Policing through social networking

Police web presence

Despite the evolutionary potential of the Internet for police-community relationships, PWP initially spurred scholarly attention in terms of law enforcement goals such as the capture and deterrence of online criminals (Giacomello, 2004; Katyal, 2003; Luckenbill and Miller, 2008; Weimann, 2005). As Whitaker (1982) demonstrated in his meta-analysis of police time-use studies, law enforcement represents a small portion of an officer’s duties. Recent PWP studies demonstrate that researchers appreciate this time-use distribution. Rather than viewing the Internet as purely another law enforcement opportunity, scholars began trying to understand both the prevalence of police department websites and the types of content offered law enforcement.

Importantly, all research on the topic of PWP is inherently fixed in a point of time, while the evolution of police practices continues. As a result, it is important to contextualise each study in terms of the development of PWP. In one of the first studies of PWP, Dykehouse and Sigler (2000) used a snowball sample to determine the national prevalence of police web pages in the USA. Due to the lack of an existing database of this type of PWP, their methods were exploratory. They relied on police web pages known to house large numbers of links to other police websites and conducted snowball sampling with these links as their initial sample. After developing a list of websites, they randomly sampled 20 at a time to obtain a total sample of 500. Among this snowball sample, Dykehouse and Sigler found that only 6% of police and sheriff’s departments had a website. With a slightly more sophisticated method in the context of the ‘world wide web’, Tully and McKee (2000) challenged this estimate in their questionnaire study. They mailed 118 surveys to 55 large American police departments, obtaining a 55% response rate. Unfortunately, their admittedly ‘unscientific’ design called for submission of multiple surveys to the same department and failed to disentangle responses. Ultimately, 67 of the 68 respondents indicated that their department maintained a web page in 2000. While this is certainly an interesting finding, their design undermines the credibility of this estimate.

Over a decade ago, Barthe and Lateano’s (2006) research of New Jersey municipal police departments utilised a further refined method for determining website prevalence. Providing an independent assessment rather than relying on survey respondents or lists of police websites, they searched for official webpages of each law enforcement agency in New Jersey. Their research established (at least in New Jersey) that municipal police web pages are not a rarity, with 40.6% of departments utilising some form of official website in 2006. While previous scholarship laid the groundwork for independent assessment of police websites, Rosenbaum et al.’s (2011) study revolutionised the subfield.

Rosenbaum et al.’s (2011) analysis provided the first independent examination of police websites on a national scale. Using the 1999 Law Enforcement Management and Administrative Statistics (LEMAS) survey, Rosenbaum et al. found that 42.5% of the 666 sample departments maintained a web page in 2008. Once again, it is crucial to highlight the age of this data, as this national estimate is almost 10 years old. Following the first phase of their study, the authors examined various component parts of PWP that held relevance to COP goals and strategies. Their study used several COP measures from the LEMAS to explain the prevalence of police websites as well as variation in utilisation of input and output opportunities. Their legitimacy-seeking model stated that police departments with a strong commitment to COP in the non-digital realm will be more likely to maintain a website, a form of online community engagement. They also hypothesised that departments with greater COP investment will be more likely to provide input opportunities to the public. Finally, they incorporated some contextual variables relevant to the use of websites, including the technological sophistication of the department, the violent crime rate (operationalising political pressure to address crime) and the size of the civilian population (as a measure of the scale of the media market).

Utilising several measures contained in the LEMAS for COP commitment, Rosenbaum et al. (2011) created a multiple regression equation that predicted the presence of various PWP components using measures from the 1999 LEMAS. While these measures provided information from nearly a decade prior to their original data collection, Rosenbaum et al. (2011) found that their COP index significantly predicted the presence of a police website. In addition, the COP index predicted whether departments would offer greater input opportunities for citizens. However, their technological sophistication index was not a significant predictor of website presence. An interaction between COP commitment and technological sophistication was significant, indicating that COP commitment held greater predictive power for departments with a level of technological sophistication. Ultimately, their study demonstrated the potential for police websites as the next frontier of COP. Future work in the PWP field largely focused on police social networking.

