3.1 Data

The study uses data from the National Panel Survey (NPS) collected by the World Bank in collaboration with the Tanzania National Bureau of Statistics within the framework of the Living Standards Measurement Study–Integrated Surveys on Agriculture (LSMS-ISA) project (link). The NPS is a nationally representative household panel survey based on a two-stage stratified cluster random sampling method. Stratification was based on two key dimensions: (i) eight administrative zones (seven in Mainland Tanzania and Zanzibar as the eighth zone), and (ii) rural versus urban clusters within each administrative zone, resulting in 16 strata. In the first stage, clusters – defined as census enumeration areas (EAs) in urban areas and villages in rural areas – were randomly selected within each stratum with a probability proportional to their population size, based on the 2002 Population and Housing Census (see Fig 2). In the second stage, households were randomly selected from each cluster using a comprehensive household listing (Himelein, Reference Himelein2014). Eligibility for the first NPS round (2008/09) and subsequent rounds required households to have at least one member aged 15 years and above, excluding live-in servants. The number of households sampled per cluster varied proportionally to the population size. Selected households were followed across all waves of the NPS to monitor their outcomes over time. Tracking was limited to households that relocated within Tanzania; those that moved to another country were not included in the follow-up. The NPS was initially designed to track households biennially. However, this schedule was not maintained during the fifth wave conducted in 2020/21. This may be due to a shift in the World Bank’s strategic priorities, among other factors.

Figure 2.

Overview of distribution of census enumeration areas.

The surveys were conducted in five rounds: 2008/09, 2010/11, 2012/13, 2014/15, and 2020/21 with household samples of 3280, 3265, 3625, 3360, and 3352, respectively. The focus is on the maize crop, as maize is the primary staple crop in Tanzania, occupying about 45% of the total cultivated area (Arce and Caballero, Reference Arce and Caballero2015; Frederick et al., Reference Frederick, Sabula, Mruma, Mzee, Mtoka, Masigo, Ndunguru and Swai2020). The sample sizes for maize farmers across the first five rounds were 1027, 1270, 1546, 1236, and 1256 households, totaling 6335 households. The analysis was conducted after addressing outliers. We considered observations below the 5^th percentile and above the 95^th percentile as outliers. Instead of trimming outliers, we replaced them using the winsor2 Stata command with the replace cut (5 95) option. This method prevents extreme values from skewing the analysis while maintaining the integrity of the dataset. The survey data include sociodemographic characteristics, household assets like mobile phone ownership, access to agricultural extension services and prices, and agricultural variables such as cultivated crops, agricultural output, and plot size, as indicated in the original questionnaire (link).

Access to extension services in this study is treated as a binary indicator, reflecting whether households received any extension services, with literature supporting this methodology despite critiques about its lack of depth in evaluating service intensity and diversity (e.g., Issahaku et al., Reference Issahaku, Abu and Nkegbe2017; Lee et al., Reference Lee, McNamara and Ho2023; Malimi, Reference Malimi2023, Nyaplue-Daywhea et al., Reference Nyaplue-Daywhea, Ahiakpa, Mensah, Annor-Frempong and Adjei-Nsiah2021). Our model assesses three types of extension services: government extension services (GES), private extension services (PES), and a general category (GE) for any extension service received. The absence of GES and PES classifies a household as having no extension service access. These services offer advice on various agricultural aspects, including production, agro-processing, marketing, pest and disease control, and pricing. While access to these services has declined over the years, Figure 3 shows that farmers with mobile phones had relatively better access than those without.

Figure 3.

Trends in access to extension services for mobile phone owners vs. non-owners over the years.

