The study was conducted in Bonthe District, located in the Southern Province of Sierra Leone. The district is characterized by extensive inland valley swamps, mangrove ecosystems, and rain-fed lowland rice production systems, making it one of the country’s major rice-producing areas. Agriculture is the principal livelihood activity, with rice serving as the dominant staple crop. The district experiences a tropical monsoon climate with distinct wet and dry seasons, annual rainfall ranging from approximately 2500 to 3500 mm, and temperatures averaging between 24˚C and 30˚C. Five rice-producing communities Torma Bum (Torma): 7.329978˚N, −12.242906˚W, Borlleh: 7.502078˚N, −12.115161˚W, Madina: 8.166667˚N, −12.066667˚W, Ngepehun: 7.375013˚N, −12.138402˚W, and Sogballeh: 7.329978˚N, −12.242906˚W, were purposively selected because of their high rice production potential and the presence of extension interventions promoting modern rice technologies.

The study employed a quantitative cross-sectional survey design. The design was appropriate because it enabled the collection of data from farmers at a single point in time to assess the factors influencing the adoption of modern rice production technologies. The approach facilitated the examination of relationships between technology adoption and extension services, biophysical conditions, and farmers’ perceptions.

The study employed a multistage sampling procedure to select respondents from rice-producing communities in Bonthe District, Sierra Leone. In the first stage, Bonthe District was purposively selected because of its significant contribution to rice production and the ongoing promotion of modern rice technologies. In the second stage, five major rice-producing sections, Torma, Koimato, Gbondubu, Fikie, and Yagbeh, represented by the communities of Torma, Borlleh, Madina, Sogballeh, and Ngepehun, respectively (Table 1), were purposively selected based on their high rice production potential and active involvement in rice farming. The lists of registered rice farmers were obtained from the district agricultural extension offices, from which individual respondents were selected using simple random sampling. The sampling frame consisted of 1720 registered rice farmers identified through the Bonthe District agricultural extension offices. Because an independent estimate of the total number of registered and unregistered rice farmers in the district was unavailable, the proportion of all rice farmers represented by the register could not be established. Accordingly, the findings and inferences of this study are limited primarily to rice farmers represented in the registered-farmer sampling frame and should not be generalized to unregistered farmers without further evidence. The primary outcome was binary technology adoption. Modern rice production technologies included improved/climate-resilient varieties, irrigation, mechanisation, soil-fertility management, improved agronomic practices, and integrated pest management. Farmers were classified as adopters (1) if they had used at least one of these technologies during the most recent rice production season and non-adopters (0) if they had used none. Thus, adoption represented the use of any one or more technologies, rather than requiring simultaneous adoption of the complete technology package. The sample size was determined using a statistical formula by Yamane [30] for a finite population:

n = N/(1 + N(e²))

where:

n = required sample size

N = total population size

e = margin of error (0.05)

where n is the required sample size, N is the total population (1720 registered rice farmers), and e is the acceptable margin of error (0.05).

Substituting the values:

N = 1720 farmers

e = 0.05

Applying the formula:

n = 1720/(1 + 1720(0.05²))

n = 1720/(1 + 1720(0.0025))

n = 1720/(1 + 4.3)

n = 1720/5.3

n = 324.53 ≈ 324 respondents

Table 1. Proportionate distribution of sample size.

Accordingly, a sample of 325 rice farmers was selected for the study. The sample was proportionately allocated according to the number of registered rice farmers in each section to ensure fair representation. This sampling approach ensured that every eligible farmer had an equal chance of being selected while maintaining proportional representation across the study sections and respective communities.

Primary data was collected using a structured interviewer-administered questionnaire. The questionnaire was administered through face-to-face interviews because many respondents had limited formal education. The questionnaire consisted of five sections: socio-demographic characteristics; access to agricultural extension services; biophysical characteristics of farms; farmers’ perceptions of benefits, risks, and constraints associated with modern rice production technologies; and adoption status of modern rice production technologies. The questionnaire was developed from previous adoption studies and adapted to the Sierra Leonean context. It was pre-tested among rice farmers outside the study communities, and necessary revisions were made to improve clarity and validity.

