1 Introduction

About 13.3% of all food produced worldwide is lost or wasted after harvest on farm, transport, storage, wholesale, and processing levels [1], featuring in the Sustainable Development Goal (SDG) of reducing hunger. The UN SDG 12 (Responsible Consumption and Production) calls for “reducing food loss in the early stages of the food supply chain and halving food waste in the retail and consumption phases by 2030” [2]. Food loss is an unintended loss in quantity of food during harvesting, postharvest handling, processing, distribution, and marketing. Food waste is “food that is of good quality and fit for human consumption, but that does not get consumed because it is discarded” [3]. Food loss is more relevant than food waste in sub-Saharan Africa (SSA). It remains a significant challenge, particularly among rural households, because they lack access to proper postharvest management technologies. They often use traditional storage practices that cannot effectively prevent storage pests, reducing food availability and market surplus. In sub-Saharan Africa, at least one in five people faces hunger daily, and approximately 140 million people are experiencing acute food insecurity [4]. The world food and agriculture organization estimate that the post-harvest loss can reach up to 20% for cereals, 30% for dairy and fish, and 40% for fruits and vegetables [5].

Maize postharvest losses (PHLs) can reach up to 35% in SSA [6]. More than a quarter of the PHLs in the maize value chain occur primarily due to storage problems (spillage, rodents, weevils, and moisture, etc.) [7]. The Food and Agriculture Organization (FAO) recently reported that food losses from harvest to processing are estimated at 21.4% in SSA [3]. The average PHL of maize per quantity stored in Tanzania is 6.9%, mainly due to insects [8].

Food loss is usually viewed as occurring only in terms of quantity. Food loss in quality deterioration (e.g., loss of nutrients) due to contamination (e.g., aflatoxin) is less emphasized as it is not easily detected through the naked eye. However, such losses are also significant [8]. Reducing PHL caused by food contamination is thus vital to increasing household food security through improved food safety and nutrition among consumers [9]. Besides food safety and nutrition, reducing PHL contributes to increased household income, as quality is directly associated with prices [10].

Food losses should not only be viewed as just loss of quantity and quality but also as a loss of all the resources that go into producing the lost and wasted food, including labor, land, water, fertilizer, pesticides, and other inputs [8]. In SSA, external inputs such as inorganic fertilizer, pesticides, and hybrid seeds are imported, suggesting that postharvest loss reduction could help the region reduce import bills. Further, it could help curb smallholder farmers’ expansion into fragile ecosystems in search of uncultivated arable land. Food waste and loss represent significant negative financial, economic, social, and environmental impacts.

Given the food security, food safety, and economic and environmental implications of PHL reduction in quantity and quality, it is critical to employ improved postharvest management technologies. Hermetic storage technologies (HSTs) have gained significant interest among smallholder farmers in Tanzania since 2014 when such technologies were disseminated [11]. HSTs protect stored grains against insect pests because they create an airtight condition, depriving the insect pests of oxygen and minimizing their chances of survival [12]. As such, HSTs reduce storage losses, insect damage, and weight [13,14,15,16].

HSTs can store the grains for months to years without impairing the quality and quantity of the stored food. HSTs helped smallholder farmers to store for extended periods, sell at higher prices, and improve food security [17, 18]. HSTs allow storing and selling of grains later at higher prices. Various studies have verified the profitability of using HSTs [19,20,21]. A cost–benefit analysis of improved storage technology by [22] shows that farmers can recover the full, unsubsidized cost of HSTs in one agricultural Season. This finding highlights the potential of HSTs in storing high-quality food and increasing smallholder farmer income’, which is crucial in reducing food insecurity and poverty. HSTs are preferred over traditional storage methods because they can be used without storage chemicals, reducing dangers to human health or the environment. HSTs maintain the quality of the stored product [23] and are a viable management tool for preventing aflatoxin accumulation in storage [24, 25].

The increased awareness of the benefits of HSTs led to increased demand for HSTs in many African countries, including Tanzania. However, empirical evidence is lacking on the drivers and societal benefits of adopting such technologies.

This study aims to address two policy research questions:

  1. i.

    Which factors influence the decision to use HSTs and,

  2. ii.

    Whether or not adopting HSTs would decrease food insecurity.

The first research question was addressed using the logit model. In contrast, the second research question was addressed using the endogenous switching regression (ESR) model based on two outcome variables—the Household Food Insecurity Access Score Scale (HFIAS) and the quantity of stored maize (QSM). The data came from 1180 maize farmers randomly selected from members of maize producer organizations (POs) in the Rukwa and Ruvuma regions in Tanzania.

This study holds significance in providing policymakers with valuable insights into policy instruments that can enhance the adoption of HSTs among farmers affiliated with POs. Unlike previous studies that focused on HSTs adoption and impacts without considering various contexts, our research specifically addresses the unique dynamics of smallholder farmers within POs. Smallholder farmers within POs exhibit distinct behaviors compared to those outside such organizations. This divergence arises from differences in access to essential resources. Farmers within POs benefit from access to input loans, storage facilities, and reliable markets or buyers. Conversely, those outside POs face constraints such as limited access to input loans. Storage facilities, and less robust market for their produce.

As a result, smallholder farmers within POs often demonstrate higher maize productivity. This is attributed to their improved access to inputs. Furthermore, they enjoy higher income levels due to the assurance of reliable markets for their agricultural products. By highlighting these contextual differences, our study contributes valuable information for policymakers seeking effective strategies to promote HSTs adoption, particularly among farmers affiliated with POs.

In addition, the study focused on maize farmers from the Rukwa and Ruvuma regions in Tanzania. This is because maize is a crucial staple crop in Tanzania, highly susceptible to postharvest losses, and also serves as a cash crop. The price of maize fluctuates significantly between the harvesting season when it is low, and the lean season, when it is high. Traditionally, the primary method for reducing these losses has been through the use of pesticides in storage. However, this approach poses health risks when consuming maize stored with pesticides. This is where HST comes in, offering a safe storage facility that helps farmers maintain both the quality and quantity of stored maize, allowing them to benefit from price arbitrage during the lean season. Furthermore, the Rukwa and Ruvuma regions are part of Tanzania's southern highland corridor, which produces about 50% of the country's maize [36]. The government relies on this corridor as a source of national food reserves. Therefore, it was relatively easy to identify and select maize producer organizations and farmers for inclusion in the study.