Table 1 provides some descriptive statistics for the sample and the key variables used in the estimations. Table 1 shows a significant drop, with access to (general) extension decreasing from 62% in 2008/09 to about 50% in 2010/11, further plummeting from 62% in 2012/13 to 40% in 2020/21. This may be explained by the decline in the public budget allocated to agricultural extension services. According to the MAFAP (2013) report, the budget for these services decreased from 10% of the total budget allocated to agriculture in 2007/08 to 5% of the total budget for agriculture in 2010/2011. Another plausible explanation for this significant drop might be attributed to the impacts of COVID-19 events on the delivery of extension services (Baffoe-Bonnie et al., Reference Baffoe-Bonnie, Martin and Mrema2021). When looking at agricultural extension sources separately, the proportion of smallholder maize farmers relying solely on government extension services has reduced from 20% in 2008/09 to 5% in 2020/21. On the other hand, the proportion of smallholder maize farmers relying on private extension services experienced a steady decline, dropping from 51% in 2008/09 to around 35% in 2020/21. However, the proportion of smallholder maize farmers who are users of mobile phones has dramatically increased from about 30% in 2008/09 to 83% in 2020/21. On average, the share of households (HH) with mobile phones to the total number of rural households at the community level significantly increased from 11% in 2008/09 to 56% in 2020/21. This supports the effectiveness of the Tanzanian Rural Development Strategy that led to the introduction of ICT in rural areas (FAO and ITU, 2022). The increased percentage of phone ownership provides an opportunity to improve the coverage of agricultural extension services in Tanzania. Moreover, there has also been a noteworthy increase in maize yields, rising from 411 kg per acre in 2008/09 to 520 kg per acre in 2020/21, indicating an effective collaboration among the Tanzanian government, international, and local organizations to enhance agricultural productivity (USAID, 2015). Table 1 also indicates a slight improvement in the education levels of household heads, accompanied by a significant decrease in the number of adults within the households from 2008/09 to 2010/11. However, the number of adults remained stable from 2010/11 to 2020/21, probably due to the National Agricultural Policy launched in 2011 aimed at reducing rural-urban migration by promoting rural development and enhancing economic opportunities in rural areas (MAFSC (Ministry of Agriculture Food Security and Cooperatives), 2013). Furthermore, the use of both inorganic fertilizers – such as phosphate, UREA, Calcium Ammonium Nitrate, and Sulphate of Ammonium – and pesticides has relatively remained steady from 2008/09 to 2020/21, while there has been a significant increase in the adoption of improved varieties of maize seeds over the same period. This reflects the effectiveness of government and local initiatives in promoting sustainable agriculture within Tanzania (USAID, 2015). Additionally, Table 1 indicates that soil quality has remained relatively stable (relatively good soil) over the years, suggesting minimal impact from climate change over the last decade. This may be attributed to the negligible changes in temperature between 2008/09 and 2020/21. Meanwhile, precipitation has shown a steady increase, rising from about 1,029 mm in 2008/09 to 1,289 mm in 2020/21.

Table 1.

Descriptive statistics of the sample

Note: standard deviations are in parentheses.

3.2 Empirical strategies

The empirical procedures encompass three steps: first, we use the correlated random effects (CRE) estimation approach to investigate the relationship between mobile phone ownership and access to extension services as well as the relationship between mobile phone ownership and maize yields. Second, considering the potential endogeneity of the mobile phone ownership variable, we employ an instrumental variables estimation to assess the causal relationship between mobile phone ownership and maize yields. Finally, we adopt a mediation technique to meticulously disentangle the indirect effects of mobile phone ownership on maize yields from its direct effects.

Additionally, we conduct two distinct analyses. Initially, we estimate the effects of mobile phone ownership assuming no heterogeneity and compare the effects of mobile phone ownership across government and private extension sources. Next, we estimate the effects of mobile phone ownership by gender to measure the heterogeneous effects of mobile phone ownership between male and female-headed farm households.

3.2.1 Correlated random effect estimation

This study employs panel data estimation models utilizing the data from LSMS-ISA data. The fixed-effects (FE) approach is favored for mitigating endogeneity bias due to unobserved heterogeneity that is constant over time. However, FE models cannot estimate the impact of time-invariant variables such as gender (Joshi and Wooldridge, Reference Joshi and Wooldridge2019; Schunck, Reference Schunck2013). Additionally, FE models may not be ideal when the treatment variable exhibits little variation over time, as could be the case for extension access since household fixed effects can absorb much of the variation. Alternatively, the random effects (RE) approach could be considered to identify the effect of time-invariant variables. However, it relies on a very strong assumption rarely fulfilled in practice. The RE approach assumes that explanatory variables included in the model are uncorrelated with the unobserved individual heterogeneity (Greene, Reference Greene2008).

