• No results found

3. Methodology

3.4 Limitations of the methodology

Similar to most studies, this paper also has methodological limitations. This section highlights the ones deemed most critical.

3.4.1 Sample

The sample of 3245 firms is meant to constitute the whole population of SMEs in Norway within the selected industries (with restrictions related to firm age). For this purpose, the Proff Forvalt database cross-referenced with a shareholder database provides a relatively comprehensive sample of the SMEs that potentially invest CVC. However, as the authors do not know the exact regularity in which these databases are updated, SMEs that made their first investment in 2020 might not have been included in the sample. Additionally, if the SME do not own their shares in the entrepreneurial venture directly, but rather

through a holding company which owns both the SME and the portfolio companies, they have not been included in the survey.

3.4.2 Exclusion by the authors

Exclusion by using the website of the SMEs could in some instances be unjustified, as the information on the website might not be up to date. Therefore, errors may have occurred where SMEs that should have been included, were excluded. Additionally, having three authors screen the SMEs also increases the chance of slight differences in the screening process. The authors attempted to reduce the likelihood of unjustified exclusions by implementing a predetermined list of exclusion criteria, as mentioned in section 3.1.1.

3.4.3 Non-response bias

Calling by phone is a method that can be used to increase the response rate, and it was deemed as both appropriate and necessary due to the uncertainty regarding how large the propensity to invest in entrepreneurial ventures is amongst SMEs. A low response rate could reduce the number of respondents to a point where a quantitative study was unfeasible. However, calling potential respondents does not come without challenges.

Issues with non-response biases can occur if there are patterns among the potential respondents that answer, and those that do not answer the phone. For instance, this can occur as some do not answer the phone if they do not recognize the number (Lavrakas, 2008). Additionally, the authors screened the SMEs’ websites for phone numbers to CEOs and executives, and it is likely that especially larger SMEs did not have phone numbers of executives listed on their website. If the phone numbers were not found through the

47

website or yellow pages, the call was made to the switch board. Therefore, there are likely patterns among the firms that transferred us to the executive employees, and the ones that did not and whose CVC-activity remains uncertain.

3.4.4 Translation

This questionnaire is predominantly based on the surveys of Berg-Utby et al. (2007) and Maula et al. (2005). The latter had to be translated from English to Norwegian, which can pose challenges to the validity of the measurements. To reduce this challenge, the

authors used a Modified Direct Translation technique, where the translations were discussed continuously with experts in the field of venture capital (Behling and Law, 2000).

3.4.5 Reliability & validity issues

This is a cross-sectional study, which excludes the possibilities of measuring test-retest reliability. A longitudinal study would be preferable, and it would improve the reliability of the data and help mitigate some of the issues related to response biases (Dikmen,

Heaton, Grant and Temkin, 1999). Three constructs posed reliability issues (see section 3.3.3), where two of these were adjusted by removing items with loadings below .3, and the last consisted of only two items. The constructs with removed items obtained

sufficient reliability, but arguably lost some validity, because they came from pre-existing and previously used survey constructs (as explained in 3.2). Nevertheless, this solution can be considered acceptable, because the values for Cronbach’s Alpha were below the minimum threshold of 0.6. The “Exploitativeness of the investment motivation” cannot be adjusted without making it into a single-item measure. This was not done, and as such this statistic is only reported as a descriptive, and not being used for inferring any conclusions.

3.4.6 Normality issues

The normality assessment showed that a number of this paper’s variables had deviations from normality. The severity of this varied, with items 4, 5, 6, 11, 14, 33, 42 and 59 being highly kurtotic and skewed in particular (see Appendix B). Parametric methods assume that the population from which the sample is collected has normally distributed scores (Pallant, 2013), making this a weakness of the paper’s analyses. However, the central limit theorem can somewhat justify the use of parametric methods, given that the sample size is large, defined as above 40 by Elliot and Woodward (2007). The central limit theorem claims that “sample means are approximately normal for sufficiently large sample sizes even when the original populations are nonnormal” (Elliott and Woodward, 2007, p. 26).

3.4.7 Likert scale

The Likert scale is widely employed in this paper’s associated survey. Researchers are increasingly becoming aware of the potential problems of assuming that ordinal level ratings like Likert scales approximate interval level scaling, even though they are commonly regarded as such (Pallant, 2013). Likert scales are supposed to represent an underlying continuous measure, and they should ideally only be parametrically analyzed

48

when they are combined into constructs that fulfill assumptions of normality and reliability, as opposed to analyzing them individually (Allen and Seaman, 2007).

Likert scale mean and standard deviation scores for both combined constructs as well as individual items are reported in this paper, because of the paper’s purpose of describing general characteristics of SMEs investing CVC. Arguably, many of the individual items are concretely and clearly defined, and they are considered as different aspects of the larger constructs rather than synonymous items (e.g. internationalization is a tangible aspect of market knowledge). Do note that caution should be taken in inferring conclusions from these values. In the paper’s inferential analysis section, the parametric analysis Pearson's r was employed, which can be considered acceptable as most of the included variables are made of constructs of several items, where reliability and normality have been assessed. However, a few of the variables come from individual items/questions, which creates a possible weakness in the conclusions inferred from correlations derived from these variables.

3.4.8 Common method variance

The research design of this paper is a cross-sectional study, where data is collected through a questionnaire at one point in time, which might reduce the quality of the dataset (Chang, van Witteloostuijn & Eden, 2010). This weakness is further enhanced as the exogenous and endogenous variables are gathered from the same respondent through self-reporting, referred to as percept-percept inflation (Crampton and Wagner, 1993). This might create issues of common method variance, where measured

correlations and variance is attributable to the method of measurement rather than the constructs the measurement is aimed at. As such, common method variance can be responsible for faulty significant correlations between variables that do not represent reality.

To reduce the risk of common method variance, questions belonging to different variables were randomized within each question. Additionally, the respondents were ensured complete confidentiality, and had the option to stay fully anonymous by not having obligatory questions related to name of company or the respondent’s role at the company (Murray, Kotabe and Zhou, 2005).

49