**6 RESULTS**

**6.2 E MPIRICAL M ODEL**

**6.2.3 Robustness Tests**

We already touched upon the association of the investment uncertainty and innovativeness when
describing Model 5. Although we detect a similar direction of the impact and a p-value of just 0.115 in
the final model, we contrastingly find no significant effect of *Uncertainty on the innovation rates of *
incumbents at a 10%-significance level. In line with our expectations, we observe a positive coefficient
for the interaction of Uncertainty and External. Due to a lack of significance, we however reject our third
hypothesis, which assumed a moderating effect of uncertainty on the relationship between the CVC
investor’s structure and the innovativeness of the incumbent firm, so that external CVC units experience
a positive impact on the innovativeness when it invest in early stage. We will closer discuss the associated
implications of this result in the next section.

In comparison to the previously presented models, we find similar results for the remaining control
variables. *Post Period, Absorptive Capacity, Size, and Slack are significantly (p<0.01) and positively *
associated with the total patent count in the post-investment period, while the investment period shows
a significant (p<0.01) but negative impact on innovativeness, as described earlier. The model building
approach reveals that most of the explanatory power of the models stems from the above-mentioned
variables, as illustrated by the Wald Chi² and Pseudo R². Most contribution comes from the control
variables leading to an overall significance of the model based on the Chi² F-test.

Results

as an alternative regression model to estimate our independent variable. Lastly, we exchange the total
number of patents for the total number of claims as our dependent variable, creating model 12.^{14}

As we can observe in Model 7 to Model 10 (Table 14), the different operationalizations of variables as
well as the change of scope of the sample lead to consistent results with our final model. We detect that
our independent variable remains insignificant throughout the models. Although the significance levels
partly differ, we can observe supporting evidence for the significance of Experience and the interaction
effect *Experience x External differing at a maximum level of 10%. In Model 8, we particularly see a *
change of significance from the 5%- to the 10%-level for both considered variables when adopting the
investor’s average investment round as a measure for uncertainty. When removing the upper 10% of the
distribution from the sample in Model 10, the significance level of Experience x External increases to
10%.

In line with our final Model 6, we cannot observe a significant result for Uncertainty and the respective interaction effect Uncertainty x External in Model 7 to 9. However, the results disclose a change in sign for some of the models compared to our original model. In Model 8 and 9, we notice a positive association of Uncertainty with the dependent variable. Further, in Model 9 the coefficient of the interaction effect becomes negative, however only to a very small degree. Since this effect shows no significance across all models, it is difficult to account for the mentioned differences. In contrast to the final model, the results of Model 10 disclose a significant (5% significance-level) and negative influence of Uncertainty.

This implies that early stage investors generate fewer patents, which can be explained by the higher risk for knowledge creation associated with such early stage investments.

In Model 11, we surprisingly not only discover significance on a 1%-level for *External, but also see *
relevant changes in some of the other variables. *Experience appears to be insignificant while the *
interaction effect becomes even more significant. Further, the sign of the interaction effect Uncertainty
*x External turns negative, as already observed and elaborated on for Model 9. We assess the *
appropriateness of the Poisson model by using the goodness-of-fit-function in Stata, which shows a high

14 In accordance with the literature, we also test for the total number of forward citations generated in the post-investment period. This test, however, shows however differing results from the final model since the distribution is not well-applicable for our conducted panel regression. This is due to the dimension of time and a considerable time-lag that comes with it, as already described earlier in this work.

result^{15} and therefore captures a bad model fit (Berenson et al., 2012). The strongly significant (p<0.01)
alphas across all other models confirm our decision to capture the analysis with a negative binomial
regression model.

While most of the influences prevail in Model 12, the significance level of Experience and the respective
interaction with External increases to 10%. Lastly, we want to refer to the control variables, which mostly
behave in line with the final Model 6. Specifically, the largest influencing factors, namely the Post Period
and *Absorptive Capacity, remain consistent across all models. When reducing the sample to the *
companies with more than one investment, we detect that the Investment Period is no longer significant
(p>0.1). This can be explained by the fact that investors with only one investment are associated with an
investment period of zero, which could highly impact the distribution. Further, we can detect that Slack
is no longer significant (p>0.1) within the Poisson model. However, we already assessed the overall
unsuitability of this model.

Based on the performed tests we can conclude on the robustness, validity, and soundness of our empirical model, specifically regarding the discovered significance for the experience effect and its moderating influence. However, we want to highlight that we do not claim to explain any causational effects through this analysis.

15 The exact numbers of the goodness-of-fit test can be found in Appendix II.

Results

**Table 14 Robustness Checks (Models 7-12)**

(7) (8) (9) (10) (11) (12)

Start-Ups Inv. Round Inv > 1 Top 10% Poisson Claims

External 0.227 0.0812 0.306 0.252 0.613*** 0.254

(0.368) (0.389) (0.402) (0.411) (0.209) (0.440)

Experience (Investments) 0.504** 0.438* 0.471** 0.655** 0.0694 0.473*

(0.242) (0.251) (0.237) (0.271) (0.395) (0.281)

Experience x External -0.843** -0.742* -0.887** -0.781* -1.158*** -0.950*

(0.416) (0.435) (0.437) (0.472) (0.304) (0.488)

Uncertainty (Early Stage) -0.472 0.217 0.0444 -0.696** -0.119 -0.512

(0.300) (0.222) (0.413) (0.349) (0.181) (0.351)

Uncertainty x External 0.376 0.332 -0.00482 0.987 -0.113 0.546

(0.576) (0.416) (0.637) (0.610) (0.325) (0.653)

Post Period (in years) 0.241*** 0.245*** 0.238*** 0.217*** 0.191*** 0.260***

(0.0207) (0.0207) (0.0211) (0.0242) (0.0223) (0.0249)

Absorptive Capacity (Patent) 0.577*** 0.568*** 0.567*** 0.503*** 0.715*** 0.570***

(0.0212) (0.0212) (0.0224) (0.0260) (0.159) (0.0247)

Investment period (in years) -0.193*** -0.190*** -0.0881 -0.193*** -0.219*** -0.176***

(0.0501) (0.0496) (0.0870) (0.0580) (0.0610) (0.0588)

Size (Employees) 0.328*** 0.327*** 0.310*** 0.161*** 0.362** 0.357***

(0.0529) (0.0529) (0.0534) (0.0619) (0.176) (0.0650)

Financial Stability (Leverage) 0.0141 0.00980 -0.000447 0.00316 -0.0299 -0.0348

(0.0566) (0.0564) (0.0594) (0.0615) (0.0470) (0.0672)

Slack (EBITDA) 0.297*** 0.337*** 0.280*** 0.355*** 0.0486 0.372***

(0.105) (0.103) (0.107) (0.115) (0.0992) (0.119)

Constant 0.320 0.220 0.201 0.989** -0.727 2.987***

(0.396) (0.404) (0.443) (0.462) (0.493) (0.473)

lnalpha 0.753*** 0.750*** 0.680*** 0.792*** 1.122***

(0.0758) (0.0759) (0.0803) (0.0835) (0.0717)

Industry Dummy Yes Yes Yes Yes Yes Yes

Observations 346 346 306 302 346 346

Wald Chi² 472.8 473.5 417.9 267.9 1033.7 373.8

Prob. > chi² 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000

Pseudo R² 0.111 0.111 0.110 0.0867 0.885 0.0641

Note: Standard errors in parenthesis, ***p<0.01, **p<0.05, *p<0.1