Literature Review: Google Analytics, My Beloved OoS
In general, researchers appear to use Google Analytics™ (GA) web analytics service as a tool for measuring web visits and, to an extent, visitor behavior. In discursive terms, GA collects and visualizes an archive of traces of user interactions with web pages. The discursive activity of visiting (and, presumably, reading) a web page is seldom referenced in research that uses GA for measurement; instead, the archival trace of the discursive activity gets captured, archived, and visualized.
Most research uses an enthymeme that reads something like this: GA data can help developers improve websites. For example, Kirk et al. (2012), in an article seeking to monitor user engagement in an Internet-delivered genetics education resource developed for nurses, report that GA “informs approaches to enhancing visibility of the website; provides an indicator of engagement with genetics-genomics both nationally and globally; [and] informs future expansion of the site as a global resource for health professional education” (p. 559). Similarly, Mc Guckin & Crowley (2012), in an article evaluating the impact of an online cyber-bullying training resource, the CyberTraining Project, report that GA data have “allowed for the project team to further understand how best to optimize the product (i.e., the Website and the eBook) for ease of access and navigation by unique and referred users” (p. 629). Focusing more specifically on GA reporting over time, Plaza (2009) notes that “GA tells the web owner how visitors found the site and how they interact with it. Users will be able to compare the behaviour of visitors who were referred from search engines and emails, from referring sites and direct visits, and thus gain insight into how to improve the site’s content and design” (p. 475). Missing from the enthymeme are assumptions that connect GA to improved websites, assumptions that can be phrased in questions about the relationship between GA, website visitors, and website developers: What data are provided by GA that can directly relate to specific improvements in website design? What user behaviors can and should be examined via GA to evaluate the success of the website? What benchmarks should developers set to measure success or failure? While these questions are not ignored in research that uses GA reporting, they are not directly or specifically addressed. As a result, readers miss out on key assumptions that researchers make about specific ways the data provided by GA reports can and will be used to make concrete changes to website design and structure.
Bruno Latour’s (2005) introduction to actor-network-theory (ANT) identifies transporters of meaning among connections as “mediators” or “intermediaries.” An intermediary “transports meaning or force without transformation” while mediators “transform, translate, distort, and modify the meaning or the elements they are supposed to carry” (p. 39). When researchers present GA as a means of measuring user interaction with websites, they generally describe GA as an intermediary. By describing GA as an intermediary, researchers ignore, potentially to their peril, the mediating potential of GA reports. For example, Dahmen & Sarraf (2009), reporting visitor analytics of an online art museum exhibition, claim that “through the use of Google Analytics, this research seeks to understand how the public used the Web representation of the special exhibition” (p. 2). Their report represents GA data as authoritative and unmediated; the GA interface that visualizes and reports visit data is accepted as accurate, without comment. Mc Guckin & Crowley (2012) take a step toward recognizing the potential mediating effects of GA reports by claiming to “ascertain the efficacy of GA as an effective resource for measuring the impact of the CyberTraining project” (p. 628), but they conclude, “Such information [provided by GA] proves valuable in the iterative development and dissemination of the project and has, directly, informed the planning of the new CT4P project” (p. 629). GA is considered a blackboxed intermediary for reporting web visits. In other words, current research offers little theoretical perspective on the potential mediating effects GA may have on the data it reports and visualizes. This blog post seeks to remedy that omission by applying both ANT and cultural-historical activity theory (CHAT) to Google Analytics and the data it provides on visitor interactions with the website of the University of Richmond School of Professional and Continuing Studies (SPCS).
An OoS on the LOoSe
One of the most interesting aspects of using GA as my object of study (OoS) is that it remains a product continually in production. Although Google does not address it explicitly, it’s become clear that Google is working to make GA a digital analytics platform that expands well beyond the measurement of interactions on websites. I’m working toward a certificate of completion for Google Analytics Platform Principles (2014) as a followup to a certificate of completion I received for Digital Analytics Fundamentals (2013), and both of these online learning modules address Google Analytics as a broad-based digital analytics platform that handles data from a wide array of sources, even non-Internet-connected applications and appliances. The result, as I’ve experienced it, is that the Google Analytics Platform (yes, that’s the proper noun) is expanding its reach and scope on a weekly, perhaps even daily, basis.