Data was coded, entered, cleaned, and analysed using R statistical software (Version 4.x). Descriptive statistics, including frequencies and percentages, were used to summarize respondents’ socio-demographic characteristics and perceptions. Chi-square (χ²) tests of independence were used to determine whether extension service indicators, biophysical conditions, and perception variables differed significantly across the study sections and communities. Binary logistic regression analysis was employed to identify determinants of modern rice production technology adoption. Odds Ratios (OR), Adjusted Odds Ratios (AOR), Wald statistics, 95% confidence intervals, and p-values were reported. Statistical significance was assessed at the 5% level (p < 0.05). Model performance was evaluated using the likelihood ratio chi-square statistic and McFadden’s pseudo-R². Because some contingency tables contained sparse cells, exact or Monte Carlo simulated tests were used where the assumptions of the asymptotic Pearson chi-square test were not adequately satisfied. Where appropriate, categories with extremely small frequencies were combined only when substantively defensible. Firth’s penalized likelihood method was used to estimate the logistic regression models to reduce small-sample and separation-related bias. Binary predictors were coded 1 = condition present/Yes and 0 = condition absent/No, with the latter serving as the reference category. Section was included as an adjustment variable in the adjusted models.

Key Informant Interviews (KIIs) were conducted among purposively selected respondents who had great knowledge about rice production and dissemination of new technologies in the study area to collect qualitative data for triangulation purposes. The respondents for this data included agricultural extension workers, rice farmers’ association leaders, community leaders, agro-input dealers, non-governmental organization (NGO) representatives and other relevant personnel from agricultural institutions. The interview was done using a semi-structured guide to examine the existence of various resources and factors that could affect adoption of modern rice production technologies such as improved seeds, credit facilities, extension services, agro-inputs, training, markets and institutions. The interview was face-to-face and was recorded with the permission of respondents and accompanied by field notes.

Content validity was ensured through expert review by agricultural extension specialists, rice production experts, and university researchers. A pilot survey was conducted before the main data collection to assess the suitability of the questionnaire for accurately capturing the intended information. Necessary modifications were made based on the pilot findings. Internal consistency of the perception items was assessed using Cronbach’s alpha coefficient. A reliability coefficient of at least 0.70 was considered acceptable for inclusion in the final analysis.

Ethical approval for the study was obtained from the relevant institutional authorities before data collection commenced. Permission was also obtained from district agricultural authorities, and community leaders/chiefs. Participation was voluntary, and informed consent was obtained from every respondent before the interviews. Respondents were assured that their responses would remain confidential and anonymous and would be used solely for academic purposes. Personal identifiers were removed during data processing, and respondents retained the right to withdraw from the study at any stage without penalty.

The demographic and socio-economic characteristics of respondents in Bonthe District reveal that rice farming is largely undertaken by middle-aged, experienced smallholder farmers who depend heavily on agriculture for their livelihoods. Male farmers constituted the majority of respondents (62.5%), with the highest proportion recorded in Ngepehun (78.9%), while Borlleh had the highest proportion of female farmers (53.3%), indicating variations in gender participation across communities. The age distribution was dominated by farmers aged 43 - 47 years (20.6%), followed by those aged 38 - 42 years (18.5%), suggesting that rice farming is primarily practiced by economically active individuals. In contrast, the youngest age group (18 - 22 years) accounted for only 2.2% of respondents (Table 2), reflecting limited youth participation in rice production.

Educational attainment was generally low, as nearly half of the respondents (49.5%) had no formal education, with Borlleh recording the highest proportion (68.3%). Conversely, tertiary education was the least represented category (1.9%), highlighting the generally low literacy level among rice farmers, which may constrain access to agricultural information and the adoption of improved technologies. Farming experience was relatively high, with the largest proportion of farmers having 11 - 15 years of experience (23.1%), followed by 16 - 20 years (20.9%). Ngepehun recorded the highest proportion of farmers within the 11 - 15-year category (42.1%), while only 2.2% had 41 - 45 years of farming experience, indicating that most farmers possess moderate to substantial practical knowledge of rice cultivation. Farm characteristics further demonstrate the dominance of smallholder agriculture in the district. The most common farm size was 2 acres (23.7%), followed by 4 acres (21.2%), with Borlleh recording the highest proportion of 2-acre farms (53.3%) and Madina the highest proportion of 4-acre farms (40.4%). Very large farms were rare, as farm sizes of 11, 12, 20, 22, and 25 acres each accounted for only 0.3% of respondents. Rice farming was the principal source of household income for the vast majority of respondents (88.3%), reaching 100% in Madina, underscoring the central role of rice production in household livelihoods and food security. In contrast, only 11.7% depended primarily on other crops, while none reported livestock or non-farm activities as their major income source.

