A multiplatform investigation of law enforcement agencies on social media

2.1 Law enforcement agencies on social media

The unique characteristics of law enforcement agencies inform and impact their day-to-day social media usage [20]. First, law enforcement agencies operate in a complex institutional milieu that is framed by federal, state, and local laws, policies, and legal guidance. Second, much of what they do requires gathering, not sharing, information. As such, law enforcement agencies often follow a “need-to-know” information sharing paradigm: agencies publicize operation updates when obligated by law in order to inform and educate the public or when in need of public collaboration to respond to public safety issues [21]. These updates usually appear in the form of press releases, after-action reports, or other formal statements, all of which go through a multi-step vetting process before being shared with the public [22, 23].

However, as they increase their use of social media, law enforcement agencies must adapt rather quickly to the instantaneous and multi-directional open-dialog environment that this new media type encourages, e.g., [24]. Meanwhile, given the potential for two-way communication, they must consider how to respond to the public [25]. Agencies were also found to be at varying stages of adopting social media in their daily practices; some already had social media policies and some were still developing theirs [26].

After investigating the top 10 U.S. municipal police departments’ social media usage over three months, Brainard et al. showed that police agencies tweeted more frequently on Twitter than they posted on Facebook or YouTube [25]. Considering Twitter, popular for information sharing [16], provides law enforcement agencies an opportunity to propagate accurate and timely information to a large Internet audience [27, 28], we propose our first hypothesis as follows:

2.2 Public relations theory and social media strategies

Social media is an efficient way to cultivate and maintain successful relationships for organizations [29]. In literature, the management of communication between an organization and its public is discussed as public relations [30]. There are four models of public relations management: (1) press agentry, one-way communication using persuasion and manipulation to influence audiences so that they behave as the organizations desire; (2) public information, another one-way communication strategy, focusing on distributing an organization’s information to the public; (3) two-way asymmetrical, aiming to collect information that can be useful for an organization; and (4) two-way symmetrical, which uses communication to enhance understanding or resolve conflicts. The excellence theory of public relations emphasizes that two-way legitimate conversation would be more effective to promote the relationship [31]. However, government agencies mainly employed one-way communication to distribute information [32].

Similar to the models of public relations, researchers also identified three social media strategies agencies commonly used in their daily practice, i.e., Push, Pull, and Networking, which correspond to one-way communication in terms of public information, two-way asymmetrical and two-way symmetrical respectively, with the intent being to use different combinations of these strategies [33, 14]. The one-way press agentry model serves a function of persuasion or propagating the belief of an organization by using misrepresented or emotional information [34], which may not be aligned with the interest of law enforcement agencies who aim to provide legitimate services.

More specifically, in a Push strategy, social media channels are used as an extension of the existing Internet properties (e.g., static websites) and traditional public relations techniques (e.g., press releases or conferences). Here, social media sites are used as an additional communication channel to get the message out. This tactic is most often used to publicize information that the organization is obliged to share with the public [14, 35].

In a Pull strategy, the organizations specifically add a call to action in posts and request that the public either directly interact with the organizations or amongst themselves. Social media applications are used to bring audiences back to an organization’s website, where the news is aggregated to avoid losing control over the information. For example, these messages will include a call to like a Facebook page, share updates with others, or even for direct input (such as information regarding an active AMBER Alert or the identification of a suspect in a robbery). Pull tactics actively involve audiences using some degree of interaction that typically results in several comments on Facebook walls and a few shares [14, 35].

A Networking strategy aims for a more informal interaction with social media users and highlights the “human” side of the organization; it also promotes trust, manages conflicts and builds up a long-term relationship between the public and agencies. Our work provides empirical evidence showing which topics the agencies post about on Facebook when using these three different strategies [14, 35].

Studies prior to 2013 found that crime related information (one-way and Push) was posted more frequently on Facebook [17] and Twitter [18] than other types of topics, whereas a more recent study [36] pointed out that there was a shift from posts containing crime related information to posts that could foster public (Networking) interactions on Facebook. This transition happened along with the effort of these agencies in building community policing [37] so as to promote legitimacy, which was defined as “the rightful exercise of authority or use of power” [38]. However, when we case studied four municipal police agencies using their Facebook posts in 2014, three of the four agencies had more Push posts than Networking posts [10], albeit the sample data was too small to claim whether the difference was significant. The finding would also have been limited to one social media platform, i.e., Facebook. In this study, we compare agencies’ application of different strategies on both Facebook and Twitter. The above literature suggested and led to us developing the following hypothesis:

H2. Agencies employ the Push strategy more frequently than Networking and Pull strategies both on Facebook and Twitter.