This makes applying activity theories like the cultural-historical remapping of rhetorical theory (CHAT) and actor-network-theory (ANT) quite comfortable. GA as OoS is itself in active flux, continually redefining (perhaps more accurately expanding) itself for a fast-changing connected world.
ChOoSing a Definition
CHAT might describe GA as a representation of practices within a laminated chronotope. As a tool that measures interactions between visitors and web pages, GA collects the results of “mediated activity:… action and cognition [that] are distributed over time and space among people, artifacts, and environments and thus also laminated, as multiple frames or fields co-exist in any situated act” (Prior et al., 2007, emphasis original). The action that gets represented as a visit in GA is loading a specific web page. Cognition gets represented in the action of following a link on a specific page to load a new page or resource. This activity is collected over time in a session, defined in GA as the time within 30 minutes a single visitor, identified by an anonymous, unique identifier and saved in a first-party cookie (“Platform Principles,” 2014) remains engaged within a surveilled website before leaving that domain or expiring the session time. GA represents all of the activity within that session in an aggregated visualization. Session data are collected over time and are the result of laminated activity among people, artifacts (like web pages) and environments (like browsers, computers, mobile devices and the like).
ANT might describe GA as traces of connections among networked actants. Actants captured in a web session might include the visitor, the technological interface (computer/mouse/monitor or mobile device), the web page content and links, the writer of the web content, the host server, the network gateways and cables, and many more too numerous to detail. ANT would likely chafe under the need to define the collecting mechanism itself, however, and suggest that GA might be an artificial data assemblage that needs to be reassembled. Specifically, since GA is a data framework that collects only preselected data points (“Tracking Code Overview,” 2012), GA might be accused of “filtering out” and “disciplining” the data collection: “Recording not filtering out, describing not disciplining, these are the Laws and the Prophets” (Latour 2005, p. 55, emphasis original). More useful might be the preprocessed data collected by Google Analytics servers; processing organizes the web session into a predefined framework, precisely the activity ANT seeks to avoid in its practice.
LOoSe the Nodes
ANT defines nodes as actors, and there are myriad actors (more precisely, actor-networks) at work in GA. From the programmed codes written and interpreted to the software and hardware mediating and displaying web pages to the visitors and writers and programmers to the network providers and databases—ANT accepts any and all of these actants as nodes with the potential of agency. Latour (2005) refers to these objects as “the non-social means mobilized to expand them [the basic social skills] a bit longer” (p. 67) and confesses that ANT will “accept as full-blown actors entities that were explicitly excluded from collective existence by more than one hundred years of social explanation” (p. 69). The implication is that all the technological hardware and software — the GA code, the wired and wireless networks (cables, routers, and servers), and the Google Analytics processes server — work together to enable the web visitor to interact with this creation of the web writer, developer, coder, and marketer. This collective is incorporated at the moment of loading a web page, and its momentary connectivity is both enabled and expanded by agency of the object actors.
Where ROoSt the Nodes?
CHAT locates nodes in hierarchical relationships with one another in the network. Prior et al. (2007) conceive of literate activity producing socialized interaction within the functional system as part of the laminated chronotope of activity in space and time (Take 2: A Cultural-Historical Remapping of Theoretical Activity). In this hierarchy, web visitors are outside the system except during literate activity, defined as interacting with the multimodal text(s) within the site. Web writers, developers, and marketers are members of the functional system where literate activity (defined as creating and instantiating the multimodal text) occurs. The website itself is the functional system; the School gives the system chronological and spatial existence while the University gives the system technological existence. GA collects traces of literate activity among nodes within the functional system of the website, visualized and reported as interactions in space (between pages) and time (within sessions).
ANT flattens the network entirely. Latour’s (2005) conception of ANT works to keep the social flat (pp. 165-172), connecting all of the actor-networks (nodes) within the activity network in a single, non-hierarchical surface. Within GA, this flatness is largely retained within the report. All actor-networks have mediated, translated experiences of web content — there are no intermediary experiences, whether visitor or writer, software or hardware. GA reports a visualization of mediated network activity in a flattened data table. The flattened data table in GA treats the visitor’s web browser or operating system as equally significant to the actor-network represented by the visitor or web writer. Relationships between actors are largely un-disciplined; they are simply reported, regardless of the inherent logic (or lack thereof) in the relationship uncovered.
