**5. Results and Discussions**

Once the six hypotheses from the model presented above were stated, we resorted to the survey method, based on the questionnaire. We had five latent variables: W, R, S, D, T, and 38 related items are described in the above Table 2. In Table 3 below the means and the standard deviations (SD = standard deviation) for the variables included in the

study are presented. For the analysis, we used both the R and R-studio software package, and the Amos software package from IBM. We used a series of indicators to determine the validity of the model, such as the coefficient of determination—CD, the standardized root mean squared residual—SRMR, the root mean square error of approximation (RMSEA), the non-normed fit Tucker–Lewis index—TLI, the goodness of fit index—GFI, the adjusted goodness of fit index—AGFI, and the comparative fit index—CFI.


**Table 3.** The mean and the standard deviations for the variables included in the study.

#### *The Latent Variables*

W. "Consumers' willingness to change their shopping habits to reduce environmental impact" was evaluated on a Likert scale with five levels, between 1 (strong disagreement) and 5 (strong agreement). For this, the survey participants answered seven questions: W1–W7. The higher the values of each question, the greater the availability of consumers to changing their shopping habits. For these items, the lowest score was 2.519 for W2, which indicates a resistance to consumer change, when we take into account the factors of convenience and especially price, which would rather translate into value-oriented consumers. The highest score, 4.131, was obtained for W6, from which we conclude that the process of paradigm change is in full swing. This aspect was reinforced by the high values from W4 and W5 (3.928 and 3.911), which reflect the awareness among respondents.

R. "Retailers' increased concentration on responsibly answering to sustainability as a personal value of consumers changing their behavior" was evaluated through six questions, R1-R6; again, we used a Likert scale with five levels: Yes, Partially true, Neutral, Rather no, and No. The closer the values are to 5, the more we can consider that retailers pay more attention to sustainability as a personal value of consumers. The lowest score of 2.882 was obtained for R5 with regard to the awareness and validation of the Japanese

initiative known as "Society 5.0". On the other hand, programs supported by European bodies scored above 3.32, which is logical for a population that is largely pro-European.

S. "Retailers' sustainability agenda, including by fulfilling consumers' sustainability demands with new products and processes" was evaluated by defining a variable with six items, evaluated on a Likert scale with five levels: Yes, Partially true, Neutral, Rather no and No. Survey participants answered questions S1–S6. The higher the values, the better retailers' sustainability agenda was outlined. For these items we obtained the highest average values, with a score of 4.602 for S1. Consumers want sustainable products, but this should be reflected as little as possible in costs. Moreover, in the case of S1 we observe the lowest value for the standard deviation, which indicates a high degree of homogeneity among respondents on this question. At the opposite pole is item S4, with a score of 3.660, which indicates a slightly above average predisposition to do research on the sustainability of the products to be purchased.

D. "Retailers' digital transformation to aid consumers to adopt more sustainable lifestyles and to make informed choices in the omnichannel world" was evaluated by defining a construct with 10 items, evaluated on a Likert scale with five levels: Yes, Partially true, Neutral, Rather no, and No. The higher values for the 10 questions (D1–D10) reflect an increased concentration of retailers on digitizing processes, including the mode of retailer–consumer communication. The lowest value of 2.747 was obtained for the item D2, which indicates a still low use of communication via mobile devices. A series of questions obtained a score over 4.20, with D6 having a value of 4.224, which explains the role of the retailer in accessing online information with regard to the everyday products' impact on the environment and shoppers' health.

T. "Retailers' need to translate consumers' uncertainty into trust, identifying risks associated with disruptive technologies and making them less severe" was also evaluated, with the help of a Likert scale with five levels: Yes, Partially true, Neutral, Rather no, and No. Survey participants answered nine questions (T1–T9), and as a means of interpretation, the higher values indicate a consumer anxiety, which should be addressed by retailers by reducing the associated risks. For this last variable, we observe the narrowest range of the mean values, between 3.834 and 3.862—values which are significantly above average for all the items considered. In this context, it would be more interesting to note the differences between the degrees of homogeneity. Thus, for T1 we record the highest value of the standard deviation and implicitly a lower degree of homogeneity regarding the inability of retailers to predict to what extent they can help shoppers to make sustainable choices or strengthen their sustainable consumption routines. At the opposite pole is T9, with the lowest value of the standard deviation, which translates into an increased degree of homogeneity concerning the digital framework built by the retailer, which can deliver sustainable value from risk and transform consumers into trusted partners.