Facebook, Twitter and YouTube all emerged as public social networking and media sites between 2005 and 2006, growing into some of the most popular Internet destinations. Worldwide, YouTube is the second most trafficked website, while Facebook ranks third, and Twitter places 14th (Alexa, 2017). YouTube and Facebook each boast over 1 billion monthly users (Facebook Newsroom, 2017; YouTube, 2013, 2017). Twitter has 313 million monthly users as of June 2016 (Twitter, 2017). As one of the initial dedicated studies of police social networking, Crump (2011) examined UK police force use of Twitter and follower statistics, and studied the prevalence of particular types of Twitter messaging (tweets), across a six-week time frame. Crump found that his random sample of five police forces was active in tweeting about information sharing (to and from the public), patrol activities and developing relationships with the public. While much of this activity clearly connects with COP, Crump was unable to analyse COP commitment as a predictor for the presence of Twitter among UK police forces. The London riots of 2011 provided the impetus behind some of the social networking research that followed Crump’s work (Panagiotopoulos et al., 2012; Procter et al., 2013), which focused on how the police used Twitter during the riots. More recently, researchers examined police use of Facebook with Project Eyewatch in Australia (Kelly and Finlayson, 2015) as well as the Toronto Police Service’s presentational strategies on Twitter (Schneider, 2016).

As with this study, several researchers addressed police social networking in the context of the USA. Brainard and Edlins (2014) examined the social networking of municipal police departments serving the 10 largest cities in the USA across a three-month period. Their research focused on the potential of police social networking to build on the COP model. However, due to the small sample size and data limitations, the authors were unable to test any relationship between police social networking and a department’s engagement in COP. Lieberman et al. (2013) provided a detailed content analysis of the 23 largest municipal departments participating in Facebook, studying the characteristics of their message content as well as the responses of citizens over a three-month period. While this new scholarship is invaluable and certainly broadened the field of PWP, this project builds on the wealth of recent research by exploring the relationship between social networking participation and COP commitment in a binary logistic regression model.

Community-oriented policing

Reisig (2011) argued that police departments commonly define and implement COP without sufficient structure, bringing any number of police practices under its umbrella. In this way, COP may function as a public relations campaign rather than an actual approach to policing. By reframing the same behaviours as community-oriented, the stamp of COP replicates the law enforcement excess and social distance of the Professional Era. However, failures in implementation are not necessarily due to a lack of structure provided by policing theorists. Despite criticising the loose interpretation and implementation of COP, Reisig (2011) cited a core viewpoint and strategic approach. Reisig (2011) used Cordner’s (1999) four dimensions of COP to provide a model with specific goals and strategies: (1) philosophical: engaging the community and requesting input; (2) strategic: encouraging face-to-face interaction with the public; (3) tactical: use of problem-solving tools; and (4) organisational: levelling the departmental hierarchy and sharing information with the public. In the field of PWP, Cordner’s (1999) work provides a useful theoretical touchstone to evaluate how different components fit or fail to mesh with a COP framework. Mastrofski et al.’s (2007) paper examining difficulties implementing COP found that funding and staffing issues plagued departmental efforts, as well as the fact that line-level officers were not always willing to embrace COP. Social networking provides an opportunity to engage with the public and share information without many of the resource-heavy requirements of tactics like foot patrol or citizen police academies.

Research questions and hypotheses

Rosenbaum et al. (2011) framed a department’s use of COP as legitimacy-seeking, arguing that maintaining a website was an opportunity to pursue COP-related goals such as community outreach and information sharing. However, COP in the digital realm represents much more than an information superhighway. Rosenbaum et al. (2011) conceptualised the link between COP and PWP, or COP-PWP, in terms of input (for example, citizen commendations) and output (for example, crime statistics) opportunities. Through this dichotomous framework, they were able to categorise many of the components of official municipal websites. However, they missed some of the potential provided by a non-dichotomous perspective. Online social interaction through social networking sites represents the key component in understanding the theoretical linkage between COP and PWP, allowing for dialogic interaction rather than separate and uninterrupted streams of information. The social networking component of PWP is dynamic instead of static, and allows for a mixture of police and the community unrivalled by any other segment of this larger phenomenon. Social networking provided a theoretically relevant context for testing Rosenbaum et al.’s (2011) legitimacy-seeking model, as a department’s identity management practices intermingle with myriad public responses. Departments willing to undertake this particular form of PWP demonstrate an openness to critique and the volatility of social networking in relation to the more controlled and static stand-alone website.

