Hypothesis Measurement Methodological reflection on Case Study 1

is likely to be very reliable because the coding rule – is it SIC code 3661 or is it not code 3661? – is precise. A high level of commitment is the “intensity of the relation” between the partners in the alliance and is determined by the type of alliance agree- ment, referring to the classification by Hagedoorn 1990. The greatest intensity of the relation can be found in joint or combined ventures, and the smallest in licensing agreements. Regarding measurement validity, it is not known whether “intensity of the relation” was as good a descriptor of level of commitment as it was meant to be. The method for measuring commitment using agreements and contracts is likely to be reliable.

5.3.8 Data presentation

All relevant data of each individual case were provided: whether the project was an incremental or a radical innovation project; why it was considered successful; and what the score of the three collaboration characteristics were.

5.3.9 Data analysis

Hypothesis-testing was straightforward: comparing the “observed” scores for the collaboration characteristics in the tables with the pre- dicted ones in the hypotheses. This test was conducted for each hypothesis and for each case separately, and each test result rejection or confirmation was evaluated on a case-by-case basis.

5.3.10 Implications for the theory

The two hypotheses on technological capabilities 2a and 2b were confirmed. This gave support to the corresponding propositions in the theory, at least for the domain of the test Nokia. The two hypotheses on collaboration history 1a and 1b were rejected. Case Study 1 does not conclude that the corresponding propositions in the theory are incorrect, but rather that they might be true for a more limited domain. The two hypotheses on level of commitment 3a and 3b were rejected. This is considered as evidence that the proposition is not cor- rect at all. Case Study 1 does not suggest a reformulation of a proposition that was not confirmed. A probabilistic proposition agrees with the test results, and it would be defensible to reformulate propositions 1a and 3a as probabilistic relations.

5.3.11 Replication strategy

Because the two hypotheses on technological capabilities 2a and 2b were confirmed for all cases it is concluded that a replication strategy should be applied in which the confirmed hypotheses are tested for cases that are very different from the ones studied here, i.e. innovation projects in other fields than telecommunication, and involving other companies than Nokia. With each new test, the researcher should put more energy into identifying and selecting a case that is less “typical” in order to increase the likelihood of a rejection of the proposition and which boils down to the same to try to get a sense of the boundaries of the domain to which the proposition applies. For the other propositions, the hypotheses were rejected. Then the researcher has two options for replication: 1. The researcher might interpret the proposition itself as cor- rect, but only in a more limited domain. Then a replication must be done with cases from a more limited domain, in which the theory points to a higher chance of its confirm- ation. This replication strategy was adopted for the two propositions on collaboration history 1a and 1b. It stated that these propositions, though not true for all projects undertaken by Nokia, might be true for a domain of innova- tion projects that does not include Nokia’s and perhaps some more projects. 2. The researcher might interpret the rejection of a hypothesis as evidence that the proposition could not be correct at all. Case Study 1 adopted this strategy for the two propositions on level of commitment 3a and 3b. A very different strategy could be to reformulate the propositions on the basis of the test results as probabilistic ones. This strategy would be defensible for propositions 1a and 3a. If such a strategy were adopted, the newly formulated propositions should be tested in a new study. If an experiment was not possible, this study could be a survey, either of newly collected data or by using the CGCP database. It is, however,