運用DEMATEL 技術改善SEM—以網路廣告效果為例

Abstract

因果分析對決策制定的效率有非常大的影響。近年來，學者通常採用SEM建立因果模式。然而，SEM 經常被誤用，在沒有理論基礎下，以資料驅使模式修正，甚至據以理論推導。本研究以決策實驗室分析法(Decision Making Trial and Evaluation Laboratory , DEMATEL)技術修正SEM。實證結果發現，透過DEMATEL方法，可以在合理的基礎下找到SEM模式的修正方向，避免上述沒有理論基礎關係存在，讓資料來決定理論，以資料來驅使(data-driven)模式修正的錯誤。研究者不會一昧地追求適配，不斷地修正模式，造成過度適配(overfitting) 的情形發生。另一方面，研究者可以重新檢視各構面間的因果關係，而不是侷限於研究者最初建置的研究假說(initially hypotheses)和路徑關係(path relation)，減少模式界定錯誤的機會。顯示透過結合DEMATEL和SEM，可以有效改善過去SEM依資料修正模式統計的各項缺失，證實DEMATEL 技術可能是一種用以作為SEM 模式修正的有效且具互補性的工具。此外，透過修正模式，找出影響網廣告影響的最重要因素，有助於管理者制定適切的行銷策略。

Causal analysis greatly affects the efficiency of decision-making. Scholars usually adopt SEM (Structural equation modeling) to establish a causal model in recent years. However, statistical data allow researchers to modify the model frequently to arrive at good model fitness, and SEM is often misapplied when the data are merely fitted to a SEM and the theory is then extended from the analytical result based on presumed hypotheses. The paper proposed SEM modified by DEMATEL (Decision making trial and evaluation laboratory) technique, taking causal model of web-advertising effects for example. Having revealed that the new model is the one that conforms to actual data and is better than initial model, the results confirm that the DEMATEL technique can be an efficient, complementary, and confident approach for reprioritization of the amended modes in a SEM model. DEMETAL provides another tool for examining the accuracy of researchers’ initial hypotheses. A model may be revised based on the analysis result of the DEMATEL technique, and a better model may be acquired. In addition, the DEMATEL technique may offer reasonable bases for modification of SEM to avoid over fitting and the above-mentioned misuses. In addition, the most important factor affecting the web-advertising effects may be found via the modified model, which benefits the manager for making strategic marketing plans.