As a result, this study extends Rosenbaum et al.’s (2011) model to test one three-pronged research question: Do a department’s practices and characteristics predict their presence on Facebook, Twitter or YouTube? I divide this larger question into specific hypotheses. First, departments that emphasise the practices and characteristics related to COP will be more likely to maintain a presence on social networking sites (Hypothesis 1). Second, departments that demonstrate a focus on COP as well as technical sophistication will be more likely to maintain a social networking presence than departments with a focus on COP but without technological prowess (Hypothesis 2). Third, departments with a higher violent crime rate will be more likely to maintain a social networking presence due to political pressure (Hypothesis 3). As stated above, all of these hypotheses mirror those of Rosenbaum et al. (2011), but applied to the practice of police social networking rather than police stand-alone websites.

Sample

Very similar to Rosenbaum et al.’s (2011) research design, this study uses LEMAS respondents as a frame for studying PWP. This research replicates several variables of interest in Rosenbaum et al.’s (2011) study to best evaluate the merits of their model. The initial sampling frame relied on the 558 self-representing municipal police forces participating in the 2007 Law Enforcement Management and Statistics Survey (LEMAS), all departments with 100 or more sworn employees (US Bureau of Justice Statistics, 2007). While this data has obvious drawbacks in terms of the seven-year gap between the data concerning the independent variables and those for the dependent variables, this was the most recent LEMAS data available during the start of the project in February 2015. However, following a large portion of the data collection and analysis, I became aware that the 2013 LEMAS data had become available (US Bureau of Justice Statistics, 2013). As a result, this study also examines a reduced sample of departments who appeared in both sets of data (n = 381), to provide a shorter latency between the key independent variables in the legitimacy-seeking model and the outcomes relating to departmental participation in these three social media sites.

Procedures

I gathered information concerning the existence of a departmental Facebook, Twitter and YouTube account for each of the 558 departments in the 2007 LEMAS sampling frame from February to October 2015. Relying on a Google search of the first 100 results with the city, state, appropriate law enforcement term and social network (e.g., ‘San Jose, CA, police, Facebook’), I located existing social networking pages associated with the department. To merit inclusion as a police social networking site, the page needed to be identity-verified with the respective social networking site or provide a link to a stand-alone departmental website. Departments with an account on one or more of these sites also needed to post at least one piece of content in 2014 to be considered active. Activity included a video (YouTube) or statement ‘tweet’ (Twitter) or ‘post’ (Facebook). This data collection provided binary measures concerning each of the 558 departments’ presence on Facebook, Twitter and YouTube. As discussed above, this variable and aspect of PWP is theoretically linked to the COP philosophy of community engagement and request for input as well as the organisational shift in providing information to the public (Cordner, 1999; Reisig, 2011).

2007 LEMAS secondary data

The 2007 LEMAS provides several COP-related questions, similarly to Rosenbaum et al.’s (2011) set of questions from the 1999 LEMAS. As it was impossible to perfectly recreate Rosenbaum et al.’s (2011) COP index due to changes in questions between 1999 and 2007, I constructed a new index from 11 dichotomous questions from ‘Section IV – Community Policing.’ I strove to be as inclusive as possible and mirror Rosenbaum et al.’s index, only eliminating questions under the LEMAS section that provided low inter-item reliability assessed through Cronbach’s alpha testing or that were not adaptable to a dichotomous framework. The questions included whether a department maintained a citizen academy, had officers patrol geographic beats, provided a mission statement with COP language, partnered with citizen groups for feedback on COP strategies, maintained a formal COP plan, included collaborative projects in evaluation of patrol officers, encouraged officers to conduct problem-oriented ‘scanning, analysis, response, assessment’ or SARA projects, conducted a survey on citizen satisfaction, crime or fear of crime, maintained a full-time COP unit, trained all recruits in at least 8 hours of COP, and trained half or more officers in-service in COP (m = 6.92, SD = 2.46, Cronbach’s alpha = .70). There were missing data for 3.76% of cases.

