The Janus-Face of Digitalization: The Relation Between Internet Use and Civic Engagement Reconsidered

Theory and Hypotheses

A plethora of studies investigates the social implications of the Internet, studying its effects on social capital formation, social relationships, community involvement, and political participation. ² Theoretical insights are divided on whether the rapid spread of Internet technologies erodes or amplifies traditional social relations. They largely build on Putnam’s (2000) seminal work Bowling Alone: The Collapse and Revival of American Community. While Putnam mainly addresses the effects of television, seeing as Internet use was not as widespread as it is today, he also extends his arguments to the new “electronic cousins” (p. 246) of television and subsequent literature follows him in that regard (e.g., Hooghe & Oser, 2015). Putnam advances a dystopian perspective, contending that digital media individualize people’s leisure time and thus lead to a decline in social capital. The causal mechanisms of time displacement and privatization are at the forefront of his line of reasoning (Putnam, 2000, p. 237). Internet use (as watching TV) is said to displace other, social capital producing activities. Each minute people spend surfing on the Internet in their leisure time cannot be used for meeting friends, being active in an association, or for other off-line community involvements. In addition, digital media leads to a privatization of the individual lifeworld. People usually use the Internet at home without face-to-face interactions with other people. Activities that previously required social contact (financial transactions, information gathering, etc.) can now be conducted privately via the Internet. As a result, the Internet encourages lethargy, and frequent users attempt to avoid further contact with the outside world. In contrast, subsequent literature contradicts Putnam’s claims that electronic media erodes social capital (e.g., Kraut et al., 2002; Pruijt, 2002; Robinson & Martin, 2010; Wellman, Quan-Haase, Witte, & Hampton, 2001). The social engagement hypothesis stands for the utopian perspective that the Internet instead strengthens social relations by creating new means to provide and distribute social and political information, communicate with social contacts, and connect with new people. In particular, social interactions become easier and faster via the Internet, lowering opportunity costs. This way, the Internet supplements or transforms, not substitutes, offline social relations (Quan-Haase & Wellman, 2002).

Most studies, however, do not content themselves with analyzing a general effect of Internet access or Internet usage but distinguish between different patterns of Internet usage, such as for informational, communicative, or entertainment purposes, or measure specific activities such as writing e-mails or reading news online (e.g., Ling, Yttri, Anderson, & DeDuchia, 2003; Purdy, 2017; Quintelier & Vissers, 2008; Reinwand, Crutzen, & Zank, 2018; Shah, McLeod, et al., 2001). Current literature relying on patterns of Internet use usually places Internet activities into distinct categories. Regarding the literature, however, and following a uses and gratifications account, we argue that specific Internet activities and broader patterns of Internet use need to be connected more systematically with the causal mechanisms relating Internet use to civic engagement (Quinn, 2016; Valenzuela et al., 2009). ³ In this vein, Internet activities can be arranged along two dimensions of broader patterns of Internet use.

The first dimension relates to the degree of social connectedness. Consumptive activities such as reading news online or shopping online are usually done in private at home without any social interactions. As such, they reinforce the privatization mechanism and should have negative effects on civic engagement. In contrast, writing e-mails or participating in online chats are social interactive activities that further communication and connectedness. In line with the social engagement mechanism, we expect positive effects on civic engagement. Several studies were able to confirm the expectation that Internet use for social connections and communication increases civic engagement (e.g., Gil de Zúñiga & Valenzuela, 2011; Kavanaugh & Patterson, 2001; Purdy, 2017) as well as the positive effect of social network site activity on civic engagement (e.g., Pasek, More, & Romer, 2009; Valenzuela et al., 2009). The second dimension distinguishes between Internet use for information (searching for information on the Internet, writing e-mails) versus entertainment (streaming media online, participating in online forums) purposes. Internet use for entertainment is inherently connected with the time displacement mechanism and should thus have negative effects on civic engagement. In contrast, using the Internet to acquire and distribute information may be time-intensive as well but at the same time “indirectly promote[s] increased political knowledge […] and awareness of civic opportunities and objectives” (Shah et al., 2005, p. 535). Indeed, statistical analyses strongly corroborate the positive effects of informational Internet use (e.g., Moy et al., 2005; Pasek et al., 2009; Shah, McLeod, et al., 2001). Admittedly, one might remark that different individuals use these interactivities in different ways. Social network sites could be used either to share political information via Twitter or to watch entertaining videos posted on Facebook. As our data do not allow to differentiate the precise purposes for which these Internet activities are used, however, we argue that our classification represents the purpose for which these Internet activities are used on average.

