**3.3. Validity and scale reliability**

**3.2. Measurement instruments**

According to the location of most network members.

**Years since creation to entrance in the network**

**Years of international** 

**experience**

**Average years elapsed = 0.28**

*Note:* \*

Years elapsed

**Average years elapsed = 0.12**

1

transform and exploit knowledge.

To measure the market orientation of the network we adapted the scale proposed by Helfert et al. [37]. These authors move away from the idea defended in previous studies of measuring the network's market orientation by simply adapting the dimensions used in the seminal scales of MO. Specifically, this scale includes a total of 12 items on the four dimensions that reflect the relational processes management of NMO: coupling (2 items), coordination

**Network size Geographical scope of the network1**

region

Years elapsed % Firms % Scope % Type %

1 13.5 4–5 19.4 Technological network 14.9 2 1.4 6–10 11.1 National 15.42 Institutional network 2.0 3 2.7 >10 13.9 Infrastructure network 5.5

**Average firms = 5.81** International 22.39 Marketing network 92.5

**International activities in the value chain\* Activity sector**

7.40

11.18

After-sales service 16.51 Services 7.2

% Employees % Activity Average% Sector %

**Average international activity = 17.63**

0 76.6 3–5 26.4 Manufacturing process 11.58 Industrial 61.2

2 2.1 12–25 23.8 Commercialisation 41.50 Commercial 31.6

development

promotion

**Type of network**

62.19 Social network 2.5

Market network 15.4

To measure absorptive capability, we use the three-item scale of Ref. [87] that evaluates the degree to which the firm's management systems encourage the ability to acquire, assimilate,

(3 items), conflict resolution (3 items) and exchange (4 items).

**Table 2.** General characteristics of the main networks of the studied firms.

0 82.4 3 55.6 Self-governing

**Total number of employees**

54 Knowledge Management Strategies and Applications

1 19.8 6–11 25.4 Research and

3 1.6 Over 25 24.4 Advertising and

Figures expressed as a percentage of total responses.

**= 28.55**

**Table 1.** General characteristics of the firms.

**Average employees** 

To refine the scales, a confirmatory factor analysis was performed using structural equations models. The analyses guarantee a measurement model consistent with the theoretical proposals, supported by scales that are reliable, valid and present a certain degree of unidimensionality.

Based on the recommendations of Jöreskog and Söbom [88], we first examined the estimation parameters. We removed those indicators with standardised coefficients (*λ*) under 0.7, significance of the Student *t* statistic under 2.58 (*P*=0.01) and *R*<sup>2</sup> under 0.49, thus ensuring that the strong and weak convergence conditions were met [89]. This process led to the removal of the indicators EXCH.3 from the NMO scale, ABS.1 from the absorption capability scale, and CACOS.3, CADIF.3 and CADIF.4 from the competitive advantages based on costs and differentiation, respectively. Several tests were then performed to verify whether or not the process of refinement of the scales had altered their level of reliability. We used Cronbach's alpha [90] to analyse internal consistency. Other complementary tests of reliability were carried out: the composite reliability of the construct and the analysis of the average variance extracted (see **Table 3**).

Finally, the convergent and discriminant validity were analysed. With reference to the former, it was sufficient to observe that the estimated value of the correlations between the dimensions configuring the scales was high and significant. The confidence interval test was performed to examine discriminant validity, verifying that '1' was not found in the intervals estimated for the correlations between each pair of dimensions (**Table 4**). The measurement model proposed is therefore reliable and valid for use in the testing of hypotheses.

Further tests were also carried out. First, we checked for signs of multicollinearity by testing the variance inflation factor (VIF) among latent variables in our proposed overall model. Values were below 10 [91], suggesting multicollinearity was not an issue in our study. Second, a *t*-test of independent means was performed on the dimensions of the variables in the proposed model. This test was conducted using the first 45 and last 45 respondents. No significant differences were found between these respondents at the 0.05 level, indicating an absence of non-response bias [92]. Third, various ANOVA were run to confirm that sample characteristics do not affect the model constructs. The following control variables were used, based on the data gathered in the questionnaire: sector of activity, international consolidation, age, international seniority, size, seniority in the network (all firm-related variables) and size of the network. Results revealed no significant differences in any of the analyses. Finally, the possibility of common method variance bias was tested with Harman's test, concluding that the bias caused by the method used was not a problem for the validity of the results obtained in the subsequent testing of the hypotheses [93, 94].


**Table 3.** Summary of the results after the definitive factor analysis.


**Table 4.** Discriminant validity analysis using the confidence interval (CI) test.
