An exploratory classification-tree analysis of crop residue allocation trade-offs in crop-livestock systems in Malawi

Background Crop residues are a critical resource in crop-livestock systems, serving competing functions as livestock feed and soil-management inputs. However, limited residue availability often creates trade-offs between supporting livestock productivity and maintaining soil fertility, making allocation decisions a persistent challenge in integrated farming systems. Understanding the factors that influence these decisions is important for promoting sustainable intensification and improving resource-use efficiency. Therefore, this study sought to identify the behavioral, socioeconomic, and human-capital factors associated with crop-residue allocation decisions among three participant groups engaged in integrated crop-livestock systems in Malawi. Groups A and B, comprising younger highly educated students and graduates/community youth with mixed livelihood activities, respectively, participated in a structured experiential-learning intervention and served as treatment groups, while Group C, consisting predominantly of older farmers, served as the comparison group. Methods An exploratory Classification and Regression Tree (CART) analysis was conducted using nine group-year observations derived from three investor groups observed over three production cycles. Residue allocation was classified into two categories: livestock feed and land management. Candidate predictor variables included risk tolerance, group income, education level, occupation, marital status, livestock ownership, and exposure to crop-livestock systems (CLS) training. Model performance was evaluated using confusion matrices, classification accuracy, precision, recall, F1-scores, and cross-validation procedures. Results The CART model identified risk tolerance, group income, and education level as the principal factors associated with residue-allocation decisions, with risk tolerance emerging as the root node and strongest discriminator among candidate predictors. Of the nine group-year observations, 55.6% were allocated to land management and 44.4% to livestock feed. Group A allocated residue to land management during the first two production cycles before shifting to livestock feed in the third year, Group B consistently allocated residues to livestock feed across all 3 years, and Group C consistently allocated residues to land management. These patterns suggest that differences in residue-allocation behavior were associated with variation in behavioral, economic, and educational characteristics across the groups. The model achieved perfect classification of the training data and a cross-validated accuracy of 77.8%. Class-specific performance metrics indicated comparable classification performance across allocation categories, with precision, recall, and F1-scores of 0.80 for land management and 0.75 for livestock feed, and macro-average and weighted-average scores of 0.78. Occupation, marital status, livestock ownership, and CLS training were not retained in the final tree structure. Conclusion The findings suggest that residue-allocation decisions within crop-livestock systems are associated primarily with behavioral, economic, and human-capital characteristics, particularly risk tolerance, income, and education. Differences observed among Groups A, B, and C appear to reflect variation in these characteristics rather than group membership itself. However, given the small sample size and exploratory nature of the analysis, the identified decision pathways should be interpreted as indicative patterns rather than robust predictive relationships. Future research using larger datasets and direct measures of learning and behavioral change is needed to validate these findings and strengthen understanding of residue-allocation decision-making in integrated crop-livestock systems.