What Insights Can Google Trends Provide About Tourism in Specific Destinations? What Insights Can Google Trends Provide About Tourism in Specific Destinations? Summary These pages explore the potential of using data on the internet search behaviour of visitors to inform strategic decision making relating to tourism destinations. In particular, they look at Google Trends which includes information back to January 2004 on searches related to tourism from the World’s most used search engine. The information here is based on a paper from the Tourism Intelligence Unit (TIU) of the Office for National Statistics (ONS) which was drafted for the 2nd International Conference on the Measurement and Economic Analysis of Regional Tourism. The website for the conference includes links to the original paper, the related presentation and a series of other papers about improving the measurement of tourism and its economic impact. The site is at: http://www.inroutenetwork.org/conference/2011/papers-presentations. Introduction Google Trends provides weekly information about internet searches with advanced features freely available at http://www.google.com/insights/search/. Users can view and download search volume patterns for one or more search term and this information is also available broken down by the location of those making the search and by the category that the search relates to (e.g. travel in general, accommodation or air transport in particular). There is also information about the top and rising searches that include the search term (and category, if used). The information from Google Trends is extremely up to date as it provides weekly figures for a period up to and including the current (even incomplete) week. The availability of data from 2004 onwards allows a time-series to be built up relating to particular search terms. Because of the timeliness of Google Trends, there has been a range of studies that examine how the data can be used to monitor economic trends as they happen, described as “nowcasting” in some papers. This avoids the time lag that is a feature of official statistical releases. As part of these pages we provide a summary of some of these studies and, in particular, highlight the findings that are of particular relevance to tourism. Our main focus is on the potential for Google Trends to provide information about visitor behaviour at the local or destination level where it is not captured by existing surveys. We include an introduction to Google Trends coupled with an explanation of its potential pitfalls and other caveats. We then compare patterns of search data from Google Trends with visitor statistics from some of the UK surveys related to tourism that do include local information, providing a justification for the possible utility of Google Trends for local and destination level tourism organisations. The core of this work, therefore, presents examples of the use of Google Trends in providing insight into the characteristics of potential visitors to destinations and of how they search for information about particular locations. Section 1: Key Facts About Google Trends In this and following sections, we explain the basic features of Google Trends. Before outlining some ‘health warnings’ that users should be aware of, we present some key facts about Google trends that are useful for those using the service: There are two web addresses for Google Trends One of these, “Insights for Search” requires the setting up of a Google account but this is free of charge and provides a full range of features. These include ways of filtering data and the ability to download information. This is the site that we will focus on here and it has the following web address: http://www.google.com/insights/search/. A more limited version of Google Trends is located at: http://www.google.co.uk/trends. The data are measures of the likelihood of searches Google “Insights for Search” analyses a portion of Google web searches to compute the number of searches that have been carried out for specific terms, relative to the total number of searches for the same term on Google over time. This analysis indicates the likelihood of a random user searching for a particular term from a certain location at a certain time. Google’s system eliminates repeated queries from a single user over a short period of time, so that the level of interest isn't artificially impacted by these types of queries. Google Trends uses relative rather than absolute volumes The data in “Insights for Search” are displayed on a scale of 0 to 100 after normalisation (see below), and each point on the graph has been divided by the highest point. Hence, if, when looking at 2009, a term was most searched for in the first week of June, the annual chart of such searches would have a peak of 100 in that week and all other weeks would be displayed as proportions of the volumes of searches for the peak week. Normalisation means that Google has divided sets of data by