Combining the two dimensions, we thus arrive at a four-field matrix (see Figure 1): On the top right, we have the social Internet use for information, such as reading or writing e-mails, social network sites activities or chatting online. The top left includes Internet activities such as reading the news online, gathering information, or using the Internet for education, which are also high on the information side of the spectrum but are usually conducted in private. On the bottom right, we have Internet activities for social recreation, such as using forums or playing online games. The last field on the bottom left consists of private Internet use for entertainment, such as streaming media or shopping online. This basic argument of this two-dimensional conceptualization of Internet use renders earlier ways of thinking more precise as well as overcoming their previous analytical deficits. Consequently, this conceptual foundation is able to produce a more appropriate model to link Internet use and civic engagement.

OPEN IN VIEWER

Figure 1.

Patterns of Internet use and specific Internet activities. Depending on the location of the activities in the matrix, we expect either a positive effect on civic engagement (+), a negative effect (−), or no effect at all (0).

We apply this framework to the study of associational participation. While civic engagement entails various activities (including more formal activities in associations as well as activities such as volunteer work, working for community projects, attending community or neighborhood meetings, raising money for charity, ethical consumerism, contacting public officials, or participating in protests), we decided to focus on the formal aspect of associational participation. Of all the measures of civic engagement discussed in literature, associational participation is most closely linked to—and can even be said to be a cornerstone of—social capital (Putnam, 2000). Its characteristics as a regular commitment that cannot be easily opted out of makes it an especially stringent test of the time displacement, privatization, and social engagement hypotheses. Furthermore, it is of particular importance for civil society because, following Tocqueville, associations are frequently heralded as schools of democracy that lead to prodemocratic socialization, increased tolerance, and political involvement (e.g., Baggetta, 2009; Freitag, 2016; Hooghe & Quintelier, 2013; Putnam, 2000; van Deth, 1997).

Our expectations as to how certain Internet activities affect associational participation are clear for the top right and the bottom left quadrant of our two-dimensional space. Here, both dimensions point in the same direction and should lead to only either social engagement effects or time displacement and privatization effects. We thus expect e-mail, social network site use, and chatting to have a positive effect on being active in an association, while streaming media or shopping online should decrease the likelihood of associational engagement. For the other two quadrants on the top left and the bottom right, we observe conflicting influences. Activities at the top left are connected to both social engagement and privatization, while activities on the bottom right result in both social engagement and time displacement. Consequently, we do not formulate any expectations for activities falling into these quadrants. Our two hypotheses are thus:

Hypothesis 1a: Social Internet use for information (i.e., writing e-mails, social network site activities, chatting the hypothesis,) increases the likelihood of being active in an association.

Following Boulianne’s (2009, p. 205) and Boulianne and Theocharis’s (2018) demand for disentangling causal effects, we also strive to assess the reverse causal direction from associational participation to Internet use in our analysis. Unfortunately, research on this topic is rare. To our knowledge, only two studies explicitly tested the reverse causal direction. While Shah, Schmierbach, Hawkins, Espino, and Donavan (2002) found no effect of civic engagement on Internet use, Jennings and Zeitner (2003) found a positive effect on Internet access. These studies do not distinguish between different patterns of Internet use, however. We argue that active members of associations are likely to distribute information about their activities through the Internet. In the wake of digitalization, civic associations slowly but increasingly adapt to the technological change and implement the Internet in their communication strategies (e.g., for publicity work, distributing information, or networking with other organizations; Wells, 2014; Yang, 2007). Following media richness theory (Daft & Lengel, 1986; Vriens & van Ingen, 2018) and the discussion on the relation between Internet communication and strong and weak ties (Gil de Zúñiga & Valenzuela, 2011), social Internet use for information, such as writing e-mails or social network site activities, are particularly useful to maintain weaker ties such as in civic organizations. For these reasons, social Internet use for information should become more prevalent with those who are active in associations. In contrast, the displacement effect (Putnam, 2000) is likely to work in both ways. In particular, Internet use for entertainment involves time-consuming leisure activities that are in direct competition with civic association activities and are thus particularly likely to be displaced.