1.1 Literature review and research gap

Empirical studies on the determinants of HST adoption highlight several influencing factors. [26] found that age, farming experience, land size, group membership, credit access, off-farm activities, education, and household head occupation significantly influence HST adoption in Kitui County, Kenya. [27] Identified educational level, gender, awareness, training, accessibility, perception of pesticide effects, social responsibility, and household income as positive determinants of HST adoption among smallholder maize farmers in northwest Ethiopia, with HST price having a negative impact [28]. Emphasized education and access to savings as the main determinants of HST adoption in Nepal.

Other studies by [29] and [20], on the adoption of HST in Nigeria, west and central Africa, and central region of Niger respectively, all using logit regression, found that the availability of HST technicians and information source such as radio were significant determinants of adoption of HST [30], who used probit regression to analyze the adoption of HST bags in Northern Nigeria, also found that the availability of HST demonstration in the village and access to information source on HST, such as radio, are determinants of HST adoption.

The literature review revealed limited information on the adoption and impacts of HST on food security. Most available studies concentrate on the performance of HST in maintaining grain quality and quantity, as well as its potential to increase marketable surplus and income [21, 31,32,33,34]. A similar study by [35] analyzed the adoption of airtight technologies, a type of HST, in Tanzania using endogenous switching regression (ESR). This study found that HST decreases food insecurity, but it focused on smallholder farmers not affiliated with cooperatives. In contrast, our analysis involves maize farmers who are members of producer organizations. This distinction is important as it allows us to explore the impacts of HST within a different organizational context.

2 Materials and methods

2.1 Study area

The study was conducted in eight districts located in two regions of Tanzania—Rukwa and Ruvuma (Fig. 1). Rukwa and Ruvuma are among the major maize-growing regions in Tanzania. According to the 2019/2020 Tanzania agriculture census, Ruvuma had the highest maize production of 498,685 tons, while Rukwa produced 338,988 tons [36]. The study focused on the HSTs with different brands (e.g., the Purdue Improved Crop Storage (PICS) and AgroZ bags).

Fig. 1
figure 1

Map of the study regions

Full size image

2.2 Sampling procedure

The study focuses on maize producers in the Rukwa and Ruvuma regions, specifically those organized into farmers-owned cooperatives known as producer organizations. These cooperatives operates under the regulation of the Tanzania Cooperative Development Commission (TCDC) and the Ministry of Agriculture (MOA). POs are similar to Agricultural Marketing Cooperative Societies (AMCOS) and are an important part of Tanzania’s agricultural sector. They are crucial in the country’s effort to increase agriculture productivity and improve e rural livelihoods. The selection criteria for POs in the study were based on their active involvement in essential activities such as providing inputs, offering loans, delivering training, and other services aimed at enhancing farming practices and increase crop yields. Members of these POs pool their resources to collectively market their maize, ensuring that fellow members receive fair prices for their maize. In addition, producer organizations have convenient access to support from the government, development partners, and extension services. The rationale for choosing POs as the sampling frame is to investigate the dynamics among members within these groups that affect the adoption of HSTs, and their impacts on food insecurity as measured by the quantity of maize stored and the HFIAS.

A two-stage sampling was applied to select the sample farmers. The first stage entailed a purposive selection of POs in each region, and the second stage entailed a random selection of households from the identified POs. In the first stage, 100 POs were identified (50 POs in each region—Rukwa and Ruvuma). In the second stage, 554 households were randomly selected from 48 POs in Rukwa, and 586 households from 50 POs in Ruvuma, comprising 1140 households (Table 1).

Table 1 Distribution of households by region and district
Full size table

The sampling frame was a roster of smallholder maize producers from the 98 POs provided in advance by the implementing partners, who compiled it with the help of POs/sub-villages/village leaders in their respective regions. The listed smallholder farmers should have grown maize crop in the previous agricultural Season of year 2020/21 season in order to be considered for the interview.

2.3 Data

The study draws on primary data collected from 1140 sample households who are members of POs in the Rukwa and Ruvuma regions of Tanzania. The data were collected by trained enumerators using a standard questionnaire designed using the SurveyCTO CAPI Software tool (v2.71.3) uploaded on tablets in the Rukwa and Ruvuma regions for the Credit Access to Scale-up the Use of Hermetic Storage in Tanzania (CASH-TZ) project. The questionnaire captured information about demographic, socioeconomic, and institutional household characteristics, HSTs adoption, QSM, and HFIAS.

Since our study involved participants to remember past production and storage data we were likely to have encountered the recall bias. To mitigate recall bias, we implemented several strategies in our survey design. The survey was designed with clear, neutral, and specific questions to reduce misinterpretation and leading questions. We pretested the survey with a small sample to identify and correct potential sources of bias. Interviewers received thorough training to ensure consistency and neutrality in administering the survey. The survey questions were designed to cover shorter recall periods (i.e. agriculture season prior to the survey), reducing the likelihood of inaccurate responses.

2.4 Variable description and measurement

2.4.1 Outcome variables

The outcome variables are household food insecurity, measured by HFIAS, and stored maize, measured by QSM in kg in the year preceding the survey year (2020/21 season). The HFIAS is a continuous measure of household food insecurity (access) in the past four weeks (30 days). It is calculated based on two sets of questions. The first question is an occurrence question (yes/no coded as 1/0), followed by a frequency of occurrence question (rarely/sometimes/often assigned codes of 1, 2, and 3, respectively). The codes for each frequency-of-occurrence question are summed to calculate the HFIAS score for each household [37]. The maximum score is 27, which indicates the highest, and the minimum score is 0, indicating the lowest food insecurity.

The score is directly related to the degree of insecure food access. A household with a higher score is more food insecure. Alternatively, a household with a lower score is less food insecure. In addition, the HFIAS can be presented as a categorical variable of four levels (i.e., food secure, mildly food insecure, moderately food insecure, or severely food insecure) using the Household Food Insecurity Access Prevalence (HFIAP). The tool used to generate the HFIAS is presented in Table 10 in the appendix. Several studies have analyzed food insecurity using the HFIAS e.g., [38,39,40,41]. In addition, [42] recommended incorporating the HFIAS for food security measurements.