a common variable to cancel out the variable's effect on the data. This ensures that the underlying characteristics of the data sets can be compared. This means that, for example, when looking at Google Trends data for two different locations, “interest” (proportion of searches) rather than “volume” is being compared. Section 2: Basic Methods of Using Google Trends There are three methods of using Google Trends to make comparisons and each of these has filters that improve the analytical capacity of the tool: i. Perhaps most basically, users can focus on one or more search term and filter by type of search, location of person making the search, date and category. ii. Alternatively, it is possible to select one or more locations of people making searches, compare the interest in a specific search term in these places and filter by type of search, date and category. iii. Finally, users can choose one or more time range (e.g. individual years or months), compare the interest in a search term in these periods and filter by type of search, location and category. Each of these methods produces a chart of “interest over time”, details of interest by area (country, sub-region or city) and lists of the top and rising search terms that relate to the selected term. The time chart compares search terms, locations of those making searches or time periods, depending on the option selected. Figures 1, 2 and 3 in the pdf download are examples of parts of the outputs from each of the three methods. Figure 1 illustrates the proportion of UK searches for all aspects of travel relating to a particular London attraction, Kew Gardens. This has been decreasing slightly in the years from 2004 to date. The chart also shows how such interest is most prevalent in the spring months. Google Trends users are given an option to include forecasts which are shown in Figure 1. These are only based upon the trends shown rather than any other, external, considerations. Figure 2 shows an analysis of the search term “Cardiff” from January 2007 onwards, restricted to the travel category and to searches from three European countries. The chart illustrates that, generally, interest has been proportionally greater in the Netherlands than in France and Germany. The most notable of the exceptions to this pattern was in the autumn of 2007 when Cardiff hosted a Rugby World Cup quarter final that included France. Other spikes in interest in France also coincide with international rugby matches in Cardiff. The spikes in interest from the Netherlands are not as easily explained by sporting events but a lot of fluctuations in such data are likely to be due to the fact that the information is based on relatively small volumes of searches. Despite these fluctuations, this particular analysis from Google Trends has clear potential for targeting and monitoring the effects of overseas marketing activity. Figure 3 again focuses on Kew Gardens, comparing each of the last four full years of the data in Figure 1. This replicates the seasonality in the proportion of searches that was shown in the previous chart but includes more detail. Patterns relating to public holidays and the periods outside of school terms are noticeable but other peaks may relate to special events, marketing campaigns or news stories and this type of chart is a useful measure of the impact of such publicity. The travel category used in these and other analyses in this pages includes sub-categories relating to hotels and accommodation, attractions and activities, bus and rail travel, air travel, car rental and taxis, cruises and charters, adventure travel and vacation destinations. Google Trends allows users to select these sub-categories and use them in the same way as categories. Figures 20 and 21, for example, chart the hotel and accommodation sub-category. Section 3: Caveats to be Considered When Using Google Trends This section highlights issues that need to be considered when using Google Trends data. Some of these are caveats that occur because of the way that the data are obtained and presented. One example of this is the use of relative rather than absolute volumes of searches. We have highlighted this in the previous section but it is worth restating and emphasising that Google Trends normalises data when comparisons are made. Normalisation means that, for example, while “Insights for Search” suggests that interest in travel to France is very similar in Wales and England, as highlighted in Figure 4 in the pdf download. However, the large differences in the relative populations lead to volumes of Google searches for this topic in England being larger. Another caveat is that Google uses IP address information to “make an educated guess” about where queries originated. In the UK this results in information based on some rather unexpected conurbations and, coupled with normalisation, this methodology means that conclusions about interest within sub-regions based on the data have to be explained