Therefore, we utilize the correlated random effects (CRE) approach also known as the Mundlak (Reference Mundlak1978) procedure to address these limitations, amalgamating the strengths of both FE and RE methods (Wooldridge, Reference Wooldridge2018). This approach allows for a correlation between time-invariant unobserved heterogeneity and explanatory variables by incorporating the time averages of the time-varying covariates (Joshi and Wooldridge, Reference Joshi and Wooldridge2019). It is also particularly effective for handling unbalanced panels (Wooldridge, Reference Wooldridge2018). The CRE method has been previously applied to similar datasets for causal inquiries (Arslan et al., Reference Arslan, Cavatassi, Alfani, Mccarthy, Lipper and Kokwe2018; Malimi, Reference Malimi2023; Mukasa, Reference Mukasa2018), and by including the time averages of the varying covariates, it effectively provides the FE estimates for these variables (Mundlak, Reference Mundlak1978). More specifically, we apply a CRE estimation based on the following models:

(1) $$S_{it}=\alpha _{0}+\alpha _{1}m_{it}+\alpha _{i}X_{it}+\delta _{t}+\varphi _{i}+\epsilon_{it}$$

$$S_{it} = \left\{ {\matrix{1 & {\rm if \;household \;{\it i} \;have \;access \;to \;agricultural \;extension} \cr 0 & {{\rm{otherwise}},} \cr } } \right.$$

(2) $$ln \;(y_{it})=\beta _{0}+\beta _{1}S_{it}+\beta _{2}m_{it}+\beta _{i}X_{it}+\rho _{t}+\omega _{i}+e_{it}$$

Where the S denotes access to agricultural extension services by the ith household at the time t, m is farmers’ mobile phone ownership, y is maize yield, X is a set of household- and plot-specific characteristics, farm inputs, and location-fixed effects, α ₀ and β ₀ are the intercepts terms, α and β the vector of parameters to be estimated, and ε _it and e _it the random error terms. Notably, φ _i and ω _i are the household fixed effects that control for unobserved time-invariant household heterogeneity. Additionally, δ _t and ρ _t are the year-fixed effects that control for standard shocks from one year to the other, e.g., changes in government policies that affect all households. This way, we control for time-varying confounders, time-invariant observable, unobservable factors, and common yearly variations to identify the causal effect of mobile phone ownership on yield.

3.2.2 Instrumental variable estimation

Mobile phone ownership measure is subject to endogeneity due to reverse causality, as farmers with higher agricultural productivity are more likely to use mobile phones (e.g., Nyagango et al., Reference Nyagango, Sife and Kazungu2023). To account for the possible endogeneity of mobile phone ownership and crop yields, we apply instrumental variables estimation (Joshi and Wooldridge, Reference Joshi and Wooldridge2019). Drawing on the research of Nie et al., (Reference Nie, Ma and Sousa-Poza2021) and Rotondi et al., (Reference Rotondi, Stanca and Tomasuolo2017), this study initially suggests the share of households owning a mobile phone at the village-year level, excluding household i, as an instrument for the mobile phone variable. This instrument is derived from existing literature that emphasizes the crucial role of peer effects and social learning in shaping households’ decisions to use mobile phones (Krell et al., Reference Krell, Giroux, Guido, Hannah, Lopus, Caylor and Evans2021; Nie et al., Reference Nie, Ma and Sousa-Poza2021; Rotondi et al., Reference Rotondi, Stanca and Tomasuolo2017; Sathye et al., Reference Sathye, Prasad, Sharma, Sharma and Sathye2018; de Silva et al., Reference de Silva, Ratnadiwakara and Zainudeen2012). This instrument has been applied in previous studies (e.g., Rotondi et al., Reference Rotondi, Stanca and Tomasuolo2017). The instrument is theoretically aligned with exclusion criteria, as farmers’ maize yields cannot be directly affected by the fraction of households having a mobile phone except through enhancing mobile phone adoption and spread of agricultural information as well through peer effects. However, building on Manski’s (Reference Manski1993) reflection principle, we suspect that the reflection problem arising from endogenous social effects might affect this instrument. Specifically, a household’s mobile phone ownership is likely influenced by the share of households owning a mobile phone in village j, and vice versa. To address this, we excluded household i when calculating the share of households owning a mobile phone in village j, thereby reducing the risk that the instrument captures intra-household peer effects. Although this helps mitigate bias, it does not fully resolve the reflection problem. To strengthen identification and enhance the efficiency of the IV estimator, we employed Lewbel’s (Reference Lewbel2012) method. By generating heteroskedasticity-based instruments from straightforward functions of the model’s existing data, this method supplements our external instrument. We used the ivreg2h Stata command to estimate the heteroskedasticity-based instrument model. Following Baum and Lewbel (Reference Baum and Lewbel2019), we applied ivreg2h to panel data using within transformation. We assessed the quality of both the generated and external instruments through the three key traditional tests – underidentification, weak identification, and Hansen J statistic. The resultsFootnote ¹ confirm that the instruments are strong and valid, with the following statistics: Kleibergen-Paap rk LM statistic= 88.7 (p=0.000), Cragg-Donald Wald F statistic= 27.47, and Hansen J statistic= 11.90 (p=0.536). We estimate the following model, namely the second-stage equation of the IVREG2H model:

