6.3 Validation and Operationalization of Multi-Item Constructs Used in the Study
6.3.2 Validation and Operationalization of Formative Constructs
As described in chapter 5.1, great care has been taken to include all facets (dimensions) of the latent variables under investigation in the definition of the constructs' content domain. Indeed, all of the formatively specified constructs are conceptualized as multidimensional (see Table 13). For example, the construct measuring a firm's degree of internationalization (DOI) comprises the performance, structural, and attitudinal attributes of corporate
internationalization.
Construct Construct dimensions / Scope of content domain
DOI Performance, structural, and attitudinal attributes of corporate
internationalization
AUTO Degree of subsidiary autonomy in strategic and operational decisions
KNOW Lateral and vertical intra-company knowledge flows
SOCON Microstructural tools (organizational anatomy), informal communication channels (organizational physiology), development of shared values and beliefs (organizational psychology)
CONFIG Concentration (number of locations) and geograhic dispersion (where in the
world) of nine value activities comprising the entire value chain
TSLACK Available slack, recoverable slack, and potential slack
Table 13: Dimensions of Formative Constructs (Source: own illustration)
The indicators used to measure each of the above latent constructs do not only fully capture the defined content domain of the respective construct but – due to their history in the international management literature– also appear to be reliable and suitable for the purpose of this study. Thus, the formative constructs fulfill the requirements regarding adequate content and indicator specification that are stipulated by Diamantopoulos and Winklhofer.490 In other words, the formative constructs used in this study appear to have adequate levels of content and face validity.
Two steps were taken prior to performing the statistical tests to further validate the formative constructs used in this study. Because the three indicators used to measure total organizational slack (ASLACK, RSLACK, and PSLACK) were on different scales, the scores on these three slack measures were normalized to a scale to 0 to 1. As stated previously, the configuration of a firm's value chain is made up of the individual
configurations of its value activities. Thus, the individual configurations of a firm's value activities function as indicators of the firm's total value chain configuration. However, the configuration of a particular value activity is measured by its degree of concentration (number of locations in which the activity is performed) and its geographic dispersion (i.e. where in the world the activity is performed). Thus, in order to obtain a single value per indicator, it was necessary to combine these two sets of information for each of the nine value activities that are analyzed for each sample MNC. This was done by multiplying the score obtained for the concentration of a particular value activity by the score obtained for its geographic dispersion. Multiplication of the two scores was chosen over simple
algebraic summation because MNCs need to score high on both dimensions (concentration and geographic dispersion of their value activities) in order to simultaneously benefit from economies of scale and the arbitrage of national differences in factor, product, or capital markets.491 The possible configurations of an individual value activity that may result from the described procedure are illustrated in Figure 15.
Concentration of value activity,
i.e., number of locations in which activity is performed
Geographic dispersion of value activity,
i.e., number of countries in which activity is
performed Score
Configuration options
• In only one location for the entire company (scored 3)
• Only domestically/ in the home country of headquarters (scored 1)
• In only one location for the entire company (scored 3)
• In only one foreign country (scored 2) • In multiple/a few locations
(scored 2)
• Only domestically/ in the home country of headquarters (scored 1)
• In multiple/a few locations
(scored 2) • In only one foreign country (scored 2) • In multiple/a few locations
(scored 2)
• In multiple foreign countries (scored 3) • In each subsidiary individually
(scored 1)
• In multiple foreign countries (scored 3)
1 2 3 4 5 6 3 6 2 4 6 3 Concentration of value activity,
i.e., number of locations in which activity is performed
Geographic dispersion of value activity,
i.e., number of countries in which activity is
performed Score
Configuration options
• In only one location for the entire company (scored 3)
• Only domestically/ in the home country of headquarters (scored 1)
• In only one location for the entire company (scored 3)
• In only one foreign country (scored 2) • In multiple/a few locations
(scored 2)
• Only domestically/ in the home country of headquarters (scored 1)
• In multiple/a few locations
(scored 2) • In only one foreign country (scored 2) • In multiple/a few locations
(scored 2)
• In multiple foreign countries (scored 3) • In each subsidiary individually
(scored 1)
• In multiple foreign countries (scored 3)
11 22 33 44 55 66 3 6 2 4 6 3
Figure 15: Possible Configurations of Individual Value Activities Captured by the Value Chain Construct (Source: own illustration)
491 Compare the recommendations made by Albers and Hildebrandt (2006), p. 13 and the approach taken by Homburg et al. (2002), p. 94.