carefully. Other considerations to be aware of when using this tool in a tourism context include the fact that users of Google may not be representative of all visitors to a destination as, for example, people from some demographic groups may be more likely to use a travel agent or may not have internet access. In addition, specific types of tourists may use search tools to a lesser extent, for example, regular visitors to a destination or those visiting friends and families. Section 4: Potential Pitfalls to Avoid When Using Google Trends An example of the care that is needed when choosing destination-based search terms relates to an investigation of the relative interest to Google users worldwide of the search terms “York” and “Edinburgh”. Interest appears, initially, to be much greater for the English historic city than its Scottish counterpart, as illustrated in Figure 5 of the pdf download. However, analysis of the most used search terms for “York” (Figure 6) indicates that a large proportion of the interest in it actually relates to New York and versions of its name in other languages. Google Trends allow users to compare search terms with related content excluded by using a minus sign. By using this method to omit “new”, “nueva” and “nova” from searches for York, a more realistic comparison of interest in Edinburgh and York is possible, with the former receiving a greater proportion of searches, as shown in Figure 7. It is possible that data still include searches for places named or including York (and Edinburgh) other than those in the UK but it is unlikely that any of these have as misleading an effect as New York. Where destinations are among those with low search volumes, Google Trends data cannot be broken down by category. In these cases it is important to consider whether searches for a particular destination may be for other purposes than tourism. For example, interest in Tintern Abbey, the historic visitor attraction in Eastern Wales, includes searches relating to a poem by William Wordsworth with the Abbey’s name in its title. Adjustments to the search term have to be made to exclude elements relating to this poem if tourism interest only is being assessed. A further consideration when using Google Trends is to take account of whether a specific location has the same name in other languages and to include these in search terms. For example, the majority of Google users in France undertaking searches for travel relating to London use its French name “Londres”. However, a significant minority (including English speaking ex-patriots) use “London”. Section 5: Comparison of Google Trends with Official Data One of the potentially useful facets of Google Trends is that it provides some information about demand or interest for destinations that are too small to be covered by national surveys. The pdf download, however, includes charts about two sets of currently available official data and compares them with related information from Google Insights for Search. Figure 8 in the pdf shows annual admissions data from Visit England (the national tourism organisation for England) for two of the UK's leading attractions; the Eden Project in Cornwall and Kew Gardens in west London. The figure indicates that the number of admissions for the former has fallen in each year from 2004 to 2010 while the pattern for the latter has been less pronounced with fluctuations and a decrease in the most recent year. The related chart in Figure 10 in the pdf shows a similar pattern in the Google chart: a more noticeable reduction in interest in the Eden Project (as measured by travel related searches in the UK) than in interest in Kew Gardens. However, by this measure, interest in the latter has also fallen slightly in recent years. The Google chart does include 2011 data which appear to show similar levels to that in 2010 for both attractions and it will be instructive to compare this with the 2011 annual admissions dataset when it is available. The chart in Figure 9 looks at the quarterly estimate of the number of visits of US residents to London from the UK’s International Passenger Survey (IPS - carried out by the Office for National Statistics). This can be compared with trends in American searches that include the term “London” and relate to hotels and accommodation (see Figure 11). The published data highlight the seasonality of visits from the USA and indicate that the total number of visits in 2010 was less than in the previous year. The Google Trends output shows a similar pattern of interest by quarter and also a falling of interest year on year which has continued into the second and third quarters of 2011, periods for which IPS data were not available when the chart was produced. Section 6: Examples of Seasonality in Google Trends This section and those following include a variety of presentations of Google Trends data to highlight how the tool can assist with analysis of aspects of tourism to a number of different