(3) $$ln(y_{it})=\phi _{0}+\phi _{1}S_{it}+\phi _{2}m_{it}^\ast+\phi _{i}X_{it}+\tau _{t}+\xi _{i}+\pi_{it}$$

Where $m_{it}^\ast$ denotes the predicted value of the endogenous variable of individual mobile phone ownership. The value of $m_{it}^\ast$ was derived automatically by estimating the first-stage equation through the IVREG2H approach as follows:

(4) $$m_{it}^\ast=\eta _{0}+\eta _{1}Z_{it}+\eta _{i}X_{it}+\lambda _{t}+\Gamma _{i}+\mu_{it}$$

where Z is the instrumental variable, which includes both the fraction of households having a mobile phone at the village level (external instrument) and heteroskedasticity-based instruments, while μ denotes the error term in the equation (4). This makes the causal claim much stronger.

3.2.3 Decomposing the direct and indirect impacts of mobile ownership

This section outlines the statistical analysis of mediation effects to assess mobile phone ownership’s direct and indirect effects on crop yields among smallholder maize farmers. Mediation occurs when the mobile phone ownership variable influences maize yields indirectly through access to extension services variable (Preacher and Hayes, Reference Preacher and Hayes2008). The direct effect is attributed solely to the farmers owning a mobile phone. For example, a farmer might use a mobile phone to get immediate market information, and weather forecasts, or consult with other farmers on best practices, leading to better decision-making and potentially higher yields (e.g., Kansiime et al., Reference Kansiime, Alawy, Allen, Subharwal, Jadhav and Parr2019; Malimi, Reference Malimi2023; Munthali et al., Reference Munthali, Leeuwis, van Paassen, Lie, Asare, van Lammeren and Schut2018; Witteveen et al., Reference Witteveen, Lie, Goris and Ingram2017). However, Lei et al. (Reference Lei, Shen, Smith and Zhou2015) point out that partial mediation is observed when both the coefficient for direct effect and the mediation are statistically significant. The indirect effect encompasses the influence of mobile phone ownership on maize yield that operates through access to extension services (Issahaku et al., Reference Issahaku, Abu and Nkegbe2017). In this pathway, owning a mobile phone helps farmers gain better access to agricultural extension services, providing valuable information and advice on improving their farming practices. This, in turn, can lead to improved maize yields. The assumption is that extension services mediate the relationship between mobile phone ownership and maize yield (mediator effect). The total effect is the sum of these two effects.

The direct, indirect, and total effects of mobile phone ownership on crop yields can be specified as:

(5.a) $$ \rm{Direct \;Effect} \;(DE)=\beta _{2}$$

(5.b) $$\rm{Indirect \;Effect} \;(IE)=\alpha _{1}\,\ast\, \beta _{1}$$

(5.c) $$\rm{Total \;Effect} \;(TE)=DE+IE=\beta _{2}+\alpha _{1}\,\ast\, \beta _{1}$$

The DE of mobile phone ownership is calculated by holding the mediator (access to extension services) constant, which stows the indirect causal mechanism. Conversely, the IE is determined by keeping the treatment (mobile phone ownership) constant, allowing us to assess the mediating impact of mobile phone ownership on maize yields (Frölich and Huber, Reference Frölich and Huber2017; Pearl, Reference Pearl2001). Following Mize et al. (Reference Mize, Doany and Long2019), we use seemingly unrelated estimation (SUEST) to retrieve the coefficients and estimate both IE and TE. To generate the indirect effects, a bootstrap with 500 repetitions was used.

In the econometric analysis, we cluster the standard errors at the household level to mitigate the potential impact of heteroskedasticity. Additionally, we apply sampling weights in the analysis to enhance estimate accuracy.