2.3 User interaction on social media

Extensive studies have been conducted to understand how different factors could impact user interaction on social media [39, 40]. For example, on Facebook, prior work found that police agencies’ Facebook posts that contain pictures received more interactions from public users [10]. On Twitter, features such as content length, URLs, hashtags, and pictures included in the message have been identified as factors influencing the received interactions from users, e.g., a tweet with URLs or hashtags is more likely to be retweeted [39, 40]. Velde and colleagues found that including URLs, mentions and hashtags or writing longer tweets increased the probability of being retweeted [41].

The impacts of content characteristics on user interaction were also investigated. Berger et al. found that the effects of online information transmission were associated not only with emotions (i.e., positive content had a greater likelihood of diffusion than negative content) but also with the extent to which the content invoked strong emotions (i.e., the content that invoked awe or anger emotions was more likely to be share than the content that invoked sad emotions) [42]. For advertising, more persuasive content (i.e., emotional and philanthropic content) on Facebook received more user engagement in terms of the number of likes and comments; however, informative content (i.e., prices, availability and product features) had a negative impact on engagement [43]. Zavattaro et al. found that the U.S. local governments who used positive tones in tweets (e.g., utilizing exclamation points, creative hashtags, retweeting positive tweets) were more likely to promote citizen participation than those who only used neutral tones when sharing information in a Push manner [44]. This was also suggested by a prior qualitative study, where an interview study was conducted to examine the purposes of different agencies’ social media strategies; Push messages were not expected to receive as many interactions as the other strategies [45]. In this paper, we apply quantitative methods to test the following hypothesis:

H3. Pull and Networking social media messages receive more user interactions than Push messages.

Public users’ interactions with agencies on social media, in terms of likes, favorites, shares, retweets or comments, are indicators of public users’ engagement and responsiveness towards police-initiated content on social media [17, 46]. Likes and favorites indicate emotional, social or even habitual engagement [47]; shares and retweets denote the level of information diffusion of police-initiated social media content; and retweets and comments represent public users’ level of conversational engagement [48]. Using 10 municipal police agencies’ social media data, more interactions were found on Facebook than on Twitter, and Facebook was more likely to trigger dialogue interaction between police agencies and residents [25]. A similar observation was made by researchers who investigated social media usage of Canadian and American public transportation agencies, where notably higher counts of likes, shares and replies were found on Facebook than their counterparts on Twitter [49]. In this study, we are able to test the following hypothesis by comparing Facebook likes with Twitter favorites, and Facebook shares with Twitter retweets received by the same message that was sent by the same agency:

H4. When the agency sends the same message on both Twitter and Facebook, the message on Facebook receives more interactions from the public.

3.1 Data collection

The data was collected using Facepager [50]. For each post and tweet, both the Facebook API and the Twitter API provided the text, the count of interactions (i.e., likes, comments, and shares for Facebook, favorites and retweets for Twitter), creation time, etc. The data was collected in February, 2016. Eventually, a total of 26,490 Facebook posts sent by the 40 agencies during the year of 2015 were collected. They received 10,767,117 likes, 715,901 comments, and 2,726,192 shares. For the same 40 agencies, we collected a total of 41,277 tweets sent in 2015. They received 607,758 favorites and 944,909 retweets.

3.2 Coding

In our prior work, we developed a code scheme for classifying police agencies’ Facebook posts [10]. The first level, post category, contains three codes (Push, Pull and Networking), which describe the general purpose of a post. The three categories were developed based on the previous work [14, 35]. The second level includes eight codes representing different post topics. Push contains 3 topics: Traffic, Crime, and Announcement. Networking contains 4 topics: Tip, Personnel, Information, and Appreciation. Since the Pull posts are about requests for various types of information (e.g., missing persons or criminals wanted) and there were very few Pull posts in total, we combined them into one topic: Request. Below we show the definitions of the codes and their sample posts.

Push: Agencies use social media as an additional broadcasting channel to perform their obligatory missions [14, 35]. Three topics are identified: (1) Crime: explicitly conveying information about a crime incident. The crime incident can be related to shootings, homicides, arrests, victims, guns, drugs, etc. (e.g., “Stabbing: [redacted], dispatched at 12:59 am. Adult male was stabbed and later died.” – BPD); (2) Traffic: information related to road conditions, such as real-time traffic, road construction alerts, and expected traffic delay alerts (e.g., “The Cleveland Division of Police is reporting that traffic signals are out in the area of Superior, Chester, Carnegie & Euclid from E. 55th to E. 89th. Cleveland Police are responding to the area in connection with traffic.” – LAPD); (3) Announcement: communicating non-Crime and non-Traffic information (e.g., “The Cleveland Division of Police First District is conducting a DUI Checkpoint in the area of Rocky River and Puritas tonight, November 11, 2014.” – CPD).

