九二一地震後臺灣學術界開始累積災難相關的調查數據，但多數研究使用有選擇性偏誤的災民樣本。因此，本文比較含括九二一相關資訊且可供量化分析之資料庫，說明其特性及其能處理的災難社會學議題，並討論各資料庫的限制與未來發展。本文發現，臺灣教育長期追蹤資料庫為中學生樣本，面臨推論至全體的挑戰；社會意向電話調查為家戶電話抽樣，可能流失受災嚴重的組合屋樣本；我們自行抽樣的家屋再造資料庫為申請九二一基金會家屋再造補助的災民資料，因為補助條件與弱勢程度重合，有待與非災民資料庫合併分析。目前相關資料庫都難以全面分析災難的社會經濟影響，因此我們建議由國家相關單位著手調查，建立完整的災民資訊，以突破研究資料不足的瓶頸。

Previous disaster studies in Taiwan are quite limited until the 921 earthquake. After the earthquake, however, most research analyzed victim-only samples suffering from selection bias. This article compares three earthquake-related surveys and illustrates their limitations. Taiwan Education Panel Survey with nationwide high school students samples is not a random sample of the national population and biased by some retrospective errors. Collected through telephone surveys, Social Image Telephone Survey may lose victims who were damaged too much to keep a household telephone. Finally, House Reconstruction Database was constructed by coding the information of the applicants for the House Reconstruction Project of the 921 Earthquake Relief Foundations. However, the requirements of the applicants may lead to the self-selection problem. These results imply that the limitations of the datasets have biased the statistics of disaster analyses. Therefore, we suggest the government and relative relief institutions shall conduct a victims' census after each disaster to support Taiwanese disaster studies in the future.