4.1 Descriptive results

Figure 4 presents a comparative analysis of the yearly trends in access to extension services and maize yields between phone owners and non-phone owners. Overall, mobile phone owners consistently showed higher proportions of access to extension services compared to their non-phone-owning counterparts. This trend holds true for government and private extension services, emphasizing the crucial role of mobile phones in facilitating exchange between farmers and extension officers. However, a significant decline in access to extension services was observed in 2020/21, largely due to the country’s heavy reliance on traditional extension approaches such as farmers’ field school, on-farm demonstrations, and farm visits, which were shown to be ineffective during the COVID-19 lockdown as public assembly restrictions and social distancing measures rendered in-person services unfeasible. Agricultural extension services had to be delivered through unconventional approaches such as mobile phones (Baffoe-Bonnie et al., Reference Baffoe-Bonnie, Martin and Mrema2021). Figure 4 also indicates a significant difference in maize yields between phone owners and non-phone owners, with phone owners consistently achieving higher yields. This suggests that mobile phones contribute positively to agricultural productivity by improving access to agricultural extension services. Additionally, Figure 4 indicates that yields did not decrease sharply despite reduced access to extension services in 2020/21, probably because the benefits from earlier extension interventions were still present. Even without prompt follow-up extension services, farmers who had previously received extension services may continue to apply the knowledge they had learned. This suggests that the benefits of extension services could be cumulative and long-lasting. Econometric models were applied to confirm the relationship between phone ownership and maize yields.

Figure 4.

Trends in access to agricultural extension services and maize yields among non-phone owners and phone owners in Tanzania.

4.2 Decomposition of the effects of mobile phone ownership on maize yields

The results from the correlated random effects model analysis and instrumental variables estimation using heteroskedasticity-based instruments (IVREG2H) are summarized in Table 2. The results for general extension are presented in columns 1-2 (CRE) and 3-4 (IVREG2H), while government extension results are shown in columns 5-6 (CRE) and 7-8 (IVREG2H), and private extension results in columns 9-10 (CRE) and 11-12 (IVREG2H). While CRE and IVREG2H estimations provide similar results, this study primarily relies on the IVREG2H model, which provides more robust estimates in the presence of potential endogeneity.

Table 2.

Correlated random effect (CRE) and IVREG2H estimates of the impact of mobile phone ownership on extension services and log of maize yields

Notes: Controls included age, gender, marital status, education, household size prime age, distance to market, distance to road, inorganic fertilizer usage, pesticide usage, improved seed variety, soil quality, temperature, precipitation, and location and year-fixed effects.

Clustered standard errors at the household level are reported in parentheses.

***p < 0.01, **p < 0.05 and p* < 0.10.

We observe a positive and significant impact of mobile phone ownership on extension services. Specifically, a smallholder farmer shifting from not owning to owning a mobile phone is 8.2 percentage points (pp) more likely to use general extension services, 3.0 pp more likely to use government extension services and 6.8 pp more likely to use private extension services. Figure 5 reports similar results, showing that owning a mobile phone significantly increases access to extension services, particularly in agricultural production and pest and disease control. These results are consistent with previous studies (Aker and Ksoll, Reference Aker and Ksoll2015; Issahaku et al., Reference Issahaku, Abu and Nkegbe2017; Khan et al., Reference Khan, Qijie, Sertse and Khan2019; Kiberiti et al., Reference Kiberiti, Sanga, Mussa, Tumbo, Mlozi and Haug2016; Witteveen et al., Reference Witteveen, Lie, Goris and Ingram2017), which have also reported a positive impact of mobile phones on access to extension services. The favorable effect of mobile phone ownership can be attributed to the efficient communication channels it offers, such as SMS, phone calls, and WhatsApp, facilitating better interactions between major actors in the agricultural sector, including smallholder farmers and extension officers (e.g., Cole and Fernando, Reference Cole and Fernando2021; Issahaku et al., Reference Issahaku, Abu and Nkegbe2017; Mtega, Reference Mtega2021). Moreover, we find that mobile phones have a more pronounced effect on accessing private services than government services. This trend underscores the private sector’s agility and innovation, potentially supported by external funding sources, which further amplifies the impact of mobile technology in agricultural extension in Tanzania (Kihoma et al., Reference Kihoma, Churi and Sanga2023; Sanga et al., Reference Sanga, Mussa, Tumbo, Mlozi, Muhichel and Haug2014). This aligns partly with Ally (Reference Ally2024) who reported that farmers in Tanzania expressed dissatisfaction with public extension services. Further, evidence across sub-Saharan Africa consistently highlights the superior quality of private extension services compared to public services, mainly due to the private sector’s effective management systems, such as well-staffed extension system with performance-based incentives (e.g., rewards and prizes), higher salaries for extension agents, adequate financial resources, and enhanced accountability among stakeholders engaged in extension activities (Kristin, Reference Kristin and Kristin2020; Moshobane and Antwi, Reference Moshobane and Antwi2022; Sylla et al., Reference Sylla, Mahama Al-Hassan, Egyir and Anim-Somuah2019; Tham-agyekum et al., Reference Tham-Agyekum, Abourden, Bakang and Juantoa2024). The pronounced impact on private extension services underscores the necessity for partnerships between government and private extension providers to enhance smallholder farmers’ access to extension services and livelihoods.

