We examined deidentified data from Yahoo! search engine queries in a 28-day period before (14-day) and after (14-day) the Sandy Hook shooting incident. Our population consisted of all users of the Yahoo! search engine located in the United States (including Puerto Rico and Mariana Islands) that queried firearm-related searches during the study period. The majority of the information consumed on the Web starts as search queries entered by the user. The choices made by the user in the form of websites they click from the list of populated search results present a much more comprehensive picture of a user’s information needs. Our goal was to use the search query data to evaluate patterns of information seeking regarding firearms and to evaluate broadly the changes in intent based on differences in the content (retail, news, education, and so on) and sources (commercial entities, noncommercial organizations, government entities, and so on) of information sought.

From the complete Yahoo! search query database, we identified all firearm-related queries from November 31, 2012 to December 28, 2012. Queries were text strings consisting of single words or phrases that users typed into the search engine; we identified these using keywords that would match partially or completely with words in the queries. Firearm-related search queries were identified by using keywords in the following categories: Gun type (gun, firearm, handgun, rifle, pistol, revolver, and shotgun), ammunition (ammunition, ammo, and bullets), law related (Brady Act, second amendment of the US constitution), and shooting. In order to choose keywords in each category, we examined Web-based trends of firearm-related search queries for December 2012 using Google Trends. We did this by first examining simpler queries (eg, handgun), and the 10 most correlated searches for these queries. This was repeated recursively with each of the correlated queries until we found no new or correlated searches. This gave us a set of 247 queries that were related to firearms. We wanted our keywords to have the ability to identify these 247 queries as well as any other searches that may be firearm related. Thus our keywords consisted of single words which could identify most firearm-related searches based on complete or partial matches with user queries. As such, our analysis included users’ actual search queries that included keywords in any of the 4 categories.

In addition, we also analyzed the uniform resource locators (URLs) that each individual user clicked from the search results generated by their search queries. First, we identified the domain for each URL that the user clicked; for example, if the user clicked the URL “http://en.wikipedia.org/wiki/Second _Amendment,” the domain was identified as “wikipedia.org.” Next, we categorized these URLs based on the top-level domain (TLD) into commercial entities (.com), noncommercial organizations (.org), government entities (.gov, .state.us, and .mil), educational institutions (.edu), and others (country specific, .pro, .tv, and so on). Including TLDs in our analysis allows us to infer the nature of the organization; for instance, TLDs such as .gov, .mil, and .edu have legal restrictions which prevent them from being used by organizations other than government, military, and educational entities. Moreover, search ranking algorithms are unlikely to place URLs from entities with erroneously used TLDs higher in the search results. These factors allow the use of TLDs to categorize the nature of organizations fairly reliably. Next, each domain was categorized as retail (websites for the purchase of guns, ammunition, and gun accessories; including gun shows), news (websites of newspapers, news channels etc), educational (websites, regardless of TLD, that host information regarding gun safety, gun laws, gun maintenance, and may include websites of gun advocacy groups), showbiz (websites of movies, television shows, music videos, and so on) or “other” which included all remaining uncategorized websites.

The TLD and the content describe different characteristics of the same website and thus examining them together provides a richer understanding of the information seeking patterns. As such we created a variable that assigns a class to each website in the dataset derived from its content category and TLD. Thus, a website with retail content hosted by a commercial entity would be classified as “retail content, .com.” Finally, we created a variable to capture all of the websites owned or affiliated with the National Rifle Association (NRA) as listed on the NRA’s website [38]. Such websites were classified as gun rights advocacy groups. The NRA website also identifies other sites that it categorizes as “antigun lobbying organizations” [39]. We categorized these websites as gun control advocacy groups.

To evaluate the association between the Sandy Hook incident and the nature of information sought, we first examined the distributions of various characteristics of the domains visited by users following the search query (category of keyword, top-level domain, category of the website’s content, and advocacy view of the websites visited). Next, we investigated differences in website characteristics in the period before and after the shooting incident using the website classes. We also examined the percentage change in website visits for each of the characteristics relative to the total websites in the before period to those in the after period. Additionally, we examined the percent change in website visits for each of the characteristics in the after period to the website visit for the same characteristics in the before period.

