ArticlePDF Available

Political contributions by American inventors: evidence from 30,000 cases

Authors:

Abstract and Figures

Political scientists know surprisingly little about the political behavior of inventors, or those who produce new technologies. I therefore merged US patent and campaign contribution (DIME) data to reveal the donation behavior of 30,603 American inventors from 1980 through 2014. Analysis of the data produces three major findings. First, the Democratic Party has made significant inroads among American inventors, but these gains increasingly come from only a few regions and flow to a relatively small number of candidates. Second, deeper geographic trends explain most of the change in aggregate donation patterns. Third, inventors do not strategically donate to candidates outside their own district and, since 2006, inventors increasingly contribute to relatively centrist employer PACs with weak ties to the Democratic Party. These findings suggest that the interaction between market-oriented policy and American electoral institutions may inhibit the formation of broad cross-regional coalitions to support the knowledge economy.
Content may be subject to copyright.
RESEARCH ARTICLE
Political contributions by American inventors: evidence
from 30,000 cases
Nicholas Short
Princeton University, Princeton, NJ 08544, USA
Email: nick.short@princeton.edu
Abstract
Political scientists know surprisingly little about the political behavior of inventors, or those who produce new
technologies. I therefore merged US patent and campaign contribution (DIME) data to reveal the donation
behavior of 30,603 American inventors from 1980 through 2014. Analysis of the data produces three major
findings. First, the Democratic Party has made significant inroads among American inventors, but these gains
increasingly come from only a few regions and flow to a relatively small number of candidates. Second, deeper
geographic trends explain most of the change in aggregate donation patterns. Third, inventors do not
strategically donate to candidates outside their own district and, since 2006, inventors increasingly contribute
to relatively centrist employer PACs with weak ties to the Democratic Party. These findings suggest that the
interaction between market-oriented policy and American electoral institutions may inhibit the formation of
broad cross-regional coalitions to support the knowledge economy.
Keywords: American political institutions; knowledge economy; innovation geography; political behavior; political parties
Introduction
Inventors, or those who produce valuable intellectual property, are central actors in the American
knowledge economy and are an equally important constituency for those elected officials within the
Democratic Party who have embraced the knowledge economy and have worked to hasten its
development.1Despite the importance of inventors in the American political economy, social scientists
know surprisingly little about the political beliefs and behaviors of those who produce intellectual
property and even less about how their behavior has changed over time. As a result, it is difficult to
determine whether the Democratic Partys attempts to cultivate the knowledge economy have allowed it
to reap electoral rewards.
Theory offers potentially competing answers to this question. On the one hand, because prominent
Democrats have publicly championed the knowledge economy since at least 1972,2we might expect
those efforts to have motivated American inventors to express deeper levels of support for Democratic
candidates over time, much in the way that the Partys positions on racial liberalism and labor
legislation cultivated deeper levels of support for Democrats among minorities and the working-class.3
On the other hand, the imperatives of divided government forced lawmakers to rely heavily on market-
based reforms, like changes to US patent law, to promote knowledge economy development 4and, as a
result, knowledge economy development to date has generally been confined to a few regions, like
Californias Silicon Valley 5. Accordingly, we might also expect that American political institutions
© The Author(s), 2024. Published by Cambridge University Press on behalf of Vinod K. Aggarwal. This is an Open Access article, distributed under
the terms of the Creative Commons Attribution licence (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted re-use,
distribution and reproduction, provided the original article is properly cited.
1Haskel andWestlake (2018), Schwartz (2022), Short (2022).
2Geismer (2015).
3Schickler (2016); Schlozman (2015).
4Short (2022).
5Moretti (2013).
Business and Politics (2024), 26,64101
doi:10.1017/bap.2023.25
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
namely winner-take-all elections in single-member districtshave concentrated the electoral payoffs to
a few Congressional Districts or states.6
To the extent existing studies address this question, they do so indirectly or by focusing on specific
groups or regions, an approach that has the potential to bias the results. Popular news stories frequently
publish headlines like Its True: Tech Workers Overwhelmingly Support Democrats in 2018,but these
analyses are often limited to a small number of companies (in this case, Amazon, Apple, Facebook,
Microsoft, and Google), or a small number of industries (in this case information technology
companies), or a small number of regions (in this case, the Silicon Valley or the Seattle suburbs).7
As indicated below, any of these choices has the potential to bias the findings. Similarly, Bramlett,
Gimpel, and Lee (2011) find that major campaign donors in both parties are increasingly concentrated
to metropolitan areas with distinct political preferences.8But not all metropolitan areas are deeply
integrated into the knowledge economy, and those that are have affluent donors in many occupations
outside of or adjacent to the knowledge economy. It remains an open question whether these trends
also apply to those directly involved in the production of new technologies.