CHAT stresses an ecological relationship among nodes, limiting that ecology to the natural and material world (Prior et al., 2007). Visitors enter into the functional system of the website and navigate through it. Web writers, developers, and marketers engender the navigation links through the system, giving visitors pathways for narrative production. The website functions as the system, enabling web visits in time and space. The School provides content for the system, while the University provides the localized instantiation of the content in the website. GA records the traces of interactions within the functional system, visualizing them in laminated chronotopes in time and space. GA does not clearly identify the human actors in the network, preferring to aggregate identities. However, GA enables web writers, developers, and marketers to examine the traces of aggregated literate activity by visitors and revise website content and structure accordingly. This provides the opportunity for dialogue among human actors.
ANT stresses incoming connections among interconnected nodes. Latour (2005) frames this according to what it means to be a “whole”: “to be a realistic whole is not an undisputed starting point but the provisional achievement of a composite assemblage” (p. 208). Nodes that have more incoming connections than others are considered more settled and blackboxed, meaning they shift from being merely actors to becoming conduits for the flow of mediators: “an actor-network is what is made to act by a large star-shaped web of mediators flowing in and out of it” (p. 217). Such a star-shaped web of mediators is immediately visible in GA reports: the page in a website that receives the most visits or page views is the most connected page. This page is generally the website’s home page, and its purpose is not to provide content but to allow mediators to flow through it — to allow visitors to find what they’re seeking and connect to it.
WhOoShing through the Network
In GA, visit data — encrypted bits and bytes, assemblages of sequenced zeros and ones — moves from the visitor’s device to the GA server for processing and reporting. The collection process leading up to this movement differs between browsers (mobile and non-mobile) and mobile apps: browsers send data collections with every page load, but mobile apps bundle visit data and send it in timed intervals to protect mobile device battery life. This too simply describes a very complex ecology of network and computer hardware and software that transmits data from web content creators to web visitors to GA servers, but I’m limiting this discussion of movement to data from visitor’s device to GA servers. See the Google Analytics (2014) Academy “Data Collection Overview” video presentation (below) for additional details.
CHAT might describe this movement as distribution in the literate activity of viewing a web page or using a mobile app. Prior et al. (2007) define distribution as “the way particular media, technologies, and social practices disseminate a text and what a particular network signifies” (Mapping Literate Activity). In this case, two distributions occur: the distribution leading to reception (by the web page visitor) and distribution leading to the assemblage of visit data collected for interpretation on GA servers.
ANT might describe this movement as the social. The assemblage of connections from hundreds of thousands of SPCS visitor pageviews flowing into the GA server could be what Latour (2005) calls “the social — at least that part that is calibrated, stabilized, and standardized — [that] is made to circulate inside tiny conduits that can expand only through more instruments, spending, and channels” (p. 241). In this case, the conduits are standardized in the GA’s preselected data points (“Tracking Code Overview,” 2012). When and if GA adds new data points for collection, these tiny conduits would be expanded. This definition also suggests that many other connections remain unsurveyed, Latour’s “plasma.” The assemblage of all connections would be the social fabric of the network.
Meaning Released from the HOoSegow?
CHAT might describe meaning as the result of literate activity in the functional system. Prior et al. (2007) map literate activity as a multidimensional process that can include production, representation, distribution, reception, socialization, activity, and ecology (Mapping Literate Activity). The results of this literate activity are recorded and transmitted from visitor’s devices to GA servers. The meaning of these data points are processed (interpreted) and reported as visualizations. That meaning becomes the basis of analysis; analysis leads to conclusions about visitor behavior, which in turn result in changes to the web content leading to new literate activities.
ANT, on the other hand, ascribes no meaning to the results of CHAT’s literate activity. Latour (2005) remains adamant into the conclusion of Reassembling the Social that the social is dynamic and active, not a substance: “the social is… detected through the surprising movements from one association to the next” (p. 246). As a result, what GA does in processing and visualizing the results of activity in the SPCS website is not about ascribing meaning, but about tracing associations. And because those associations (connections) are mediated by the limited data points collected, the processing done by the GA servers, and the visualizations available, the reassembled social of GA is likely too limited to trace the plasmatic connectivity of the visitor’s web browsing experience.