QSM is defined by the amount (in kilograms) of maize stored by the household in the 2020/2021 agriculture season. QSM allows households to have food and cash throughout the year. As maize is the main staple crop in the study area, QSM can capture the stability dimension of household food security by providing a future source of food (via decreasing PHLs) and income (via a sale at higher prices later in the Season).

2.4.2 Treatment and control variables

The treatment variable is HSTs adoption. The HSTs adoption variable was measured based on whether households stored their harvested maize using hermetic storage technology such as the PICS bag, AgroZ bags, metal silos, GrainPro, and Super grain bags. Households who used the hermetic bags in the year preceding the survey are treated as adopters. Households who did not use hermetic bags are treated as on-adopters. Table 2 describes the explanatory variables used in the model to explain HSTs adoption. HSTs adoption is hypothesized to depend on demographic, socioeconomic, and institutional characteristics [43].

Table 2 Definition and measurement of model variables
Full size table

Demographic characteristics: Four demographic characteristics are hypothesized to influence the farmers’ decision to adopt HSTs. These are the age, gender, and educational level of the household head and household size. As a proxy for risk aversion, age may influence the decision to adopt HSTs. Older household heads tend to be more reluctant to adopt new technologies such as the HSTs. Hence, the coefficient of age is expected to be negative. As a proxy for gender roles, the gender of the household head may influence the decision to adopt new technologies through institutional and cultural limitations linked to women’s roles. Male-headed households tend to have more access to extension information and hence adopt new technologies such as the HSTs compared to female-headed households. Given the information-intensive nature of the technology adoption process, household heads with a higher level of education may be able to obtain and process the information and become well aware of the benefits and risks associated with HSTs adoption. The coefficient of education is therefore expected to be positive. As a proxy for family labor availability, household size may influence the decision to adopt new technologies. The larger the household labor force, the higher the likelihood of using the HSTs. The coefficient of household size is therefore expected to be positive.

Socioeconomic characteristics: Three socioeconomic characteristics are hypothesized to influence the farmers’ decision to adopt HSTs. These are livestock income, off-farm income, motorbike ownership, and land size. As HSTs are relatively more expensive than traditional storage bags, households with a relatively higher income can afford to buy them and are, therefore, more likely to adopt HSTs. The coefficient of off-farm income is, therefore, expected to be positive. As a proxy for wealth status, households with more household assets (motorbikes) are likelier to adopt the HSTs. As the production level depends on farm size, farmers with a relatively large plot size tend to produce more and store more and longer. Hence, they are more likely to adopt the HSTs. The coefficient of plot size is, therefore, expected to be positive.

Institutional characteristics: Two institutional characteristics are hypothesized to influence the farmers’ decision to adopt HSTs. These are access to training and availability of vendors in the community. Households with access to credit, vendors, and training on HSTs practices tend to adopt HSTs:

Technological components: Four socioeconomic characteristics are hypothesized to influence the farmers’ decision to adopt HSTs. These are improved varieties, herbicides, crop spraying, and hired labor. As the production level depends on using a package of modern technologies, households using crop protection technologies tend to produce more and store more and longer. Therefore, the coefficients of using complementary technological components are expected to be positive. Considering the intensity of labor (number of hours of labor per acre) and number of acres at the farm level on farms using improved technological components, the coefficients of hired labor are expected to be positive.

Regional locations: Regions are included to capture the influence of regional differences in using HSTs.

2.5 Empirical models

The study employs two empirical models to address the above research questions—the logit and ESR models.

2.5.1 Logit model

The decision on whether or not to use the HSTs is considered under the general framework of utility maximization. Within this framework, maize farmers decide to use the HSTs if the benefit of using the technology is more significant than without it.

Following the random utility theory in [44], the conditional probability of adopting HSTs \(p_{i}\) can be given as:

$${P}_{i}=\frac{1}{1+exp\left(-{Z}_{i}\right)},$$
(1)

where the right-hand expression is the logistic distribution function; zi is the conditional odds of using the HSTs.

Rearranging Eq. (1), the log odds of using the HSTs can be given as:

$$Ln\left({Z}_{i}\right)=Ln\left[{P}_{i}/\left(1-{P}_{i}\right)\right]={X}_{i}{\beta }_{j}+{\epsilon }_{i},$$
(2)

where \(Ln\) is natural log; \({X}_{i}\) is a vector of the ith household’s demographic, socioeconomic, institutional, and technological characteristics hypothesized to influence the household’s decision to use the HSTs; \({\beta }_{j}\) is a vector of parameters representing the change in the log odds due to a unit change in the values of the predictors, \({X}_{i}\) and \({\epsilon }_{i}\) is the error term.

2.5.2 ESR model

Given that our measures of food insecurity are continuous variables (HFIAS) and (QSM) and the sample farmers were not randomly assigned to HSTs adoption, we apply the linear endogenous switching that takes account of the potential self-selection problem. The advantage of estimating the ESR model is that it accounts for unobserved heterogeneity and endogeneity in the covariates [45]. When an unobserved characteristic such as risk-taking behavior is omitted from the model, its effects are pooled into the error term, which will correlate with adoption and induce endogeneity; hence, it will be difficult to determine if the difference in HFIAS and QSM between adopters and non-adopters is due to HSTs adoption or the underlying difference in other characteristics (e.g., risk-taking behavior) that already exists between them. The failure to account for the potential self-selection to adoption may result in biased and inconsistent parameter estimators.

