*3.2. Variables*

Since none of the defined constructs is directly observable, measurement scales consisting of a number of indicators were developed. Table 3 presents the indicators chosen for each construct.

#### 3.2.1. ICT Adoption

In order to measure how ICT are adopted in the hospitality industry, a latent variable with five indicators adapted from previous research was created [54–56]. These indicators were measured on a scale ranging from 1 (minimum importance) to 5 (greatest importance).

#### 3.2.2. CSR

In line with Gallardo-Vázquez et al. [57], CSR was assessed by a latent variable with seven indicators formulated from the most current and important theories relating to CSR's social, economic, and environmental activities [58–60]. The indicators of the CSR dimension were measured on a scale ranging from 1 (absolutely disagree) to 5 (absolutely agree).


**Table 3.** Constructs and dimensions used in the research.

The indicators in italics were not included in latent variables due to convergent and/or discriminant criteria of PLS path modeling. All the measures were Likert-type scales.

#### 3.2.3. Future Expectations

This time, hotel managers were directly asked to rate their confidence and expectations about the immediate future on a scale ranging from 0 (very bad) to 10 (very good).

#### 3.2.4. Performance

Hotels performance was evaluated with a scale created from previous research [61–63]. We have considered the financial dimension (three items) and the non-financial dimension (five items), in which the company's position with respect to its competitors is contrasted, which allows us to measure business success better than with accounting information [62]. A scale that ranges between 1 (absolutely disagree) and 5 (absolutely agree) was used to measure the items of the two established dimensions.

#### *3.3. Statistical Procedure*

This study adopted a confirmatory and explanatory approach [64]. For this purpose, using SmartPLS 3.3.2 software (SmartPLS GmbH, Boenningstedt, Germany) [65], the statistical technique of partial least squares (PLS), a variance-based structural equation modelling (SEM) [66], was used to validate the hypotheses developed in our model.

PLS-SEM was chosen for the following reasons: First, this model contains first-order composite type A, and a definitional relationship between the latent variables and their items is assumed in this model [64]. For this reason, PLS-SEM is considered the most appropriate static method to be applied when the latent variables are composites [67]. Second, this technique is the most appropriate to apply in a theory approach, such as that in the present research. The reason is based on the possibility of estimating multiple relationships between the variables [68], especially if they involve mediation. Moreover, it accounts for measurement errors in the constructs [69]. Third, PLS-SEM is also recommended in situations where a large sample size is not available [70]. As recommended by Henseler et al. [71], a bootstrapping technique with 10,000 subsamples was used to verify the hypotheses.

#### **4. Results**

We have assessed our PLS model in three stages: (1) Overall model, (2) measurement model and (3) structural model, in line with [72].