Finally, it is possible that observed changes in information-seeking behavior over time may be due to the presence of secular or temporal trends and not as a result of the Sandy Hook shooting incident. For example, given that our study period overlapped with the holiday shopping season, one might expect an increase in Web-based shopping activity that can include increases in firearm-related searches, independent of the Sandy Hook incident. To differentiate the shopping activity related increase in search activity from that related to the shooting incident, we included a control query that would be agnostic to the trend due to the Sandy Hook incident but sensitive to the temporal trends of the holiday season. Thus, a query for “bicycle” (and related synonyms) was used as a control search term.

The following limitations must be noted. First, given that Yahoo! search accounted for about 12% of the US search engine market share in December 2012 [40], we recognize that caution must be used before generalizing to the entire US population. Additionally, the Web pages visited by the users may also be associated with result-ranking algorithms which vary by search engines. Since 2011, Yahoo! search is powered by Bing [41] and whereas the exact algorithms are proprietary, evidence suggests that Bing emphasizes keywords (search strings) in ranking search results [42]. Second, our analysis was focused on the query-level (ie, website visited after each search) and not the user level which may include several queries in a given search episode. Third, approximately 30% of all observations consisted of a large number of unique domains occurring with a low frequency and thus could not be classified. Nevertheless, these domains individually accounted for less than 1% of all observations and thus their effect on the findings is likely minimal. Finally, our work represents an exploratory study to examine whether search data can be used for a new purpose. Thus, the existing body of literature provided little guidance on the methods or approaches to analyzing such data. We recognize that future studies may identify additional techniques for analyzing similarly complex data.

One of the key findings of our analysis was that firearm-related searches more than doubled immediately after the Sandy Hook shooting incident in contrast to the control searches for “bicycle” which showed a small change with a decrease in the number of searches in the after period. This finding suggests that Web-based user search queries capture the immediate change in public interest following events of the nature of Sandy Hook shooting and can thus potentially serve as real-time indicators of the public psyche.

Overall, retail websites were the most visited websites following searches for gun types and ammunition. A salient finding was that gun type and ammunition searches had a 2-fold to 3-fold increase after the shooting incident. Furthermore, although it may seem natural to expect a greater interest in news articles following a major societal event, retail website visits had the highest and second highest increase after the shooting incident for gun type and ammunition searches, respectively. This finding may possibly be the result of a heightened interest in purchasing firearms and/or ammunition for one’s protection against the apparent public safety concerns raised by the mass shooting [43]. Additionally, it is possible that some individuals may anticipate an increase in regulatory control over access to firearms as a ramification of the Sandy Hook incident and as such prompt purchase of firearms before any such legislative action is passed.

Furthermore, there was a 6-fold increase in law-related firearm queries in the period immediately after the shooting. Importantly, these were the least likely searched terms in the before period and noted the greatest percent increase in the after period. This increased interest may be due—in part —to the purchase-related search or inquiry conducted by the potential firearm and ammunition buyers discussed above, not to mention the renewed interest in the gun-policy debate after Sandy Hook. Interestingly, most users seeking law-related information were interested in educational information and chose websites of noncommercial organizations, commercial entities, or educational institutions. From the advocacy perspective, more people visited websites of gun rights groups than did the gun control groups. However, despite gun control websites forming a lower proportion of all websites supporting an advocacy stance, they experienced the greatest percent increase from before to the after period. This trend was seen in all categories of firearm-related searches, with Web traffic to gun control advocacy groups exhibiting between almost 4-fold and 6-fold increase in the after period.

In addition to the trends discussed previously, a key feature of user searches and the subsequent URL clicks was that in all categories users were far less likely to choose content from a government entity. For example, even though the majority of the law-related searches are directed toward educational content, users are more likely to choose noncommercial organizations (including gun control or gun rights advocacy groups), commercial entities, and educational institutions as their preferred sources of information. The nature of advocacy groups is such that they exist to influence stakeholder decision to align with their agenda and therefore, the resulting conflict of interest may be an impediment to providing unbiased information. Thus, it is also likely that users seeking information about gun laws may obtain this information from websites of advocacy groups.