Once again, due to changes in questions between the 1999 and 2007 LEMAS, I also constructed a technological sophistication index from 14 dichotomous questions from ‘Section VI – Equipment’ and ‘Section VII – Computers and Information Systems’. As before, I tried to replicate Rosenbaum et al.’s (2011) index as closely as possible and include all technology-relevant questions in my index. I eliminated questions that significantly weakened the index in terms of inter-item reliability or would not modify to a dichotomous framework. The questions include whether a department provided automated booking, used gunshot detectors, or relied on computers for the purposes of dispatch, community problems analysis, crime analysis, crime mapping, fleet management, hotspot identification, in-field communications, internet access, crime investigations, personnel records, records management and resource allocation (m = 10.43, SD = 2.46, Cronbach’s alpha = .75). Following index construction, I excluded three outlier values for this variable that were more than three standard deviations from the mean. In addition, the 2007 LEMAS data provided a civilian population estimate from US Census data used to group departments as ‘large’ (200,000 or more citizens), ‘medium (75,000–199,999 citizens), or ‘small’ (74,999 or less).

2013 LEMAS secondary data

The 2013 LEMAS also provided a wealth of questions concerning COP and technological sophistication, but did not mirror the 2007 questions, so complete replication of the 2007 indices was not feasible. I constructed a new COP index from seven dichotomous questions from ‘Section E – Community Policing’ concerning whether a department provided a mission statement with COP language, trained all new recruits in eight or more hours of COP, trained half or more of officers in-service in COP, encouraged officers to conduct problem solving, used collaborative projects as part of officer evaluations, regularly assigned officers to the same geographic beats, and utilised data from a community survey concerning crime, fear of crime or satisfaction (m = 4.50, SD = 2.26, Cronbach’s alpha = .55). There were missing data for 4.46% of cases. I also constructed a technological sophistication index from nine dichotomous 2013 LEMAS questions from ‘Section F – Technology and Information Systems’ concerning content available on the departmental website (as all sample departments stated they provided a website): jurisdiction-wide crime-related information, crime statistics by beat, street-level crime maps, sex-offender maps, crime reporting for civilians, feedback or questions for civilians, agency/officer complaints for civilians, crime reporting or reporting other issues through email or texting, crime or other information available to civilians through email or texting (m = 5.37, SD = 2.32, Cronbach’s alpha = .70). There were missing data for 5.25% of cases. As with the 2007 survey, the 2013 LEMAS data provided a civilian population estimate from US Census data used to group departments as ‘large’ (200,000 or more citizens), ‘medium (75,000–199,999 citizens), or ‘small’ (74,999 or less).

Uniform Crime Reports 2010–2012

As the Uniform Crime Reports’ operationalisation of rape changed in 2013 to provide a more inclusive and gender-neutral definition focusing on the act of penetration (FBI, 2014), I focused on 2010–2012 UCR violent crime rates. The US Bureau of Justice Statistics (2010) provides an online data tool that includes violent crime rates for each department. Using this data tool, I averaged the violent crime rate across each of the available years for each department. There were missing data for 5.38% of cases.

Binary logistic regression analyses

In order to predict the presence or absence of police social networking, I used the predictor variables of the civilian population, the UCR violent crime rate average 2010–2012, the COP index, the technological sophistication index and an interaction term of COP index x technological sophistication index. I used SPSS Version 22 (IBM Corp, 2013) to analyse the data and provide the descriptive and inferential statistical analyses. I provide the binary logistic regression equation used for each form of police social networking below:

Logit (p) = b_o + b₁ x₁ (small population) + b₂ x₂ (medium population) + b₃ x₃ (violent crime average) + b₄ x₄ (COP index) + b₅ x₅ (technological sophistication index) + b₆ x₄ * x₅ (COP index * technological sophistication index)

Descriptive findings

The vast majority of departments (88.35%) maintained a social networking account and were active on at least of the three social networking sites in 2014 (see Table 1). Lieberman et al. (2013) found that 32 of the largest 61 departments (52.46%) in the USA utilised Facebook. In this more contemporary and expansive analysis, 69.53% of the 558 departments maintained a Facebook account in 2014, with 64.70% on Twitter, and only 13.08% on YouTube. To my knowledge, no previously published research examined a variety of American police social networking participation with a large, nationally representative sample. Importantly, these figures are likely in flux as departments begin to embrace the digital age and this particular form of communication.

OPEN IN VIEWER

Table 1.

Frequency and percentage of departments using social media sites (n = 558).