Membership in farmer organizations was nearly evenly distributed, with 49.5% belonging to farmer groups and 50.5% not affiliated with any organization. Borlleh exhibited the highest membership rate (98.3%), suggesting stronger opportunities for collective action, access to extension services, training, and agricultural inputs. Conversely, Sogballeh recorded the lowest membership level (12.3%), with most farmers (87.7%) operating outside formal farmer organizations. The findings indicate that rice farming in Bonthe District is dominated by middle-aged, moderately experienced farmers with low levels of formal education, operating predominantly small-scale farms and relying heavily on rice cultivation as their main livelihood source. The observed differences in education, organizational membership, and farm resources across communities may have important implications for agricultural productivity and the adoption of improved rice production technologies.

Table 2. Percentage distribution of respondents’ demographic characteristics.

The chi-square analysis showed that all extension service indicators were significantly associated with section location (p < 0.001), indicating substantial differences in extension service delivery across the study sections (Table 3) [31][32]. The strongest variation was observed in the frequency of extension agent visits (χ² = 470.2, df = 12), where Madina recorded the highest proportion of farmers receiving very regular visits (89.5%), while Sogballeh recorded the highest proportion reporting no visits (98.2%). Similarly, access to agricultural extension services (χ² = 365.9, df = 12) and ease of access to extension services (χ² = 314.8, df = 12) showed large disparities, with Borlleh and Madina recording the highest access levels, whereas Sogballeh consistently reported the lowest access (Table 3) [31][33].

Table 3. Role of extension services in adoption of modern rice technology.

Note. Values are n (%). Chi-square statistics test association between each extension indicator and section; empty rows/columns were dropped before testing. All tests were significant at p < 0.001.

Other indicators, including follow-up support after training (χ² = 281.9, df = 12), information on modern rice production technologies (χ² = 274.9, df = 12), and overall satisfaction with extension services (χ² = 267.6, df = 12) (Table 3), also revealed significant sectional differences [34]. Madina generally recorded the highest levels of detailed information, regular support, and positive perceptions of extension services, while Sogballeh recorded the highest proportions of respondents lacking information, support, and satisfaction. Likewise, influence of extension on adoption (χ² = 222.7, df = 16), perceived clarity of information (χ² = 216.7, df = 12), and training attendance (χ² = 209.2, df = 16) demonstrated that farmers in Madina and Borlleh benefited more from extension interventions than those in Sogballeh [35][36].

The lowest chi-square value was recorded for the most effective extension method (χ² = 121.6, df = 16, p < 0.001), suggesting comparatively smaller differences among sections regarding preferred communication approaches [37]. Farm visits were identified as the most effective extension method overall (54.8%), whereas radio-based extension was the least preferred (0.9%). Hence, Madina and Borlleh consistently recorded the highest positive extension indicators, while Sogballeh recorded the lowest values across most variables. These findings indicate that stronger extension service delivery is closely associated with greater farmer awareness, participation, satisfaction, and adoption of modern rice [38][39].

In Table 4, it can be seen that the logistic regression model demonstrated a good overall fit, with the section-adjusted model explaining 44.2% of the variation in modern rice production technology adoption (McFadden R² = 0.442) and a significant likelihood ratio statistic (LR χ² = 195.1, df = 13), indicating that extension service variables collectively influenced farmers’ adoption decisions [40]. Among the extension service variables, attendance at training and demonstration programmes was the strongest positive determinant of adoption, with farmers who participated being over 14 times more likely to adopt modern rice technologies (OR = 14.35, Adjusted OR = 7.01, p < 0.001) [41]. This variable also recorded the

Table 4. Logistic regression of adoption on extension services.