Hypothesis 2a: Being active in an association increases social Internet use for information (i.e., writing e-mails, social network site activities, chatting).

Method and Data

Our modeling strategy is an autoregressive cross-lagged panel design. Such designs were developed in developmental psychology (e.g., Kenny, 1975; Selig & Little, 2012; Watkins, Lei, & Canivez, 2007) in order to estimate how two variables are causally related to each other. It requires that these two variables are measured simultaneously at least two points in time. Each variable is regressed on its lagged value, lagged values of the other variable, as well as lagged control variables. In our case, these variables are the associational participation of an individual respondent in a given year A_it , a set of Internet use and activities measures I_it as well as a range of control variables X_it . We thus arrive at the following two equations:

By including a lagged-dependent variable, the cross-lagged parameters β I i t − 1 and β A i t − 1 are thus effectively predicting change in the dependent variable (Finkel, 1995; Hamaker, Kuiper, & Grasman, 2015, p. 104). Since associational participation is binary, we employ logistic regression models for Equation 1 to test our hypotheses. The Internet use measures are in part metric, ordinal, and binary, we thus use ordinary least squares (OLS) regression, ordered logit regression as well as logistic regression, respectively. Our data are nested in three levels: the year, the individual, and the household. We include year-fixed effects to account for common time trends, and following the recommendation of Cameron, Gelbach, and Miller (2011), we use cluster-robust standard errors at the highest level of aggregation, the household, as the most conservative approach.

We apply this modeling strategy to data from two annual panel surveys. The first is the LISS Panel (CentERdata, 2018), which interviews a representative sample of households in the Netherlands since 2007 via online questionnaires. Each wave comprises around 7,000 individuals. Our key variables, associational participation as well as Internet use and activities, are available for all waves. We thus use the 31,308 observations of respondents in Waves 1–10 (2008–2017), for which data on all variables were available. The second is the SHP (Tillmann et al., 2018), which interviews a random sample of households in Switzerland, mainly by telephone. Depending on the wave, the SHP includes in between around 10,000–20,000 respondents. While associational participation is available for all panel waves, questions related to Internet use were only asked in Waves 12, 15, and 18 (2010, 2013, and 2016). ⁴ We thus use the 17,948 observations of individuals in these waves who also took part in the subsequent panel wave and for which data on all of the variables were available. Unfortunately, this means that we have to use 3-year lags instead of 1-year lags to estimate the reverse causal effect from associational participation to Internet use. The variables were operationalized as follows: a full list of all variables, their operationalization as well as descriptive statistics can be found in Tables A1 and A2 in the Online Appendix.

LISS

Our first analytical part examines the Dutch LISS Panel data. In a first step, we look at the overall trends of the key variables over the observation period. Table 1 shows the (untransformed) mean values of our measures of associational participation and Internet use in the first and the last year that was included in our model. ⁸ Associational participation has declined by almost 13 percentage points throughout the observation period. While in 2008, 38.7% of the respondents participated in the activities of at least one organization, this number dropped to only 26.0% of the respondents in 2017. At the same time, average Internet use went up by roughly 6 min. This development varies between the different online activities, however. In particular, playing online games and streaming media have surged in popularity, while spending time on forums and chatting even saw a decline in use. Even though an increase in the average Internet use thus coincides with a decrease in the level of associational participation on the aggregate level, this does not have to be true on the individual level. For this reason and to clarify the causal direction, we turn to our regression models next.

OPEN IN VIEWER

Table 1.

Development of Associational Participation and Internet Use Over Time Longitudinal Internet Studies for the Social sciences Panel.

Variables	Available Period	Start	End	Difference
Associational participation	2008–2017	.387	0.260	−.127***
Internet usea	2008–2017	.94	1.03	+.09**
E-maila	2008–2017	.55	0.65	+.10***
Social network sitea	2012–2017	.49	0.49	−.00
Chata	2008–2017	.53	0.40	−.13***
Newsa	2008–2017	.25	0.33	+.08***
Informationa	2008–2017	.34	0.35	+.01
Online gamesa	2008–2017	.43	0.64	+.21***
Forumsa	2008–2017	.30	0.19	−.11***
Streaming mediaa	2008–2017	.26	0.45	+.19***
Shoppinga	2008–2017	.28	0.30	+.02†

^aIn hours per day, untransformed. The decimal places can be transformed into minutes via the formula (*60/100).