Other aspects of the political behavior of knowledge economy workers are also under-studied. Even
if inventors have developed stronger attachments to the Democratic Party over time, it is an open
question whether they have done so by virtue of their status as inventors or whether they are simply
caught up in deeper political currents shaping the partisan preferences of certain industrial sectors or
certain regions, like the growing ruralurban divide.9This question cannot be answered in the absence
of longitudinal data measuring behaviors across time. News sources also routinely emphasize that the
Democratic Party leverages affluent donors who reside in relatively safe states or districts to fund more
competitive campaigns across the country, as indicated by ubiquitous news references to California as a
political ATM.10 This hypothesis has some empirical support, though it appears that both political
parties are adept at leveraging donor networks in affluent metropolitan areas.11 It is an open question
whether knowledge economy workers engage in strategic behaviors, like donating heavily to outside
campaigns, which might increase their influence within the Party and act as a countervailing force to
their spatial concentration.
To test these hypotheses, I developed a unique data set containing ideology scores and information
on the donation behavior for 30,603 American inventors across 18 election cycles. Specifically, I used
the research data sets provided by the US Patent and Trademark Office to identify US residents listed as
named inventors on a US patent applied for on or after January 1, 1979. I then merged the inventor data
with campaign contribution data from the Database on Ideology, Money in Politics (DIME)12 to
capture campaign donations and the common-factor ideology scores imputed from those donations
among US inventors for every election cycle from 1980 through 2014. Finally, I linked the self-reported
donor employer names to organizations in the Capital IQ database to obtain unique employer
identifiers and industry data (4-digit SIC codes), where available. With such data, it is possible to
analyze changes in aggregate donation patterns and their geographic expression; it is also possible to
determine whether American inventors are unique in their behavior after controlling for things like
gender, geography, and place of work (firm or sector). I briefly describe and motivate the construction
of the data set in Section Construction of the Dataset(and more details are in Section A.1 of the
Appendix).
Analysis of the data set confirms that, while the Democratic Party has made significant inroads
among American inventors in terms of garnering higher shares of donors and donations, the vast
majority of those benefits have increasingly come from only a few regions and have flown to a relatively
small number of candidates, as shown in Sections Inventors Have Strengthened Their Attachment to
6Bramlett, Gimpel, and Lee (2011); Gimpel, Lee, and Kaminski (2006); Rodden (2019).
7Pearlstein (2018).
8Bramlett, Gimpel, and Lee (2011).
9Rodden (2019).
10See, e.g., Scola (2014).
11Gimpel, Lee, and Kaminski (2006).
12Bonica (2016).
Business and Politics 65
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
the Democratic Partyand The Democratic Partys Electoral Rewards Are Geographically
Concentrated. I also find that, while American inventors who contribute to Democrats have become
much more liberal over time, on average, this development seems to be driven by changes in political
geography, as shown in Section Changes in Inventor Partisanship and Ideology Arise from Geographic
Trends. The average ideology scores of Democratic inventors are not substantively different from
those of their peers (those of the same gender who work at the same firm and live in the same
Congressional District), and the large observed decline in ideological variance among Democratic
inventors has been significantly driven by similar declines in the average ideology scores across the
districts in which American inventors reside. Finally, the results suggest that, in recent elections,
inventors do not exhibit strategic behaviors and they differ from their peers only in the extent to which
they have forgone donating to Democrats in favor of donating to their employers corporate political
action committee (PAC).
Taken together, the results suggest that American political institutions have limited the electoral
payoffs for the Democratic Party, that American inventors who donate to campaigns increasingly live in
liberal enclaves of similar ideological persuasion, and that knowledge economy participation motivates
regional rather than individual differences in political behavior. They also suggest that, despite growing
somewhat more liberal over time as a group, inventors increasingly choose to funnel their political
contributions to employer PACs that give to both parties.