Figure 5.

CRE estimates of the impact of mobile phone ownership on diverse extension services.

In terms of the effect of mobile phone ownership on maize yields, we find a positive direct effect. Notably, maize yields increase by around 40% across general, government, and private extension models when a farmer transitions from not owning to owning a mobile phone. Although the coefficients differ slightly, the overall conclusion is that mobile phone ownership enhances maize yields. These results are consistent with previous research (Issahaku et al., Reference Issahaku, Abu and Nkegbe2017; Quandt et al., Reference Quandt, Salerno, Neff, Baird, Herrick, Terrence McCabe, Xu and Hartter2020). Interestingly, Quandt et al. (Reference Quandt, Salerno, Neff, Baird, Herrick, Terrence McCabe, Xu and Hartter2020), using data from Tanzania, provide a plausible explanation that mobile phone usage in farming reduces both costs and time investments, which can be efficiently reinvested in improving productivity. Moreover, mobile phones help smallholder farmers access a diverse range of extension sources, including more effective private extension services (e.g., Sylla et al., Reference Sylla, Mahama Al-Hassan, Egyir and Anim-Somuah2019).

The effect of access to extension services on yield is statistically significant, with farmers experiencing maize yield increases of 9% and 7.1% after gaining access to general and private extension services, respectively. This confirms the positive impacts of extension services. These results align with those of Malimi (Reference Malimi2023), who also observed the positive effects of extension access on smallholder farmers’ maize productivity in Tanzania. Similarly, when looking at the impact of extension services on smallholder farmers’ productivity in northern Ghana, Danso-Abbeam et al., (Reference Danso-Abbeam, Ehiakpor and Aidoo2018) found strong evidence that agricultural extension services led to increased productivity in maize plots. Our findings are consistent with the broader literature suggesting that smallholder farmers with access to farm inputs, including agricultural extension services, tend to be more productive. This is because extension services enhance knowledge about farming practices and greater ability to predict weather patterns, leading to more informed decision-making in farming activities (Kassem et al., Reference Kassem, Shabana, Ghoneim and Alotaibi2020; Kansiime et al., Reference Kansiime, Alawy, Allen, Subharwal, Jadhav and Parr2019; Mtega, Reference Mtega2021; Ortiz-Crespo et al., Reference Ortiz-Crespo, Steinke, Quirós, van de Gevel, Daudi, Gaspar Mgimiloko and van Etten2021). However, government extension services have limited effects, probably due to quality issues and others (Ally, Reference Ally2024).

Turning our attention to the indirect effect, a positive and significant indirect effect of mobile phone ownership on maize yields is revealed in the estimation. We find that mobile phone-induced access to general or private agricultural extension services mediates the causal relationship between mobile phones and maize yields. Specifically, mobile phone-induced access to general or private agricultural extension services accounts for only 1.3 to 2% of the total effect of mobile phone ownership on maize yields. This might be ascribed to farmers’ limited smartphone use, which restricts their access to more advanced mobile agricultural extension services (FAO and ITU, 2022). These results are consistent with the study of Issahaku et al. (Reference Issahaku, Abu and Nkegbe2017), who identified extension services as one of the key channels through which mobile phones affect farmers’ productivity in Ghana. However, following Lei et al., (Reference Lei, Shen, Smith and Zhou2015), we can conclude that extension services only partially mediate the causal relationship between mobile phone ownership and maize yields in Tanzania. Mobile phone ownership influences yields through multiple mechanisms, including agricultural extension services, adoption of modern agricultural technologies, market participation, mobile money, e-vouchers, among others (Issahaku et al., Reference Issahaku, Abu and Nkegbe2017; Tabetando et al., Reference Tabetando, Matsumoto and Fani2022). Consequently, when considering the total effect of mobile phone ownership, significant increases in maize yields are observed, with effect sizes of around 50% among smallholder farmers using general, government, and private agricultural extension services. These findings underscore the crucial role of mobile phone ownership in enhancing agricultural productivity and emphasize the benefits of aligning with effective extension services.

