Figure 1 shows our study process. We searched for policies, laws, and regulations promulgated or amended in Taiwan during the period between January 2004 and November 2019. We then identified 2 working conditions related to the aforementioned policy changes: (1) overwork, which has no official definition in Taiwan but is understood to describe both working too long (too many hours) and as working too hard (physically or psychologically); and (2) long working hours, which are regulated through both regular working time and overtime policies. We further identified a third working condition related to the 2004–2019 policy changes but had no direct quantitative measure: (3) high job stress. We treated overwork and long working hours as the main topics of interest, and high job stress as a reference topic (relevant to overwork but not addressed in the amendments of Labor Standards Act). To measure these 3 working conditions, we identified a keyword for each condition.

We used the keywords to extract search volumes from the Google Trends platform [18], which is used to record big data related to search volumes. The data on search volumes were scaled on a range of 0-100, with the scaled number shown as relative proportional to all searches. The website has 4 filter options: region, time, category, and search type; for these filter options, we selected Taiwan, January 2004–November 2019, all categories, and web search, respectively.

We noticed the variability of search volumes while we repeatedly searched on the Google Trends platform. The Google Trends sampled data on searches correspond to billions of Google searches per day. However, the algorithms behind the platform were not available to the general public [19]. We, therefore, developed our search strategy, repeated data extraction, and smoothed data by taking averages. We extracted one search volume data set every 5 minutes and repeated five times for each keyword. We then calculated the average of search volumes for each keyword and analyzed search volume trends using the Joinpoint Regression Program [20]. Next, we identified the turning points and peaks of search volumes that followed policy implementations. For points or peaks that did not follow policy changes, we used Google’s search engine to explore related events before the turning points and peaks.

Our study period was from January 2004 to November 2019. January 2004 was the earliest data we could access from the Google Trends platform, and 2004 was also the year when Taiwan’s Recognition Guidelines for Overwork-Related Diseases were first amended.

First, we calculated the average monthly search volume for each keyword. Second, we used the Joinpoint Regression Program to estimate the web search volume trends over time [20]. The Joinpoint Regression Program allows us to analyze trend data and test joinpoints (or apparent changes or peaks) across the whole study period. Data captured as 0 were adjusted to 0.5 to fit all data on a log-linear Poisson regression model. Bayesian information criterion was used to identify turning points and select the best model. The monthly percentage change of search volumes of each segment during the study period was estimated by the following equation:

where log(V_t) is the natural log of the average number of relative search volumes at time t; β₀ indicates the intercept of the model; β₁ is the slope of the model, which was later used to estimate the monthly percentage change of search volumes of each segment [20]; and t is the time variable, indicating the month (from January 2004 to November 2019, indicated as 1 to 191).

Over the past 15 years, Taiwanese authorities have revised and implemented several policies to prevent overwork-related health problems. By treating search volumes for overwork-related keywords as proxy indicators of public awareness, our findings suggest that increasing public awareness of the importance of overwork prevention and shorter working hours was observed after Taiwan’s recent policy changes related to overwork prevention. Such increased awareness is expected to contribute to better attitudes and practices regarding overwork prevention among the general public [9].

Public awareness of overwork significantly increased (P<.001) following the 2010 revision and implementation of the Recognition Guidelines for Overwork-Related Diseases, as well as the frequent media reports in 2009–2011 of suspected overwork-related incidents. Public awareness of working hours increased from 2013 and peaked twice, in the end of 2015 and early 2017, along with the two amendments (12th and 16th) of the Labor Standards Act.

The search volumes for overwork increased from June 2010 to March 2011, a period which saw frequent reports of suspected overwork-related incidents prompt public realization of the seriousness of the problem. During the same period, more overwork cases were recognized by the 2010 Recognition Guidelines for Overwork-Related Diseases, which encouraged public confidence in overwork-related policy. After the revised recognition policy was implemented in 2010, the average number of working hours per month of those employed in the industry and service sectors in Taiwan decreased by 9.1 hours from 178.7 hours in 2011 to 169.6 hours in 2016 [21]. The amendment to the recognition guidelines followed the sudden death, in January 2010, of an engineer in Taiwan by cardiogenic shock and hypertrophic cardiomyopathy. Although some of the disease cases could not be officially recognized as related to overwork right out, they exerted influences because a nongovernmental organization (Taiwan Labour Front) and a legislator (Ms Sue‐Ying Huang) organized several press conferences on behalf of families of deceased to promote social awareness via social media, and finally intervene in the decision making of the Ministry of Labor in Taiwan [6]. These activities had aroused public awareness about the strenuous working conditions of elite image engineers. When the 2010 Recognition Guidelines for Overwork-Related Diseases were implemented, cerebrovascular and cardiovascular diseases related to working conditions with long-term excessive overtime became the first recognized overwork disease [6]. Social media together with efforts of Legislator Huang and nongovernmental organizations appeared to have a strong influence on the policy changes in Taiwan [6,22]. Based on this understanding, one way to significantly enhance public awareness of overwork is to cooperate with legislators and nongovernmental organizations and to encourage more media reports on policy changes via social media.

Public awareness about working hours significantly increased (P<.001) from September 2013 and reached a peak in January 2017, which included 2 relevant policy amendments to the Labor Standards Act. Among all search volume peaks in this period, the highest were observed in January 2017, aligned with the start of implementation of the 16th amendment, which called for “1 fixed and 1 flexible day off per week.” That this policy change would call forth such a high search volume might be due to presidential influence. The theme of overwork was a hot issue during Taiwan’s 2016 presidential election. Among all presidential candidates, one candidate (Dr Ing-Wen Tsai, who later won the election) was most committed to preventing overwork problem in Taiwan and even claimed to have prepared related policies that would improve working conditions. This candidate went on to get most of the labor unions’ support. Along with their candidate’s victory in the presidential election, union workers now look forward to the corresponding policy changes [23], which they expect to provide better working conditions. Since President Tsai was inaugurated in May 2016, the Labor Standards Act has been revised seven more times between November 2016 and May 2019 [24].

