influence household decisions on land allocated, rice supply, and external income. Table 4 shows that experience in rice farming influences positively the allocation of land for rice production. The results show that the more experienced the farmer, the larger the land allocated for rice farming. They also show that the more experience the farmer has, the less land is allocated to SRI (0.63 acre) compared to TFIT (0.78 acre).

Moreover, experience in rice farming influenced rice supply negatively for households practicing SRI and positively for households practicing TFIT. Each increment in years of experience may cause a decrease of about 162 kg in rice supply from households practicing SRI and an increase of about 97 kg of rice from households practicing TFIT.

Table 4 shows that a unit price increase in rice would cause a decline in quantity demanded for rice by 0.623 and 2.11 kg for SRI and TFIT among farming households. This negative relationship between price and quantity demanded of rice in both groups is consistent with the demand theory for any rational consumer (Mankiw, 2016). The increase in rice price led to a decline in quantity demanded for rice as food in farming households belonging to both irrigation technologies. However, the degree of responsiveness in quantity demanded between the two groups differs. Such a difference might be attributed to the fact that, unlike for SRI farmers, rice is the leading staple for TFIT farmers, so a unit change in the price of rice might have led to a larger change in rice quantity demanded for farmers belonging to TFIT than SRI. There was a small change in rice quantity demanded for SRI farmers because they consumed less rice and more of other market goods, so they are less affected by rice price changes.

Nonetheless, Table 4 shows that land rent has a significant negative effect on both technologies. As land rent increases, SRI and TFIT farming households decide to reduce land allocated to each technology by 0.6 and 0.5 acre, respectively. This negative relationship is attributed to the fact that the price for land in irrigation schemes is relatively higher than outside the scheme where TFIT is largely applied.

Equally important, remittance was found to be a significant predictor of the decision to allocate land for rice farming among household heads. Table 4 shows that remittance and land allocated for rice production had a positive relationship, whereby as remittance increases, land allocated for rice production increases, because remittance increases the ability of the farming household to cover land acquisition expenses. More importantly, farming households in the TFIT group (1.13 acre) were found to benefit more from an increase in remittance on land allocation than those in SRI (0.93 acre). This can be attributed to the fact that SRI plots are more expensive to acquire, hence given the same increase in remittance, SRI households can afford to acquire a relatively smaller area in the irrigation scheme than their counterparts (TFIT households), who acquire plots at cheaper prices.

5 Discussions

Results from SUR are largely interconnected as conceptualized previously. Specifically, the figures in labor to land allocation reveal that as household land size increases, SRI technology demands more man-days than TFIT. This indicates that SRI is more labor intensive than TFIT; SRI involves additional activities, such as alternate wetting and drying (AWD) of land, that require additional man-days of working in the field. According to Kumar and Nayak (2013), an increase in land size will cause the need for more man-days in SRI than TFIT.

Also, from the results, positive response in rice yield from a land increase is also attributed to factors other than land size increase, such as good land management practices like fertilizer application and timely weeding. More importantly, a farmer practicing SRI is keener in observing these plant management practices than a farmer practicing TFIT, and that is why there is more productivity under SRI than TFIT. In addition, farmers practicing SRI are more exposed to scientific and recommended ways of growing rice than those practicing TFIT. SRI comes with a training package from the extension officer on the frequency of AWD, transplanting, and planting spaces, all of which make the rice plant to reproduce more. Similar findings are reported by Thakur et al. (2023) in India, that an increase in the size of land leads to a higher increase in rice yield grown under SRI irrigation than TFIT.

There is another implication from the results that as household size increases, demand for food and other market goods also increases; therefore, if rice production is one of the major sources of income to a household, more area will be needed to grow more rice. The results further tell that households rely more on land size under TFIT than SRI. More land is allocated to TFIT to make it produce a yield equivalent to SRI. In addition, more use of land in TFIT makes it less efficient in land utilization as far as environmental conservation is concerned. More land allocation to TFIT means more methane emission which further intensifies heat stress to rice plants. This result aligns with the conclusion drawn by Okello et al. (2019) in Uganda, that as household members increase, land allocated for conventional rice irrigation also increases.