The present study applies the ESR model to estimate the average effects of HSTs adoption on HFIAS and QSM. The ESR model consists of the HSTs adoption (Eq. 3) and two separate linear outcome equations (Eq. 4a, b) of HFIAS and QSM for adopters and non-adopters. The ESR model can be identified through nonlinear functional forms and some exclusion restrictions.

$${Y}_{i}={X}_{i}^{\prime}\beta +\vartheta {M}_{i}+{\varepsilon }_{i},$$
(3)

where \({Y}_{i}\) is the HFIAS or QSM (transformed into its natural logarithm to smooth the convergence of the maximum likelihood estimators) or; \({X}_{i}\) = vector of covariates; \(\beta\) = a vector of parameters of the outcome equations to be estimated; \({M}_{\text{i}}\) = HSTs adoption; \(\vartheta\) = the parameter associated with HSTs adoption; \({\varepsilon }_{i}\) = the error term assumed to be normally distributed with mean zero.

$$\text{Regime}\,1: \quad {Y}_{1i}={\beta }_{1}{X}_{1i}+{e}_{1i} \quad \text{if}\,\, {M}_{i}=1,$$
(4)
$$\text{Regime }2: \quad {Y}_{2i}={\beta }_{2}{X}_{2i}+{e}_{2i} \quad \text{if}\,\,{M}_{i}=0,$$
(5)

where \({Y}_{i}\) is as defined above in each regime; \({X}_{i}\) represents a vector of exogenous variables expected to affect the outcome variable; \({e}_{1i}\) and \({e}_{2i}\) are random errors. The error terms of Eqs. (3) and Eq. 4a, b are assumed to have a tri-variate normal distribution, with mean zero and non-singular covariance matrix given as:

$$cov\left({e}_{1i},{e}_{2i},{u}_{i}\right)=\left(\begin{array}{ccc}{\sigma }_{u}^{2}& {\sigma }_{{{\varvec{e}}}_{1}{\varvec{u}}}& {\sigma }_{{{\varvec{e}}}_{2}{\varvec{u}}}\\ {\sigma }_{{\varvec{u}}{{\varvec{e}}}_{1}}& {\sigma }_{{{\varvec{e}}}_{1}}^{2}& .\\ {\sigma }_{{{\varvec{u}}{\varvec{e}}}_{2}}& .& {\sigma }_{{{\varvec{e}}}_{2}}^{2}\end{array}\right),$$
(6)

where \(var\left({u}_{i}\right)={\sigma }_{u}^{2}\), \({var\left({e}_{1}\right)=\sigma }_{e1}^{2}\), \({var(e}_{2})={\sigma }_{e2}^{2}\), \(cov({u}_{i}{,e}_{1})={\sigma }_{e1u}\) and \({cov\left({u}_{i}{,e}_{2}\right)=\sigma }_{e2u}\).

The covariance between the error terms in the outcome equations is undefined because the outcome variables \({Y}_{1i}\) and \({Y}_{2i}\) cannot be observed simultaneously [45]. In other words, the same household cannot be observed with and without HSTs adoption simultaneously. This suggests that the anticipated values of \({e}_{1i} \text{and} {e}_{2i}\) given the sample selection are non-zero. This is because the error term in the selection equation is correlated with the error terms in the outcome functions specified in Eqs. (6) and (7).

$$E\left[{e}_{1i}|{{M}}_{i}=1\right]={\sigma }_{e1u}\frac{\varphi (\widehat{M})}{\Phi (\widehat{M})}={\sigma }_{e1u}{\lambda }_{1i},$$
(7)
$$E\left[{e}_{2i}|{{M}}_{i}=0\right]={-\sigma }_{e2u}\frac{\varphi (\widehat{M})}{1-\Phi (\widehat{M})}={\sigma }_{e2u}{\lambda }_{2i},$$
(8)

where \(\varphi (.)\) is the standard normal probability density function, \(\Phi (.)\) the standard normal cumulative density function, and \({\lambda }_{1i}=\frac{\varphi (\widehat{M})}{\Phi (\widehat{M}))}\) and \({\lambda }_{2i}=\frac{\varphi (\widehat{M})}{1-\Phi (\widehat{M})}\) represent the inverse mills ratio evaluated at \(\widehat{M}={\text{Z}}_{\text{i}}\upgamma\) in the selection (adoption) equation where \(\widehat{M}\) is the predicted probability of adoption \({M}_{\text{i}}\).

Therefore, the inverse Mill’s ratio components from the logit model are incorporated into the outcome equations (Eqs. 9 and 10) to mitigate potential selection bias, treating it as a missing variable issue in the ESR model.

$${Y}_{i1}={X}_{i1}{\beta }_{1}+{\sigma }_{e1u}{\lambda }_{i1}+{\xi }_{1i},$$
(9)
$${Y}_{i2}={X}_{i1}{\beta }_{2}+{\sigma }_{e2u}{\lambda }_{i1}+{\xi }_{2i},$$
(10)

The presence of endogenous switching is identified when the statistical significance of the covariance between the error terms in the selection equation and the outcome equations (\({\upsigma }_{{\text{u}}\varepsilon_{1}}\) and \({\upsigma }_{{\text{u}}\varepsilon_{2}}\)) is established. Although the above equations can be estimated through a two-stage procedure, the Full Information Maximum Likelihood (FIML) estimator, as suggested by Lokshin and Sajaia [47], is considered more efficient.

[46], the expected values of the outcomes of adopters and non-adopters in actual and counterfactual scenarios can be computed using the following equations (Eqs. 11, 12, 13, 14):

For adopters;

$$E\left[{Y}_{i1}|{M}_{i}=1\right]={X}_{i1}{\beta }_{1}+{\sigma }_{e1}{\lambda }_{i1}.$$
(11)

Had they not adopted (counterfactual)

$$E\left[{Y}_{i2}|{M}_{i}=1\right]={X}_{i1}{\beta }_{2}+{\sigma }_{e2}{\lambda }_{i1}.$$
(12)

For non-adopters;

$$E\left[{Y}_{i2}|{M}_{i}=0\right]={X}_{i2}{\beta }_{2}+{\sigma }_{e2}{\lambda }_{i2}.$$
(13)

Had they adopted (counterfactual)

$$E\left[{Y}_{i1}|{M}_{i}=0\right]={X}_{i2}{\beta }_{1}+{\sigma }_{e1}{\lambda }_{i2}.$$
(14)

Then, the impact of HSTs adoption on maize storage quantities and the HFIAS for households that used HSTs can be calculated as the difference between Eqs. (11) and (12):

$$E\left[{Y}_{i1}|{M}_{i}=1\right]- E\left[{Y}_{i2}|{M}_{i}=1\right]={(\beta }_{1}-{\beta }_{2}){X}_{i1}+({\sigma }_{e1}-{\sigma }_{e2}{)\lambda }_{i1}.$$
(15)

