1 Introduction

There have been few systematic studies of the propensity to patent by innovators.Footnote 1 Three studies are particularly noteworthy, but each approached the propensity to patent by innovators from a different perspective. Jones (2009) discussed the role of teamwork in developing an innovation suitable for patenting. Link and Ruhm (2011b) focused on prior business education/experience as a correlate with own patenting activity. And, from an alternative and more general perspective, Nicolaou et al. (2008) argued that genetic factors should be considered in explanations for why individuals engage in innovative and entrepreneurial activity.Footnote 2 This paper contributes to this literature, albeit in an exploratory manner.Footnote 3,Footnote 4

Motivating our analysis is the literature related to life course perspectives.Footnote 5 Simply, a life course perspective refers “to the social patterning of events and roles over a person’s life span, a process shaped by the interaction of individuals’ behaviors and changing historical contexts” (Aldrich and Kim 2007, p. 36). Based on this framework, intergenerational behavior has been studied from various perspectives. Scholars have investigated, among other things, the tendency of children to choose occupations that are similar to the occupations of their parents.Footnote 6 Much of the research in economics (e.g., Blanchflower and Oswald 1998; Lentz and Laband 1990) and in entrepreneurship (as reviewed by Storey 1994) has focused on the occupational inheritance of children of parents that are self-employed under the premise that the occupational pursuits of parents influence the development of the resources and capabilities of children. To the extent that the general behavior of a parent—in particular the behavior of a parent to patent—shapes the resources and capabilities of offspring, then the tendency of offspring to patent is related to similar behavior by a parent.Footnote 7,Footnote 8

In Section 2, we present an empirical model for testing the relationship between the patenting behavior of an innovator and how that behavior is related to similar behavior by a parent. In Section 3 we describe our database. It was developed from a survey of the most inventive young individuals in the world, as acknowledged by MIT’s Technology Review. Our empirical findings and robustness tests are presented in Section 4. Section 5 concludes the paper with a discussion of the implications from our empirical findings and directions for future research.

2 Empirical model

We hypothesize that the propensity to patent by innovator (i) is related to the same behavior of his or her parents:

$$ Patent_{i} = {\text{F}}({\mathbf{X}}_{i} + u_{i} ) $$
(1)

where Patent is a measure of the propensity of an innovator to patent, X a vector of parental patenting activity and other characteristics, and u i is an error term that incorporates both unobserved individual characteristics and other determinants of patenting behavior.Footnote 9 We will examine several specifications of innovative activity including the probability that the individual has one or more patents, the number of patents both conditional on some such activity and for all individuals, including both those who have and have not patented. For example, if F(.) is the cumulative normal distribution function and the u i  ~ N(0,1), the probability of patenting can be estimated as a probit model specified by:

$$ Patent_{i} = {\text{F}}({\mathbf{X}}_{i} + u_{i} > 0) $$
(2)

We estimated these models using a novel, and previously unexamined database of international inventors, as acknowledged by MIT’s Technology Review.

3 Technology Review database

To commemorate the 100th year of continuous publication of MIT’s innovation magazine, Technology Review, 100 young international inventors (under age 35 at the time of nomination) from universities, large and small businesses, and government laboratories, who have the potential to make major contributions in fields related to technology in the decades ahead, were identified in the November/December 1999 issue of the Review (Benditt 1999). TR100 inventors received this distinction in 2002, 2003, and 2004. In 2005, and in every year thereafter, the TR100 became the TR35.Footnote 10

All TR winners—arguably among the most inventive young individuals in the world from 1999 through 2009 and thus not representative of all creative individuals—are the population for our survey-based study. Although the criteria for selection as a TR winner is not public, the editors of Technology Review receive nominations from a panel of judges charged with providing advice to them on candidates “who are opening up new possibilities in technology.”Footnote 11 Thus, TR winners represent a particularly interesting group for analysis.

Using biographical and descriptive information in each year’s Technology Review, we were able to obtain e-mail addresses and contact 341 of the 575 winners over this time period. Sixty-three, or 18.5 percent, of those returned surveys and so were included in this study; this response rate is on par with other studies of innovative behavior.Footnote 12 See Table 1.