4.3 Results with heterogeneity by gender

Table 3 presents the effect of mobile phone ownership on male-headed and female-headed farm households estimated using IVREG2H. The results for general extension are summarized in columns 1 – 4, government extension in columns 5 – 8, and private extension in columns 9 – 12.

Table 3.

Effect of mobile phone ownership on extension and log of maize yields across gender

Notes: Controls are the same as in Table 2.

Clustered standard errors at the household level are reported in parentheses.

***p < 0.01, **p < 0.05 and *p < 0.10

The analysis reveals that transitioning from not owning to owning a mobile phone significantly facilitates access to general and government extension services for both male and female-headed households, suggesting that mobile phones offer an opportunity to bridge the knowledge gaps between male and female farmers. Further comparison between male and female-headed households indicates that the coefficient of the mobile phone ownership variable in the government extension model is greater for female-headed households, suggesting a higher likelihood of using government extension services. This might stem from the Tanzanian government’s adoption of the Beijing Platform for Action and the 2030 Agenda for Sustainable Development, which prioritize strengthening gender equality policies and enhancing rural and marginalized women’s access to agricultural extension servicesFootnote ² . However, the analysis also demonstrates a positive and significant impact of mobile phone ownership on the male-headed farm households’ likelihood of using private extension services, but no significant impact for female-headed farm households. This evidence supports the assertion that mobile phone ownership has a stronger effect on improving male-headed farm households’ access to private extension services in comparison to their female counterparts, revealing a disparity in access to (digital) agricultural extension services between male and female smallholder farmers. These results are consistent with Krell et al.’s (Reference Krell, Giroux, Guido, Hannah, Lopus, Caylor and Evans2021) findings, indicating that female farmers in Kenya are less likely to use mobile mobile phone-based agricultural services compared to their male counterparts. This digital divide has been attributed to a number of factors, including female smallholder farmers’ lower awareness and knowledge about how to use mobile services (Krell et al., Reference Krell, Giroux, Guido, Hannah, Lopus, Caylor and Evans2021), less mobile phone ownership, and societal gender norms (McCormack, Reference McCormack2018). The gender-based disparity in access to extension services can also be attributed to the absence or ineffectiveness of gender-sensitive measures in the region (McCormack, Reference McCormack2018; Meinzen-Dick et al., Reference Meinzen-Dick, Quisumbing, Behrman, Biermayr-Jenzano, Wilde, Noordeloos, Ragasa and Beintema2011).

In terms of the effect on maize yields, male-headed farm households shifting from not owning to owning a mobile phone demonstrate significantly higher productivity in maize yields when compared to their female counterparts. Gaining access to extension services has the same effect as owning a mobile phone. Additionally, gaining access to general, government, and private extension services assumes a mediating role in the causal link between mobile phones and maize yields in male-headed households. However, this mediating effect is not observed in the context of the relationship between mobile phones and maize yields in female-headed households. The main conclusion from these results is that significant heterogeneity across gender in the impact of mobile phone ownership on smallholder farmers’ outcomes exists. Sanga (Reference Sanga2018) provides a plausible explanation for the limited impact of mobile phones on female smallholder farmers’ outcomes, pointing out that information transmitted through phones, which are mostly owned by males might not be shared or when transmitted to women’s phones, may lead to conflict. This is because cultural norms in various parts of Tanzania prohibit women from exchanging SMS with strangers (McCormack, Reference McCormack2018; Sanga, Reference Sanga2018). Restricting female smallholder farmers from freely exchanging ideas and information with unfamiliar extension officers could hinder effective communication. Consequently, the information deficit could have detrimental effects on female smallholder farmers’ agricultural output.