^† p < .1. *p < .05. **p < .01. ***p < .001.

Figure 2A presents the results of our regression models for Equation 1 regressing associational participation on Internet use and activities, while Figure 2B displays the coefficients for the reverse regression model with Internet use and activities as the dependent variables (Equation 2). For ease of presentation, we only display the coefficients of our key variables, confidence intervals (CIs) are set to a 95% significance level. Detailed regression models with coefficients for all of the controls can be found in Table A3 in Appendix. Each line in the figures represents a separate model—since the number of missing cases varies heavily between the Internet activities asked in the LISS panel, we cannot include them all in the same model. All Internet activity models also include general Internet use as a control variable in order to control for the general effect of time spent online. We thus estimate the effect of how the time spent online is distributed between different patterns of Internet activities, that is, the effect of the time spent on an Internet activity independent of general Internet use. All models also include the aforementioned controls and the lagged-dependent variable. In Figure 2A, the coefficients display the effect of each respective Internet use and activities variable on associational participation, in Figure 2B, the coefficients show the reverse effect of associational participation on each respective Internet use and activities variable.

OPEN IN VIEWER

Figure 2.

Regression coefficients of the relation between associational participation and Internet use and activities Longitudinal Internet Studies for the Social sciences Panel. Shown are regression coefficients with 95% confidence intervals. A. Effect of Internet use and activities on associational participation. B. Effects of associational participation on Internet use and activities.

Overall, we find no general effect of Internet use (see Figure 2A)—the coefficient is negative but insignificant and only turns significant in a few of the Internet activity models (but this is mostly due to missing data changing the sample). Looking at the effect of different Internet activities on associational participation, only one of them achieves conventional significance levels. As predicted by Hypothesis 1a, social Internet use for information in the form of writing e-mails has a clear positive impact on associational participation significant at the 0.1% significance level. Doubling the number of hours spent writing e-mails per day is associated with an increase of 2.5% in the predicted probability of becoming or remaining active in an association in the subsequent year—an effect size that is comparable to other predictors of associational participation that we control for. A similar effect of social network site use or chatting on associational participation cannot be found, however. In contrast to Hypothesis 1a, Hypothesis 1b cannot be supported at all by the LISS panel data. Although the coefficients are negative, as predicted, the uncertainty of these coefficients is too high to reject the null hypothesis. As the other Internet activities, no hypotheses were formulated due to conflicting influences between the social-private- and the information-entertainment-dimension. Indeed, none of these Internet activities reaches conventional significance levels. Finally, most of the control variables are consistent with previous results. While gender, age, income, and having children show no significant effects, respondents with a higher education, married respondents and students are more likely to become or remain active in an organization. Contrary to expectations, unemployed respondents are also more active in organizations, but the effect is not robust throughout the models. In addition, associational membership is associated with a higher interest in politics, meeting friends more often as well as watching less television.

With regard to reverse causality, Figure 2B shows that there is indeed some evidence. While we do not find an effect of associational participation on general Internet use, respondents who participate in associations spend, in the following year, on average around 1 min more per day writing e-mails than nonparticipants and around 1 min less per day playing video games online. The effects are significant at the 1% and 5% significance levels, respectively. We thus find some support for Hypotheses 2a and 2b but only for a subset of the activities.

To test whether the results are dependent on the inclusion of certain panel waves, we perform jackknife robustness checks, leaving one wave out at a time. Our conclusions hold in most models (173 of 180 models). Against our expectations, chatting becomes significantly negative related to associational participation when the newest panel wave is excluded. The effect of shopping on associational participation becomes significantly negative in two models. In two models, associational participation has a significant positive effect on searching for information online, albeit only marginally. Additionally, the effect of associational participation on streaming media becomes significantly negative in two models. These results thus provide additional tentative, but nonrobust support for Hypotheses 1b and 2b.

