A Tale of Three Technologies: A Survival Analysis of Municipal Adoption of Websites, Twitter, and YouTube

The advent of the Internet and other ICT components have been critical to the deployment and spread of e-government [25]. In this respect, the implementation of ICT communication can be thought of as a series of innovations adopted by service providers over time [26]. For this reason, the adoption of newer tools for communication in an e-government context has frequently been connected to the diffusion of innovations (DOI) theory [27]. According to DOI theory, organizations can be classified into five adopter categories based on when they adopt the innovation in relation to other organizations in their social system [67, 78]:

Organizations adopt at different times, and organizations in each of the adopter categories are thought to have different perceptions of the innovation's general organizational attributes. One element of this is that adoption is influenced by the innovation's relative advantage over current practice [28]. The previous section describes how the benefits of the website format were not yet clearly outlined up to the early 2000s. Therefore, it seems acceptable to expect a slow adoption pattern for websites in the 1990s. On the other hand, as in the period when social media were on the rise, the benefits of using Internet tools for communication and interaction with stakeholders were already much more visible. The perceived benefits of engaging in timely and direct end-consumer contact at a low cost thus might have resonated easier with organizations than with websites [1], suggesting a comparatively faster adoption rate.

In terms of compatibility, a technology is more likely to be used when it is consistent with the existing infrastructure, culture, values, and preferred work practices of the organization [27]. During the early stages of website adoption, not all municipalities possessed the infrastructure to facilitate website use. This was not the case prior to the adoption of free social media platforms such as Facebook, Twitter, and YouTube, which allow organizations to reach stakeholders more easily than websites [76], resulting in lower start-up costs for municipalities.

Due to Twitter's responsive characteristics, local governments are expected to adopt this platform earlier than YouTube [5, 22]. YouTube is geared toward multimedia exchange and allows users to share, comment on, and rate videos [76]. Although some research points toward different demographic groups represented on YouTube [29], municipalities might be quicker to adopt Twitter, as YouTube adoption entails video content creation, which requires more resources than text-based messaging that are less compatible with existing work practices [7]. Therefore, I formulate the first hypothesis as follows:

In the organizational context, prior experience with different social media might be an important driver. Top management is primarily viewed in the literature on innovation adoption as the agency responsible for transforming norms, values, and culture, allowing other organizational members to adapt to the new technical artifact [1, 30]. Proclivity to engage in projects with uncertain outcomes or high profits and losses is linked to the risk-taking element [31]. In line with this, a municipality's willingness to adopt innovations can be a stimulus for social media adoption. Preexisting capabilities could also constitute resources that organizations can mobilize in pursuit of goals of facilitating online corporate dialog, as suggested by resource mobilization theory [32, 33].

At the same time, path dependence and soft determinism suggest that “one historical event did not rigidly prescribe certain subsequent technological developments, but at least made sequences of technological improvements in one direction easier” ([34], p. 19, see also [35]). In the context of social media adoption, it could be argued that as soon as an organization perceives presence on one online social network as sufficient in terms of outcome, this might limit the possibility of adopting other forms of social media.

Translating this to our context, YouTube adoption before Twitter adoption could indicate options for resource mobilization, as well as an organizational propensity for innovation and risk-taking, stimulating adoption of alternative social media. However, Twitter adoption prior to YouTube adoption could point more toward path dependency and soft determinism, especially given the higher entry costs of adopting YouTube (i.e., the needed resources and skills to make video content). I thus formulate the following hypotheses:

[36] argue that even if a company lacks the financial resources to invest in a variety of advertising or public relations strategies, it can still reach a large audience by making effective use of its web technologies. This has recently expanded to include social media [5, 33]. Conversely, several authors claim that while creating social media accounts is free, effective management and monitoring of those accounts necessitates time and well-trained employees, emphasizing the need for resources before adopting an innovation [7, 37]. In light of this mixed evidence, I put forward the following hypothesis for both platforms:

