我國工作生活品質內涵的發展、題項縮減及初步效度檢驗：機器學習取向

Development, Item Abbreviation, and Initial Validation of Quality of Work Life in Taiwan: A Machine Learning Approach

本研究考量個人與環境的互動關係，立基於需求滿足的觀點，旨在發展我國企業組織工作生活品質的架構內涵與相應的測量工具，希冀能捕捉現今員工對於工作生活品質的期待與需求。首先，透過文獻評析掌握主體架構，再透過歸納法確認主要向度與子面向內容，發展初步的量表題目。其次，採用發展組與驗證組樣本分別進行其試題分析與心理計量特徵檢驗，以確認工作生活品質量表的信、效度。最後，為了增進工作生活品質量表的實徵應用價值，引入機器學習中的基因演算法進行自動化量表題數縮減程序，以發展簡版工作生活品質量表，並檢驗其跨樣本效度。本研究發現，工作生活品質包含了五大需求向度，分別是：休閒健康、經濟保障、同事合作、主管領導、自我實現，並據以發展具有良好心理計量特徵的初步量表。同時，透過基因演算法，自動化發展能反應與原量表相似的良好信度與建構效度、與原量表具有高輻合效度、且表現出與原量表相似的效標關聯效度之簡版測量工具。

This study explores the interactive relationship between individuals and their environment based on the needs satisfaction approach. The objective of the study was to develop a conceptual framework and corresponding measurement tools to assess the quality of work life among employees in Taiwanese organizations. The study aimed to capture today’s employees’expectations and needs for quality of work life. The approach taken involved three steps. First, the main structure was understood through a literature review, and the primary dimensions and content considerations were identified using inductive methods to develop preliminary scale items. Second, a sample of development and validation groups was employed to conduct item analysis and psychometric characteristic testing to confirm the reliability and validity of the quality of work life scale. Finally, to enhance the empirical application value of the scale, a genetic algorithm from machine learning was introduced to automatically reduce scale items. This was undertaken to develop a concise version of the quality of work life scale. Its cross-sample validity was also examined. This study found that the quality of work life encompasses five major dimensions of needs: leisure and health, economic security, coworker cooperation, supervisor leadership, and self-actualization. Based on these findings, a preliminary scale with sound psychometric characteristics was developed. Using a genetic algorithm, an automated process was developed to create a concise measurement tool that demonstrated robust reliability and construct validity, similar to the original scale. The abbreviated tool also exhibited high convergent validity with the original scale and similar criterion-related validity.