In the social sciences, although a number of measurement methods of internal consistency reliability have been proposed, such as Omega [176], GLB (Greatest Lower Bound) [177], GLB.fa, or GLB.a [178], the most frequently used is Cronbach's alpha index, still proposed since 1951. The standardized calculation formula is as follows:

$$\alpha = \frac{N\overline{c}}{\overline{v} + (N-1)\overline{c}}$$

where *N* represents the number of items, *c* is the average inter-item covariance among the items, and *v* is the average variance.

At least in the case of items that follow a normal distribution, the use of alpha is recommended, as it avoids overestimation problems. Another aspect that is intensely debated is related to the acceptable lower limit of Cronbach's alpha value. Cho and Kim [179] have reservations about the application of arbitrary or automatic cutting criteria, and suggest that the minimum accepted values should be determined individually based on the purpose of the research, the importance of the decision involved, and/or the research stage (i.e., exploratory, basic, or applied). Nunnally advances the idea of a threshold of 0.5

for exploratory stages, as well as a threshold of 0.9 for applied research [180]. In principle, a higher value (close to 1) is considered better, and the lower limit most often circulated is 0.7 (it can take values between 0 and 1; if the scores are not allocated correctly, alpha can also take negative values, although this is an exception). However, there is also the reverse of the medal; a value that is too high can highlight redundant questions. The alpha value can also be manipulated: Cortina [181], for instance, showed that variables containing at least 20 items will have a coefficient greater than 0.7, even if the intercorrelations between them are very small.

George and Mallery [182] suggest a multi-level approach consisting of the following aspects: "≥0.9—Excellent, ≥0.8—Good, ≥0.7—Acceptable, ≥0.6—Doubt, ≥0.5—Weak, ≤0.5—Unacceptable". It should be noted that reliability refers to data and not to scale or unit of measurement.

For calculating Cronbach's alpha indices, we used the "ltm" and "DescTools" packages developed under the R software [183,184]. With the help of these packages, we could calculate the potential values of alpha in case we eliminated any of the items. Thus, adjusting latent variables has become much simpler. In Table 4 below, we present the results obtained. We note on the one hand the value of 0.904 (excellent) obtained for the construct T. "Retailers' need to translate consumers' uncertainty into trust, identifying risks associated with disruptive technologies and making them less severe", as well as 0.824 for the construct D. "Retailers' digital transformation to aid consumers to adopt more sustainable lifestyles and to make informed choices in the omnichannel world". We also record acceptable values for R. "Retailers' increased concentration on responsibly answering to sustainability as a personal value of consumers changing their behavior" (0.785) and for S. "Retailers' sustainability agenda, including by fulfilling consumers' sustainability demands with new products and processes" (0.763). The only construct that raises questions but is still very close to the lower limit of acceptability is the construct W. "Consumers' willingness to change their shopping habits to reduce environmental impact", with a value of 0.682.


**Table 4.** Scale reliability.

The model generated with the help of the Amos software package produced by IBM is presented in Figure 3 below.

**Figure 3.** The model generated with the help of the Amos software package produced by IBM.

Five of the six working hypotheses were validated, because the *p*-value is less than 0.05. In the case of the sixth hypothesis, between "Retailers' digital transformation to aid consumers to adopt more sustainable lifestyles and to make informed choices in the omnichannel world" ("D") and "Retailers' need to translate consumers' uncertainty into trust, identifying risks associated with disruptive technologies and making them less severe" ("T"), there is a direct link, but the associated risk is approximately 16%. The output also generated by the Amos software package produced by IBM is presented in Table 5 below.


**Table 5.** The output generated by the Amos software package produced by IBM.

Ideally, the loading factors should be greater than 0.7, but we can consider that values greater than 0.5 are acceptable. Two of the coefficients are less than 0.6 (0.57 and 0.59) and seven are less than 0.7. The remaining 29 meet the optimal criterion. In the last phase, it is useful to evaluate the accuracy of the model by checking the fit indices. In the research we analyzed the four categories: absolute fit, incremental fit, residual-based fit, and predictive fit. By looking at these values, we reconstructed the model several times, and we will now discuss the final values which we consider to be appropriate.