Model fit: unadjusted LR chi-square = 133.8 (df = 9), McFadden R-square = 0.303; section-adjusted LR chi-square = 195.1 (df = 13), McFadden R-square = 0.442. n = 325; adopters = 182. Note. B = penalized log-odds coefficient; OR = odds ratio; Wald = squared z-statistic. Predictors are binary (1 = condition present). Adj. OR (p) is the odds ratio from the section-adjusted model. p < 0.05, p < 0.01, p < 0.001.

largest Wald statistic (33.23), highlighting the critical role of practical training and demonstrations in promoting technology adoption [42]. Regular extension agent visits also showed a relatively strong positive effect (OR = 5.47, Adjusted OR = 6.28), although the relationship was only marginally significant (p = 0.056; adjusted p = 0.053).

Conversely, strong extension influence exhibited the strongest negative association with adoption (OR = 0.07, Adjusted OR = 0.11, p < 0.001), making it the least favourable predictor in the model [43]. Other variables, including easy access to extension services (OR = 1.26, p = 0.755), information clarity (OR = 0.86, p = 0.739), service satisfaction (OR = 0.73, p = 0.607), follow-up support (OR = 1.54, p = 0.451), useful information received (OR = 2.49, p = 0.116), and access to extension services (OR = 0.37, p = 0.227) were not statistically significant predictors of adoption [44][45].

Table 5 indicates that dominant soil type showed the strongest association with the study sections (χ² = 159.2, df = 12, p < 0.001), making it the most significant biophysical factor differentiating adoption environments [46]. The loam soil was the predominant soil type (73.2%), particularly in Madina (100.0%), Borlleh (96.7%), and Sogballeh (98.2%), while mixed soil types were concentrated in Ngepehun (77.2%). This suggests substantial spatial variation in soil characteristics that may influence the suitability and adoption of modern rice technologies [47][48]. Regarding farming systems, the dominant rice cultivation system recorded the lowest but still significant association among the biophysical variables (χ² = 28.6, df = 12, p = 0.004). Across all sections, rain-fed lowland farming overwhelmingly dominated (93.5%), indicating that rice production largely depends on rain-fed lowland conditions, whereas fully irrigated systems (0.9%) and both irrigated and rain-fed systems (0.6%) were rarely practiced [49][50].

Table 5. Chi-square tests for biophysical conditions.

Note. Values are n (%). Chi-square statistics test association between each biophysical indicator and section.

Other biophysical conditions also showed highly significant differences across sections (p < 0.001). Very high soil fertility was the most common fertility category (56.6%), especially in Sogballeh (89.5%) and Borlleh (78.3%). Occasional flooding was the predominant flooding pattern (53.8%), while occasional drought stress was the most frequently reported drought condition (48.3%). In contrast, frequent drought (0.6%), never experiencing flooding (0.9%), and clay soils (0.6%) were the least reported conditions. These findings demonstrate that significant differences in soil characteristics, fertility, flooding, and drought conditions exist across the sections and are likely to influence the adoption of modern rice technologies [51][52].

Table 6 shows that drought exposure was the strongest positive biophysical predictor of modern rice production technology adoption, with farmers experiencing drought being three times more likely to adopt modern technologies (OR = 3.00, Wald = 14.45, p < 0.001). This relationship remained significant after adjustment (Adjusted OR = 3.19, p = 0.002), suggesting that drought serves as a major driver encouraging farmers to adopt technologies that improve resilience and productivity [53][54].

In contrast, loam soil was the strongest negative predictor in the unadjusted model (OR = 0.37, Wald = 9.06, p = 0.003), indicating that farmers cultivating loam soils were significantly less likely to adopt modern rice production technologies [55]. However, this association disappeared after controlling for section-level effects (Adjusted OR = 1.05, p = 0.915), implying that the initial effect was influenced by geographical differences rather than soil type alone. The least influential variable was irrigated cultivation, which showed no significant association with adoption (OR = 0.39, Wald = 0.78, p = 0.376; Adjusted OR = 0.30, p = 0.267). Similarly, high soil fertility and flood exposure were not significant predictors in the unadjusted model, although flood exposure became significant after adjustment (Adjusted OR = 0.36, p = 0.018), suggesting lower adoption among farmers exposed to flooding when section-level differences were considered [56].