Similarly, the impact of HSTs adoption on maize storage quantities and food insecurity for households that did not use HSTs can be calculated as the difference between Eqs. (13) and (14):

$$E\left[{Y}_{i2}|{M}_{i}=0\right]- E\left[{Y}_{i1}|{M}_{i}=0\right]=\left({\beta }_{2}-{\beta }_{1}\right){X}_{i2}+\left({\sigma }_{\varepsilon 2}-{\sigma }_{\varepsilon 1}\right){\lambda }_{i2}.$$
(16)

2.6 Impact assessment procedure

The procedure for assessing the impact of adoption on HFIAS and QSM involved two steps. First, we estimated the ESR model of HFIAS and QSM using the Full Information Maximum Likelihood (FIML) estimator in the movestay command of Stata 16. Then we generated the distributions of expected HFIAS and QSM. For adopters, we generated the observed (Eq. 13) and counterfactual (Eq. 14) distributions using the ESR model parameter estimates and data for household characteristics observed in the sample. Then, we plotted the cumulative distributions and checked their stochastic dominance. If, for example, the observed QSM distribution lies to the right of the corresponding counterfactual, we can suggest that HSTs adoption increases QSM. This will, however, be concluded using the Kolmogorov–Smirnov statistic for first-degree stochastic dominance. A statically significant Kolmogorov–Smirnov statistic rejects the hypothesis that the two distributions (observed and counterfactual) are the same, concluding that HSTs adoption increases QSM.

3 Results

3.1 Descriptive results

Means for all model variables (outcome, treatment, and control) and mean differences between adopters and non-adopters are presented in Table 3. Results show that about 39.62% of all the surveyed households adopted HSTs. Most households in this study had male heads (89%) with an average age of 47 years. They had completed primary school education (74.5%) and had an average of 8 years of schooling. The households had an average of 5 members. Regarding farming characteristics, on average, they cultivated a plot size of 3.33 acres, with about 82.8% using improved maize varieties. Crop spraying for pest/insect control and herbicide use was moderate, with 31.9% and 47.8% of the households applying these inputs to their maize plots, respectively.

Table 3 Descriptive summary of model variables
Full size table

Regarding socioeconomic characteristics, 76.9% of the households used hired labor for maize production. About 42% earned income from livestock, 34.7% earned income from off-farm activities, and 30.5% owned motorbikes. The study also found that only 27.3% of households had HSTs vendors available in their villages, and around 60% used pesticides for maize storage. The average length of maize storage was 5.13 months, and about 67.3% of the households included in this study received training on HSTs. In addition, the average distance to the village market from home was 1.16 km.

Results showed statistically significant differences between adopters and non-adopters in most demographic and socioeconomic characteristics (Table 3). Adopters of HSTs were significantly older than non-adopters at the 5% significance level. A large percentage of non-adopters had only attained primary school education, while adopters had significantly more years of schooling than non-adopters. In terms of socioeconomic characteristics, adopters of HSTs had high percentage of ownership of motorbikes, earning income from livestock, and access to off-farm income. This suggests that adoption of HSTs is accompanied by good socioeconomic status of the households. In addition, a large percentage of adopters of HSTs resides in villages with HSTs vendors and received training on storage pest protection than non-adopters.

Large percentage of HSTs adopters were shown to have used improved maize varieties and herbicides than non-adopters, and the differences were statistically significant. Interestingly, many non-adopters significantly used pesticides for storage compared to adopters. This result implies that adopters of HSTs are less likely to use pesticides for storage. In addition, HSTs adopters significantly stored more maize at harvest and had lower food insecurity (significantly lower HFIAS value), while non-adopters stored a low amount of maize at harvest and had high food insecurity.

Table 4 presents the mean differences in responses to the questions used to construct the HFIAS and the HFIAS. Except for questions 3 and 4, there was a statistically significant mean difference in responses to the HFIAS questions. HSTs adopters had lower values of HFIAS, indicating that HSTs adopters were less food insecure than non-adopters. The differences in the magnitude of the HFIAS value between adopters and non-adopters of HSTs were statistically significant. However, we cannot make causal inferences based on these descriptive statistics results. The observed differences in maize storage quantities and the HFIAS between adopters and non-adopters might not be solely caused by HSTs adoption. Other factors not accounted for might have compounded the effects of HSTs adoption on the outcome variables.

Table 4 Descriptive summary of the HFIAS indicators
Full size table

3.2 Determinants of the adoption of HSTs

Table 5 presents the ML parameter estimates of the determinants of adopting HSTs using the logit model. The logit model was correctly fitted with the low log-likelihood ratio of − 524.99, and the Wald χ2 (22) test value 437.93 was significant at a 1% level, indicating that the explanatory variables together influenced the probability of adoption of HSTs among the maize farmers in the study area. Results show that adoption of HSTs is strongly associated with the gender of the household head, primary school education, number of years of schooling, off-farm income, presence of HSTs vendors close to farmer’s residence, herbicide usage, storage pesticides, crop spraying, and training on HSTs.

Table 5 The maximum likelihood estimates of the logit model
Full size table

Results show that the gender of the household head positively influences the adoption of HSTs at a 5% significance level. Male household heads are 12.4% more likely to adopt HSTs than female household heads. This finding is also revealed by [27]. Education, in terms of primary school education and years of schooling, was also found to be significant and positively related to the adoption of HSTs at a 10% and 1% level respectively. Households with primary school education and more years of schooling are 10.6% and 4.5% more likely to adopt HSTs. Households with off-farm income can easily access HSTs vendors, influencing their adoption decision by 7.1%. The presence of HSTs vendors in the community was found to have a significant positive relationship with the adoption of HSTs at a 1% level. Households residing in villages with the availability of HSTs vendors were 28.3% more likely to adopt HSTs. Households who received training on HSTs were also more likely to adopt HSTs by 34.3%.