The first thing we analyze is related to absolute fit indices that do not help to approximate the amount of variance that can be explained by the proposed model. The most important are the chi-square (which we want to be as small as possible) and the matching goodness index (or the adjusted matching goodness index). However, the chi-square, like the *p*-value and GFI, may be affected by the sample size. For example, in this case we would want a *p*-value greater than 0.05, which we did not get (*p*-value = 0.48). Given the large size of the sample, however, we consider that the model is good, looking further at the ratio between chi-square and the degrees of freedom (degrees of freedom): CMIN/DF = 1604. This ratio should be as low as possible, with some authors recommending values <5 as allowed in certain circumstances and values <3 as very good [185]. We can thus consider that the model is suitable for the covariance matrix.

Incremental match indices show the second block of information and try to compare the model built with a basic model. Most authors recommend values greater than 0.9 for each of these indicators. We take a closer look at the value of the standard matching index (NFI), which may also be affected by the sample size. Comparative fit indices (IFC) = 0.942; the Tucker–Lewis index (TLI, which is in fact a non-normed fit index) = 0.936; and the relative fit index (RFI) = 0.906 indicates a good fit.

Residue-based matching indices, as the name suggests, analyze the differences between the observed covariates and the estimated variants. Meyers et al. [186] show that values of these indicators higher than 0.1 indicate a poor match, between 0.08 and 0.1 a moderate match, less than 0.08 a reasonable match. An average approximation error (RMSEA) = 0.033 indicates a model that fits well. Moreover, the confidence interval does not include zero.

Predictive matching indices such as the Akaike information criterion (AIC), the Akaike constant information criterion (CAIC), and the Bayesian information criterion (BIC) are used only to choose the best of several models. The one with the lowest values is preferred.

#### **6. Conclusions**

#### *6.1. Summary*

It is highly relevant and urgent now to consider the discrepancy between consumers' attitudes towards sustainable consumption and their purchasing behavior with respect to sustainable products, and to better understand the link between retailers' physical strategies and the sustainable smart store of the future. Consumers' preferences and shopping patterns have never changed as quickly as they have in the context of the COVID-19 pandemic. The findings of this study can be of interest to scholars researching consumers' decision-making impacted by both their perspective towards sustainability and their willingness to participate in it, and may enable them to make better and more informed choices in their omnichannel journey. These findings are based on a retail sector with a particular configuration, which may have an impact upon the study's generalizability:


by Romanians AliExpress/Alibaba Group ranks second, in front of the e-commerce giant Amazon.

As mentioned earlier, since the obvious impact of the COVID-19 pandemic on consumers' behavior is global, a more or less unconscious shift appears to have taken shape in consumers towards sustainability and purpose-driven brands within a phygital retail landscape, with consumers taking into account the sustainability of retailers' entire supply chain process ensuring greater visibility and traceability. Retailers need increased concentration on answering in a responsible manner to sustainability as a personal value of consumers (changing their behavior) and must continue digital transformation (considering the complementarity of sustainability and digital technology) to aid consumers to adopt more sustainable lifestyles and to make informed choices in the omnichannel world. Sustainability is seen across countries, product categories (which can influence consumers' feelings with products' purpose-driven element), and consumer segments through a different lens, but beyond global consumers' similarities and differences (including generational and cohort differences) the sustainability-driven consumers seem to be purpose-driven, but impacted by their satisfaction with new and improved, real and immersive experiences resulting from the relationship between purposeful retail and purposeful shopping. On the other hand, purpose-driven positioning also means that retailers must manage consumers' emotions, with the difference in consumers' perceived value being created with the help of phygital marketing strategies (across all channels, touchpoints, and micro-moments) focused on sustainable brands' (having points of parity and differentiation) and products' value proposition and based on both sustainable consumers' and sustainable prospects' insights. This means that retailers need to listen to all consumers (including considering the risks associated with consumers' uncertainty and anxiety within the context of the COVID-19 pandemic and recession crisis) and improve their experiences, enabling hybrid shopping and making them choose sustainable smart stores (in-store, online, mobile, BOPIS/Click & Collect, BORIS, Curbside pickup, etc.).

In developing our research hypotheses based on a consideration of the existing evidence beyond our own prior work, practitioner experience, and the literature review including gaps identified by us and presented in each area of the literature, we payed attention also to key research directions suggested by a recent systematic review of consumers' motivations to make green purchase decisions, and have both avoided consumers' subjectivity in answering the questionnaire and considered lessons to learn from cross-cultural research.

Here are the key findings on the relationship between the major shifts in sustainable consumer behavior on the Romanian retail market and retailers' priorities in agilely adapting:

