This section is split into four subsections. In the first three, we focus on specific constructs: the overall environmental orientation of digital users, their digital expertise, and the digital-environmental habitus, which encompasses both awareness and behavioural dimensions. The fourth subsection presents the results of the path structural models.
4.1. Awareness and Behavioural Component of the Digital-Environmental Habitus
To explore the intersection of the habitus’ digital and environmental dimensions, we have separately examined awareness and behaviours aimed at mitigating the negative environmental impacts of digital technology use [
2]. The first step was to create an index that would capture users’ awareness of both online behaviours and their uses of digital technologies, considering the impact of digital tools throughout their lifecycle, from material extraction to the end of their functionality [
79].
We conducted a factorial analysis (FA) to explore respondents’ awareness of how digital technologies and online activities affect the environment (
Table 2). The items were included according to the main findings emerging in the literature that show how the operation of the Internet network and the production of hardware produce digital pollution (especially in terms of electricity demand, extraction of material, and waste production). By contrast, the dematerialisation and the “smartness” of automated systems have been identified as beneficial to the environment (for a review, see [
79]).
One component was extracted, as the second component only included two items, which also positively contributed to the first component. This component was named “digital-environmental awareness” (DEA) and accounted for 41% of the variance (
Table 2).
Note that the Kaiser–Meyer–Olkin (KMO) test indicates that the partial correlations between the items belonging to the construct assume each have relatively high values (a value above 0.8 is considered good by Keiser and Rice [
80]). Also, the Bartlett’s test refuses the null that the items are not correlated. Therefore, we can conclude in favour of the convergent validity of the scale. Regarding internal consistency, the Cronbach’s alpha is equal to 0.682. Generally, a value above 0.7 is considered good, however, Nunnally and Bernstein [
81] suggested 0.6 as the minimum threshold for acceptability.
Behaviours were explored using a set of items, as detailed in
Table 3, where respondents were asked to express their level of agreement on a scale from 0 (totally disagree) to 7 (totally agree) concerning the consumption and use of digital technologies and their impact on the environment. The items were selected as proxies of environmentally informed digital behaviours following the findings in the literature that identify how some individual behaviours can reduce individual environmental impact (e.g., [
82,
83]). An FA was conducted to combine the items. In addition to the items explored by Ruiu et al. [
2] related to the online experience, we included statements concerning the tangible elements of digital technologies that contribute to the digital experience. Specifically, the digital components’ manufacturing process and the e-waste generated are recognised as integral aspects of digital pollution [
82]. Two components were extracted based on an eigenvalue higher than 1, together explaining 47% of the variance. The first component reflects uses of digital technologies that are eco-centric-oriented, explaining 31% of the variance. We labelled this component “eco-centric digital-environmental behaviour” (EDEB). The second component includes pro-environmental behaviours offering additional practical benefits, explaining 16% of the variance. We labelled this variable “benefit-oriented digital-environmental behaviour” (BDEB). BDEB includes items related to both the material aspects of digital tools and online behaviours. The material components involve using digital technologies until they are no longer functional (which can offer financial benefits) and disposing of discarded devices at recycling centres (which can free up space at home). Online behaviours include unsubscribing from automatically generated newsletters (which can prevent memory overloads) and condensing content in emails/messages (which can save time). Additionally, these behaviours contribute to reducing energy demand and CO2 production [
84].
4.3. Digital Expertise
We conducted an FA to summarise a series of items to assess respondents’ digital competencies. Following the model proposed by Ragnedda, Ruiu, and Addeo [
34], this set of questions explored users’ perceptions of their digital know-how and how it enhances their daily lives. Respondents were asked to rate their level of agreement (on a scale from 0 to 10) regarding various aspects of their digital competencies. These aspects included their proficiency with digital devices, their ability to connect online, the length of their digital experience, their capacity to identify trustworthy sources and protect their privacy, their ability to express opinions and create content on appropriate digital channels, and their skill in resolving technical issues (see
Table 5) [
34]. The FA resulted in extracting a single component, labelled digital expertise (DE), which explains 42% of the variance. This choice was made, as the second component only presented low score loadings. All the included items in the retained component had a positive impact, indicating that this component effectively captures the respondents’ digital expertise.
The KMO test and Bartlett’s test indicate that there are not problems with convergence validity. The Cronbach’s Alpha is 0.79, indicating good internal consistency.
As a final test for the validity of our constructs we implemented the Harman’s test for common method bias (i.e., the inflation of covariation among different constructs caused by using a set of items with similar characteristics). In particular, we ran a factorial analysis using all the 36 items, we then evaluated the variance explained by the first extracted factor. If the amount of explained variance is more than 50%, then it should be concluded that the analyses are affected by common method bias. In our case, the variance explained is 20%.