This article makes three contributions to the literature. The first is to add to a small and growing
literature on the ways in which American political institutions have shaped knowledge economy
development in the United States,13 especially the debate about whether agglomerationor the
tendency for innovation to increasingly concentrate in a small number of regionsis making the
knowledge economy politically unstable.14 I do so by addressing a series of empirical questions raised in
this debate about whether the Democratic Party has succeeded in attracting more knowledge economy
workers to its coalition over time; if so, whether the spatial concentration of innovation limits the
Partys electoral benefits; and, if so, whether strategic behaviors could plausibly surmount those
limitations.
The second contribution is to supplement what we know about the political behavior of knowledge
economy workers, an under-studied group in American politics, by considering how their partisan
preferences have changed over time and whether those changes are rooted in broader sectoral or
geographic trends. Our dominant understanding of this group is based a single survey conducted in
2017 which compared the policy preferences of technology entrepreneurs to those of partisan donors
and the general public.15 I build on this groundbreaking study by analyzing changes in partisanship and
ideology over time and by using regression analysis to compare the political behavior of inventors to
those who live and work in the same place.
The third contribution is to offer a tentative theoretical critique of neoliberalism that applies outside
of the articles specific setting. I use the term neoliberalism to refer to policies (like patent reform) that
induce changes in individual or firm behavior to produce a desired outcome (like more private sector
spending on technological development) in lieu of alternatives (like heavier federal spending on
research and development) that give the state a more prominent role in shaping economic affairs. Such
policies are used in many contexts and their proliferation since the 1980s has led one scholar to describe
them collectively as a submerged state.16 By focusing on those who develop intellectual property, this
study analyzes changes in political behavior among the direct beneficiaries of the political movement to
strengthen US and foreign patent law which unfolded between 1980 and 1994.17
I argue that, irrespective of whether neoliberal policies produce ideal economic outcomes, their well-
documented (and in many cases expected) tendency to exacerbate inequality creates a special problem
13Soskice (2022); Barnes (2022); Gingrich (2022).
14Iversen and Soskice (2019); Short (2022).
15Broockman, Ferenstein, and Malhotra (2019).
16Mettler (2011).
17Short (2022).
66 Nicholas Short
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
in a political system with winner-take-all elections in single-member districts. Specifically, American
institutions have the potential to create a political form of double marginalizationin economic
policymaking. By first constraining the economic policy choice set to those policies that exacerbate
geographic inequalities18 and then impeding the formation of cross-regional coalitions that might
advocate for a more equitable geographic distribution of resources, American institutions may doom
many such reforms to marginal (and highly unequal) success. I comment on this possibility and other
implications in the Conclusion.
Prior work and hypotheses
Is the American knowledge economy politically sustainable? Prior work suggests contrary answers.
Iversen and Soskice argue that, in the 1980s and 1990s, elected officials in advanced capitalist
democracies including the United States promoted the knowledge economy by enacting a set of policies,
like increased investments in higher education and reduced trade barriers, demanded by middle-class
voters. They did so to spread the economic benefits of the knowledge economy transition to a majority
of votersand expand the size of the political coalition supporting a further deepening of these
policies.19
In this theory, knowledge economies are politically stable and resilient in the face of economic
shocks and rising populism. Their underlying policies have a strong electoral connection to a broad
political coalition. And the tendency for innovation to concentrate in a small number of highly
productive regions (or agglomeration) limits capital mobility and restrains the power of business
interests to influence policy in other directions.20
Short argues, in contrast, that after the economic turmoil of the 1970s, the two main political parties in
the United States staked out unique positions on the macroeconomy: a Republican approach based on
market fundamentalism and a Democratic approach that sought to promote economic development
through technological innovation, but which also envisioned a significant role for the government in this
process. The imperatives of divided government in the 1980s and 1990s then forced elected Democrats to
abandon many policies demanded by the middle class, like investing in higher education, and to instead
pursue market-based policies, like patent reform, to hasten the knowledge economy transition.21
In this theory, the American knowledge economy is politically unstable. It relies heavily on market-
based or neoliberalpolicies that are more firmly rooted in the demands of business managers at
companies that invest in technology rather than middle-class voters, thus weakening the electoral
connection. And those policies are known to exacerbate geographic disparities in a political system
where the main political parties must win a majority of districts and states (irrespective of whether they
win a majority of voters) in order to govern.