Pull: Agencies leverage social media platforms to acquire information from the public [14, 35]. All posts sent with this strategy are classified into one topic, Request: asking the public to provide information about urgent issues, such as identifying a crime incident, finding a missing person, or obtaining critical feedback (e.g., “Help us identify a person of interest in a robbery investigation: Jan 14th [redacted]. Call [redacted] w/info.” – BPD).

Networking: Agencies try to build long-term relationships and benign interactions with the public [14, 35]. Four topics are identified. (1) Tip: communicating suggestions to improve community members’ lives and well-being (e.g., “Very high temperatures expected throughout the region this week. Remember to properly care for pets!” – Cleveland Police); (2) Personnel: information that addresses individual names of police department employees (e.g., “My name is Officer Brown. I’ve worked for the BPD for 24-years. I’m currently assigned to the Southern District.” – BPD); (3) Appreciation: information that expresses gratitude and appreciation (e.g., “In honor of all who have served and those who continue to serve our country. We take pause to reflect on your commitment and dedication. Thank you” – CPD); (4) Information: including all other non-Tip, non-Personnel, and non-Appreciation information (e.g., “Plz Support/Follow our partners at @LAPDHQ” – BPD).

We then applied a similar coding process and annotated the Twitter data. We first randomly selected 500 tweets from the collected tweets. We reviewed each tweet and classified the 500 tweets into three categories (i.e., Pull, Push, and Networking). Then within each category, we extracted the topics to develop the second level of code. We analyzed emerging topics and created a list accordingly. Finally, we merged similar topics and produced a final list of mutually exclusive topics from each category. As a result, the final coding scheme contains the same 8 topics as we had developed using the Facebook data.

Once the coding scheme was finalized, we hired 3 coders and trained them to understand the coding rules and concepts. After the training, two coders started to independently code the rest tweets. They first labeled these tweets into the first coding level. We computed the Cohen’s Kappa to evaluate the inter-coder agreement, defined by Viera et al. [51]. The results showed almost perfect agreement (0.81 ∼ 0.99) for each category, and topics had substantial agreement (0.61 ∼ 0.80). We then asked the third coder to label the posts that the first two coders had disagreed on independently and resolved the disagreement with the first two coders together using the “majority rule” approach. Three coders then labeled the posts into the second coding level using the same methodology. The Cohen’s Kappa value for each code is above 0.80.

Using 6,823 manually annotated Facebook posts, we executed the SVM algorithm in Scikit-learn method to classify 26,490 posts sent by 40 agencies, and the accuracy was 76.66%. We also manually annotated 9,837 tweets by using the same code scheme, and classified 41,277 tweets sent by 40 agencies; the accuracy was 80.35%.

After classifying all the agencies’ messages, we employed different statistical methods to analyze the data for different purposes. First, we performed a Grubbs’ test to examine whether there were any outlier posts in terms of likes, comments, shares, favorites and retweets that may skew our analyses [52]. We then excluded the outlier posts/tweets before conducting subsequent analyses.

4.1 Agencies’ posts and tweets

As an overview, Table 1 shows the agency names, the average number of daily posts and tweets each agency made on Facebook and Twitter.

Before we examined whether Pull and Networking social media messages received more user interactions than Push messages and the message on Facebook received more interactions than Twitter, we performed Grubbs’ tests to find out whether there were any outliers. The Facebook posts contained 1,938 like outliers, 1,720 comment outliers and 2,334 share outliers, each of which accounts for about 8% of the whole dataset. Similarly, for the Twitter tweets, we also performed Grubbs’ tests, and found 2,948 favorite outliers and 2,806 retweet outliers, each of which accounts for about 7% of the whole dataset. We removed the outliers when conducting the following comparisons.

5.1 Reflections on the findings

Agencies were significantly more active on Twitter than on Facebook (Finding 1), but the average interactions received on Twitter were significantly less than on Facebook (Finding 4). It is possible that the negative impact of high frequency posting contributed to the lower user interaction rates [54] that Twitter received compared to Facebook.

Our findings also indicated that agencies turned to Facebook for much longer posts with richer information than they would tweet on Twitter, and those posts and tweets that they shared on both platforms were confined by the character count constraint imposed by Twitter. We took a good sample of non-paired (i.e., unique on one platform) and paired (i.e., shared on both platforms) posts and tweets sent by each agency. There was more rich information in the non-paired posts and tweets than the paired ones.