Table 1 Data reduction process
Full size table

The primary empirical specification of Eq. (1) is parsimonious owing to the limited survey information and relatively homogenous nature of TR winners. For example, 83 percent of the sample holds a terminal degree (i.e., PhD, MD, or JD). As well, the age range of TR winners is small by intent of Technology Review; the range of ages is 26–35. We do, however, test the robustness of the results to several alternative models and samples, as discussed below.

4 Empirical findings

The variables used to estimate Eq. (1) are defined, and descriptive statistics for them are provided, in Table 2. Data for each variable except Year came from our survey.

Table 2 Definition and descriptive statistics for variables relevant to TR entrepreneur (n = 63)
Full size table

We first estimated patenting behavior using a two-part model. Regarding the probit results in column (1) of Table 3 for the full survey sample of n = 63, creative innovators with fathers who patented are nearly 26 percentage points more likely to patent themselves compared to a similar innovator whose father did not patent, ceteris paribus.Footnote 13 Our arguments above apply, in principle, to both parents of an innovator, but our empirical analysis is delimited by the fact that none of the TR winner’s mothers held a patent as reflected in the survey responses.

Table 3 Econometric estimates of the determinants of patenting behavior (standard errors in parentheses)
Full size table

Also, those of Asian descent and those with a graduate degree in science or engineering are relatively more likely to patent than other innovators. Finally, males are more likely to patent than females but the difference does not quite reach conventional statistical significance (p value = 0.12).Footnote 14

The second part of our estimation involves identifying correlates with the natural log of the number of patents received conditional on patenting. As seen in column (2) for the sample of n = 29 who patented, those with a patenting father patent more, ceteris paribus. There is also evidence that age is a factor in determining the number of patents received rather than the per se propensity to patent, but this finding may only represent the fact that receipt of a patent is time intensive. Nationality is not significant among those who patent, but field of study is.

Finally, in column (3), we treated patenting as a count process and so estimated a negative binomial model. The results confirm the positive predicted effect of one’s father having patented. Males and Asians also have higher patent counts, but field of study is less relevant reflecting the previous evidence that scientists and engineers are more likely to patent but in smaller numbers among those who do so.

Response or selection bias is an obvious potential concern, particularly since just 18.5 percent of persons receiving surveys provided responses. Although we cannot fully deal with the issue of selection, we did partially address it by estimating Eq. (1) as a maximum likelihood model with a correction for selection. Specifically, we modeled for non-response as a function of the award year, Year, under the argument that the earlier in time the award the less likely the awardee would respond to the survey. First-stage probit results confirmed this. However, when estimated simultaneously with the probability of patenting model, the correlation between the error terms in the first-stage non-response model and the second stage patenting equation was not significant. Separately, we also estimated the model underlying the results in column (1) of Table 3 with Year as a supplementary regressor. The estimated coefficient on Year was not significant thus supporting the conclusion that this variable could reasonably be excluded from the patenting probit model. In combination, these results indicate that we do not have evidence of bias due to selection, although a more in-depth analysis of this issue (which is not possible using our data) would be needed to have confidence that this null hypothesis is confirmed.

To test for the robustness of our main result that TR award winners were more likely to patent if their fathers have also done so, we considered several alternative specifications or samples. These results are summarized in Table 4, row (a) of which repeats for ease of comparison the estimates from the Basic Model in Table 3.

Table 4 Additional estimates of the father’s patenting on TR winners patenting behavior (standard errors in parentheses)
Full size table

Bias due to the failure to control for potentially important confounding factors is conceivable, given the relatively limited set of controls available for analysis. While we cannot eliminate this possibility, we did examine whether the addition of observable characteristics that we could control for had much effect on the estimates. Specifically, in row (b) of Table 4 are results from a Sparse Covariates Model that holds constant only gender and age, in addition to patenting by the father.

The third row presents the results from the Rich Covariates Model in which we include all of the covariates that are available to us—those from the Basic Model in Table 3 plus a quadratic in age, award year, and a dichotomous variable measuring whether the respondent and his or her father holds a PhD.