Another requirement for e-government adoption could be a willingness to change [38]. Municipal amalgamations could be an example of a momentum in which e-government innovations could be implemented, as they are frequently designed to take advantage of economies of scale, as well as a chance for a quality impulse in public service delivery [37, 39, 73]. At the same time, some authors note that the positive effect of amalgamations does not result in more efficient management, and that they generate an increase in coordination and management costs due to more complex bureaucratic structures [40, 41, 71]. The existing evidence thus suggests that it is appropriate to expect a null hypothesis for both platforms:

As public organizations must be politically legitimate, this necessitates that they follow social norms regarding organizational structure and function [14, 42, 43]. These norms could be grouped under the environmental context of e-government adoption.

Most literature on the digital divide assumes that people of a higher age are less digitally engaged, less likely to trust e-government services, and thus less likely to adopt forms of e-government [44, 45]. The reverse is said of people of a younger age. At the same time, contradictory evidence exists that digital skills do not predict or relate to channel choice, implying that the assumed digital divide has changed in recent days [46]. Furthermore, despite the fact that social media is said to be a universal platform for social interaction for adolescents and young adults [47, 48], no relationship has been observed with municipal adoption of e-government for a larger presence of this group, as they are less likely to use municipal products [73]. I align with the observed mixed results of the digital divide with the following hypothesis for both platforms:

According to research, stakeholders with a better socioeconomic position are more inclined to support ICT advances because they use the Internet more frequently and seek higher-quality public services [49, 50]. Recent studies also show that sophisticated social media use is likely to be found in municipalities with a larger presence of people with higher education levels [37, 48, 51]. For this reason, I will put forward the following hypothesis for both platforms:

Because of their professional background, IT professionals may have more positive intentions toward using e-government services, as new innovations are easily compatible with existing values, beliefs, experiences, and needs [37, 52, 69, 75]. IT professionals are already accustomed to using digital devices and incorporating technology into their daily routines. This familiarity can be viewed as a beneficial prerequisite for using e-government. Therefore, I formulated the following hypothesis for both platforms:

Finally, as the general number of technology adopters among citizens grows, expectations for organizational practices begin to emerge, and lagging organizations feel compelled to adopt to avoid losing legitimacy [65]. In line with the relative advantage of technologies, a larger online community encourages organizations to make better use of content that allows them to reach a larger share of their constituency through social media [1, 5]. Next to this, trust in e-government generally comes with trust in ICT tools in general [53], which suggests that as particular communication technologies become more trusted, communication with local governments through these technologies becomes more trusted as well.

A particular dynamic in this context is media coverage, which lends legitimacy to the voices, frames, and subjects that are chosen to be featured [54, 68]. Furthermore, media coverage of a particular topic increases user engagement with that topic [55]. In this respect, I propose the final hypothesis:

In order to test the hypotheses from the organizational and environmental context, a dataset was constructed which covered the adoption timespans of Twitter and YouTube.³ Table 2 provides an overview of the independent variables, their sources, and the time period for which the data could be retrieved. Subsequently, Table 3 provides the descriptive statistics of the independent variables. Due to the evidence for population size and population density in digital government contexts [5, 12, 44], these were entered as control variables in the models.

LexisNexis Academic was used to measure media coverage. This database performs full-text searches on thousands of news articles, and it searches national sources and local newspapers across the country for news articles. Time-based media coverage measurements using LexisNexis Academic have been employed in other contexts [54, 66]. Accordingly, the search strings “website,” “Twitter,” and “YouTube” were used to find articles that were published in every month of the respective timespans. One study [66] has analyzed the amount of press coverage of the social media platform Twitter between 2006 and 2009. Based on this methodology, I restrict the search results for this study to coverage in newspapers and magazines from Netherlands-based sources during the adoption timeframes.⁵ Our dataset included a total of 10,581 news articles using “website” during the time periods of interest; 208,688 news articles using “Twitter,” and 80,567 news articles using “YouTube”. In order to verify that a later moment in time would not by definition yield more search results (due to possible higher data availability in the database over the years), the control search term “newspaper” was checked for all timespans of adoption. As this did not lead to possible issues, I used the results for the original search terms as a reference. The highest number of news articles per month for each search term was used as a benchmark for all other observations, generating a value between 0 (no media coverage of the technology) and 100 (maximum media coverage of the technology).