To facilitate the reading of subsequent analyses,
Table 6 provides a summary of all the acronyms used for our main variables.
4.4. Results of the Path Structural Models
To test the first hypothesis, the relationship between “eco-centric orientation” (EO) and “digital-environmental awareness” (DEA), as well as “benefit environmental orientation” (BEO) and DEA, were explored while controlling for sociodemographic characteristics. The results are reported in
Table 7, while
Table 8 reports the results for the test of the mediating effect of DEA (Hyp. 2).
Both EO (b = 0.309; t = 17.341; and p < 0.001) and BEO (b = 0.466; t = 20.998; and p ≤ 0.001) were found to have a significant effect on DEA. However, the direct effect of sociodemographics on DEA was not found to be strongly significant for any of the sociodemographic variables used, indicating that they have a minimal or inconsequential influence on the model.
Moreover, the direct effect of EO on “eco-centric digital-environmental behaviour” (EDEB), and of BEO on EDEB, was explored. In this case, EO, controlled by sociodemographic characteristics, had a positive and significant direct effect on EDEB (b = 0.588; t = 24.444; and p < 0.001), whereas BEO had a significant but negative effect on EDEB (b = −0.400; t = −12.156; and p < 0.001). DEA also had a positive and significant effect on predicting EDEB (b = 0.120; t = 3.275; and p = 0.001). The direct effect of sociodemographic variables on EDEB was found significant only for age, and with a negligible positive effect (b = 0.075; t = 1.997; and p = 0.046).
Additionally, the effects of EO and BEO on “benefit-oriented digital-environmental behaviour” (BDEB) were explored. Both EO (b = 0.107; t = 5.732; and p < 0.001) and BEO (b = 0.443; t = 18.257; and p < 0.001) had a positive and significant effect in predicting BDEB. DEA also had a significant positive effect on BDEB (b = 0.272; t = 10.054; and p < 0.001). The direct effect of sociodemographic variables on benefit-oriented digital-environmental behaviours (BDEB) was found to be positive and significant for age (b = 0.100; t = 3.6646; and p < 0.001).
The model’s fit indices, as shown in
Table 7, fall within an acceptable range: CMIN/df = 0.346, goodness-of-fit (GFI) = 1 (see [
88]), Tucker and Lewis index (TLI) = 1.012, confirmatory fit index (CFI) = 1 (see [
87]), standardised root mean square residual (SRMR) = 0.0012, and root mean square error approximation (RMSEA) = 0.000 (see [
89]).
The square multiple correlations were as follows: 0.41 for “digital-environmental awareness” (DEA), 0.47 for “eco-centric environmental behaviour” (EDEB), and 0.49 for “benefit-oriented digital behaviour” (BDEB). To assess the significance of the mediation of DEA (H2), we used a bootstrap technique with a bootstrap sample of 5000 and 95% of bias-corrected confidence intervals [
90].
In summary, the results show that DEA partially mediates the relationship between overall environmental orientation (both eco-centric oriented and benefit oriented) and digital-environmental behaviour (both eco-centric-oriented and benefit-oriented) (H2). The indirect effects of both EO (b = 0.037 and
p = 0.003) and BEO (b = 0.056 and
p = 0.002) on EDEB and of EO (b = 0.084 and
p = 0.003) and BEO (b = 0.127 and
p = 0.003) on BDEB (
Table 8) are statistically significant. However, DEA has a complementary mediation role in the relationship between EO and both types of digital-environmental behaviours. The effect ratio (between the indirect and total effect) suggests that DEA mediates 5% of the total effect. The proportion of the total effect mediated by DEA in the relationship between EO and BDEB is around 44%. DEA also has a complementary mediating effect between BEO and BDEB, mediating around 22% of the total effect. DEA has a competitive impact on the relationship between BEO and EDEB. This suggests that while overall benefit-oriented environmental predispositions reduce the likelihood of adopting eco-centric-oriented digital behaviours, digital-environmental awareness positively mediates this effect. The fit indices for this model (
Table 8) fall within an acceptable range: CMIN/df = 0.025, GFI = 1, TLI = 1.010, CFI = 1, SRMR = 0.0004, and RMSEA = 0.000. The square multiple correlations were 0.42 for “digital-environmental awareness” (DEA), 0.47 for “eco-centric environmental behaviour” (EDEB), and 0.49 for “benefit-oriented digital behaviour” (BDEB).
