Before turning to multivariate analysis, we briefly explore some simple associations between the ranking of the team, that is, the level of team success, and the variables that presumably have an impact on success in order to give some intuition on the findings and demonstrate their robustness. We measure the success of each team by computing the share of cases in which the team was ranked among the top four teams, the upper quintile. This share variable can be interpreted as the team's likelihood of reaching a certain cutoff level (the top 20%), which would (hypothetically) lead to an invitation to meet with a VC.6
Table 4 lists the teams and their characteristics in the order of the share of top quintile rankings achieved in our conjoint design. Since we use a reduced conjoint design, the “dream team” configuration, that is, the theoretically best profile, will not necessarily be among the 20 profiles presented to the interviewees. Team 10, which receives top quintile rankings in 96.1% of all cases, is therefore the most preferred team in the choice set according to our success variable, but not necessarily the theoretically optimal team configuration. While Table 4 shows that the top quintile share decreases quickly among the first 10 teams, it is difficult to extract clear information on the relative contribution of the various team characteristics from the simple ranking performed here. However, there appears to be a positive relationship between (favorable) ranking and industry experience, leadership experience, and the age of team members. It is more difficult to derive clear statements with respect to the other variables from the aggregate ranking information.
Table 4. Descriptive Statistics on Team Characteristics
|Team number||Share of top quintile rankings (%)||Relevant industry experience||Field of education||Leadership experience||Acquaintance among team members||University degree||Age of team members||Prior job experience|
| 8||25.5||All||All management||All||Professional||None||25–45||Mixed|
| 6||23.5||Some||All management||Some||Professional||Some||25–35||Corporate|
| 5||11.8||Some||All engineering||None||Professional||All||25–45||Start-up|
| 2||7.8||Some||All engineering||Some||Private||None||25–35||Mixed|
| 4||0.0||None||All engineering||All||Brief||Some||25–35||Start-up|
| 1||0.0||None||All management||Some||Brief||All||25–45||Corporate|
|17||0.0||None||All management||None||Private||Some||35 to 45||Mixed|
Whereas Table 4 shows complete team profiles, Table 2 presents the “success information” treating the parameter values of the team characteristics as fully independent. This table allows us to get a clearer impression of which team characteristics and which parameter values are likely to be important. For example, in 6 of our 20 team descriptions all team members have industry experience. Given 51 interviews, this yields 306 observations, of which 110 (35.9%) were ranked among the top four teams.
For the attributes “industry experience,”“field of education,”“acquaintance,” and “age,” we find a clear preference for a particular parameter value in that the distance from the respective next-best parameter value is larger than 10%. Preferred teams are those in which all members have industry experience, their educational background is mixed (some engineering, some management expertise), founders have known each other for a longer time professionally, and members are older (aged 35–45).
For the remaining three characteristics, a somewhat less transparent picture emerges: With regard to university training, prior job experience in corporate or start-up environments, and leadership experience, the best and second-best parameter values do not differ greatly when evaluated according to the share of top quintile rankings.
Note that Table 2 summarizes seven bivariate relationships—it is therefore not a substitute for a multivariate analysis. Nor does this table give us the opportunity to generate inference results. Hence, while Tables 2 and 4 provide some indication of which team characteristics are particularly important, a multivariate treatment of the data is required in order to arrive at a more structured response to our research questions.
Discrete Choice Analysis—Model Specification
The results of estimating the rank-ordered logit model are presented in Table 5. In specification (1), we use only the team characteristics as explanatory variables, while in specifications (2) to (5) we introduce interaction terms with the dummy variable Δi, which indicates whether the rater is an experienced VC. In essence, the upper half of columns (2) to (5) (i.e., those coefficients shown in the first part of Table 5) describe the choice behavior of less experienced VCs, while the lower half describes the difference between the preferences of more and less experienced raters.
Table 5. Rank-Ordered Logit Results
|Explanatory variables: Team characteristics. In spec. (2) to (5), coefficients refer only to inexperienced VCs||(1) No interactions,all 20 ranks||(2) With interactions,all 20 ranks||(3) With interactions, top 16 ranks||(4) With interactions, top 12 ranks||(5) With interactions, top 8 ranks|
|Experience in relevant industry—all team members||1.986***||1.980***||1.992***||2.278***||2.767***|
|Experience in relevant industry—some team members||1.614***||1.519***||1.476***||1.649***||1.706***|
|Field of education—all engineering||0.265**||0.462**||0.488**||0.653**||0.860**|
|Field of education—some engineering, some mgmt.||1.113***||1.194***||1.269***||1.497***||2.031***|
|Leadership experience—all team members||0.725***||1.001***||1.029***||1.165***||1.498***|
|Leadership experience—some team members||0.704***||1.012***||1.078***||1.129***||1.650***|
|Acquaintance—for a long time, professionally||0.585***||0.300*||0.321**||0.408**||0.831***|
|Acquaintance—for a long time, privately||0.247**||−0.034||−0.033||−0.048||0.082|
|University degree—all team members||0.912***||1.505***||1.577***||1.432***||2.144***|
|University degree—some team members||1.003***||1.332***||1.363***||1.213***||1.530***|
|Age of team members between 25 and 45||0.191***||0.128||0.096||−0.011||0.148|
|Age of team members between 35 and 45||0.517***||0.517***||0.397***||0.237||−0.270|
|Prior job experience—some large firm, some start-up||0.221**||0.176*||0.181||0.084||0.053|
|Prior job experience—mostly start-up||0.246***||0.273**||0.204||0.217||−0.090|
|Explanatory variables: team characteristics interacted with dummy variable Δi (Δi = 1 if rater is experienced)|| || || || || |
|Δi × experience in relevant industry—all team members|| ||0.258||0.345||0.392||0.223|
|Δi × experience in relevant industry—some team members|| ||0.326||0.436||0.470||0.624|
|Δi × field of education—all engineering|| ||−0.397||−0.330||−0.289||−0.103|
|Δi × field of education—some engineering, some mgmt|| ||−0.170||−0.098||−0.050||−0.084|
|Δi × leadership experience—all team members|| ||−0.487**||−0.456*||−0.531*||−0.538|
|Δi × leadership experience—some team members|| ||−0.558***||−0.607***||−0.668**||−0.885**|
|Δi × acquaintance—for a long time, professionally|| ||0.635**||0.811**||1.035**||0.769|
|Δi × acquaintance—for a long time, privately|| ||0.587**||0.676**||0.983***||0.582*|
|Δi × university degree—all team members|| ||−1.127***||−1.262***||−1.249***||−1.903***|
|Δi × university degree—some team members|| ||−0.644***||−0.799***||−0.667***||−0.864***|
|Δi × age of team members between 25 and 45|| ||0.187||0.172||0.206||−0.167|
|Δi × age of team members between 35 and 45|| ||0.071||0.210||0.439**||0.591|
|Δi × prior job experience—some large firm, some start-up|| ||0.216||0.232||0.451**||0.472*|
|Δi × prior job experience—mostly start-up|| ||−0.037||0.077||0.092||0.275|
Before interpreting the results, we need to discuss whether our findings are consistent with the assumption that our subjects have provided us with full rankings of the alternatives. There is considerable doubt in the literature that this assumption is always justified (Hausman & Ruud, 1987). What might have happened—and comments from our interviewees provide some evidence to this effect—is that subjects do spend effort on the upper ranks but pay less attention to the lower ones. In this case, heteroscedasticity will be introduced, which (in this model) will lead to inconsistent estimates if the full ranking information is used. For this reason, we present several specifications that differ with respect to the number of ranks taken into account. In columns (1) and (2) of Table 5, we present rank-ordered logit estimates which take the full rankings at face value. In specifications (3)/(4)/(5), in contrast, only the top 16/12/8 ranks are taken into account, while the residual ranks are treated as noninformative.