types of destination. The pdf download includes three charts from Google Trends of interest over time for each year from 2007 to 2010. These reflect UK-based travel-related searches for a coastal resort (Skegness), an inland holiday destination (the Lake District) and a major city (London). The fourth chart in the pdf investigates the seasonality of travel-related interest in Glasgow and Edinburgh, the two largest Scottish cities. Figure 12 in the pdf indicates that the period where there is the greatest travel-related interest in Skegness coincides with July and August, the months that include school holidays. There are also spikes of interest in April, which usually includes an Easter school break. The chart for the Lake District (Figure 13) has peaks that are less pronounced, although the periods of greatest interest are the same as for Skegness. It also highlights that interest in travel to the area has fallen during the past four years, in contrast to the trend for the coastal resort. The equivalent chart for London in Figure 14 shows comparatively little seasonal variation in Google searches and, again, seemingly falling interest from 2007 to 2010. The patterns over time for travel-related interest in Edinburgh and Glasgow in the Figure 15 are very similar. The former is the subject of a higher proportion of searches from UK-based Google users but in both cases there are peaks of interest each year at the start of the third quarter and another upturn in interest at the turn of the year. Possible explanations for the patterns in the chart would include the internationally renowned Edinburgh Festival each summer and the New Year’s Eve celebrations. The resemblance between the trends for each city may suggest that the festival prompts searches relating to Glasgow as well as to Edinburgh. Section 7: Examples of the Impact of Major Events in Google Trends A major event that has an effect on levels of travel-related interest for a specific location is the Glastonbury Music Festival which takes place in June of most years. Figure 16 in the pdf download illustrates that searches relating to all aspects of travel and “Glastonbury” peak dramatically at around the time of the festival. Smaller peaks are likely to relate to announcements of when tickets go on sale or of announcements of which acts are performing. The absence of a festival in 2006 also gives an indication of the levels of travel-related internet searches relating to the Somerset town for purposes other than the festival. Figure 17 focuses on the small town of Castle Cary which is the location of the nearest railway station to the festival, about eight miles away. Interest relating to travel to this town peaks in June of each year that the festival takes place but the relative high levels of searches in 2006 and other parts of other years suggest that Google users are also interested in non-festival related aspects of the town. The Google Trends tool also allows analysis of the amount of interest that smaller, non-music based events prompt for specific areas. The example in Figure 18 is the Welsh town of Abergavenny which hosts a food festival in the third weekend of September each year. Because of the comparatively small number of searches relating to this location the analysis is not limited to a specific category. In each of the three years covered by the chart, interest in “Abergavenny” peaks in September and the degree of interest in the town at the time of the festival, as measured in this way, is similar. Section 8: Examples of the Impact of News Stories in Google Trends Positive News Stories In the summer of 2010, Rhossili Bay on the western tip of the Gower Peninsula in south Wales was voted best British beach. Figure 19 in the pdf download highlights how this award prompted an increase in interest in the beach at about the time the news story was released. However, the chart also indicates a higher level of interest in the summer of 2011 than in 2008 or 2009, perhaps suggesting that the award has increased the profile of the beach. The comparator in the chart is a neighbouring beach, Port Eynon, which received the same honour in the summer of 2011. Major News Stories Riots in London and other English cities took place in early August 2011 and there was widespread international coverage of these. There was a lag until the data that show any effects on the number of tourism visits were due to be released but Google Trends was a useful up to date indication of any impact on the level of interest in searching for travel information for London. Figures 20 and 21 in the pdf feature two charts of accommodation-related searches that include the term “London”. These relate to US and UK residents, respectively. In both cases there is a distinct fall in the level of interest in the second half of August 2011. As the tables include more than one year’s data, we can establish that the recent drop in interest does not mirror what happened in 2009 and 2010 and