These competing theories raise multiple empirical questions that have yet to be studied in American
political science. Four such questions are addressed in this article.
First, is it true that knowledge economy workers have developed stronger ties to the Democratic
Party over time? Many popular accounts support this hypothesis but often confine their analysis to
tech workersat a small number of companies or a small number of regions.22 Survey evidence
suggests that Democratic donors believe technology entrepreneurs will have increasing influence on
elected Democrats in the future, but these studies tell us little about historical trends.23 At the same
time, existing work also suggests that technology entrepreneurs have idiosyncratic preferences that do
not make them obvious allies with the Democratic Party on all issues, especially when it comes
to business regulation.24 The inventor-donor dataset allows us to explore this question by examining the
18Short (2022).
19Iversen and Soskice (2019, p. 138).
20Iversen and Soskice (2019, p. 159).
21Short (2022).
22See, e.g., Pearlstein (2018).
23Broockman, Ferenstein, and Malhotra (2019, Fig. 2).
24Broockman, Ferenstein, and Malhotra (2019).
Business and Politics 67
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
historical political behavior of those who invent new technologies in many sectors across the United
States.
Second, if knowledge economy workers have developed stronger ties to the Democratic Party, are the
Partys electoral rewards increasingly confined to a small number of regions? Prior work suggests the
answer should be yes. Economists have documented that technological production has become
increasingly concentrated, geographically, over time.25 And political scholars have documented a
similar trend in campaign contributions, generally.26 The inventor-donor dataset allows us to
determine whether this remains true of knowledge economy workers without biasing the results by
focusing on only a few technological domains or regions.
Third, is any observed affinity between knowledge economy workers and the Democratic Party a
byproduct of their status as producers of new technologies (an individual characteristic) or of broader
sectoral and regional trends? To my knowledge, no prior work has addressed this question because
there is no longitudinal nationwide dataset of knowledge economy workers whose preferences or
behavior can be compared to those who work in the same industry or reside in the same region. The
inventor-donor dataset enables such an analysis.
Fourth, does monetary surrogacy,or the ability of knowledge economy workers to strategically
invest in competitive elections nationwide, act as a countervailing force to agglomeration in the
American political system? Prior work has documented a growing share of contributions from donors
outside of the recipients district or state27 and that the competitiveness of a race is a significant
predictor of receiving outside donations.28 But prior work also suggests they, even if inventors behave in
this manner, they would still face significant obstacles to influencing member policy,29 especially when
directing donations to outside competitive districts.30 The inventor-donor dataset allows us to
qualitatively assess whether knowledge economy workers exhibit these behaviors.
Construction of the dataset
The process for creating the inventor-donor data set involved three main steps: (1) identify all inventors
(first and last name, firm, and city and state of residence) listed on US patents that were applied for on
or after 1 January 1979 and who resided within the United States using research datasets provided by
the US Patent and Trademark Office; (2) identify the subset of these US inventors that also appear in
the DIME database using fastLink31 and acquire data on their contribution history and imputed
ideology; and (3) match the self-reported employer names from the DIME database to organizations in
Capital IQ to generate unique identifiers for these organizations plus other information, like SIC codes,
where available. More details on each of these steps and statistics characterizing the aggregate dataset
are provided in Section A.1 of the Appendix.
The merge between inventor and donor data (steps 1 and 2) was executed using fastLink.32 In the
final dataset, I set the posterior probability of a match to a high threshold (0.99) to minimize false
positives, but the results are robust to using a lower threshold (0.9) which increases the sample size by
about 17 percent (from 30,603 inventor donors to 35,778). The merge between donor employers and
Capital IQ firms was executed using Capital IQs lookup algorithm, but I manually audited the links for
over two thousand firm names representing almost 75 percent of all inventor campaign contributions
and, for the remaining firms, retained only those links that also had a high degree of string similarity.
