Fireworks Work Safety Incidents Reported by Media, 2011- 2016

Data were collected by using the search engine of Xinhua, the country’s official press agency. It is the most influential media outlet in China, encompassing national and local news from 23 provinces, 4 direct-controlled municipalities, 5 autonomous regions and 2 special-administrative regions (Xinhua). A search using Xinhua’s news engine will run through all national, district and county-level news agencies’ databases, thereby offering a comprehensive data source. Keywords ‘firework explosion’ (Chinese: yanhua baozhu baozha), without quotation mark (‘‘), were typed into the entry to search ‘in all fields’. The duration of this study is from Feb 20th, 2011 to Feb 19th, 2016, spanning five years. To simplify the process, searches were done in order of decreasing relevance by month, and results were browsed through manually to pinpoint relevant reports. Quotation marks were not included in the search because this would have only yielded articles with that exact phrase (Barker, unknown), and the goal is to capture all possible reports that has keywords switched order or separated by other information. Each accident is characterized by the following six variables: timestamp, location, severity, legality, level of administration, and the stage at which an accident happened. A ‘timestamp’ is a footprint of the year, month and day when the event happened. ‘Location’ is determined by province, city and county/district. ‘Level of administration’ describes whether the location is urban, suburban or rural. ‘Severity’ has two components to represent the number of people ‘injured’ or ‘dead’ as a result of the explosion. ‘Legality’ represents the presence or absence of legal allowance of operation granted to the firework manufacturers. ‘Stage’ is the manufacturing step at which an accident happened; it can be during production, storage, transportation, retail, disposal or others. ‘Others’ is a separate category to subsume rare causal factors such as lightning-initiated incidents. The breakdown of each variable in more detail can be found in Table 1. An arbitrary numeric value is assigned to each variable for subsequent statistical analysis. In some cases, when certain information cannot be obtained from the report, missing entries are allowed. This is due to the heterogeneity of news reports, and the fact that different media companies emphasize on differing aspects of an incident.

本数据集的数据源自中国官方通讯社新华社（Xinhua）的搜索引擎。作为国内最具影响力的媒体机构，新华社覆盖了全国23个省、4个直辖市、5个自治区及2个特别行政区的国家与地方新闻资源（Xinhua）。依托其新闻引擎进行检索时，系统将遍历所有国家级、区级及县级新闻机构的数据库，因此可提供覆盖全面的数据源。

本次检索以英文关键词"firework explosion"（中文：烟花爆炸）为检索项，未使用引号，在"全字段"范围内开展搜索。本次研究的检索时间跨度为2011年2月20日至2016年2月19日，共计五年。为简化检索流程，本次检索按照相关性从高到低按月进行，并通过人工浏览筛选出相关报道。本次检索未使用引号，原因在于若使用引号将仅能返回包含精确短语的文章（Barker，未注明日期），本研究旨在捕获所有关键词顺序颠倒或被其他信息分隔的相关报道。

每起事故通过以下六个变量进行表征：时间戳（timestamp）、事发地点、事故严重程度、合规性、行政层级，以及事故发生阶段。其中，时间戳记录事件发生的年、月、日信息；事发地点由省、市、县/区三级行政区划确定；行政层级用于描述事发地点属于城区、郊区还是乡村；事故严重程度包含两个维度，分别表征爆炸事故造成的受伤人数与死亡人数；合规性用于标识烟花生产企业是否获得合法经营许可；事故发生阶段指事故发生时烟花所处的生产、储存、运输、销售、处置或其他环节，其中"其他"类别用于涵盖罕见的致灾因素，如雷击引发的事故。各变量的详细分类说明可参见表1。

为便于后续开展统计分析，为每个变量分配了任意数值编码。部分情况下，若无法从新闻报道中获取某类信息，则允许存在缺失值。这是由于新闻报道本身存在异质性，且不同媒体机构对事件的报道侧重点各有不同。