Type of social networking site	Frequency	Percentage of departments
YouTube	73	13.08%
Twitter	361	64.70%
Facebook	388	69.53%
One or more sites	493	88.35%

Binary logistic regression findings

The overall models, with the control variables of civilian population and average official violent crime rate, as well as COP index, technological sophistication index and an interaction term of COP index x technological sophistication index provided Nagelkerke pseudo-R² values including .06 (Facebook), .09 (Twitter) and .11 (YouTube) for the 505 cases with complete information (9.50% of cases with missing data on at least one variable). The small population variable was a statistically significant predictor across all three models, with the medium population variable only significant in the YouTube and Facebook models at alpha = .05. This replicates Rosenbaum et al.’s (2011) findings about the significance of large media markets in predicting whether departments will rely on PWP. There is no support for Hypothesis 3, as the violent crime average was not a statistically significant predictor in any of the three models.

Moving to the variables of particular interest to Rosenbaum et al.’s (2011) model, neither the technological sophistication index nor the COP commitment index were statistically significant predictors of any of the three types of police social networking (See Table 2). Because social networking and media sites like Facebook, Twitter and YouTube are free-to-join and designed for popular use, it appears that a lack of technological sophistication did not greatly undermine departments who want to provide an online social presence. Not surprisingly, the interaction terms were also non-significant across the three models. As a result, this study rejects both Hypothesis 1 and Hypothesis 2. Rosenbaum et al.’s (2011) model for predicting PWP with departmental practices and characteristics does not receive empirical support through this extension.

OPEN IN VIEWER

Table 2.

2007 LEMAS logistic regression results (n = 505).

	YouTube	Twitter	Facebook
	Odds ratio	Odds ratio	Odds ratio
Small population	.15***	.26***	.25***
Medium population	.32***	.58	.34**
Violent crime rate average	1.00	1.00	1.00
COP index	1.02	1.08	1.02
TECHSOPH index	1.02	.97	1.02
COP x TECHSOPH	.73	1.11	1.01
Cox and Snell R2	.06	.07	.04
Nagelkerke R2	.11	.09	.06

*p ≤ .05. **p ≤ .01. ***p ≤ .001.

As a safeguard against issues concerning the time gap between the information in the 2007 LEMAS and 2014 departmental behaviour, I also pared down the sample to departments present in both the 2007 and 2013 LEMAS datasets. A good deal of the sample and some statistical power was lost in this process. However, I am left with 324 cases, which provides more than sufficient power to test Rosenbaum et al.’s (2011) model (Soper, 2015). Shortening the latency between the independent variables and dependent variables did not significantly alter the findings in terms of individual predictors or the amount of variability predicted by the models (see Table 3). None of the predictors specified in Rosenbaum et al.’s (2011) model receive empirical support with the 2013 LEMAS data, and I reject all three hypotheses. As before, the control variables concerning the size of the civilian population demonstrate significance and remain the only source of valuable predictors.

OPEN IN VIEWER

Table 3.

2013 LEMAS logistic regression results (n = 324).

	YouTube	Twitter	Facebook
	Odds ratio	Odds ratio	Odds ratio
Small population	.25**	.35**	.23***
Medium population	.41*	.65	.35**
Violent crime rate average	1.00	1.00	1.00
COP index	1.43	1.25	.56
TECHSOPH index	1.07	1.05	1.10
COP x TECHSOPH	1.15	1.08	.89
Cox and Snell R2	.04	.05	.07
Nagelkerke R2	.08	.07	.10

*p ≤ .05. **p ≤ .01. ***p ≤ .001.

Study limitations

This study is limited in a number of ways. As discussed above, the data is inescapably a snapshot of police social networking participation, and one prior to several controversial use of force incidents. In addition, I focus on three of the most popular social networking/media sites. However, I did not also include other sites such as Instagram, Snapchat, Tumblr or Google+. While this is the first study to examine a breadth of police social networking beyond a single site, the findings may vary with an even more inclusive perspective on police social networking. Additionally, due to data collection limitations and the large sample size, this study does not include a measure of the amount of engagement between citizens and the department, focusing on whether a department maintained a presence. Rosenbaum et al.’s (2011) model may perform better when operationalising the dependent variable as level of engagement. Lastly, the sample size reduction when examining departments in both the 2007 and 2013 LEMAS is not ideal. While the more recent data provided the same significant and non-significant relationships as the 2007 data, the model using the 2013 LEMAS data should be viewed with additional caution.