Google Trends interrogation and data collection
Google Trends (http://www.google.com/trends), based on Google Search that analyzes how often a particular query-term is entered relative to the total search volume across the world. Instead of reporting the absolute, raw search figures, Google Trends does this by presenting relative search volume (RSV). Each data point on the graph is divided by the total searches of the geography and time range it represents and multiplying by 100. The data points can be obtained in Comma Separated Values (CSV) format and the results are normalized and scaled on a range of 0 to 100. The value of 100 represents the term had the highest number of searches within a selected region and time frame on that day, while a score of 0 means that the term is below 1 percent of its peak popularity [16,17]. To make the selection more reliable, the system automatically excludes the repeated searches over a short period of time by the same person.
There are two options for searching keywords in Google Trends tool, searches can be divided into two sections: by “search term”, which enables to search the exact keywords; or by “search topic”, a broader search that containing the particular keywords or the terms that related. In this study, the later search option (searching by entering the particular keywords or terms that related) was used, because there are no definite technical terms available. Among all the relevant search choices, we choose the following search terms “quit smoking”, “stop smoking” and “smoking cessation”, which are the most searched. Those multiple terms can be searched in combination with a plus sign (+) that means “OR” and excluded with a “-” sign. In order to keep in line with the current standards for reporting Google trends data, weekly data from 2014/01/01 to 2018/12/31 were downloaded in CSV files for the United States of America (USA), the United Kingdom (UK), Canada and Ireland as well as New Zealand and Australia [18]. For statistical analyzing, those countries were grouped in relation to the northern or southern hemisphere. The searches were conducted on 6th March 2019, and before the analysis, the accuracy of the data was evaluated by two separate individuals crosschecking the data.
Dynamic series analysis
Dynamic series analysis means a series of statistical indicators are arranged in sequence according to a certain time order to observe and compare the change and development trend of a specified object in time. There are three statistical indicators contained in the dynamic series analysis model, the absolute, relative and average numbers, which were used to describe the object. The ratio of fixed base and link relative, based on the relative comparison, were used in this dynamic series analysis. Unify the index of a certain time as the cardinal number, the fixed base ratio can reflect the development of the observing object by dividing the index value of each reporting time with this cardinal number. While the link relative ratio, define the index of the previous time as the cardinal number, reflects the development direction of the object by dividing the index value of the next reporting time with the cardinal number. The commonly used dynamic series analysis indicators are the absolute growth, development speed and increment speed, average speed of development and the average speed of growth.
Statistical analysis
To identify seasonal variations in the search volume of quit smoking from 2004 to 2018, we utilize the seasonal decomposition of time series model to represent the trend and seasonality in a time series by decomposing these data into a trend component, a remainder component and a seasonal component.
Besides, in order to investigate the systematic seasonal variations of quit smoking, the strategy of cosinor analysis was used to test whether there was a significant seasonal variation in the volume of Internet searches for the term “quit smoking + stop smoking + smoking cessation”. The method and the program used to implement the cosinor analysis were presented in detail by Barnett et al [19]. In short, the cosinor analysis, based on the sinusoidal patterns which fitted to an observed time series, is a common parametric seasonal model that hinged on the following sinusoid: (see Equation 1 in the Supplemental Files)
in which A indicates the amplitude of the seasonal effect and explains the size of seasonal changes, P indicates the phase of the seasonal peak, c indicates the length of the seasonal cycle, which established at 12 for monthly data, t indicates the time of each data point, and n indicates the number of observations.
Since there are two seasonal components exist in this linear model: sine and cosine, the threshold of significance was adjusted as p<0.025 to correct for multiple comparisons. In addition, in an effort to quantify the magnitude of seasonal peaks and troughs for countries that demonstrating significant seasonality, we calculate the percent change in search volume from winter months (USA, UK, Canada and Ireland from the northern hemisphere: December, January and February; New Zealand and Australia from the southern hemisphere: June, July and August ) to summer months (USA, UK, Canada and Ireland: June, July and August; New Zealand and Australia: December, January and February), which was similar to the process in several previous studies [20]. Besides, the conformance of the seasonal variations was emphasized by time series plots.
The "season” Package in R version 3.5.1 was used to perform all the data processing and analyses.