Google Trends was used in order to measure user engagement. Google Trends sends website traffic data to the analytics server via a tracking code embedded on the website that is activated when a visitor views a page on a website [72]. Accordingly, it presents the monthly volume of Google searches for a given keyword, providing a normalized value as an output similar to the output that was used for LexisNexis Academic. [59] was followed and the search terms “Twitter” and “YouTube” were entered into Google Trends in the Dutch territory between 2006 and 2018.⁶ The level of interest is expressed as a percentage, with 100 being assigned to the month with the most searches for a given term. All other values are ranked in order of importance in relation to that date. No numerical data is available on the absolute number of searches. For the survival model, media coverage and user engagement were converted to variables expressing three-month moving averages during the month of adoption by a municipality.⁷

A multiple-record survival dataset was constructed. The month of the first registration of the website domain, Twitter account, and YouTube account were treated as the first month of adoption of the innovation, and all other dates were offset against the first adoption month. One yearly observation was generated for every municipality, including the data for the independent variables in the adoption timespan, assessing if the municipality had ‘survived’ (0) or if it adopted the technology (1). If the municipality had adopted the technology, the exact month of adoption set against ‘patient zero’ (i.e., the first adopter of the technology) was recorded, and also regarded as the final observation for the specific municipality in the dataset.

First, Kaplan-Meier analyses were used to investigate variations in adoption rates. To account for multiple predictor variables at the same time, Cox Proportional Hazards models and parametric survival models were used [13, 60]. All variables were tested in order to see if the proportionality assumption was met, which is a requirement for Cox Proportional Hazards. Media coverage and user engagement did not meet the proportionality assumption, but they did meet the exponential distribution assumption. Therefore, separate exponential proportional hazard models were computed for these variables.

After conducting normality tests, all variables except the dichotomous variables were converted into their natural log form. In order to rule out multicollinearity, correlation coefficients were scrutinized, which led to no possible issues. Model fit for all models was evaluated by conducting Wald \(\chi\) ² tests. Furthermore, robust variance estimators were used for all survival models [61], which use the efficient score residual for each subject in the data for the variance calculation.

Two earlier studies [56, 67] have analyzed corporate website adoption by classifying their observations into the five adopter categories based on the number of months that their organization had an operational website. Inspired by this methodology, Figure 1 and Table 4 compare the DOI distribution as suggested by [27] with the Dutch municipal adoption of websites and social media.

This suggests that the three adoption patterns differ significantly from what DOI theory suggests. First, website adopters are relatively overrepresented in the Late majority and Laggards categories, in support of H_1a. Twitter has a very large proportion of the municipalities falling into the Early majority category and few municipalities in the remaining categories, providing support for H_1b. To a lesser degree, YouTube is overrepresented in the Early majority category, in support of H_1c.

In order to gain more insight into municipal adoption of the technologies, Figure 2 and Table 5 show the results of the Kaplan-Meier estimates for the municipal adoption rates. In Figure 2, the X-axis represents the ‘survival duration’ for the particular interval. The cumulative probability of adoption during a given time is visible on the Y-axis.

The incidence rate – i.e., the ratio of the number of municipalities adopting the technology during the study period to the analysis time – confirms a relatively fast adoption rate for Twitter by Dutch municipalities (H_1a). Although website adoption takes longer than Twitter adoption (supporting H_1b), it also shows that the YouTube adoption rate is even slower than the municipal adoption rate for websites, providing support for H_1c.

Table 6 shows the results of the Cox regression models for municipal Twitter and YouTube adoption.

