This dataset is used to support findings in the paper "Explore the potential of using social media crowdsourcing for earthquake damage assessment." In this paper, we investigated six earthquake events to explore the potential of using social media crowdsourcing for earthquake damage assessment. These cases include the 2014 Iquique earthquake, the 2015 Napa earthquake, the 2015 Nepal earthquake, the 2019 Ridgecrest earthquake, the 2021 Fukushima earthquake, and the 2021 Haiti earthquake. For the first three cases, tweet ID information was collected from the publicly available website CrisisNLP (https://crisisnlp.qcri.org/). For the last three cases, we collected “near-real-time” tweets with the key search term “earthquake.” For the 2019 Ridgecrest earthquake, the search date range was from July 4 to 10, 2019. For the 2021 Fukushima earthquake, the search date range was February 13 to 17, 2021. For the 2021 Haiti earthquake, the search date range was August 14 to 19, 2021. It should be noted that restrictions apply to the availability of tweet data due to the policy from Twitter Inc. Tweet data showing user profiles or text content is not allowed to be shared publicly, but tweet IDs are available to download from this dataset. In addition, we attached our processed results to this dataset, including, Damage estimation. The damage estimation for each event is based on the average estimated level of all damage-related tweets identified by text classification models, as documented in our paper. Temporal dynamics of damage estimation. We conducted the temporal analysis to explore the convergence of damage estimates over time, in which data were first binned based on the number of damage-related tweets in each 2-hour interval. If you need the training and testing datasets or have any further questions regarding the dataset or paper, please do not hesitate to reach us.