Further, household size increases with demand for other market goods. This concurs with previous results that households practicing SRI earn more income than those practicing TFIT, which gives them the capacity to buy other food products they prefer other than rice. This is the opposite for households practicing TFIT: they earn less income than their counterparts practicing SRI and therefore have to rely on what they produce, which is rice. There is a close similarity between this finding and that reported by Ramdhanie et al. (2017) in North Trinidad, that household size positively determines the quantity of food and other market goods demanded by smallholder farmers, as they are more interested in quantity than quality of food.

Furthermore, in gender and technology returnability, better rice yield in female-headed households than in male-headed ones was also reported by Thabiti (2014), who conducted a study in Kyela, Mbeya. Furthermore, gender was found to be an issue with respect to external income: female-headed households were found to earn less external income than male-headed households under SRI. This can be attributed to the fact that SRI is labor intensive, so females have to devote much of their time to farm work rather than other economic activities, hence earning less from these sources. However, the situation is different under TFIT: female-headed households earn more external income than male-headed households because TFIT is less labor intensive, so female household heads have more time to engage in off-farm economic activities and hence earn more external income.

As expected, years spent in education showed a positive contribution to external income generated by the farming household. This is because more education means the household head is more likely to obtain a job outside farming activities, hence there would be an increase in external income accrued from paid salaries. A similar result was obtained by Tran et al. (2023) in Thailand and Vietnam, that an increase in a farmer's years spent in education went with an increase in external income generated, ceteris paribus. This study, however, found that SRI farmers' external income is more elastic to a change in education years, probably because they are more attentive in grabbing external opportunities to earn extra income. Finally, SRI farming households are probably in relatively higher need of extra income than TFIT households because SRI farmers need to cover new farming expenses (highlighted earlier) that are not incurred by TFIT farmers. The stronger education elasticity of external income among SRI compared to TFIT households suggests that SRI adopters are more responsive to human capital accumulation in both farm and non-farm domains consistent with Hypothesis 1.

Furthermore, the negative effect on rice supply shown by education under TFIT indicates that most households practicing TFIT are less educated and hence have low productivity, which makes them unable to produce a surplus to supply to the markets. Similar findings are reported by Tusekelege et al. (2014), who conducted their study in Dakawa, Morogoro, and pinpointed the significance of education in explaining the productivity of SRI irrigation among surveyed farmers. Equally important, as noted earlier, educated households have more opportunities to earn external income than less educated ones, making them able to afford more other market goods than their counterparts, so the difference in the effect is also attributed to this fact. Moreover, Miassi et al. (2022) made a similar observation in West Africa, that as a farmer becomes more educated, his or her demand for market goods increases, holding other things unchanged, due to increased awareness of dietary foods.

On the farmer's experience to smaller land allocated to SRI, according to Kasase et al. (2022), this can be attributed to the observation that SRI is a relatively newer technology and people do not have much experience compared to TFIT, hence more land is allocated to TFIT than to SRI. Moreover, the finding reveal that more experienced rice farming households will remain/shift to TFIT since the technology is less labor intensive and hence has lower production costs than SRI. This finding contradicts that of Karki et al. (2020), where farmers with more experience are more likely to go for climate-change-resilient technologies like SRI irrigation, which offers higher produce. Such inconsistency is probably because their study was too generic to observe site-specific findings like those of the Kilombero sub-basin (KSB).

Findings from Table 4 indicate that as experience years in farming increase, external income for the farming household decreases. This is because in the short run the farming household will decide to continue rice farming rather than going for off-farm economic activities to earn external income, as also reported by Layani et al. (2020) in Marvdasht County; this, however, may change in the long run when a farmer gets older and may switch to off-farm economic activities. Therefore, an increase in experience years may lead to an increase in external income in the long run. The market price of rice was found to be statistically significant in explaining variation in household demand for rice as food and demand for other market goods.

Furthermore, given the same increase in land rent, farming households practicing SRI are affected more than households practicing TFIT. This is why households practicing SRI reduce more of the area allocated to rice than households practicing TFIT. Largely, this finding agrees with the report by Qiu et al. (2020), that an increase in land rent reduced land acres allocated for modern rice production in rural China.

