深度學習在各個領域展現出優異的效能，但其模型往往過於複雜且缺乏可解
釋性，本研究旨在了解結合了深度學習的彈性與線性迴歸的可解釋性的半結構化深度迴歸模型，以協助在實證研究中取得更準確、穩健的推論。
本研究透過模擬實驗，探討了半結構化深度迴歸模型在不同資料結構和模型
配置下的效果估計和理論測試表現，研究結果顯示，相較於傳統線性迴歸模型，結合線性迴歸、深度學習與正交化技巧的半結構化深度迴歸模型在估計線性關係上更具優勢，不管是在處理複雜的交互作用關係、內生性或殘差異質性問題上皆有改善，應用在多層次資料上亦有良好的估計表現。然而，也發現了模型在處理單一變數線性和非線性加總後的關係時，仍可能出現係數估計偏誤。本研究為半結構化深度迴歸模型在實證研究的應用上提供了實際的估計與測試案例，有助於學者了解其優缺點並判斷合適的使用情境。

Deep learning has demonstrated remarkable performance in various domains, but its models are often overly complex and lack interpretability. This study aims to understand
semi-structured deep regression models, which combine the flexibility of deep learning with the interpretability of linear regression, to assist in achieving more accurate and
robust inference in empirical research.
Through simulation experiments, this study investigates the performance of semistructured deep regression models in effect estimation and theory testing under different data structures and model configurations. The results show that compared to traditional linear regression models, semi-structured deep regression models that integrate linear regression, deep learning, and orthogonalization techniques have advantages in estimating linear relationships, particularly when dealing with complex interaction effects, endogeneity, and heteroscedasticity. The model also performs well when applied to multilevel data. However, it is also found that the model may still exhibit coefficient estimation bias when dealing with relationships involving the summation of linear and non-linear terms for a single variable. This study provides practical estimation and testing cases for applying semi-structured deep regression models in empirical research, helping scholars understand their strengths and weaknesses and determine appropriate usage scenarios.

Antonakis, J., Bastardoz, N., & Rönkkö, M. (2021). On ignoring the random effects assumption in multilevel models: Review, critique, and recommendations. Organizational Research Methods, 24(2), 443–483.
Chou, Y.-C., Chuang, H. H.-C., Chou, P., & Oliva, R. (2023). Supervised machine learning for theory building and testing: Opportunities in operations management. Journal of Operations Management, 69(4), 643–675.
Choudhury, P., Allen, R. T., & Endres, M. G. (2021). Machine learning for pattern discovery in management research. Strategic Management Journal, 42(1), 30–57.
Dorie, V., Perrett, G., Hill, J. L., & Goodrich, B. (2022). Stan and bart for causal inference: Estimating heterogeneous treatment effects using the power of stan and the flexibility of machine learning. Entropy, 24(12), 1782.
Enders, C. K., & Tofighi, D. (2007). Centering predictor variables in cross-sectional multilevel models: A new look at an old issue. Psychological methods, 12(2), 121.
Mundlak, Y. (1978). On the pooling of time series and cross section data. Econometrica: journal of the Econometric Society, 69–85.
Rudin, C. (2019). Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nature machine intelligence, 1(5), 206–215.
Rügamer, D., Kolb, C., & Klein, N. (2024). Semi-structured distributional regression. The American Statistician, 78(1), 88–99.
Shmueli, G. (2010). To explain or to predict? Statistical Science, 25(3), 289–310.
Shrestha, Y. R., He, V. F., Puranam, P., & von Krogh, G. (2021). Algorithm supported induction for building theory: How can we use prediction models to theorize? Organization Science, 32(3), 856–880.
Sun, L., Lyu, G., Yu, Y., & Teo, C.-P. (2020). Fulfillment by amazon versus fulfillment by seller: An interpretable risk-adjusted fulfillment model. Naval Research Logistics (NRL), 67(8), 627–645.