The study revealed a significant positive relationship (at 1%) between the application of herbicide and the adoption of HSTs. Households that applied herbicides were 7.1% more likely to adopt HSTs. Households that applied crop spraying on their maize plots were 7.1% less likely to adopt HSTs, and those who applied storage pesticides were 42.7% less likely.

3.3 ESR model estimation results

3.3.1 Model diagnostics and parameter estimates

We utilized the FIML method proposed by [47] in Stata Version 16 (Stata Corp, College Station, TX, USA) to estimate the ESR models, assessing the impacts of HSTs on QSM (Table 11 presented in the appendix) and HFIAS (Table 12 presented in the appendix). The Wald test of independent equations rejects the null hypothesis of joint independence between the selection and outcome equations for adopters and non-adopters. As HSTs adoption is identified as an endogenous variable in the outcome equations, acknowledging and correcting for selection bias becomes crucial for obtaining an unbiased and consistent estimation of the impacts of HSTs adoption. This validates the selection of ESR model over the OLS estimation, providing evidence of statistically significant differences in the second-stage parameter estimates of the control variables in the outcome equations between adopters and non-adopters, justifying the need for a more nuanced analytical approach.

When estimating the ESR models, we included herbicide use and training on HSTs as instruments for causal identification of the maize storage and HFIAS impact estimations, respectively. Farmers who applied herbicides to their maize plots were likelier to adopt HSTs. This shows that households that use herbicides have a positive attitude toward technologies that maximize productivity. Hence, they are more likely to adopt HSTs to reduce storage losses. However, herbicide usage is not expected to affect the maize storage outcome variable directly. Similarly, farmers who received training on HSTs are more likely to adopt HSTs because of the knowledge they acquired [20, 29, 48]. In addition, training received on stored product protection is not expected to directly affect food insecurity as measured by HFIAS.

We conducted a falsification test on the validity of the instruments following [49]. The test results showed that the application of herbicides directly influences the decision to adopt HSTs (chi2 = 6.36, p = 0.0117) but has no effect on maize storage (F = 1.61, P = 0.2049). The test results for training received showed that receiving training on HSTs had a significant positive effect on HSTs adoption (chi2 = 87.15, p = 0.000) but had no effect on HFIAS (F = 2.67, P = 0.1026). This indicates that the instruments are not directly correlated to the outcome variables (maize storage and HFIAS).The correlation coefficient between the error terms in the equations for HSTs adoption and QSM for non-adopters is both negative and statistically significant (ρe2u − 0.410; P < 0.01). Conversely, the correlation coefficient between the error terms in the HSTs adoption equation and the HFIAS equation for adopters is positive and statistically significant (ρe1u = 0.249; P < 0.001). These findings indicate the presence of selection bias in HSTs adoption. Hence the choice of the ESR model was important to remove this bias. This is based on the null hypothesis of no sample selection bias in impact estimation, which was rejected and controlled through the ESR model.

The correlation coefficients between error terms of the selection equation and the outcome equation for maize storage are characterized by negative signs (same signs) for both adopters and non-adopters, indicating a consistent pattern of hierarchical sorting in adoption decisions. This means that adopters are likely to have adopted HSTs triggered by expected benefits such as increased food storage [50]. Similarly, non-adopters are likely to have not adopted HSTs because they may not have expected to benefit. Therefore the actual adopters, had they not adopted (counterfactual), would have less food storage than what is actually observed. Similarly, if non-adopters had adopted the HSTs, they would have had more food storage than was observed. Furthermore, the correlation between the HSTs adoption and the HFIAS outcome equations for both adopters and non-adopters exhibits opposing signs, suggesting that their decision to adopt HSTs is influenced by their comparative advantages over other storage technologies.

3.3.2 Effects of HSTs adoption at different levels of outcome variables for adopters

The observed and counterfactual cumulative distributions of the QSM and HFIAS for adopters are displayed in Figs. 2 and 3, respectively. The position of the observed distribution of the outcome variables relative to the corresponding counterfactual at different levels suggests the range of outcomes over which HSTs adoption has an effect. Figure 2 shows that the observed distribution of QSM lies predominantly to the right of the corresponding counterfactual distribution at all QSM levels, suggesting that HSTs adoption has a higher probability of increasing the QSM at all levels. In contrast, Fig. 3 shows that the observed distribution of HFIAS lies predominantly to the left of the counterfactual distribution at all HFIAS levels, suggesting that HSTs adoption has a higher probability of reducing HFIAS at all levels.

Fig. 2
figure 2

Observed and counterfactual cumulative distributions of QSM for adopters (KS test 0.9613***)

Full size image
Fig. 3
figure 3

Observed and counterfactual cumulative distributions of HFIAS for adopters (KS test = 0.5672***)

Full size image

Using the Kolmogorov–Smirnov test for first-degree stochastic dominance, we reject the null hypothesis that the two distributions (observed and counterfactual) are the same, concluding that HSTs adoption increased QMS (KS statistic = 0.9613 statistically significant at 1%) and decreased HFIAS (KS statistic = 0.5672 statistically significant at 1%).

3.3.3 The average effects of HSTs adoption for adopters

The average effects of HSTs adoption on QSM and HFIAS for adopters are presented in Table 6. The results show that adopters stored an average of 1169.54 kg of maize. However, if they had not adopted HSTs, they would have stored only 701.55 kg. This means that HSTs has increased maize storage quantities by 467.97 kg, representing a 40% increase accrued to HSTs adoption. The observed HFIAS for adopters was 2.52. The counterfactual HFIAS was 4.43 points, equivalent to more than 43% decrease in HFIAS accrued from HSTs adoption. This result suggests that if the adopters had not adopted the HST, they would have been more food insecure by 1.91 units.

Table 6 ESR-based average impacts of HSTs for adopters
Full size table

3.3.4 Effects of HSTs adoption at different levels of outcome variables for non-adopters

The observed and counterfactual cumulative distributions of the QSM and HFIAS for non-adopters are displayed in Figs. 4 and 5, respectively. The position of the observed distribution of the outcome variables relative to the corresponding counterfactual at different levels suggests the range of outcomes over which HSTs adoption has a potential effect. Figure 4 shows that the counterfactual distribution of QSM at its levels beyond 1000 kg lies to the right of the corresponding observed distribution, suggesting that HSTs adoption would potentially have a higher probability of increasing QSM for households storing more than one ton. In contrast, Fig. 5 shows that the counterfactual distribution of HFIAS predominantly lies to the left of the observed distribution, suggesting that HSTs adoption would potentially have a higher probability of reducing food insecurity at all levels.