SHP

Next, we perform the same analytical steps for Switzerland using the SHP data. Table 2 again compares descriptive data for the development of associational participation and Internet use between the first and the last year of the observation period. With an associational participation rate of 52.4%, club participation starts at a substantially higher level when compared to the Netherlands and declines only marginally, albeit significantly, to 49.8% in 2017. At the same time, average levels of Internet use are lower in Switzerland than in the Netherlands but increase significantly by 10 min throughout the observation period. Looking at the individual Internet activities, all except for Internet use for educational purposes have increased significantly over time—in particular reading news online and watching Internet television or listening to Internet radio have risen considerably.

OPEN IN VIEWER

Table 2.

Development of Associational Participation and Internet Use Over Time Swiss Household Panel.

Variables	Available Period	Start	End	Difference
Associational participation	2010–2017	0.524	0.498	−.026**
Internet usea	2010–2016	0.59	0.76	+.17***
E-mail	2010–2016	3.66	3.92	+.27***
Social network site member	2013–2016	0.440	0.475	+.035***
Chat	2010–2016	1.66	1.96	+.30***
News	2010–2016	2.74	3.47	+.73***
Education	2010–2016	1.69	1.67	−.01
Streaming media	2010–2016	1.70	2.16	+.46***
Shopping	2010–2016	1.83	2.12	+.29***

^aIn hours per day, untransformed. The decimal places can be transformed into minutes via the formula (*60/100).

^† p < .01. *p < .05. **p < .01. ***p < .001.

We now turn to the results of the regression analyses, presented in Figure 3 for Equation 1 and Figure 4 for the reverse causality Equation 2 (detailed results can be found in Table A4 in the Online Appendix). As with LISS, we run one model with Internet use only and then models including each Internet activity separately. In contrast to LISS, the Internet activity variables are not measured continuously in hours per day. Social network site use is measured dichotomously as being a member or not, the other Internet activities form categories from “never” to “every day.” We thus enter them into the model as factor variables excluding “never” as the base category. Given that the coefficients of categorical variables depend on the base category, we present predicted probabilities of associational participation for all values in Figure 3 to show the effect of the Internet use and activity measures on associational participation. The dashed grey line signals the average predicted probability of associational participation. CIs are again shown at the 95% significance levels.

OPEN IN VIEWER

Figure 3.

Predicted probabilities of associational participation by different values of the Internet use and activities variables Swiss Household Panel. Shown are predicted probabilities of associational participation with 95% confidence intervals.

OPEN IN VIEWER

Figure 4.

Regression coefficients of the relation between associational participation and Internet use and activities Swiss Household Panel. Shown are regression coefficients with 95% confidence intervals.

While Internet use has no effect on its own in the LISS panel, all SHP models show a significant negative effect on associational participation. Judging from the base model without any Internet activities included, a doubling in the number of hours spent online reduces the predicted probability of becoming or remaining active in an association in the next year by 1.4%—an effect that is significant at the 1% level but weaker than most of the other common predictors of associational predictors. Turning to the Internet activities, we again find support for Hypothesis 1a. Going from never writing e-mails to writing e-mails daily significantly increases the predicted probabilities of being or remaining active in an organization from 46.7% to 51.7%. In particular, respondents writing e-mails less than once a week have a significantly lower predicted probability, while respondents writing e-mails every day have a significantly higher predicted probability of being or remaining active in an organization. ⁹ Additionally, the effect of social network sites is significant at the 5% level in the SHP. ¹⁰ Being a member of a social network increases the predicted probability of becoming or remaining active in an association by 1.8%. For chatting, the effect is mixed: Infrequent chatting boosts, but daily chatting dampens associational participation. Hypothesis 1b can be soundly rejected again. We even find, albeit small, evidence that infrequent streaming of media online and purchasing or selling items online has a positive effect on associational participation. As to the other activities, for which no hypotheses were formulated, no effect can be found for reading news online or online education. Finally, we find similar results for our control variables as in the LISS panel: Respondents with a higher education and students are more likely to become or remain active in an organization, and associational membership is associated with a higher interest in politics, meeting friends more often as well as watching less television. In addition, we find that men are more likely to be active, and age has the expected inverted U-shaped effect on associational participation. Confirming our expectations, retired respondents and respondents staying at home are more likely while unemployed respondents are less likely to be active in organizations. Finally, as expected, Swiss nationals, respondents living at their place of residence for a longer time as well as German-speaking respondents are more active in organizations.

