"background": "Accurately measuring the adoption rates of agricultural innovations by smallholder farmers is critical for evaluating development interventions and informing policy. In Ethiopia, where smallholder systems dominate, methodological rigour in forecasting adoption remains a significant challenge, with a proliferation of time-series models applied without systematic evaluation of their comparative efficacy. ", "purpose and objectives": "This systematic review aims to critically evaluate the methodological approaches, specifically time-series forecasting models, used to measure and predict technology adoption rates within Ethiopian smallholder farming systems. It seeks to assess model specifications, data requirements, and predictive performance. ", "methodology": "A systematic search of peer-reviewed literature and grey sources was conducted following the PRISMA framework. Studies employing quantitative time-series models to forecast adoption rates were included. Data were extracted on model types, covariates, validation techniques, and performance metrics. A quality appraisal of methodological rigour was performed. ", "findings": "The review identified a predominant reliance on autoregressive integrated moving average (ARIMA) models, often specified as yt = \ + \1 y{t-1 + \ + \ yt-p + \1 -1 + \ + \ -q + \, which accounted for approximately 60% of the models reviewed. A key finding is that models incorporating climate covariates and market access variables demonstrated superior predictive accuracy, with forecast errors reduced by an estimated 15-20% compared to baseline models, though confidence intervals for adoption projections frequently exceeded ±10 percentage points. ", "conclusion": "Current methodological applications are heterogeneous and often lack standardised validation, leading to high uncertainty in adoption rate forecasts. While advanced models integrating socio-ecological drivers show promise, their implementation is inconsistent, limiting the reliability of evidence for policy. ", "recommendations": "Future research should prioritise hybrid modelling approaches that combine classical time-series methods with machine learning techniques, mandate the reporting of robust standard errors for all forecasts
Mekonnen et al. (Fri,) studied this question.