A primary advantage of this dataset is that it allows us to study the political behavior of the people
and organizations that produce new technologies while remaining agnostic as to the boundaries of what
25Moretti (2013).
26Bramlett, Gimpel, and Lee (2011); Gimpel, Lee, and Kaminski (2006).
27Gimpel, Lee, and Pearson-Merkowitz (2008); Grenzke (1988).
28Gimpel, Lee, and Pearson-Merkowitz (2008).
29Canes-Wrone, Brady, and Cogan (2002).
30Canes-Wrone and Miller (2022).
31Enamorado, Field, and Imai (2019).
32Enamorado, Field, and Imai (2019).
68 Nicholas Short
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
constitutes technology,which can bias the results of any political analysis. US patent law places very
few restrictions on what constitutes patent eligible subject matter,33 and so, subject to certain disclosure
requirements and an examination of prior art, the Patent and Trademark Office generally issues patents
for any new and non-obvious invention, broadly construed. Accordingly, the technologies that are the
subject of this study are not limited to the computer and internet technologies that tend to dominate the
news cycle but also include new drugs, nanotechnology, genetically modified crops, and many other
lesser-known domains of invention, like the design (look and feel) of new sneakers. While this may
seem over-inclusive to some, it is important to cast a broad net to avoid the bias inherent in individual
judgments about what constitutes technology.
Table 1illustrates this point. To generate the table, I identified the primary technological domain of
each inventor-donor using the classification scheme developed by the National Bureau of Economic
Research, and then tabulated the total dollar amount of campaign contributions across all election
cycles within each domain. The table presents the top 7 technology domains with the highest share of
donations going to Democratic candidates and committees (High Dem Share) and the top 7 with the
highest share of donations going to Republican candidates and committees (High Rep Share). The
table shows that inventors in computing (computer hardware and software, computer peripherals, and
semiconductor devices) and some other areas like optics and genetics give quite heavily to Democratic
candidates and committees. At the same time, inventors in other technological domains, including
those related to agriculture and resource extraction, donate quite heavily to Republican candidates and
committees. All of these inventors are arguably working at the technological frontier within their
respective industries and are therefore participating in the knowledge economy. But an exclusive
focus on those who work in computer and internet technology would suggestinappropriately in my
Table 1. Donations by technology classes show political bias
NBER subcategory Dem share (%) Rep share (%) Total (Mil USD)
High Dem share
Optics 66.84 29.72 3.73
Computer hardware & software 63.29 31.40 57.73
Computer peripherals 60.33 23.07 3.45
Semiconductor devices 58.92 28.64 3.60
Information storage 50.19 23.45 19.36
Resins 48.91 38.93 2.98
Genetics 48.23 35.25 0.91
High Rep share
Pipes & joints 5.54 93.06 3.83
Heating 7.29 89.13 4.05
Misc. mechanical 14.21 81.61 16.99
Gas 14.79 81.27 1.92
Agriculture, husbandry, & food 11.86 79.93 14.35
Earth working & wells 16.74 78.89 9.32
Motors, engines, & parts 10.29 77.55 3.59
Note: Each row shows the aggregate contributions made by inventor-donors in certain technological subcategories (in millions of
2019 dollars) as well as the share of that total going to Democratic and Republican candidates and committees. There are
37 technology subcategories in the NBER scheme but only 14 are presented in the table, reflecting the top 7 domains with the
highest share going to Democrats and the top 7 domains with the highest share going to Republicans.
33Diamond v. Chakrabarty, 447 US 303, 309 (1980) (The Committee Reports accompanying the 1952 [Patent] Act inform us
that Congress intended statutory subject matter to include anything under the sun that is made by man.’”).
Business and Politics 69
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
viewthat commanding majorities of all knowledge economy workers have a strong partisan
attachment to the Democratic Party. An analysis of all inventor-donors helps avoid this bias.