6.3 Validation and Operationalization of Multi-Item Constructs Used in the Study 155 As can be seen, the configuration options with the highest scores are those that
simultaneously allow for the realization of economies of scale and the exploitation of arbitrage opportunities (options 2 and 5). This is also illustrated in Figure 16, which
classifies the possible configuration options according to their potential to realize these two distinct benefits of corporate internationalization in a 2x2 matrix. Because the combined scores used to measure the individual configurations of a firm's value activities are
obviously aligned with the hypothesis developed in chapter 4492, they can subsequently be used as indicators of the firm's total value chain configuration.
Opportunity to arbitrage location differences Low High High Low Opportunity to realize economies of scale & scope
1 2 3 4 5 6 Configuration options Opportunity to arbitrage location differences Low High High Low Opportunity to realize economies of scale & scope
1 2 3 4 5 6 Opportunity to arbitrage location differences Low High High Low Opportunity to realize economies of scale & scope
11 22
33 44 55
66
Configuration options
Figure 16: Reflection of the Benefits of Corporate Internationalization in the Possible Configurations of Individual Value Activities (Source: own illustration)
The results of the statistical tests used to further assess the validity of the formative constructs employed by this study are presented in Table 14. As can be seen, for each of the constructs the maximum VIF of the indicators is far below the commonly used threshold value of 10. This result does not only indicate that there is no concern about indicator collinearity but also lends support to the notion that each formative indicator contributes a unique facet to the total meaning of its respective latent construct. Thus, based on the assessment of the collinearity among their indicators, the formative constructs under investigation appear to be reliable and valid. Moreover, except for the construct measuring total organizational slack (TSALCK), all test statistics obtained from the
nonparametric bootstrap tetrad test and the Bollen-Stine bootstrap test are statistically significant (p-values < 0.05). This finding also provides strong support for the validity of the formative specification chosen for the five constructs measuring DOI, AUTO, KNOW, SOCON, and CONFIG.493
Bollen-Stine Bootstrap Test
T1 statistic p-value p-value FSTS 1,772 FATA 1,623 OSTS 1,491 PDIO 1,409 INTOR 1,225 C1 2,275 C2 2,335 C3 1,373 C4 1,544 C5 2,101 C6 1,201 C7 1,412 D1 1,296 D2 3,472 D3 1,237 D4 3,349 E1 1,580 E2 2,849 E3 2,219 F1 1,732 F2 1,380 F3 1,465 G1 1,240 G2 1,736 G3 1,829 Ha 1,186 Hb 1,184 Hc 1,196 Hd 1,510 He 1,307 Hf 1,456 Hg 1,914 Hh 1,464 Hi 1,413 ASLACK 1,654 RSLACK 1,015 PSLACK 1,635
TSLACK n/a n/a n/a
CONFIG 5,283 0,042 0,031
SOCON 7,382 0,002 0,007
KNOW 0,123 0,024 0,009
AUTO 1,872 0,040 0,048
0,001
Construct Indicators VIF Tetrad Test
DOI 0,641 0,001
Table 14: Results of Reliability and Validity Tests – Formative Constructs (Source: own illustration)
Because the slack construct is comprised of only three indicators, neither the bootstrap tetrad test nor the Bollen-Stine bootstrap test could be performed.494 Therefore, the validation of the slack construct largely has to rely on conceptual considerations. At this juncture, the prevailing conceptualization of organizational slack provides valuable
493 See chapter 6.1.2.
494 Taking a fourth indicator from another latent variable as suggested by Bollen and Ting (2000), p.10 was also no feasible strategy in this particular case because this indicator should come from the same nomological network as the construct under investigation and, within the pool of indicators used in this empirical study, no such indicator was available.
