Questions about the method

The variables at issue cannot be measured directly. Nobody can read off leadership efficiency or customer value; they are captured through several test statements. A structural equation model (SEM) estimates in one step how well the statements represent their construct and how the constructs act on one another. A correlation looks only at pairs of values. The SEM tests an effect structure derived in advance from theory as a whole. Regression and path analysis work only with directly measured variables. The QMC model, the quality management model for the communication industry, contains seven latent constructs and one mediating latent variable, communication quality. For this set-up, the thesis chose the SEM and computed it in AMOS 21.

In a reflective model, the construct comes first. The indicators reflect it without describing it completely. In a formative model it is the other way round: there, the construct is composed of its indicators. The thesis chooses the reflective form because a literature review never finds all components of a construct. For example, the model lacks an employee-related success parameter that could well count towards communication success. So that such gaps and measurement errors are absorbed, the model contains additional error variables.

The categories of the initial model come from the literature. Most studies there consider only individual parts, and on quality management in communication agencies itself the search found hardly any studies. Before contractors rate the categories, practitioners were to check whether the terms are understandable and correctly assigned. Action research involves the participants in this check. This fits the basic assumption of the thesis that communication quality arises in dialogue, and eases the later application of the model. Habermas objected to action research that it neglects theory. The pre-test therefore built on the literature-based model and restricted itself to checking its categories.

Executives from two internationally active communication agencies rated each of the 26 categories on a scale from 0 to 3; categories clearly below the total mean of 1.70 were dropped. Six were discarded, 20 went into the survey as indicators (key figure of the pre-test).

Contractors are suited in a double role. They know the agency from working together, and they see the effect of the projects in their own company, in the reactions of their customers, in public perception and in their own business figures. That is why in the questionnaire they first rated the agency and then the consequences of the project for their own firm.

1,331 companies were addressed that had given one of two agencies a completed order of more than €100,000. 298 responded, 89 of them broke off early. 209 complete questionnaires were evaluated, 15.7% of those addressed. The thesis follows the rule according to which an SEM needs at least 10 cases per indicator and at least 200 cases overall. With 20 indicators, that makes 200. The 209 data sets lie just above this.

The survey ran anonymously via an internet-based survey system. The researcher knew neither names nor other personal data, position and answers were stored separately, and the questionnaires were destroyed after the evaluation (Method).

The thesis regards a path as significant if its p-value is below 0.10. This corresponds to an error probability of at most 10%. For an SEM to have a unique solution, each construct needs a fixed reference. That is why one path per construct is set to the value 1. Such reference paths have no p-value of their own and are not tested individually. This also affects two of the four accepted core hypotheses, the path from leadership efficiency to communication quality and the one from communication quality to customer value.

They are standardized weights from the final model. Standardized means that all variables are converted to the same unit, so that the strength of different paths can be compared directly. They are not percentages. 0.98 applies to the path from leadership efficiency to communication quality, 0.80 to the path to customer trust, 0.14 to employee quality and reputation gains. The two largest values belong to reference paths that were not tested individually. All paths with estimate and p-value are set out under Results.

No indicator of financial value was significant, neither stock value/market value nor return on equity (ROE) nor Tobin’s q. The path from communication quality to financial value, too, missed the threshold, with a p-value of 0.24. The thesis therefore discarded the area in the first amendment; its reasons are set out under Communication success.

In the initial model, several indicators contributed nothing to the explanation, and some even showed negative weights. The thesis sees the cause in the selection and assignment of the indicators. It therefore removed step by step what was not significant and re-estimated the model after each step. The final model retains leadership efficiency, employee quality and communicational strategy, captured through seven indicators, and, on the success side, customer trust and reputation gains. Nine variables are still measured directly in it. Resource integration, financial value and the constructs social value and customer value are no longer included in it.

The study considered a preliminary exploratory factor analysis dispensable because the pre-test had already checked the assignment of the categories in conversation. In retrospect, the thesis acknowledges that such an analysis could possibly have grouped the indicators more efficiently.

Contractors of two internationally active agency networks were surveyed. Their companies are spread across manufacturing, services, sales and other branches as well as across small, medium-sized and large firms. The findings initially describe this group of contractors. How long the relationships hold the thesis leaves open. A structural change in the business, further growth of web-based communication or a changed way of perceiving quality and measuring success could shift them. Open points are collected on the page Limitations and outlook.