*Sourcing Innovation in the Digital Age DOI: http://dx.doi.org/10.5772/intechopen.111707*

Because of the geographic and industry diversity in our sample, we do not restrict answers to a particular economic dimension, lest we force a company whose most important innovation is in one area (e.g., cost-cutting) to report on a less-important project in another area (e.g., revenue generation). While this choice comes at the cost of additional variation in the type of importance meant by respondents, we argue that this reflects true industry variation in where innovation is most important and is thus the right empirical choice if the desired outcome is to produce broad claims about the sources of the most important innovations.

The survey also asked about a second innovation project, one that came from a different source than their first project. This innovation source was randomly chosen for respondents from amongst those they were working with. Gathering data about a second innovation project provides two main benefits. First, it allows us to understand how innovation sources are being used even when they are not producing the mostimportant innovation. Second, our construction allows us to use compound lottery sampling ([28], p 169) to make statistically valid inferences about their answers had we asked them about a project from a random innovation source (survey testing confirmed that respondents typically called to mind the most important innovation that they had done using this secondary source, and had no trouble providing concrete details about the "enablers, features, and outcomes of business innovations."). Project-level questions that were gathered for both the most important project and the second project included:




#### **Table 1.**

*Innovation sources.*

• The extent to which the project gave the company an enduring advantage over competitors (**Appendix A**, Exhibit 3)

The definition of innovation provided to respondents was "the ways that companies create new products, services, business models or improve their existing ones". Our initial interviews and survey testing confirmed that respondents had no difficulty interpreting this, which is unsurprising since so many work directly in the innovation field. We also reinforced our intended meaning through examples. The innovation sources included in the survey are shown in **Table 1**.

### **3.2 Survey administration and sampling**

The survey was launched in August 2018 and lasted six weeks. The survey company, Phronesis Partners, administered the survey respondents via telephone (predominantly) and/or through an online platform (when needed).

The survey was conducted across seven industry sectors: Consumer Products and Retail, Manufacturing, Automotive, Financial Services (banking, insurance), Pharmaceutical & Life Science, High-Tech, and Utilities. It also covered eight countries: US, France, Germany, UK, Australia, China Mainland/Taiwan/Hong Kong, Japan, and South Korea. This allows us to provide a more global view of open innovation than previous work. It also allows us to examine country and sector differences (which we do in another paper).

Respondents were allowed to either answer in English or via a translated questionnaire in French, German, Mandarin, Japanese, or Korean. Only large firms were targeted for the survey, with one-third having annual revenues of USD \$500 m-\$1bn, and the remainder having revenues greater than USD \$1bn.

Companies were sampled in order to be *statistically representative*, meaning that (up to sampling error, which was minimized through substantial samples) the conclusions from our sample would be expected to be quite close to those if *all* firms in that group had been surveyed. This was done as follows. For each country, a list was compiled of all companies in the target industries that had revenues greater than \$500 m using Dunn and Bradstreet/Hoovers. Each country's list was then sub-divided into lists for each industry and then each of two revenue size categories (\$500 m-\$1bn, and \$1bn+) Then, *by random draw* from each of these sublists, firms were offered to participate in the survey. Because of this random sampling, each of these subcategories is representatively sampled. The success of this approach in producing a representative sample is shown below.

Sufficient numbers of firms were sampled such that, after response rate and screening were taken into account, respondent counts would meet target levels. The response rate to the survey was 34% (e.g., in contrast with 20–24% for large firms for [20]).

Of those who agreed to participate, 7.4% were screened out for having too little knowledge or responsibility for innovation within their company to participate in the survey. Post-survey, we matched firms to Capital IQ to gather their financial information.

#### **3.3 Target respondents**

For each company, the target respondents were innovation leaders that held managerial or higher rank and who had a holistic view of the firm's innovation activities. The respondent's seniority was determined from their job title and job description in LinkedIn Navigator, ZoomInfo, or the survey company's internal database. Job target titles included: Chief Innovation Officer, Head of R&D, Head of Open Innovation, etc., which is broadly similar to the respondents reached by the European CIS surveys [16]. To ensure our survey reached sufficiently senior innovation leaders, the target ratio of Executives, Directors, and Managers was 30–40–30%. To ensure that respondents had sufficient knowledge and responsibility for innovation within the company, early survey questions explicitly screened for these characteristics and excluded respondents that did not meet them.

#### **3.4 Sample summary statistics**

**Table 2** shows the summary statistics for our sample. In order to maintain the anonymity of the firms in our sample, we present these only in aggregated form, at the level of industry or country. The column "# Companies: Sample" shows the number of observations per country and industry, which met our targets almost exactly. The sample for seniority level was also met closely, with 89 executives/CXOs, 119 Directors, and 90 Managers interviewed, as was the sampling of firm sizes, with 98 with \$500 m-\$1bn in yearly revenue, and 202 with \$1bn + .

To compare the representativeness of our sample with the underlying populations, we use the sector and country categories in Capital IQ. This is an approximation of the population used for actual sampling, since the sample group originated in D&B/ Hoover's. But approximating with Capital IQ allows us to compare revenue, profitability, growth rates, and firm sizes (one exception here is the South Korean banking industry for which only a small number of firms reported the number of employees, thus we omit that comparison to maintain anonymity). To match the sample composition, we weight the population values as one-third, two-thirds based on firm sizes of \$500 m-\$1bn and \$1bn + .

Our results in **Table 2** show that our mean sample values are similar to the population mean values. Per the statistical guidance in Imai, King, and Stuart [29], we do not commit the fallacy of running a t-test on these variables, since that conflates sampling accuracy with test power and therefore erroneously favors small samples. Instead, we do a Q-Q plot (**Appendix C**), which shows that, if anything, our deviations are slightly more centrally clustered (and therefore more representative) than one would expect from a normal distribution.