Though the lack of comparable databases makes it difficult to benchmark descriptive statistics, the
database can be used to replicate prior findings in ways that provide some confidence that it is soundly
constructed. For example, Rodden reports that Democratic presidential vote share was not correlated
with measures of patent output (patents per thousand people on the log scale) as recently as 1996, but
the two variables have become strongly correlated since then.34 Data from the inventor-donor dataset
produces similar findings, albeit with respect to donor rather than vote shares. For each of four election
cycles (1980, 1996, 2004, and 2012), Figure 1shows the share of all patents applied for by inventors
located in each Congressional District against the share of all inventor donations to Democratic
candidates and committees by inventor-donors located in that same district. The blue line shows the
results of regressing Democratic contribution shares on patent shares. The figure shows that, from 1980
through 1996, the patent share of a Congressional District was not significantly associated with the
share of total donations to Democratic candidates or committees by inventor-donors. But since 1996
that relationship has grown more positive. In this way, changes in regional donation patterns from the
inventor-donor dataset mirror changes in regional voting patterns reported in other studies.
Analysis
Inventors have strengthened their attachment to the democratic party
If the Democratic Partys commitments to the knowledge economy have produced electoral payoffs, we
would expect to observe the Democratic Party earning larger aggregate shares of either inventor
2004 2012
1980 1996
1e−04 1e−03 1e−02 1e−01 1e+00 1e−04 1e−03 1e−02 1e−01 1e+00
1e−04
1e−02
1e+00
1e−04
1e−02
1e+00
Share of Patents
Dem. Share of Inventor Contributions
High Patent Districts Give More to Democrats Since 1996
Figure 1. For each of four presidential elections (1980, 1992, 2004, and 2012), each panel shows the share of all patents applied
for by US inventors in each Congressional District against the share of all contributions to Democratic candidates and
committees from inventor-donors in the same District. Patents with more than one inventor were counted as a fractional share
(1 divided by the number of inventors) accruing to each inventor. Congressional District boundaries are based on the 1990
Census. Districts that produced no patents or no campaign contributions are treated as missing data. The blue line shows the
best linear fit given the data (i.e., a regression of contribution share on patent share).
34Rodden (2019, Fig. 3.1).
70 Nicholas Short
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
donations or inventor donors, the latter of which neutralizes any potential bias from a small number of
donors who contribute exceptionally high amounts to political campaigns. We would also expect to see
higher shares of inventors change their party affiliation from Republican to Democrat than vice versa.
And we would expect new donors to have more liberal ideology scores on average and to exhibit less
ideological variance around the party mean over time.
For each election cycle from 1980 to 2014, Figure 2shows the total amount of political contributions
(in millions of 2019 dollars) that American inventors made to each of the two major parties (left panel)
as well as the total number of inventors that donated to each of the two major parties (right panel).
Though Republicans attracted about 73.3 percent of inventor donations in the 1980 election cycle, the
parties were almost at parity in the 2008 election cycle, and though Republicans still held an advantage
in the 2014 cycle, it was significantly smaller than in prior years (58.5 percent of donations in a cycle
where 67.7 million dollars was raised by the two parties).
Democratic gains among inventors are even more significant when considering the share of donors
rather than donations: though 68.5 percent of inventors contributed to Republicans in the 1980 election
cycle, 62.9 percent of inventors contributed to Democrats in the 2014 cycle. This suggests that, between
1980 and 2014, the Democratic Partys commitments to knowledge economy development effectively
reversed the Republican Partys commanding advantage in the number of knowledge economy workers
who contribute to political campaigns.
The aggregate results in Figure 2combine the effect flowing from changes in the sample (as more
inventors donate to campaigns over time) as well as the effect flowing from changes within individuals
(as some inventors who previously contributed change their partisan attachments). To assess the
strength of the latter channel, I limited the dataset to those donors who only contributed to one of the
two main parties in any given election cycle and used the party affiliation of the recipient to impute a
Total Contributions (Millions 2019 USD) Total Donors (Thousands)
1980
1984
1988
1992
1996
2000
2004
2008
2012
1980
1984
1988
1992
1996
2000
2004
2008
2012
0.1
0.3
1.0
3.0
1
3
10
30
Recipient Part
y
Dem Rep
Democrats Have Erased Early Republican Advantages Among Inventor Donors
Figure 2. The left panel in this figure shows total contributions by American inventor-donors in all federal elections from 1980-
2014 broken down by recipient type: Democratic candidates and PACs (blue line) and Republican candidates and PACs (red
line). The contribution amounts are reported in millions of 2019 dollars. The right panel shows the total number of American
inventor-donors that contributed to each recipient type for each election cycle from 1980-2014. Inventor-donors are political
donors who reside in the United States and are listed as an inventor on any United States patent applied for on or after 1
January 1979. Both vertical axes are on the logarithmic scale.