Figure 4 presents the results for the reverse causal relation. Here, we again plot the coefficients as was done with the LISS panel because the key independent variable (associational participation) is dichotomous. The regression model, however, changes in between the different variables. We use OLS regression for the continuous variable Internet use in hours per day, logistic regression for the dichotomous variable social network site membership, and ordered logit regression for the other Internet activity measures that are measured as ordinal categories. ¹¹ As the Internet use and activity measures were only collected every 3 years, we use a 3-year lag structure for all independent variables instead. Associational participation has a marginally significant negative effect on general Internet use. Participating in associational activities reduces the time spent online by roughly a minute per day over 3 years. In addition, we find similar support to Hypothesis 2a as in the LISS panel: Being active in associations increases the frequency of writing e-mails. It has no impact on social network site activity or chatting, however. Besides, we find no support for Hypothesis 2b—associational participation has no effect on any of the other Internet activities either.

We again performed jackknife robustness tests for the panel waves. The general effect of Internet use was not robust to the exclusion of the Panel Wave 15. In fact, the size of the coefficient varies widely between the three panel waves (Wave 12: −.067, Wave 15: −.396, Wave 18: −.048) and a significant effect can only be found in Wave 15 but not in the other two waves included in the analysis. The reverse effect of associational participation on Internet use shows the same pattern. In addition, the already weak and inconsistent effects of chatting and streaming media were not robust between the panel waves, while news and Internet education showed inconsistent effects in some of the models. Lastly, the effect of social network site membership also varies between the two panel waves: They are only significant in Wave 15 but not in Wave 18. Overall, only the effect of writing e-mails proves to be very robust.

Finally, in the SHP, Equations 1 and 2 were tested on different samples because we had to employ a 3-year lag for the reverse causality regressions. In Figure A1 and Table A5 in Appendix, we make them more comparable by also using a 3-year lag structure for Equation 1. All in all, our conclusions hold, but the nonrobust effects of Internet use and social network site membership on associational participation lose significance altogether.

Conclusion

In this article, we strived to reassess the relation between Internet use and civic engagement, that is, associational participation. Theoretical arguments are divided between an utopian perspective that emphasizes the social connection building effects of the Internet and a dystopian account arguing that Internet use leads to an erosion of the traditional social fabric. To date, statistical analyses of the effect of Internet use on civic engagement show mixed results. Our goal in this article was to expand on previous literature in two ways. First, our theoretical innovation was to connect the causal mechanisms relating Internet use to civic engagement with broad patterns of Internet use and specific Internet activities to overcome the previous analytical deficits. We argued that these can be lined up on two dimensions: social versus private Internet use and Internet use for information versus Internet use for entertainment. Following this distinction, we hypothesized that social Internet activities for information, such as writing e-mails, should further civic engagement, while private Internet use for entertainment, such as streaming media online, should contribute to the erosion of civic engagement.

Second, our methodical innovation was to address the issue of endogeneity directly by using an autoregressive cross-lagged panel design. With both lagged-dependent and lagged-independent variables, we are able to estimate how Internet use affects subsequent changes in associational participation without the endogeneity issues that plague cross-sectional designs. In addition, we are able to estimate directly the reverse causal effect from associational participation to Internet use. Panel data from the Dutch LISS household panel as well as the SHP offered the opportunity to employ this approach.

Regarding the overall effect of Internet use independent of how the time online was spent, we find no significant effect in the LISS panel and only nonrobust evidence for a negative effect in one panel wave of the SHP. A closer look at how respondents spend their time on the Internet provides more nuanced results. In support of Hypothesis 1a, a clear robust positive effect of writing e-mails on associational participation can be observed in both countries. Additionally, social network site use leads to higher associational participation in the SHP, but the effect could again only be found in one panel wave and thus lacks robustness. For the other Internet activities, we could not find a clear and robust effect on associational participation. Following our theoretical framework, passive Internet use for entertainment should be associated with lower civic engagement. This hypothesis could not be supported by our empirical analysis. Finally, our analyses of the reverse effect of associational participation on Internet activities show evidence that reverse causality is an issue that needs to be addressed—a purely cross-sectional model would thus lead to biased estimates. Being active in an organization leads to a subsequent increase in e-mail writing activities in both countries. In addition, active participants of associations spend less time playing online games.