Networks Emerge, Networks VamOoSe
CHAT and ANT will agree on this: actors initiate, grow, and dissolve networks. Prior et al. (2007) and Latour (2005) build their arguments on the social activities of actors. CHAT engages those actors in literate activity, while ANT engages those actors as connected actor-networks. Only activity on the part of actors can cause the network to emerge. For CHAT, only the activity of web content creators, web developers, database administrators, marketers, and web visitors can generate the first packet of data to flow across the network from visitor device to GA server. For ANT, the list of actors can extend much farther into non-human actants, but the principle remains the same: actors must initiate the network. Actors can grow the network through more visitor sessions — by many measurements, adding visitor sessions and growing session length is my primary professional objective as web manager — and actors can also dissolve the network by removing a web page (authors) or no longer visiting the website (visitors).
GA itself is a fairly limited network. Its boundaries could easily be drawn around the connection between the GA code on the web page or in the mobile app and the GA server. Any other activity that either leads up to the connection or follows the connection — namely writing and viewing a web page or viewing and interpreting GA visualized data — could be seen outside the network. Except that CHAT and ANT seek to problematize such limited perspectives of networks by addressing the activity that enlivens connectivity. So for these two theories, I found myself widening the focus to include the biological (CHAT and ANT) and non-biological (ANT) nodes in the network. This perspective turns into an ecology whose various members are only momentarily connected at the moment of accessing a web page or mobile app. But in that moment, myriad connections reveal actors and build a remarkably complex assemblage of networked components. As a result, I found few limits in CHAT or ANT to addressing GA as my OoS — other than the shortage of meaningful English words that contain the character string “-oos”.
Dahmen, N., & Sarraf, S. (2009). Edward Hopper Goes to the Net: Media Aesthetics and Visitor Analytics of an Online Art Museum Exhibition. Conference Papers — International Communication Association, 1-28.
Digital analytics fundamentals [Online course]. (2013, October). Retrieved from Google Analytics Academy https://analyticsacademy.withgoogle.com/explorer
Google Analytics platform principles [Online course]. (2014, March). Retrieved from Google Analytics Academy https://analyticsacademy.withgoogle.com/explorer
Kirk, M., Morgan, R., Tonkin, E., McDonald, K., & Skirton, H. (2012). An objective approach to evaluating an internet-delivered genetics education resource developed for nurses: Using Google Analytics™ to monitor global visitor engagement. Journal of Research in Nursing, 17(6), 557–579. doi:10.1177/1744987112458669
Latour, B. (2005). Reassembling the social: An introduction to actor-network-theory. Oxford, UK: Oxford University Press. Clarendon Lectures in Management Studies
Mc Guckin, C., & Crowley, N., (2012). Using Google Analytics to evaluate the impact of the CyberTraining Project. CyberPsychology, Behavior & Social Networking, 15(11), 625-629. doi:10.1089/cyber.2011.0460
Platform principles: Website data collection [Video transcript]. (2014, March). Google Analytics Platform Principles. Retrieved from Google Analytics Academy https://analyticsacademy.withgoogle.com/course02/assets/html/GoogleAnalyticsAcademy-PlatformPrinciples-Lesson2.2-TextLesson.html
Plaza, B. (2009). Monitoring web traffic source effectiveness with Google Analytics: An experiment with time series. Aslib Proceedings, 61(5), 474-482. doi:http://dx.doi.org/10.1108/00012530910989625
Prior, P., Solberg, J., Berry, P., Bellwoar, H., Chewning, B., Lunsford, K. J., Rohan, L., Roozen, K., Sheridan-Rabideau, M. P., Shipka, J., Van Ittersum, D., & Walker, J. R. (2007). Re-situating and re-mediating the Canons: A cultural-historical remapping of rhetorical activity [Multimodal composition]. Kairos, 11(3). Retrieved from http://kairos.technorhetoric.net/11.3/binder.html?topoi/prior-et-al/index.html
Google Analytics. (2014, March 11). Google Analytics Platform Principles – Lesson 2.1 Data collection overview [Video file]. Retrieved from http://youtu.be/qQdPXouWeJE
Tracking code overview [Web page]. (2012, October 29). Google Analytics. Retrieved from Google Developers https://developers.google.com/analytics/resources/concepts/gaConceptsTrackingOverview#howAnalyticsGetsData
[Header image: I’m a Google Analytics Geek: Screen capture of the Google Analytics Academy]