The fact that these unique and longer posts on Facebook received more interactions suggests that agencies should not restrict their posts to follow the norm of social media, where posts are typically short. They should provide more details if necessary. For example, a paired Pull post from the Cleveland police agency wrote “WANTED for Bank Robbery: Demar Bradford, black male age 32. Know this guy? Call Fifth District investigators at 216-623-5518,” whereas the agency’s non-paired Pull post provided more detailed information about the victim, e.g., “MISSING FEMALE ENDANGERED the Cleveland Division of Police Domestic Violence Unit is currently seeking the public’s help in locating a 20 year old female. This female is a victim of Domestic Violence and was last seen on October 25, 2015. She was reported missing on November 9, 2015. Angela Lochan is described as a 20-year-old female of West Indian decent. She is 5’9” tall with brown hair and brown eyes. Anyone with information regarding her whereabouts is asked to contact detectives at 216-623-5864.”

Similarly, a non-paired Traffic post from the Columbia Police Department not only informed the traffic condition but also gave suggestions: “***TRAFFIC ALERT*** The Columbia Police Department is on the scene of a broken gas line on Nashville Hwy between Bear Creek Pike/Hwy 43 intersection and Burt Dr. Motorists are encouraged to seek alternate routes and avoid this area if possible. Expect heavy traffic and detours in this area. PLEASE USE CAUTION WHEN DRIVING THROUGH RESIDENTIAL AREAS as DETOURS!!! Drive the posted speed limit, CHILDREN at PLAY. UNKNOWN when traffic flow will return to normal but we will update this post at that time.” On the other hand, a paired traffic post stated: “TRAFFIC ALERT UPDATE – Nashville Hwy is back open to normal traffic flow. Thank you for your patience. 7:25 pm.”

We examined the law enforcement agencies’ social strategies through the lens of public relations theory. Our early work of case studying four municipal police agencies [10] showed that Pull strategies were effectively in receiving interactions from the public. This study of 40 agencies’ social media messages re-confirmed the finding across two platforms (Finding 3). However, most of the agencies sent very few Pull messages on either of the platforms, and some agencies did not apply Pull strategy at all (Finding 2). This empirical evidence suggests that agencies’ under-use of social media for one-way asymmetric communication [30].

Agencies made more Push posts on Facebook and more Networking tweets on Twitter (Finding 2). This was a surprising finding since previous studies found that Facebook was regarded as a social network platform for image promotion and Twitter was more news-oriented and valuable for informing the public and for urgent warnings [17, 18, 55, 56]. A practical implication of our finding for police agencies could be to balance their use of social media strategies by sending more Networking messages on Facebook and Push tweets on Twitter.

Our findings also have theoretical contributions. According to Milgram’s famous obedience experiment, people tend to be persuaded by others with status and authority [57]. A legitimate source of authority could make this persuasion even stronger [58]. Our findings show that posts and tweets containing officers’ names received significantly more likes on Facebook and favorites on Twitter, and were shared and retweeted more than many other topics, could be explained in part by these theories. For instance, NYPD posted “Winter Storm Information and Safety Tips Follow @NYPDNews, @NYCMayorsOffice, @NotifyNYC and @NYCoem on Twitter for the latest updates on the expected winter storm. – Drivers are urged to stay off New York City roads – Call 911 for emergencies, 311 for non-emergencies and loss of heat – Use only portable heating equipment that is approved for indoor use More Safety Tips: http://www.nypdnews.com/2015/01/winter-storm-information-and-safety-tips.html” and received 4,288 likes, 170 comments. The post was shared 516 times. Posting and tweeting about officials, detectives or staff members, even simply including their names, might help promote positive sentiment and legitimacy on social media.

5.2 Limitations and future work

The data used in this study is a snap shot of agencies located in large cities at a certain time. It is likely that police agencies altered their social media posting practices over time for different purposes. Our data can be used to examine if certain strategies received more interactions, however the data cannot be used to claim causal relationships. To fully examine the causality, we would need to apply a triangulation approach, e.g., by interviewing the agencies, collecting their policies, organizational factors as well as other environmental factors such as cultural differences.

In this study, we focused on the top level posting and did not analyze replies to the posts. Public users’ responsiveness, especially the conversational responsiveness, indicates the potential for further collaboration between police and citizens on social media [25]. In the future, we will expand our comment analysis to further understand conversations between the agencies and the public on social media platforms. Studies on how social media is used differently during daily operation versus crisis situations could be developed by integrating our codes with the code themes for crisis management [59, 60].