Appendix Respondent Pool, Statistical Tests, and Presentation of Results
The potential pool of respondent firms has two characteristics: (1) the firms are research intensive; and (2) access is available via a third party to an R&D manager who is familiar with decisions on the placement of R&D facilities. The “third” parties for the second characteristic are industrial research groups that not only provided contact information on R&D managers who are members of the organization, but also sent letters introducing the survey and encouraging participation. Without letters of introduction, response rates would be very low. Membership in the organizations who aided our survey is self-limited to research-intensive firms and, in general, to large firms.
It is appropriate to have multiple respondents from a single firm. If decisions on R&D site locations are made independently by multiple entities in a single firm, then each entity is an appropriate respondent. For example, if decisions are made at the level of business units, then each business unit could potentially provide a response. A total of 418 firms were contacted, and responses from 203 firms were received. This is a 48.6 percent response rate. One reason for the high response rate is that we had multiple potential contacts for many firms. Thus, a non-response from one contact from a firm might be negated by a response from another contact at that same firm. As noted, it can be appropriate to have multiple responses from the same firm so long as each respondent is responding for a different decision-making unit.
From the 203 firms there are 250 responses. Each respondent was asked for the name of the unit for which they were responding so that a check could be made that multiple responses were not being received from the same decision-making unit. Of the 250 responses, 76 (30.4 percent) were from business units and 174 (69.9 percent) were responding for a corporate R&D unit. There is no way to determine whether a non-respondent would have answered for a business unit or for a corporate R&D unit.
Results of statistical tests are presented in the text and below the figures. With the exception of the test reported for the comparison of the data in Figure 5, all tests are standard tests of equality of means. It is assumed that variances are unequal and observations are not paired. Alternative tests would consider whether distributions of responses are significantly different. Since the comparisons in the figures and the discussion in the text are for differences of means, the tests are tests of equality of means rather than distributions. The statistical test reported for the data in Figure 5 is Pearson’s chi-square test of whether the distributions of years for sites inside and sites outside are significantly different.
The level of significance reported for tests is the largest significance level at which the null hypothesis of no difference is rejected. Thus, if it is reported that a group of comparisons are significantly different at the 5-percent level, then every test statistic has a p-value of 0.05 or less. Some might be less than 0.01 (that is, a 1-percent significance level). If differences are reported as not being significantly different, then a significance level of 5 percent has been used.
The number of respondents answering questions is presented below most figures. This is typically a range since the figures are drawn from a number of questions. Thus, for example, below Figure 4 it is reported that there are 223–230 respondents, indicating that a minimum of 223 respondents answered each of the seven questions in Figure 4 and for some questions there were 230 respondents.