Fig. 4
figure 4

Observed and counterfactual cumulative distributions of maize storage for non-adopters (KS test = 0.5381***)

Full size image
Fig. 5
figure 5

Observed and counterfactual cumulative distributions of HFIASS for non-adopters (KS test = 0.1689***)

Full size image

Using the Kolmogorov–Smirnov test for first-degree stochastic dominance, we rejected the null hypothesis that the two distributions (observed and counterfactual) are the same, concluding that HSTs adoption would potentially have increased QSM (KS statistic = 0.5381 statistically significant at 1%) and decreased the HFIAS (KS statistic = 0.1689 statistically significant at 1%).

3.3.5 The average potential effects of HSTs adoption for non-adopters

The average potential impacts of HSTs adoption for non-adopters on QSM and HFIAS are presented in Table 7. The actual average QSM by non-adopters was 1051.12 kg. However, if they had adopted the HSTs, they would have stored 1265.47, meaning they could have stored an additional 214.18 kg. The average observed HFIAS for non-adopters was 3.19 points. However, the average counterfactual HFIAS would have been 2.90, suggesting that non-adopters would have been food secure if they had adopted HSTs. Therefore, adopting HSTs would be beneficial for farmers who did not adopt the technology, as evidenced by the increase in QSM and decrease in HFIAS. This result implies that non-adopters are missing out on the benefits of HSTs.

Table 7 ESR-based average impacts of HSTs for non-adopters
Full size table

3.3.6 Impacts of HSTs adoption by gender of the household head and use of storage pesticides

Table 8 presents the impacts of HSTs adoption on QSM and the HFIAS disaggregated by the gender of the household head. The results show that the MHHs and FHHs had more QSM under observed conditions than counterfactual conditions, suggesting they benefited from HSTs adoption. However, the two groups had no significant difference in the QSM, suggesting that the impact of HSTs adoption on QSM is not associated with the gender of the household head. In other words, HSTs adoption benefitted the FHHs as equally as the MHHs.

Table 8 Average impacts by gender of the household head
Full size table

Likewise, the impact of HSTs adoption on HFIAS is not associated with the gender of the household head. The HFIAS for the FHHs decreased by 1.88 units compared to 1.91 for the MHHs. The differences between the two are not statistically significant, suggesting that both MHHs and FHHs benefited equally.

Table 9 presents the impacts of HSTs adoption on QSM and the HFIAS disaggregated into users and non-users of storage pesticides. The results show that while the users and non-users of storage pesticides benefited from HSTs adoption, the users had more QSM, suggesting they gained more from HSTs adoption than non-users, suggesting that the impact of HSTs adoption on QSM is associated with storage pesticide usage. HSTs adoption benefitted the users in terms of QSM more than the non-users.

Table 9 Average impacts by application of storage pesticides
Full size table

Regarding HFIAS, it is revealed that the users and non-users reduced HFIAS, suggesting they benefited from HSTs adoption. The two groups had significant differences in the HFIAS, suggesting that the impact of HSTs adoption on HFIAS is associated with the application of storage pests. In addition, HSTs adoption benefitted households who did not use storage pesticides more than the users.

4 Discussion

4.1 Discussion of the findings

The results reveal a slightly higher adoption rate in Tanzania compared to previous studies, indicating an improvement in adoption rate. A more recent study by [35] shows that the adoption rate of HSTs in Tanzania is 34.5%. The increased adoption of HSTs suggests positive outcomes resulting from promotional and scaling efforts of the HSTs undertaken by various stakeholders. Additionally, the increased adoption identified in this analysis may be attributed to membership in producer organizations. Joining such organizations enables farmers to access loan that can be utilized for acquiring the HSTs. Some producer organizations even assist members in procuring the HSTs during planting season, allowing them to repay after the harvest, hence enhancing their adoption decision. The study by [51] demonstrated that farmers join producer organizations to access credit, and that members of POs have higher income and lower indebtedness. This particular finding aligns with the research conducted by [52], which suggests that producer groups/organizations should be made accessible to farmers to enhance financial inclusion, an essential factor in improving their welfare.

Determinants of HSTs adoption among members of PO’s as presented in Table 5 show that education was among the factors that influence HSTs adoption. Households with educated heads are more likely to adopt HSTs because they are more exposed and positive to new technologies and development. This finding is consistent with many other studies that revealed the importance of education in technology adoption [50]. Another important result is that adoption of HSTs is influenced by off-farm income. This is because the technology is too expensive for a smallholder farmer who is used to purchasing an ordinary woven bag at 0.2$ to 0.4$. After all, one HSTs bag (i.e. PICS bag) is sold at a minimum of 2$. Therefore, reliance on farming only as a source of income may limit farmers to purchase the HSTs. Furthermore, [35], argued that off farm income plays a crucial role in facilitating farmers’ ability to make payments seamlessly before acquiring the HSTs. This is particularly relevant as HSTs are typically procured through cash payments rather than in kind exchanges or credit arrangements.

The presence of HSTs vendors close to farmers’ residences is important in influencing the adoption of HSTs as it removes the transportation constraint. This finding is consistent with [11], who found that vendors play a vital role in smoothing the supply of HSTs, and for effective provision of the HSTs, the vendors should be located at the village centers which are the closest location to farmers. This shows that an effective supply chain for HSTs is crucial for enhancing the adoption of HSTs [53]. Training on HSTs was found to affect the adoption of HSTs positively. This means that households who received training on HSTs are exposed to the benefits and effectiveness of the technology in protecting the stored product and hence easily convinced to adopt the HSTs. This finding emphasizes the importance of training for technology adoption as also found by [20].

Application of herbicides increased the probability of adopting HSTs. This suggests that households that use recommended inputs are more likely to adopt HSTs to protect the food produced from storage pests/insects. Since these households are willing to increase productivity by using inputs such as herbicides, they are likely to be sensitive to the need to protect the large amount of food produced after harvest. Furthermore, [54] also found that households who used HSTsare more likely to adopt modern inputs such as hybdrid maize varieties. Our finding compliment this by indicating that households that used HSTs are more likely to use herbicides, another form of modern input.