Business and Politics 71
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
partisan attachment for each donor.35 I then calculated the share of single-party inventor donors in each
election cycle who changed their party affiliation (compared to the last election cycle) from Democrat to
Republican or from Republican to Democrat.
Figure 3shows the results. Except for the 1996 election cycle, higher shares of inventors switched
their partisan affiliation from the Republican to Democratic Party (than vice versa) from 1980 to 2004,
but party defections were about the same starting around 2006 and have remained at low levels since.
In short, the Democratic Partys commitments to the knowledge economy did not simply induce more
liberal inventors to start donating to political campaigns. It also seems to have induced donors with
established records of contributing to change their partisan commitments at relatively high rates.
Similar results emerge if we focus on donor ideology. Figure 4depicts how inventor ideology has
changed over time. For each election cycle from 1980 through 2014, it shows the average ideology score
(left panel) and the variance in ideology scores (right panel) for two sub-populations: those who
contributed to Democratic candidates and committees (blue line) and those who contributed to
Republican candidates and committees (red line).
As shown, the average ideology score among Republican donors remained relatively stable at about
0.75 until the 2006 election cycle, when it increased a bit. This suggests that inventors who contribute to
Republicans were fairly conservative to begin with and have become slightly more conservative since
2006. In contrast, the average ideology score among Democratic donors remained constant and close to
zero (at about 0.08) through 1990 but then dropped dramatically over the next twelve election cycles,
reaching a low of 1.23 in the 2012 election cycle. This suggests that inventors who contributed to
Democrats were a relatively moderate group to begin with but became much more liberal beginning
with the election of 1992. Because ideology scores are fixed and do not change over time, these trends
arise solely from changes in the sample, meaning those inventors who began donating to Democratic
campaigns in 1992 were consistently more liberal than those inventors who had donated to Democratic
campaigns in the 1980s.
0.05
0.10
0.15
0.20
0.25
1984
1988
1992
1996
2000
2004
2008
2012
Part
y
Dem Rep
Higher Shares of Inventors Switched to Dem. Party Until 2006
Figure 3. The figure shows the share of single-party donors in each election cycle who switched from being a Republican to a
Democratic donor (blue) and from being a Democratic to a Republican donor (red).
35I thank an anonymous reviewer for proposing this analysis.
72 Nicholas Short
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
Similarly, the variance or spread in ideology scores for Republican donors was quite small from the
beginning and appears to have slightly increased over the course of the entire time series. In contrast,
inventors who gave to Democrats appeared to be relatively moderate, on average, because they had
widely varying ideology scores in early election cycles. But from roughly 1992 through 2012, the
variance in ideology scores for Democratic donors dropped dramatically so that in recent elections,
Democratic donors have been as tightly distributed about their mean as Republican donors were in
1980 and 1982. In short, American inventors that contribute to political campaigns have become more
polarized, as we observe both higher separation between average ideology scores and lower variances
around those means, but that polarization arises mostly from ideological changes that took place among
inventors who contribute to Democrats.
In sum, the Democratic Party has made tremendous gains among knowledge economy workers
since 1980. It effectively reversed a strong Republican advantage in the number of inventors who donate
to each party and is competitive in terms of total contributions. For more than 20 years, from 1982 to
2004, it consistently recruited higher shares of inventors to switch their partisan affiliation. And
inventors who contribute to Democrats have become much more liberal as a group, on average, and
much more homogenous in their ideological leanings.