In essence, we discard information in columns (3) through (5) and should thus expect the precision of our estimates to decrease as more and more rankings are discarded. Indeed, even a cursory glance at the results shows that standard errors increase monotonically from column (2) to column (5). Moreover, the estimates show a second well-known pattern—the coefficients increase in size as we discard more of the lower ranks in our estimate. Hausman and Ruud (1987) argue that this phenomenon is consistent with the lower ranks being evaluated less carefully than the upper ones.7 Still, while the coefficients increase overall, their relative size remains largely stable.
Discrete Choice Analysis—Pooled Results
We start by discussing the pooled results for all respondents (specification ) before addressing the differences due to the rater's level of experience. In discussing the pooled results, we first analyze the relative importance of the various team characteristics and then address the benefit contribution of the various parameter values for each characteristic. Finally, we consider trade-offs between different parameter values for different characteristics.
We define the “importance” of a characteristic as the difference between the benefit contributions (i.e., the estimated coefficient) of the most and least preferred parameter values, normalized such that the sum of all importance values yields 100%. In other words, the importance of a characteristic is that share of the value difference between the best and the worst possible team that can be attributed to this characteristic.8 Given that the reference parameter value, by construction, has a benefit contribution of zero for most characteristics, the importance is essentially the (normalized) benefit contribution of the most preferred parameter value.9
As Figure 2 illustrates, industry experience is by far the most important characteristic (32.2%). While this in itself is not a new insight, our approach allows a meaningful comparison of characteristics beyond a mere ordering of their relative importance. In particular, we find that industry experience is 1.8 times—i.e., almost twice—as important as the field of education, which ranks second in overall importance (18.0%). Third comes academic education with 16.2%, meaning that it is about half as important as industry experience. Less importance is attributed to leadership experience (11.7%), the team members' mutual acquaintance within the team (9.5%), and age (8.4%). The type of prior job experience ranks last at 4.0%.
We now delve deeper into the benefit contributions of each characteristic's parameter values. To begin with, we find that the marginal benefit contribution of having more team members with industry experience decreases strongly. When only some team members have relevant experience, the benefit contribution (1.61) is about 80% of that attained when all founders know the industry (1.99). Hence, while having no industry experience seems to be a conditio sine qua non (knock-out criterion) for a VC evaluating a venture team, it will often be sufficient to have some industry insiders on board.
For the field of education, the relative benefit contribution of the various parameter values confirms the insight that a heterogeneous team comprising technical and management skills is much desired (benefit contribution 1.11). A management-only team is clearly not viable (benefit contribution 0), which was to be expected given the technical nature of our business model. Despite the model's technical nature, however, teams consisting entirely of engineers also fare so badly that this parameter value (benefit contribution 0.27) seems like a disqualifier for advancing to further stages in the evaluation process.
For the team's level of education, we find that an academic background is essential, but that it hardly makes a difference whether some or all team members have an academic background. While a team with only some university graduates is slightly preferred, the difference between the two coefficients in Table 5 is insignificant. This could mean that VCs see the participation of founders with university degrees as a positive signal—which, however, does not improve further when the number of graduates in the team increases from “some” to “all”; in fact, it decreases. Alternatively, an “all university” team may mean a higher average level of human capital, while a mixed team offers (desirable) heterogeneity. When these two effects are of equal size, we should observe (as we do) equal benefit contributions for both parameter values.
For leadership experience, we find a pattern similar to the one identified for industry experience. Having no members with leadership experience (benefit contribution 0) is likely to be a knock-out criterion in the evaluation process. However, the benefit contribution of “some team members with leadership experience” (0.70) is nearly identical to that of “all team members” (0.73). This is a rather plausible finding since not all members in a venture team can assume a leadership role. Note, however, that this is only true in the early stages of the start-up, whereas after successful expansion all founders might find themselves in leading positions and thus need leadership experience.
With regard to mutual acquaintance, we find that the type of acquaintance is just as important as its duration. Being acquainted for a long time is less than half as valuable (benefit contribution 0.25) when based on private relationships than when it is based on professional collaboration (benefit contribution 0.59).
