Measure validation

The fundamental objective of measure validation is to produce observable scores as close to the true scores as possible. Moreover, eliminating measurement error is considered a critical step in producing this score. The true scores, especially for the subjective state questions, are not known to the researcher who must merely rely on inferences. The quality of inferences would depend directly on the procedures that were used to develop the measures and the evidence supporting their worth.

5.5.1 Reliability

Reliability is the degree to which measures are free from error and therefore yield consistent results (Peter 1979). Thus, reliability is the extent to which measures are repeatable in different situations by different respondents (Nunnally 1978:191). Reliability is a necessary condition for the validity of measures be-cause unreliable ones may decrease the correlation between them. To increase reliability, the use of multi-item measures has been suggested in literature, as they allow measurement errors to cancel out each other, thus increasing the reliability of the scale (Peter 1979).

5.5.2 Validity

Validity refers to the degree to which instruments truly measure the constructs they are intended to measure (Peter 1979; Nunnally 1978:86). A measure is valid when the differences in observed scores reflect true variations in the characteris-tics one is attempting to measure. Validity is divided into internal and external validity. Internal validity reflects the study’s overall ability to measure what it was intended to measure. External validity, on the other hand, refers to the gener-alizability of the research findings across other firms in the population. Internal validity of a study is further divided into content validity, criterion-related validity and construct validity.

Content validity is the extent to which indicators provide adequate coverage of the topic under study (Emory 1980:129). The content validity for this study was ensured by carefully choosing the literature related to the subject matter. The items were selected from previous studies and were further supported by the ana-lytical findings of the preliminary case studies. Thus, a representative collection of items was chosen. Further, several questions related to a scale were posed to

the respondents. As mentioned in Chapter 5, pre-testing of the instrument was carried-out to further enhance content validity.

Criterion-related validity (or predictive validity) reflects the success of measures used for certain empirical estimation purposes (Emory 1980). This refers to the extent to which the theoretical prediction of the outcome or existence of some concurrent behaviour can be assessed. It can also be seen as the extent to which a measure predicts the answers to another question or a result to which it ought to be related. Thus, predictive validity is used to estimate an important form of be-haviour that is external to the measuring instrument itself. This kind of validity is divided into predictive and concurrent validity. These sub-divisions only seem to differ from each other from a time perspective. It is suggested that any criterion must be judged in terms of relevance, freedom from bias, reliability and availabil-ity. Predictive validity is determined by the degree of correspondence between the two measures involved. The relevancy of the criterion can be increased by prop-erly defining the measure and its scores. For example, the success of export ex-pansion in this study was measured by relative as well as absolute measures.

Construct validity refers to the measurement of the abstract characteristics of the construct for which no empirical validation is possible. This kind of validity re-lates to what construct or concept is measured, and aims to assess the degree of generalization of the constructs. Nunnally (1978) describes discriminant validity as a closely related approach to assessing the construct validity of the measures. It is the extent to which a group of respondents is being measured, who are sup-posed to and indeed do differ, in regard to their answers. Churchill (1979) ex-plains this as the extent to which the measure is certainly novel, and not simply a reflection of another variable. Thus, scales that correlate too highly may be meas-uring the same rather than different constructs. Discriminant validity is indicated by predictably low correlations between the measure of interest and other gauges that are supposedly not measuring the same variable or concept.

To ensure construct validity in this study, the main theoretical constructs were developed from the theoretical approaches which were closely linked to the re-search questions. The construct of market knowledge competence was developed in the theoretical framework of the study, taking into consideration the key ideas of the theory which appeared to relate to the dependent variable. Then, hypothe-ses were developed to analyze different components of the construct, such as a given set of firm capabilities like new product development, marketing planning and implementation, alliance learning and management capabilities. Thus, a fo-cused approach was followed from the beginning to develop the construct and its variables. Further, construct validity was ensured by multi-item measures,

select-ing the sample of the study from the same industry and usselect-ing the preliminary case study findings in the survey. The aim was to ensure the generalization of the pre-liminary findings on a larger sample.