could therefore be a reaction to the riots. Another major news story in London during 2011 was the royal wedding at the end of April. Interest in London accommodation among Americans has a particularly noticeable spike for the week containing the wedding day, the only week where such interest was higher in 2011 than in the previous two years. The peak was not particularly striking in the chart for UK residents but a noticeable trend over 2011 is that interest among UK residents is generally higher than in 2009 and 2010. This could be a reflection of a greater likelihood to investigate domestic holiday options rather than overseas equivalents for economic reasons. Section 9: Implications of Previous Research The availability of indices analysing the queries made using the world’s most popular search engine has prompted many researchers and analysts to examine the extent to which Google Trends can be used as an indicator of economic activity. This section introduces and summarises recent reports that focus on the tool and highlights the research within these that has implications for tourism. Full references of these and other papers are in a later section. One of the key introductory papers on these topics was published by two economists affiliated to Google. Predicting the Present with Google Trends (Choi and Varian, 2009) investigated whether search queries could help "predict" current economic activity, in other words, report it without a lag. The report aimed to familiarize readers with Google Trends and to illustrate some simple forecasting methods that use its data. The paper included analysis of Hong Kong visitor data and found that Google searches on `Hong Kong' were positively related to arrivals. In addition, the paper concluded that arrivals in the destination in the previous month were positively related to arrivals in the current month, as were arrivals 12 months earlier. The authors also explained that during the Beijing Olympics travel to Hong Kong decreased. A second Google paper, On the Predictability of Search Trends (Shimshoni, Efron and Matias, 2009), provided the analysis that resulted in a basic forecasting capability being introduced into Google Insights for Search (as shown in Figure 1). The paper found that over half of the most popular Google search queries were predictable using the method that the authors selected. The analysis included an assessment of the predictability of 10 broad search categories and concluded that, of these, only health was more predictable than the two tourism-related categories of food (and drink) and travel. The ONS Economic and Labour Market Review (ELMR) in the UK followed up the above analyses with a detailed investigation of the use of Google Trends data for various search categories: Googling the Present (Chamberlin, 2010). The article looked at the correlation with official data of over 30 categories of retail sales, property transactions, car registrations and foreign trips. In terms of tourism, the research found that none of the numerous relevant Google Trends categories were significant in a regression with the numbers of foreign trips. However, the article did conclude that the 'Travel' category showed similar seasonal movements to this statistic. Two Ruhr Economic Papers from Germany introduced private consumption indicators based on search query time series provided by Google Trends. Forecasting Private Consumption: Survey-based Indicators vs. Google Trends (Schmidt and Vosen, 2009) used information from US-based searches and compared this to the two most common American survey-based indicators. A second paper, A Monthly Consumption Indicator for Germany Based on Internet Search Query Data (Schmidt and Vosen, 2010), makes similar comparisons between German search data and European Commission confidence indicators. In both cases the new search data indicator outperformed the survey-based indicators. Within the UK, the Bank of England raised the profile of the discussion of the possible uses of Google Trends by highlighting the topic in its Quarterly Bulletin. The article, Using Internet Search Data as Economic Indicators (McLaren, 2011) focussed on the UK labour and housing markets and includes a useful summary of the potential benefits and problems of internet search data. It also referred to analyses elsewhere, including the Bank of Israel paper Query Indices and a 2008 Downturn (Suhoy, 2009). This tested the hypothesis that Google query indices may be helpful in drawing inferences about the state of current economic growth and confirmed that Israeli data supported the hypothesis. As well as focussing on the whole economy, there have also been papers that concentrate on how Google Trends data can be used within a specific part of the economy, including tourism or tourism industries. Forecasting Tourism in Dubai (Saidi, Scacciavillani and Ali, 2010), for example, uses traditional empirical methods as well as analysis of Google Trends. It concludes that