In our study, we observed higher search volumes around the months when policies were revised and implemented. Revision and implementation are the 2 phases wherein the public can participate in the policy change process (comprised from invention, estimation, selection [revision], implementation, evaluation, to termination) [25]. High search volume observed during policy revision and implementation periods may be attributable to the healthy democracy in Taiwan, which allows for public participation in the policy change process, promotes mutual understanding among stakeholders, and emphasizes the legitimacy of governance [26].

Public awareness on both overwork and working hours was enhanced along with the related policy changes and frequent reports of suspected overwork-related events. However, the search volume trends on overwork and working hours differ. The trend of increasing search volumes on overwork (June 2010 to March 2011) lasted for only 10 months, following the reporting of suspected overwork-related events. By contrast, the trend of increasing search volumes on working hours (September 2013 to October 2016) lasted for 38 months, following the amendments of the Labor Standards Act. Our results suggest that public awareness caused by events is short lived, whereas public awareness induced by policy exists for a longer time, likely because social media continues to provide new information and discussion on the topic [27,28]. Therefore, we suggest that developing overwork-related policies that clearly address risk factors contributes to increasing the public awareness of overwork prevention.

As a reference outcome, public awareness of job stress was lower than public awareness of overwork and working hours over the past 15 years. Given the fact that high job stress is one of the major reasons for overwork [29], we suggest that policies should also target overwork caused by job stress, similar to European policies for psychosocial risk management in the workplace [30].

There are some limitations to our study. First, we identified one keyword with the highest search volume as the representative keyword for each working condition. The general public might use other unpopular keywords (synonyms) to search for more knowledge on each working condition. That is, each working condition could be described using different keywords. Thus, our choice of high-volume keywords might underestimate the effect of policy changes on public awareness.

Second, we could not get an absolute value for search volume on Google Trends. Data obtained from Google Trends are a relative search volume (search volume range from 0 to 100). For cross-keyword and overtime comparisons, we could only compare search volumes for up to 5 keywords at once. Our choice of 3 keywords ensured the comparability of search volumes in our results. Future studies that attempt to expand the number of keywords compared will warrant the development of advanced adjustment methods.

Third, search volumes on Google Trends have been used to explore and predict behaviors on the outpatient visits and people’s awareness of health care [16,31-33], but have been shown to be less predictive of public opinions on elections and referendums in low-internet-use or low-freedom-of-speech regions [34]. Our study focuses on searches in Taiwan, where over 80% of the population uses the internet and people have a high degree of freedom to speak their personal opinions [35,36]. The advantages of using Google Trends should also be highlighted, including reducing the time and cost associated with the traditional measures of public awareness and acting as a supplement and complement to traditional polls or surveillance systems [37], especially in countries with high internet penetration and freedom of speech. Although Google is the most commonly used search engine in Taiwan, there are other types of search engines that may be used by different sectors of population, thus may generate different search patterns. Yet, comparative analyses of search patterns across different search engines are beyond the scope of this study. A separate study involving the collection and comparison of data from different search engines can help explore the diverse patterns.

As described in a 2014 review about big data’s features [38], the 4V concept—related to the high volume, velocity, variety, and veracity of information—was developed, along with the understanding that transforming data into value requires specific technology and analytical methods. Search volumes obtained from Google Trends are relative, and some might doubt if these search volumes should be considered big data. However, Google Trends data fit the aforesaid 4 features, and some earlier studies have set precedents for considering Google Trends data as online big data [39]. Moreover, the novel way in which Google Trends presents data largely reduces equipment requirements for big data analysis and can be used in many fields. Google Trends can in fact help government officials quickly clarify fast-changing events that raise public concern [40,41] and collect large volumes and a variety of past data in a short period, which is hard to achieve through traditional public polls.

What our study contributes to the global society, particularly labor policymakers, is that a clear policy will lead to rising public awareness, and such an increase in public awareness will reinforce policy change—leading to a virtuous circle. In addition, study findings from Taiwan, a democratic society, highlight the importance of communication capacity in informing and engaging citizens. We believe that our approach enabled a reasonable representation of countries with similar internet access, freedom of participation in political and public affairs, and government responsiveness; by contrast, countries with different political environments warrant separate analyses. We encourage the country’s citizens to have more access and freedom to engage in public affairs, particularly labor and public health issues.

Overwork-related health problems due to long working hours, heavy workloads, hectic work pace, and insufficient rest time are major occupational health concerns in many East Asian countries, and in certain occupational groups of Western societies. Regulations concerning working hours, overtime pay, paid leave, and rest time have long been central issues in labor and health protection policies. While in Taiwan, the labor and health authorities have made efforts by revising laws and implementing new policies to improve work quality and to reduce overwork-related health risks, there are still controversies and debates with respect to their regulatory designs and enforcement. Analyses of substantive policies are beyond the scope of this study.

Yet, this study demonstrates rising public awareness of overwork prevention, particularly as it relates to overwork and working hours, following recent overwork-related policy changes. Public awareness of policy contents and citizens’ supports are essential for effective policy implementation. Our findings suggested that in Taiwan, certain components of overwork prevention policies were conveyed to the public more effectively while the others were conveyed less effectively. Because our findings indicate that the public are less aware of the role of job stress in overwork prevention, we suggest implementing interventions to address this phenomenon. Ultimately, to enhance public awareness of overwork prevention, national overwork-related policies that clearly address risk factors are necessary.