6.3 Validation and Operationalization of Multi-Item Constructs Used in the Study 157 insights. Bourgeois and Singh conceive organizational slack as being composed of three interrelated but conceptually distinct dimension (available slack, recoverable slack, and potential slack) which differ in the ease or quickness with which the slack resource could be recovered for potential redeployment. Thus, failing to include one of these dimensions, i.e. dropping on of the three indicators used in this study, would significantly change the conceptual and empirical meaning of the latent construct. Along the same lines, Bourgeois and Singh propose to create an overall measure of organizational slack by summing the measures obtained for a firm's available, recoverable, and potential slack. 495 Together with the low VIFs found for the three indicators of total organizational slack (see Table 14), these considerations provide sufficient confidence to also deem the formative specification of the slack construct as reliable and valid.
As described in chapter 6.1.2, discriminant validity between formative constructs can only be assessed using their composite scores. Composite scores (indices) of the formative latent constructs used in this empirical study were calculated by summing the scores obtained for their respective indicators. For example, a single composite score (index) of a firm's total value chain configuration (CONFIG) was calculated by summing the scores obtained for the individual configurations of its value activities496 (see Figure 17).
Score on number of locations - activity A - Score on number of foreign countries - activity A - Score on the configuration
of activity A x Score on number of locations - activity B - Score on number of foreign countries - activity B - Score on the configuration
of activity B x Score on number of locations - activity I - Score on number of foreign countries - activity I - Score on the configuration
of activity I x … … … … +
Index value of the total value chain configuration of one particular sample company
Score on number of locations - activity A - Score on number of foreign countries - activity A - Score on the configuration
of activity A x Score on number of locations - activity B - Score on number of foreign countries - activity B - Score on the configuration
of activity B x Score on number of locations - activity I - Score on number of foreign countries - activity I - Score on the configuration
of activity I x … … … … +
Index value of the total value chain configuration of one particular sample company
Figure 17: Construction of the Index Measuring a Firm's Total Value Chain Configuration (Source: own illustration)
The procedure used to create composite scores (indices) of the formative constructs used in this study follows the recommendations of Rossiter and Albers & Hildebrandt for the 495 Cf. Bourgeois and Singh (1983), pp. 43-44.
496 For each sample company the configurations of nine particular value activities are examined. These value activities are able to describe the complete value chain of a company because they encompass upstream and downstream as well as primary and secondary activities (see chapter 5.1.3).
treatment of formative indicators.497 Moreover, "the summated rating scale is one of the most frequently used tools in the social sciences."498 Thus, once items have been selected following reliability and validity tests, "it is common practice to combine them to generate overall (i.e., aggregate) measures of the construct(s) of interest."499
To assess the discriminant validity of the formative constructs under investigation, Pearson correlations between their composite scores were calculated. As can be seen in Table 15, the correlation coefficients between all pairs of formative constructs are far below the suggested threshold level of 0.8. This result clearly indicates the absence of strong linear associations between the constructs and therefore provides strong support for their discriminant validity. That is, the formative constructs employed by this study are sufficiently different from each other to be considered legitimate.
DOI AUTO KNOW SOCON CONFIG TSLACK
DOI 1,000 AUTO -0,097 1,000 KNOW 0,009 0,163 1,000 SOCON 0,126 -0,053 0,455 1,000 CONFIG -0,001 -0,245 -0,045 0,047 1,000 TSLACK -0,170 -0,240 -0,035 0,017 0,029 1,000
Pearson Correlations of Latent Variable Constructs
Table 15: Assessment of Discriminant Validity between Multi-Item Constructs (Source: own illustration)
Overall, the conceptual considerations and results of statistical tests presented in this section provide sufficient support to deem the formative latent constructs used in this empirical study as reliable and valid. Therefore, their composite scores can be readily used in the subsequent analysis of the internationalization-performance relationship and tests of the developed hypotheses.
497 Cf. Rossiter (2002), p 315; Albers and Hildebrandt (2006), p. 13 and pp. 25-29. 498 Spector (1992), p. 1.