the latter is helpful for improving the long-term forecast for guest nights but has less short-term usefulness. It also highlights the effectiveness of air travel searches in forecasting arrivals at Dubai airport. An earlier tourism-related paper examined the behaviour of users of the Dogpile.com search engine rather than of Google. This paper, An Analysis of Travel Information Searching on the Web (Jansen, Ciamacca and Spink, 2008), concluded that, in 2005 at least, about 6.5% of Web queries were for travel searching and that geographical information accounted for nearly 50% of this. Other recent research includes a UK study (Judge and Hand, 2010) that found clear evidence that Google Trends data on searches relevant to cinema visits could increase the accuracy of cinema admissions forecasting models. Another paper for the Journal of Travel Research (Fesenmaier and others, 2011) proposed and evaluated a three stage model to examine how online travellers use search engines and how aspects of the travel planning process shape this use. Finally, Google Trends is not the only aspect of internet usage that has prompted research. In terms of tourism, a good example of further analysis is a paper (Milano and others, 2011) that measures the impact that the social media sites Facebook and Twitter have on the popularity of tourism websites. More generally, a paper from the University of Illinois (Xin Jin and others, 2010) discussed the potential of using the photo-sharing site Flickr for forecasting. Conclusion & References Google Trends provides an effective measure of levels of interest in one or more topics and of changes over time in this interest. These pages highlight how useful this can be for those involved in analysing tourism, particularly at a local or destination level. They include examples that match patterns in internet search terms with aspects of tourism. Finally, we have also summarised research papers that discuss how Google Trends data can act as a proxy for economic measures and, in particular, provide more up to date information than even the most current data sources can supply. References A Monthly Consumption Indicator for Germany Based on Internet Search Query Data Torsten Schmidt, Simeon Vosen (2010) Rheinisch-Westfälisches Institut http://ideas.repec.org/p/rwi/repape/0208.html An Analysis of Search Engine Use for Travel Planning Daniel R. Fesenmaier, Zheng Xiang, Bing Pan, Rob Law (2011) http://sb.cofc.edu/academicdepartments/hospitalitytourism/facultyandstaff/pan-bing.php An Analysis of Travel Information Searching On The Web Bernard J. Jansen, Christopher C. Ciamacca, Amanda Spink (2008) http://academic.research.microsoft.com/Publication/6113425/an-analysis-of-travel-information-searching-on-the-web Do Google Searches Help in Nowcasting Private Consumption? Konstantin A. Kholodilin, M. Podstawski, B. Siliverstovs (2010) Deutsches Institut für Wirtschaftsforschung http://ideas.repec.org/p/diw/diwwpp/dp997.html Forecasting Private Consumption - Survey-based Indicators vs. Google Trends Torsten Schmidt, Simeon Vosen (2009) Rheinisch-Westfälisches Institut http://ideas.repec.org/p/rwi/repape/0155.html Forecasting Tourism in Dubai Dr. Nasser Saidi, Dr. Fabio Scacciavillani, Fahad Ali (2010) Dubai International Finance Centre http://www.difc.ae/publications Google Econometrics and Unemployment Forecasting Nikos Askitas, Klaus F. Zimmermann (2009) Deutsches Institut für Wirtschaftsforschung http://ideas.repec.org/p/diw/diwwpp/dp899.html Googling the present Graeme Chamberlin (2010) ONS Economic & Labour Market Review http://www.ons.gov.uk/ons/rel/elmr/economic-and-labour-market-review/no--12--december-2010/index.html On the Predictability of Search Trends Yossi Matias, Niv Efron, Yair Shimshoni (2009) Google Labs, Israel http://googleresearch.blogspot.com/2009/08/on-predictability-of-search-trends.html Predicting the Present with Google Trends Hal Varian, Hyunyoung Choi (2009) Google Research Blog http://googleresearch.blogspot.com/2009/04/predicting-present-with-google-trends.html Query Indices and a 2008 Downturn - Israeli Data. Tanya Suhoy (2009) Bank of Israel http://www.bankisrael.gov.il/deptdata/mehkar/papers/dp0906e.htm Searching for the picture - forecasting UK cinema admissions making use of Google Trends data Guy Judge, Chris Hand (2010) University of Portsmouth Dept of Economics http://eprints.port.ac.uk/4820/ The Effects of online social media on tourism websites Roberta Milano, Rodolfo Baggio, Robert Piattelli (2011) International Conference on IT and Travel & Tourism http://www.iby.it/turismo/ The Wisdom of Social Multimedia - Using Flickr for prediction & forecast Xin Jin, Andrew Gallagher, Liangliang Cao, Jiebo Luo, Jiawei Han (2010) ACM Conf. on Multimedia http://www.ifp.illinois.edu/~cao4/publications.html Using internet search data as economic indicators Nick McLaren, Rachana Shanbhogue (2011) Bank of England http://www.bankofengland.co.uk/publications/quarterlybulletin/m11qbcon.htm