The substantive effect size, however, warrants further discussion, because with such a high number of observations, even small effect sizes might turn out as significant (Lin, Lucas, & Shmueli, 2013). Indeed, especially the sizes of the reverse causal effect are meager, with associational participation only leading to an increase in the time spent writing e-mails by 1 min. It has to be stressed, however, that by including a lagged-dependent variable in our models, we are effectively explaining 1-year changes in the dependent variable instead of its overall level (Hamaker et al., 2015, p. 104). Naturally, due to the lower variance of 1-year changes, independent variables tend to have smaller effect sizes in such models, but such small effects can sum up over time. For instance, the 2.5% difference in the predicted probabilities of becoming or remaining active in associations within 1 year that results from doubling the time spent writing e-mails might not seem like much, but over a longer time span of 5 or 10 years, such an effect can accumulate and lead to substantial differences. And when compared to our control variables, writing e-mails is one of the strongest predictors of becoming or remaining active in associations. The same does not hold true for the reverse causal effect, however. Here, the effect of being active in associations is one of the weaker predictors of changes in e-mail usage. While the effect size may still accumulate over time (for instance, on average 10 additional min spent writing e-mails each day after 10 years of associational participation), it is indeed small enough to question whether associational participation substantially impacts Internet use.

In addition to the weak effect sizes, our study faces certain data limitations. In particular, our analysis is limited to the Netherlands and Switzerland, and it remains unclear how our findings relate to other nations. With regard to our variables of interest, the Netherlands stands out as a country with both high levels of Internet use and civic engagement in comparison to other European countries, while Switzerland is below average in terms of Internet use and with significant differences between the cultural, linguistic regions of Switzerland in terms of civic engagement (European Social Survey ERIC 2016; Freitag & Ackermann, 2016). While the Netherlands could be regarded as a vanguard, following the late Stein Rokkan (1970), Switzerland can be thought of as a microcosm of Europe because of its cultural, linguistic, religious, and regional diversity. In addition, meta-analyses comparing studies from different countries have not discovered differences between advanced economies of different regions in the world so far (Boulianne, 2018). For these reasons, we believe that our conclusions are likely to be valid for other advanced economies as well. Still, it is essential to examine other countries using panel data in order to ascertain whether our results are generalizable and also hold for lesser developed countries with lower levels of Internet access and civic engagement.

While we were able to test (most of) our hypotheses in two countries using two different data sets, a strict comparison of the results is made difficult by the slight differences in how the questions were asked between the two surveys. Ideally, we would have preferred a truly comparative panel, in which the same survey questions are asked in various countries. Unfortunately, such data are not yet available. In addition, our Internet activity measures might be more precise than just generalizing from Internet use. Still, different respondents might use certain Internet activities in different ways. Streaming media can be done either to watch documentaries or informational shots or to watch pure entertainment program. Similarly, news read online may vary between online versions of quality newspapers and tabloid press. In further research, more precise measures for our two dimensions would be desirable. Finally, while or research design addresses the issue of causality more directly, omitted variable bias may still be an issue. Although we control for many variables discussed in literature, e-mail use may still coincide with, for instance, a respondent’s role as a professional manager in particular in the LISS panel, which does not distinguish between e-mail use for work and for leisure.

Despite these limitations, our empirical analysis was able to provide evidence for a virtuous circle (Norris, 2000; Shah, Schmierbach, Hawkins, Espino, & Donavan, 2002), albeit contingent on how users spend their time online. Social Internet use for information—in the form of writing e-mails—increases and reinforces civic engagement in organizations, while civic engagement at the same time also enhances social Internet use for information. This suggests, as other studies have shown (Boulianne & Theocharis, 2018; Gil de Zúñiga, Barnidge, & Scherman, 2017), that the online and off-line forms of social engagement complement and enhance each other. Informational exchange is facilitated via the Internet and leads to more opportunities to take notice of civic engagement options, and subsequently civic engagement promotes further informational exchange via the Internet.

Supplemental Material