6 Conclusions

The motive of this paper has been to evaluate how farmers arrive at choosing a rice irrigation technology in Kilombero District. To achieve this objective, household reduced-form equations were derived to understand how household heads come to decide on the irrigation technology to use. The results from this research are vital in understanding the reasons why most rice farmers in the KSB are yet to move from TFIT to the latest SRI irrigation technology.

The principal novel contribution of this study is the demonstration that SRI adoption is driven by a constellation of mutually reinforcing factors for instance education, male headship (through land access), and remittance availability which simultaneously affect not only land allocation but also labor demand, rice supply, consumption patterns, and off-farm income.

Specific results from the household reduced models show that the household head's years spent in education and gender were most influential in determining various decisions such as land allocation and labor supply across irrigation technologies. Nevertheless, it was found that education contributed more to decisions in favor of SRI irrigation than TFIT, but SRI irrigation was found to be more labor intensive than TFIT, hence the continuation of farming households with TFIT. Household head experience in farming was found to influence decisions in favor of TFIT over SRI irrigation.

Generally, from the findings, low education has led farming households in the studied sites of the KSB to continue with TFIT. Socially, men were found to have a greater chance of accepting SRI irrigation since they have more of the prerequisite resources for practicing it; in particular, they are more entitled to land ownership and allocation than women. Also, the labor-intensive nature of SRI has triggered many farming households in the KSB to continue practicing TFIT. Further, farming households in the surveyed areas are still using TFIT because they face a shortage of income, especially from off-farm economic activities, to cover the additional costs arising in SRI.

6.1 Recommendations and policy implications

• Although higher labor requirement in SRI favors the selection of TFIT by the surveyed households, still SRI irrigation remain essential in reducing methane emission and lowering heat stress compared to continuous flooding which by itself favors methane emission. The higher labor requirement in SRI irrigation could be compensated by reinforcing carbon credit markets to induce more farmers to SRI
• The key policy implication from these results is that government intervention through directing an indicative rent for SRI plots is important to induce the shift from TFIT to SRI irrigation.
• Newly introduced technologies like SRI should be incorporated into agriculture-related subjects or topics taught in primary schools to familiarize the young generation with the technology earlier.
• Information on diversified sources of income should be provided to farming households, because through external income farming households can afford SRI, which has been found to be relatively more costly than TFIT, hence increasing the number of SRI adopters.
• As SRI irrigation is more labor intensive than TFIT due to added activities such as alternate wetting and drying, this calls for more research and development (R&D) by the Tanzania Agricultural Research Institute (TARI) centers with rice as their mandate crop (TARI-Dakawa and TARI-Ifakara) to experiment with and disseminate technologies with minimal labor requirements in SRI irrigation. Experimenting direct seeded rice (DSR) with AWD could be ideal in this case.
• Finally, extensive campaigns should be carried out in the studied sites by gender development practitioners with an emphasis on increasing women's entitlement to resources such as land, since women might be willing to practice SRI irrigation but obtain no irrigable land for rice production, as such entitlement is largely held by men.

6.2 Limitations of the study

Several limitations are acknowledged by this study. First, the relatively modest sample of 100 households from formal irrigation scheme villages limits the external generalizability of the findings to non-scheme farming systems and to the broader Kilombero Valley population. Second, the SUR estimator cannot fully account for the self-selection of households into SRI vs. TFIT systems; unobservable determinants of technology choice may introduce residual bias in the estimated coefficients. Three-Stage Least Squares (3SLS) or endogenous switching regression approaches, requiring suitable instrumental variables, would provide additional robustness and are recommended for future research. Notwithstanding these limitations, the RFHM framework developed here offers a transferable analytical template for examining irrigation technology decisions in other smallholder contexts where production, consumption, and labor supply decisions are jointly determined.

Generative AI statement

The author(s) declared that Generative AI was used in the creation of this manuscript. Claude was used for proofreading the paper.

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

Publisher's note

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

Footnotes

Edited by: Anwarul Abedin, Bangladesh Agricultural University, Bangladesh

Reviewed by: Vicente Montano, University of Mindanao, Philippines

Imade Yoga Prasada, Universitas Putra Bangsa, Indonesia