Finally,
Table 9 focuses on hypothesis 3. Hypotheses H3a and H3b were tested using a second path structural model, where we introduced the product terms for overall “eco-centric-oriented dispositions” and “digital expertise” (EO × DE) and “benefit environmental orientation” and “digital expertise” (BEO × DE). The interaction terms are the product terms of z-scores for both types of environmental orientations (EO and BEO) and “digital expertise” (DE). To simplify this second model, we excluded the sociodemographic variables that were insignificant in the first model or had a negligible effect. We retained only the variables essential for testing H3. In particular,
Table 9 shows that digital expertise itself does not significantly predict digital-environmental awareness. However, DE has a negative and significant effect on EDEB (b = −0.116; t = −4.674; and
p < 0.001) and a positive and significant effect on BDEB (b = 0.076; t = −3.111; and
p = 0.002). This suggests that individuals with more digital expertise are likely to engage in behaviours that benefit both themselves and the environment.
Regarding the interaction terms, EO × DE is positively and significantly related to DEA (b = 0.056; t = 2.860; and p = 0.004) and BDEB (b = 0.066; t = 3.569; and p < 0.001). This suggests that digital expertise strengthens the relationship between eco-centric dispositions, digital-environmental awareness, and benefit-oriented digital-environmental behaviours. However, the interaction term is negatively related to EDEB (b = −0.085, t = −4.522; and p < 0.001), suggesting that digital expertise weakens the relationship between EO and EDEB.
The interaction term BEO × DE is not significant for predicting both DEA and BDEB, but it has a positive and significant effect on EDEB (b = 0.032; t = 2.228; and p = 0.026).
In summary, digital expertise moderates the relationship between eco-centric orientations and digital-environmental behaviours, strengthening the relationship with digital-environmental awareness and benefit-oriented behaviours while weakening the relationship with eco-centric-oriented behaviours. For benefit orientations, digital expertise strengthens the ties with eco-centric-oriented behaviours.
To better understand the effects of the moderators a “pick-a-point approach” was used [
91], creating two new variables—low-level moderator (“low digital expertise”, LDE) and high-level moderator (“high digital expertise”, HDE)—by adding and subtracting the standard deviation from the standardised variable of digital expertise, respectively. The moderation analysis was conducted by introducing the product of the standardised variables for EO and BEO and the new low-level moderators (EO × LDE and BEO × LDE) and high-level moderators (EO × HDE and BEO × HDE).
The results reported in
Table 10 show that when users possess lower levels of digital expertise, the relationship between overall eco-centric orientations and digital-environmental awareness is not significant. In contrast, the relationship is significantly reinforced when they have higher expertise. The original test had an unstandardised regression weight of 0.380 (
p < 0.001), and now, under higher levels of DE, this is 0.989 (
p < 0.001). This suggests that as digital expertise increases, it also strengthens the relationship between possessing an overall eco-centric disposition and developing a digital-environmental awareness.
However, exploring the moderation of DE on the relationship between EO and EDEB, this relationship is significantly strengthened at lower levels of digital expertise with an unstandardised coefficient of 1.503 (
p < 0.001). This suggests that individuals with lower digital expertise tend to adopt more eco-centric-oriented digital behaviours when they have overall eco-centric dispositions. At higher levels of DE, the moderation effect becomes negative and not significant. At the mean level of DE, the unstandardised coefficient was 0.586 (
p < 0.001) (see
Table 10). This suggests that possessing lower levels of digital expertise reinforces the relationship between overall eco-centric dispositions and adopting eco-centric-oriented digital behaviours.
Exploring the relationship between EO and BDEB through the moderation of DE, possessing higher levels of digital expertise strengthens the relationship significantly, whereas possessing lower levels of digital expertise weakens it. The original test had an unstandardised regression weight of 0.128 (p < 0.001), whereas, under higher levels of digital expertise, the coefficient is 0.844 (p < 0.001) and under lower levels is −0.589 (p = 0.004).
To summarise, users with an eco-centric disposition tend to possess higher digital-environmental awareness when they have higher levels of digital expertise. However, when their digital expertise is lower, they are more likely to adopt digital behaviours that are eco-centric-oriented. By contrast, users with higher digital expertise are more likely to adopt benefit-oriented digital behaviours.
Exploring the moderation of DE on the relationship between BEO and EDEB, at lower levels of DE this relationship is significantly weakened with an unstandardised coefficient of −0.642 (p < 0.001) whereas, at higher levels of DE, the moderator is no longer significant and negative. At the mean level, the unstandardised coefficient was negative but not significant. The analysis revealed that the interactive term BEO × DE did not significantly affect benefit-oriented digital-environmental behaviour (BDEB) and digital-environmental awareness (DEA). As a result, the moderation of different levels of digital expertise between BEO and BDEB and between BEO and DEA was not tested.