A negative effect was observed with respect to prior Twitter use in the YouTube model, and prior YouTube use in the Twitter model was positively significant, providing support for H₂ and H₃. Strong negative significance could be observed for recently amalgamated municipalities in the Twitter model, not in support of H₅. Negative significance could be observed in the Twitter model for IT professionals, not supporting H₈. No results were found for financial health, age, and higher education, supporting H₄ and H_6ab but providing no support for H₇. Finally, the strongest significance and effects could be observed for the control variable population size.

In this section, I assess the relation between the municipal adoption of the technologies and media coverage of the technologies, as well as user engagement with the technologies. Figure 3 illustrates how adoption rates, media coverage, and (for Twitter and YouTube) user engagement with the topic have changed over time. The adoption rates correlated highly with the assessed variables for all three technologies, although for Twitter and YouTube it could be noticed that user engagement correlated somewhat higher than media coverage (see Appendix A).⁹

In order to test the effects of media coverage, Table 7 shows the regression models for municipal platform adoption with media coverage and user engagement, controlling for population size and population density. All models show strong significance for media coverage, and its effects outperform those of the control variables. The strongest effect can be observed with website adoption, followed by Twitter and YouTube. The same is true for user engagement, which for both Twitter and YouTube appears to show an even stronger effect than media coverage, in line with the correlation coefficients. This provides strong support for H_9ab.

The findings in this study have several theoretical implications. First, despite context and adoption studies being mainstream for research in social media and government [81], the longitudinal nature of this study forms an addition to the literature on technology adoption by municipalities that consider a variety of points in time. Second, with respect to the diffusion of innovations (DOI) literature, the findings underline the importance of the temporal and the environmental context. The comparison of adoption patterns shows that although there are parallels (for example in terms of the importance of media coverage), the context of technology adoption has changed greatly overtime: website adoption happened during a period when the merits of the Internet were not yet clearly outlined, although this changed after the turn of the century. In the context of Twitter, many users already embedded the Internet into their daily web routines, creating a different playing field for local authorities to see the benefits of social media adoption – although the balance between costs and benefits was perceived differently for Twitter and YouTube. Finally, in terms of the e-governance of public organizations, the results in this study might point to the possible role of path dependence, which up to date has been addressed only to a limited extent in the perspective of social media adoption (one exception is [63]).

At least two important implications for practice can be derived from the findings. First, the initial slow adoption rate of websites provides a clear illustration of how perceptions of the relative advantage of technology adoption change over time. This also ties into the risks of path dependence and soft determinism and underlines the importance of openness to new technologies by managers and the need to critically assess the potential benefits of adoption.

Second, the comparatively slow adoption of YouTube is understandable from the vantage point of local authorities in deciding on the limited relative advantage of adoption. At the same time, it could be important for municipalities to have an eye for the different demographic groups and how they could be reached. Looking at Twitter and YouTube, it seems clear that the Dutch userbase is considerably smaller on Twitter (9%) than on YouTube (20%) [64], although municipalities have been considerably quicker in adopting the former. There is value for local authorities to investigate different platforms and their specifics, in order to stay in touch with all their constituents as a service-providing organization.

The findings in this study are subject to some limitations. First, innovation adoption has only been studied in a Western-European country, and as such, it draws from a specific administrative tradition. More research is needed to see if these findings can be replicated in countries with different political-administrative contexts. Future research should also consider additional factors such as gender (see e.g., [80]), which were beyond the scope of this study. It could also be informative to conduct similar research with newer social media platforms such as Instagram, which is mainly focused on the sharing of pictures rather than videos, and whose adoption has grown steadily during the last few years (see e.g., [79]). Next to this, it remains an open question if the perceived relative advantage of Twitter and YouTube for local authorities is in line with the actual outcomes of their use. Therefore, it remains essential to supplement adoption-focused research with research that helps to understand how social media can be used effectively [5]. Finally, as important as it is to look at e-government adoption, it is becoming more important to also take into consideration abandonment, as social media platforms are no static entities and their userbase can collapse rapidly [70]. Future research needs to consider these limitations.