Table 6. Logistic regression of adoption on biophysical conditions.

Model fit: unadjusted LR chi-square = 36.9 (df = 5), McFadden R-square = 0.084; section-adjusted LR chi-square = 156.7 (df = 9), McFadden R-square = 0.355. n = 325; adopters = 182. Note. B = penalized log-odds coefficient; OR = odds ratio; Wald = squared z-statistic. Predictors are binary (1 = condition present). Adj. OR (p) is the odds ratio from the section-adjusted model. p < 0.05, p < 0.01, p < 0.001.

Table 7 indicates that farmers generally held positive perceptions of modern rice production technologies, particularly regarding their ability to increase rice yield, improve grain quality, and enhance farm income. The greatest perceived benefit was increased rice yield, with 69.8% of respondents strongly agreeing, followed by improved rice quality (62.8%) and increased farm income (58.5%) [57][58]. Ngepehun and Sogballeh consistently recorded the highest levels of strong agreement for these benefits, while Madina showed the lowest levels of strong agreement, although most respondents there still agreed that the technologies were beneficial. Among the perceived risks, high financial risk was the most significant concern, with 38.2% strongly agreeing that adopting modern rice production technologies requires substantial financial investment [59]. In contrast, crop failure was the least perceived risk, as 48.3% of respondents disagreed that modern technologies increase the likelihood of crop failure, while environmental concerns were also relatively low, with 40.3% disagreeing that the technologies are environmentally harmful [60].

Regarding adoption constraints, limited access to credit emerged as the greatest barrier (56.6% strongly agree), followed by inadequate extension support (50.2%), high input costs (47.7%), and limited technical knowledge (47.4%) [61]. Ngepehun consistently reported the highest perceptions of these financial and technical barriers, whereas Madina recorded the lowest. Overall, the findings suggest that although farmers strongly recognize the productivity and economic advantages of modern rice production technologies, improving access to affordable credit, reducing input costs, strengthening extension services, and enhancing farmer training remain critical for increasing technology adoption across Bonthe District.

Table 7. Farmers’ perceptions of benefits, risks and constraints.

Table 8 shows that perceived environmental harm was the strongest positive perception influencing the adoption of modern rice production technologies. Farmers who perceived environmental benefits were almost twice as likely to adopt modern technologies (OR = 1.92, Wald = 30.58, p < 0.001), and this relationship remained highly significant after adjustment (Adjusted OR = 1.69, p < 0.001). This finding indicates that environmental considerations play a major role in motivating technology adoption [62][63].

Among the negative predictors, perceived income increase exhibited the strongest significant reduction in adoption (OR = 0.52, Wald = 10.17, p = 0.001), and this effect persisted after adjustment (Adjusted OR = 0.52, p = 0.007). Similarly, yield benefit (OR = 0.46, p = 0.022), crop-failure risk (OR = 0.75, p = 0.008), credit-access barrier (OR = 0.63, p = 0.020), and extension-support barrier (OR = 0.57, p = 0.002) were significant in the unadjusted model, although some lost significance after controlling for section-level effects. The least influential variable was high input cost, which showed virtually no association with adoption (OR = 0.97, Wald = 0.07, p = 0.797), followed by grain quality benefit (OR = 0.82, p = 0.448) and technical-knowledge barrier (OR = 1.20, p = 0.289), all of which were statistically non-significant.

Table 8. Logistic regression of farmers’ perceptions of benefits, risks, and constraints.

Model fit: unadjusted LR chi-square = 177.5 (df = 10), McFadden R-square = 0.402; section-adjusted LR chi-square = 250.3 (df = 14), McFadden R-square = 0.567. n = 325; adopters = 182. Note. B = penalized log-odds coefficient; OR = odds ratio; Wald = squared z statistic. Perception items are scored 1 (strongly disagree) to 6 (strongly agree); OR expresses the change in odds per one-point increase. Adj. OR (p) is from the section-adjusted model. p < 0.05, p < 0.01, p < 0.001.