Large language model-based artificial intelligence for improving personal finance in higher technical institutes

This dataset contains individual pretest and posttest measurements from 77 students of a technical higher education institute in the La Libertad region, Peru, who participated in a single-group quasi-experimental study on the use of a locally executed large language model for personal finance management, conducted during the 2025-II academic term. Measurements were obtained through the CFPE-IMFP questionnaire, a 39-item instrument organised in three sections: financial knowledge (K, 15 multiple-choice items), financial behaviour (C, 16 frequency-scale items) and financial outcome (R, 8 indicators of economic situation). Each section is transformed to a standardised 0-100 scale, and the Personal Finance Improvement Index is computed as IMFP = 0.30K + 0.40C + 0.30R, weighting behaviour most heavily given its mediating role between declarative knowledge and observable economic outcomes. The posttest was administered after 7 to 14 days of system use. The dataset includes raw item-level responses for both measurement points, the derived standardised scores, and the scoring keys required to recompute those scores from scratch. Participants are identified by sequential codes (EST-001 to EST-077) that match across all files, enabling the paired analyses reported in the associated article. No personal identifiers are included and all participants provided written informed consent. Note on scoring: the maximum score for section R is 20, not 24, because four of the eight indicators are scored 0-2 rather than 0-3. Using 24 as the denominator will not reproduce the published results. These data come from a design without a control condition, with a short intervention period and self-reported measures. Effect sizes should be read as preliminary evidence of feasibility and acceptability rather than as unbiased estimates of a causal effect.本数据集涵盖秘鲁拉利伯塔德地区某技术类高等教育院校77名学生的前测与后测个体测量数据，这些学生参与了一项于2025-II学期开展的单组准实验研究，主题为使用本地化运行的大语言模型（Large Language Model）进行个人财务管理。 测量采用CFPE-IMFP问卷完成，该问卷共包含39个条目，划分为三个维度：金融知识（K，15道单项选择题）、金融行为（C，16道频率量表条目）与金融结果（R，8项经济状况指标）。各维度得分均被转换为标准化0-100分制，个人理财改善指数（IMFP）的计算公式为：IMFP = 0.30K + 0.40C + 0.30R。鉴于行为在陈述性知识与可观测经济结果间发挥中介作用，因此对行为维度赋予了最高权重。后测于系统使用7至14天后施测。 本数据集包含两次测量的原始条目级响应数据、衍生的标准化得分，以及可从零开始重新计算上述得分的评分密钥。参与者以连续编号（EST-001至EST-077）进行标识，所有文件中的编号均保持一致，可支撑关联论文中报告的配对分析。本数据集未包含任何个人识别信息，且所有参与者均已签署书面知情同意书。 评分说明：R维度的最高得分为20而非24，原因是8项指标中有4项采用0-2分制计分，而非0-3分制。若以24作为分母进行计算，将无法复现已发表的研究结果。 本研究数据采用无对照组的研究设计，干预周期较短且测量方式为自我报告。效应量应被视为可行性与可接受性的初步佐证，而非因果效应的无偏估计值。