Our analysis of user search query data presents several key implications from a policy perspective. First, as stated above, user search queries present a valuable real-time indicator of the attitudes of the population as shown by the effect of the Sandy Hook shooting incident. In fact, the spike seen 6 days before the Sandy Hook event corresponds to 2 news stories: one on December 7, when supermarket employees found a handgun in frozen meat [44] and another on December 9, when a 7-year-old boy was fatally shot in the parking lot of a gun store [45]. Similarly, the spike seen around day 12 corresponds with the much publicized advocacy speech given by a prominent American sportscaster on television [46]. These spikes highlight user search queries as a timely measure of the public’s reaction to societal events. The time period immediately after a major event is characterized by heightened awareness and information-seeking behavior that may not be representative of public action during normal states (eg, buying firearms at twice the regular prices [43]). Indeed, Oh et al note that “rumormongering” is common after major societal events including shooting events [32]. On the one hand, this may indicate that policymakers should consider the timing of their actions noting that while a societal event can trigger interest in a topic, it ironically may not the best time to debate major tenets of policy change. On the other hand, some observed behavior may be due to fears arising from misinformation. For example, the increased purchase-related queries in our findings corroborate increased firearm sales due to fear of increased gun control legislation [43].

Second, it is possible that people are accessing information sources with either commercial or advocacy-related interest, at the same time being far less likely to choose content from government and educational institution websites. This may be because websites of government and educational institutions rank lower in the search results compared with those of commercial and advocacy interest groups. Although search engine optimization (SEO) may play a role in the higher ranking of commercial and advocacy interest websites, it is also possible the information presented by government and educational institutions may be less accessible. This may be due to suboptimal website design, jargon-filled language, poor SEO, lack of up-to-date information, and so on. Policy efforts should focus on providing reliable information as well as improved dissemination of this information by government institutions. Government entities may collaborate with educational institutions toward the creation of information portals focused on dissemination of accurate, timely, and high-quality information that is easy to understand. Furthermore, resources allocated toward making the public aware of these portals as well as on SEO may ensure that these websites rank higher in search results and thus visited more often.

Finally, the increased interest generated by the shooting incident appears to start tending toward normal levels around day 10, eventually returning to the levels before the shooting. This indicates that the increased interest generated due to incidents such as Sandy Hook presents a short window in which to form the public’s opinion. As discussed previously, this may not present the best opportunity to engage in public debate due to the increased anxiety and fear following these events. Whether this fear was driven by the need to protect oneself or the possibility of losing the right to purchase a firearm, it is unlikely that political sentiment for policy change will be easy to accomplish when fear is driving some stakeholder’s perspectives. Instead, policymakers should consider preemptively addressing some of the anticipated fears by implementing targeted campaigns that focus on specific groups of individuals. A recent US study reported that 3 percent of the US population owns nearly half of all firearms in the country with an average of 17 firearms each [43,47-49]. The median firearm ownership, however, remains at 1 to 2 firearms per owner. These individuals are likely to indulge in firearm purchases [43] after events such as the Sandy Hook shooting. Furthermore, given that personal protection against other people remains the most prevalent reason for firearm ownership in the US [47], mass shooting events may also motivate those on the fence to purchase firearms. As such, targeted campaigns that focus on these groups of individuals in order to allay fears and reduce reactionary purchase of firearms may help achieve some policymaker’s goals of lower rates of firearm ownership.

Our findings enabled us to identify directions for future research; web browsing choices and attitudes toward firearms may be affected by numerous other factors. As such, it may be valuable to examine the differences between attitudes toward firearms based on state characteristics such as political affiliation, socioeconomic status, and gun ownership. It may also be interesting to look at ordered queries nested within each deidentified user based on the order in which the user clicked each URL to provide richer data on users’ search intent. Search query data presents a valuable infodemiology metric of near real-time analysis of peoples’ attitudes and responses to major societal events. We believe future studies can employ the use of other search query datasets possibly with active user participation to examine the impact of society events over a longer period of time.