Results of the impacts of HSTs adoption reveal that HSTs increase maize storage quantities by 40% and reduces food insecurity by 43%. In other words, the increased maize storage is shown to assist in reduction in food insecurity. These findings are consistent with previous studies which found that HSTs adopters had lower food insecurity than non-adopters [32, 55]. A study by [56] on the impacts of hermetic storage bags supply in Tanzania on food security showed that it decreased the HFIAS by 30.09%. A more recent study by [35] who analyzed the adoption and impacts of airtight storage (another form of HSTs) in Tanzania found that it reduced food insecurity by 51% arguing that HSTs address both product quality and quantity during storage. Using a unique sample of maize producers in producer organizations, our result show a significant increase in food storage (40%) and decrease in HFIAS (43%).

Analysis of the impacts of HSTs by usage of storage pesticides (chemicals) showed that households that used pesticides are observed to have higher impacts in terms of maize storage quantity than those that did not. This result is interesting because HSTs was expected to reduce the use of storage pesticides, and therefore, the impacts of HSTs for adopters by application of storage pesticides were expected to be low. In contrast, the results suggest that HSTs adopters still used pesticides for storage, and as a result, were able to store more. Moreover, this result indicates that some farmers combine HSTs and non-HSTs for maize storage, as suggested by the high HSTs impacts on storage quantities for adopters who used storage pesticides. This denotes that the application of storage pesticides for HSTs adopters was for maize stored using non-HSTs, which explains why the impacts of HSTs by application of storage pesticides was high for households that used pesticides. These findings are consistent with the results of the study [48], which showed that farmers use a combination of different storage technologies to store their grains to minimize losses during storage.

This paper stands out due to its exclusive focus on producer organizations in the analysis. Members of producer organizations exhibit different behaviors compared to smallholder farmers, largely due to their access to input credits and output markets. The examination of the adoption and impacts of hermetic storage technologies on food insecurity presented in this paper provides crucial evidence for governments, donors, and other stakeholders advocating for the institutionalization of the agriculture sector through the formation of farmers' associations, organizations, and groups. However, similar factors explored in previous studies that did not specifically consider producer organizations, such as HST supply chain development and training on HSTs, remain relevant. These findings build upon previous research and highlight the need for interventions aimed at improving the development of the supply chain for HSTs and providing training programs to enhance adoption and the resulting impacts on food security.

5 Conclusion and implications

5.1 Conclussion

This article identified the determinants of HST adoption among members of producer organizations using the logit model. It also assessed the impacts of HST adoption on reducing household food insecurity using the ESR model based on two indicators—QSM and HFIAS. The study revealed factors that affect adoption of HST among members of POs was gender, education, off-farm income, training on HSTs, presence of HST vendors in the village, herbicide usage, storage pesticides, crop spraying, and hired labor. These parameters present potential differences among members of producer organizations which have implications in their adoption decisions and their impacts. The availability of HST vendors is a basic requirement for enhancing the adoption of HST. Periodic training on HST proper usage is important for improving their knowledge. The adoption of HST increased QSM by about 467.97 kgs and decreased the HFIAS by 1.91 points for adopters. Impacts are also observed for non-adopters, had they adopted HST their QSM would potentially have increased by 214.18 kgs. HFIAS would also have decreased by 0.29 units. This suggests that they would have more food available and less food insecurity.

The intra-household distribution of the impacts by gender did not show any significant differences between male and female-headed households. The distribution of the impacts by usage of storage pesticides denoted a tendency to combine both improved and local technologies during storage. This signifies the importance of training to emphasize the reduction of storage pesticide usage by stressing the health effects of consuming maize stored with chemicals. It is important to note that the data used to generate results from maize producers among producer organizations. The HSTs adoption rate and its determinants presented in this study may not be applicable for typical smallholder farmers, hence it should be used with caution. Generally, the results support promoting HSTs as part of household food insecurity-reduction strategies, particularly among the current non-adopters constrained by knowledge of storage pest control and missing out on the benefits of adopting HSTs. These suggestions are consistent with the country’s recently adopted 10-year National Postharvest Management Strategy (NPMS) of 2019–2029 that aims to build postharvest actors' institutional capacity to reduce postharvest losses and increase income, food, and nutrition security.

Therefore, given the findings of this study, we recommend the following:

  1. 1.

    To ensure timely supply of HST and eliminate stock out problems during harvest time, distributors and vendors should have access to capital so they can be able to order the HST early prior to harvest. Additionally, farmers facing financial constraints can be offered the HST on credit and pay after selling the stored maize under a simple agreement.

  2. 2.

    Improving the distribution of HST bags can be achieved by partnering with Polypropylene woven bags distributors. This can be done by combining the Polypropylene woven and HST bags during distribution. Distributors can offer various options to include HST such as offering a free woven bag with the purchase of one HST bag, or creating bundles of HST and woven bags at different combinations and prices. This strategy can improve the efficiency of supply chain and result in better prices for HST bags.

5.2 Limitations

Our study's findings are based on a specific population of maize producer organizations in Tanzania. The socio-economic factors influencing HST adoption within these POs may differ from those affecting maize producers who are not part of such organizations. For instance, POs often have better access to input loans and markets, which can significantly influence adoption behavior. Consequently, the generalizability of our results to other settings or populations with different characteristics may be limited. Additionally, while we employed several strategies to mitigate biases in self-reported data, some degree of bias may still be present, potentially affecting the broader applicability of our findings.

5.3 Implications and future research

Despite these limitations, our study offers valuable insights into the factors influencing HST adoption within Tanzanian POs. Future research should aim to replicate this study across diverse geographical regions and different types of agricultural organizations to enhance the generalizability of the findings. Larger and more diverse samples, including individual farmers not affiliated with POs, would help capture a wider range of influences on HST adoption. Furthermore, incorporating objective measures alongside self-reported data could improve data accuracy and reliability. This approach would provide more robust evidence for developing policy recommendations that are applicable to a broader range of contexts.