*2.4. Research Model*

Figure 1 shows the research model of the study. In the model, Attitude Toward Online Purchase (ATOP), Subjective Norms (SN), Perceived Behavioral Control (PBC), Consumer Ethnocentrism (CE), and Country of Origin (COO) determine online consumer purchase behavior (OPB). Additionally, CE determines COO.

**Figure 1.** Research model.

#### **3. Materials and Methods**

Colombian consumers aged 18 and above from three major cities, Bogotá, Medellín, and Bucaramanga, participated in this research. The survey was distributed through email blasting and using Survey Monkey services. The email outlined the study's goals and

invited recipients to participate in an online poll in which they would remain anonymous. Every other week, we sent out email notifications to all the prospective participants. The data-collecting period lasted 20 weeks. The researchers received 398 questionnaires. However, 104 had to be discarded since they lacked crucial data for this investigation. Two hundred and ninety-four questionnaires were approved, yielding a response rate of 73.9 percent.

Regarding age, 23.1 percent of the participants identified as 18–24 years old, while 17.6 percent identified as 45–54 years old. In terms of gender, 51.7 percent of the participants were female, and 48.3 percent were male. Regarding educational attainment, 30.2 percent reported some college studies, 25.9 percent a bachelor's degrees, 9.2 percent some graduate studies but no degree, and 29.9 percent a graduate degree, respectively (see Table 1).


**Table 1.** Demographic Characteristics.

The questionnaire was created using scales that had been established in earlier research. The questionnaire included questions that were scored on a seven-point Likert scale. CETSCALE was proposed by [27], and it was used to develop the CE scale. The COO was determined using the scale of [51]. The TPB was determined using the model for the Internet buying environment proposed by [77]. Four latent variables were included, adapted from the TPB. Four items evaluated attitudes toward online purchases. In addition, two items assessed subjective norms, and two items evaluated perceived behavioral control. To verify semantic equivalence in the translation of the scales from English to Spanish, a pilot study with 50 participants was performed. A single question assessed online consumer purchase behavior: How much would you say you spend on Internet purchases each month? The questionnaire is exhibited in Table A1.

#### **4. Analysis and Results**

Partial Least Squares Structural Equation Modeling (PLS-SEM) was employed to analyze the data. The selection of the technique is due to the exploratory nature of the hypothesized relationships in this study. The PLS analysis consists of two phases, examining the validity of the measurement model and evaluating the structural model. The results of each stage will then be reported.

#### *4.1. Measurement Model*

The guidelines for the evaluation of the model were applied in accordance with the procedures described in the literature. Table 2 shows the measurement models, outer loadings exceeding 0.707 met technical recommendations, with the only exception being SN1 with a value of 0.676. Furthermore, composite reliability and average variance extracted (AVE) exceeded 0.83 and 0.59, respectively. Therefore, the convergent validity has been proven.

**Table 2.** Cronbach's Alpha, Composite Reliability, AVE, and Factor Loadings.


Discriminant validity was assessed based on the Fornell–Larcker criterion and the Heterotrait–Monotrait Ratio (HTMT) approach. Table 3 indicates these results, the square roots of the AVEs for the latent variables being higher than the correlations between them. Furthermore, HTMT values were all below 0.50, offering full support for discriminant validity.


**Table 3.** Discriminant validity analysis.

#### *4.2. Structural Model*

Table 4 and Figure 2 present the analysis of the structural model. The assessment of the structural models started with the evaluation of the Standardized Root Mean Square Residual (SRMR) for the estimated models. As a result, the model has an SRMR of 0.058, an appropriate fit based on the cut-off value of 0.08. Overall, both OPB R<sup>2</sup> and COO R2 indicate lower effects. Similarly, the Q<sup>2</sup> predict values show the lower predictive accuracy for the model.

**Table 4.** Path coefficients and indexes of structural models.


Notes: \*\*\* *p*-value < 0.001, \* *p*-value < 0.05, ns non-significant.

Next, we examined the hypothesized relationships. For the research model, beta coefficients are positive and significant for the relationships between CE and COO, PBC and OPB, and ATOP and OPB but not for the other relations. The analysis supports the beta coefficients being significantly different from zero between the independent variables PBC, ATOP, and the variable OPB, and, in the same way, between the independent variable CE and the dependent variable COO. This same analysis cannot support beta coefficients being significantly different from zero between the independent variables CE, SN, and COO and the dependent variable OPB. This result indicates that with an increase/decrease in the PBC, ATOP variables' values implies an increase/decrease in the OPB variable. These effects follow the theory of planned behavior. On the other hand, it is not supported that the variation of the SN values impacts OPB. One explanation is that subjective norms are perceived as the same for these consumers, and in a pandemic context, and there is not enough variation for the model to detect it. In the case of the relationship between CE and COO, an increase/decrease in the values of the CE variable implies an increase/decrease in the COO variable, a situation following the idea of social identity in the context of consumer behavior. Finally, the COO variation does not affect the OPB variation. A possible reason is that most products bought online for the consumers surveyed in a pandemic context do not include the country of origin as an essential attribute. Consequently, hypotheses H2, H4, and H5 are supported for the model.

**Figure 2.** PLS results. Notes: \*\*\* *p*-value < 0.001, \* *p*-value < 0.05, ns non-significant.