The estimated marginal effects of the father’s patenting behavior do not change materially when moving from the sparse to rich set of controls [row (b) to row (c)]. Specifically, the effect of the father’s patenting on the likelihood that the respondent patents—from the probit models—becomes marginally stronger, while that on the number of patents, conditional on some such activity, decreases slightly and the negative binomial estimate on the total number of patents is essentially unchanged. None of the differences between the results in row (b) and row (c) are statistically significant. In combination, these results suggest that the results are not sensitive to the inclusion of controls for observable components of individual heterogeneity, although it remains possible that there are differences due to unobserved (to us) factors.

It is possible, although unlikely in our opinion, that the positive coefficient on DadPatent is positive because parents work together with their children on the same inventions.Footnote 15 Although interesting in its own right, this would not indicate a role for parental guidance or family structure per se, the mechanisms mentioned previously. To focus on cases where it seems unlikely that children are working together with their parents, we estimated alternative specifications using samples restricted to TR winners with university or government email addresses (n = 46). The rationale for this partitioning is that parents would be more likely to work together with their children if they are in private companies (presumably the same one). The results, shown in row (d), suggest a somewhat weaker father patenting effect on the probability of patenting, but a strong one on the conditional patenting equation and little difference in the expected number of patents from the negative binomial model. As expected, given the smaller sample sizes, the standard errors increase, and we are unable to reject the possibility that the marginal effects are the same as in the Basic Model.

Patenting behavior among the TR winners is highly skewed. In our sample of 63 respondents, 34 have not patented as of the time of our survey, 12 hold one patent, and 5 hold 2 patents. At the other extreme, the top 3 patent-holders have 20, 19, and 16 patents, respectively.Footnote 16 Such a skewed distribution of patenting raises the possible concern that a small number of extreme observations are unduly influencing our results. As a first step in examining this possibility we show, in row (e) of Table 4, the results from models with the 3 top patent-holders excluded from the sample. This exclusion attenuates the estimated effect of having a father who patents by 20–50 percent and renders it statistically insignificant. However, while interesting, we think it is more likely that these individuals represent valid observations rather than outliers that should be deleted. Specifically, the data indicate that each of the 6 individuals with 5 or more patents has a father who has patented and that there is an almost monotonic increase between the number of patents held and the likelihood that the father has also patented.Footnote 17 This finding suggests that any trimming of the data will lead to attenuated coefficients because it is likely to eliminate relatively strong, genuine effects.

5 Discussion

Our primary finding is that the patenting behavior of award-winning innovators is related to the environment in which they were raised, as proxied by the probability that their father has also patented.

In our main specification (Table 3), the fathers having patented is associated with a 26 percentage point increase in patenting by offspring, which is large relative to the sample average of 46 percent. While this pattern suggests a variety of fascinating possibilities, we have not identified mechanisms for the patent inheritance relationship, and doing so represents an important subject for future research given more comprehensive data. For instance, one possibility is that individuals who patent are also particularly productive in other ways and that their family background has provided them with a strong overall work ethic. Conversely, they may have been raised in households where particular value is placed on the commercialization of technologies rather than in the pursuit of basic knowledge. Or, it could be that locality characteristics (like the quality and orientation of the schools and the composition of local industries) are of key importance but that these are correlated with the family and individual characteristics that we control for. To help resolve these uncertainties, future efforts to collect primary data on innovators and their behavior might, for example, consider the research background of the innovator’s major professor and his or her orientation toward basic versus applied research.

Beyond these caveats, we caution against generalizing from our patent-specific findings that observed parental behavior is related to other dimensions of creativity and innovativeness. Our sample of TR winners is unique; and, our analysis, while broadening the foundation for the inquiry to the literature on intergenerational behavior and creativity, is yet exploratory in structure and scope. Moreover, patents represent one but certainly not the only important form of innovation. That said, our findings might suggest that intergenerational variables be considered in greater detail in other studies of innovative behavior.

Lastly, we have examined the propensity to patent in this paper through the lens of several bodies of literature. Perhaps this approach will be of more general relevance to future studies of innovative behavior.