As for age, we find that having only young team members (aged 25–35) on board yields the lowest evaluation (zero). This result is consistent with anecdotal evidence from VCs who had negative experiences with “boy groups” during the e-commerce boom. What is surprising is that having some more senior people in addition to young members on the team only partly remedies the problem: A mixed team with members aged between 25 and 45 (benefit contribution 0.19) still fares much worse than a team consisting exclusively of older founders (35–45, benefit contribution 0.52).
Finally, for the type of prior job experience, we find similar positive benefit contributions for heterogeneous teams (i.e., those whose members have experience partly in large firms, partly in start-ups) (0.22) and teams in which members have only start-up experience (0.25). However, even though both coefficients are significant, their size shows that VCs seem to care comparatively little about this team characteristic.
Discrete Choice Analysis—Effects of VC Experience
We now explore whether VC experience has a significant moderating effect on the evaluation of start-up teams. Overall, we find that both more and less experienced raters attach the highest importance to industry experience and the lowest to the type of prior professional experience. However, our analysis also reveals some key differences. The level of academic education ranks second for less experienced VCs (importance: 22.1%) and only fourth for their more experienced colleagues (10.8%). Leadership experience is ranked fourth (14.8%) by novices and sixth (8.1%) by experienced raters. The latter, in turn, attach more importance to mutual acquaintance within the team (ranked third at 14.7%) than less experienced VCs (ranked sixth at 4.9%).
Table 5 provides more detailed insights into the ratings of novice and experienced VCs. As specifications (2) to (5) show, we consistently find significant differences between the preferences of novice and experienced VCs for each parameter value of the following three characteristics: leadership experience, mutual acquaintance, and academic education. In addition, heterogeneous prior job experience (some start-up, some large firm) receives significantly higher ratings from experienced raters in specifications (4) and (5), as does a higher age (35–45) in specification (4). As the results of specifications (2) to (5) are identical in qualitative terms, and as we seek comparability with the basic model (1), the following discussion will focus on specification (2).
Figure 3 displays the coefficients of the interaction terms as given in the second part of Table 5. Coefficients significantly different from zero on the 1% level are rendered as solid black bars, those significant on the 5% level as hatched bars. We find the largest and most significant differences between novice and experienced raters in the perceived benefit contribution of a university degree. All team members having a university degree leads to a benefit contribution of 1.51 for a novice VC and of only 0.38 (i.e., 1.51–1.13) for experienced VCs. While the latter value is still positive and significantly different from zero (1% level), it is only a quarter of the size of the value for novices. We obtain similar results for the benefit contribution of “some team members having a university degree”: 0.69 for more vs. 1.33 (1% level) for less experienced VCs, a difference of −0.64 (see Figure 3). Furthermore, the preference order between purely academic and mixed teams is reversed for experienced raters: With a difference of 0.21, they significantly (1% level) prefer mixed teams, while their less experienced colleagues (insignificantly) prefer, by a margin of 0.17, teams in which all members have a university degree.
Figure 3. Difference in Benefit Contributions between Experienced and Novice Raters (Specification ). Reading Example: Experienced VCs Rate “Mutual Acquaintance for a Long Time, Professionally” 0.64 Points Higher
Download figure to PowerPoint
Leadership experience is also valued significantly less by experienced raters. Novices value leadership experience with a benefit contribution of 1.0 and attach little importance to whether all or some team members have such experience. In both cases, the benefit contributions perceived by experienced VCs are smaller by a value of roughly 0.5. While they are still highly significant (1% level), they are only about half as large as the values we obtained from less experienced raters.
The one characteristic for which we find a significantly higher valuation among experienced raters is mutual acquaintance within the team. If team members have known each other for a long time professionally, senior VCs perceive a benefit contribution that is 0.64 higher than their younger colleagues (0.94 vs. 0.30). Given a long-standing private acquaintance, the difference is 0.59, with novices perceiving no benefit contribution at all (−0.03, not significant) in that parameter value.