The Democratic Partys electoral rewards are geographically concentrated
In a majoritarian political system with single-member districts, the tendency for knowledge economy
work to cluster (or agglomerate) in a handful of regions with strong pre-existing advantages might limit
the electoral payoffs to be earned from supporting the knowledge economy transition.36 Figure 5
illustrates this trend. It shows that the share of patents coming from the top 1 percent of counties, based
on patent production, has steadily increased since the mid-1980s. While the most innovative counties in
the US produced 25 percent of all patents in 1980, by 2014, their share of all patents was more than 45
percent. We would therefore expect to see patterns of political behavior that reflect patterns of
Mean Contributor CF Score Variance in Contributor CF Score
1980
1984
1988
1992
1996
2000
2004
2008
2012
1980
1984
1988
1992
1996
2000
2004
2008
2012
0.2
0.3
0.4
0.5
0.6
0.7
−1.0
−0.5
0.0
0.5
1.0
Recipient Part
y
Dem Rep
Inventors Have Become More Polarized Over Time
Figure 4. This figure shows the average (left panel) and variance (right panel) of the ideology scores for those inventor-donors
who contributed to Democratic candidates and committees (blue line) and those who contributed to Republican candidates
and committees (red line) in each election cycle from 1980 through 2014.
36Moretti (2013); Rodden (2019).
Business and Politics 73
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
economic behavior, with an increasing geographic concentration in innovation translating to an
increasing geographic concentration in inventor donations.
Figure 6indicates that this has taken place. For each of five presidential election cycles (1980, 1988,
1996, 2004, and 2012), the figure shows, for each party, the share of inventor donations coming from
30
35
40
45
1980
1984
1988
1992
1996
2000
2004
2008
2012
Agglomeration: Share of Patents from Top 1% of Zip Codes Has Increased
Figure 5. The figure shows the share of all patents coming from the top 1 percent of zip codes.
Share from Top 1% Zip Codes Share to Top 1% of Candidates
0 20 40 60 0 20 40 60
1980
1988
1996
2004
2012
Party Democrat Republican
Inventor Donations and Receipts Have Also Grown More Concentrated
Figure 6. The figure shows the share of all donations from American inventors flowing from the top 1 percent of zip codes, by
donation amount (left panel), and the share flowing to the top 1 percent of political candidates, by donation receipts (right panel).
74 Nicholas Short
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
the top 1 percent of zip codes, by donation amount (left panel), and the share of donations going to the
top 1 percent of candidates (right panel). Campaign contributions, by this measure, have become even
more concentrated than inventions. The Democratic Partys share of donations flowing from the top
1 percent of zip codes has grown, for example, from about 25 percent in the 1980 election cycle to more
than 50 percent in the 2012 election cycle. Also, the top 1 percent of Democratic candidates who receive
the most inventor donations has increased their share of donations from about 16 percent to about
67 percent. Inventor contributions to Republican have experienced similar trends, though the levels of
concentration are not quite as severe as for Democrats. And while growing concentration in receipts is
dominated by donations to presidential candidates, the trends still exist in midterm elections.37
Changes in inventor partisanship and ideology arise from geographic trends
While Figure 2suggests that the Democratic Party has made significant progress in courting inventors
as a constituency, it is amenable to multiple interpretations. Importantly, changes in aggregate donation
patterns do not reveal whether inventors, as a class, have begun to favor Democrats by virtue of their
status as producers of new technologies or whether inventors increasingly reside in metropolitan areas
that have acquired stronger attachments to the Democratic Party over time. The same is true of sectoral
trends. If the Democratic Party has taken policy positions that are consistent with the interests of some
industries, like solar power companies, but opposed to others, like conventional carbon-based energy
companies, the patterns documented above may suggest that inventor behavior reflects a deeper
sectoral realignment in American politics.
To disentangle the effects of geography, industry, and inventorship, I matched inventor donors to
non-inventor donors who have the same (imputed) gender, work at the same organization, and reside
in the same Congressional District.38 For each election cycle, I then regressed each of two dependent
variablesa continuous ideology score and a binary variable indicating whether the donor contributed
to Democratic candidates or committeeson a binary variable indicating whether the donor is an
inventor. The evolution in the coefficients on the inventorship variable, over time, reveal whether
inventors have become more liberal or developed a stronger propensity to contribute to Democrats
after controlling for differences arising from gender, place of work, and place of residence.
The regression output is reported in Section A.3 of the Appendix, but the main results are illustrated
in Figures 7and 8.39 As shown in Figure 7, from 1980 through 2006, inventors were just as likely as their
peers to donate to Democrats, but since the 2008 election cycle, they have become slightly less likely
than their peers to donate to Democrats. These results are consistent with Broockman et al. to the extent
they suggest that inventors have somewhat unique political preferences and may be more conservative
than their peers in certain dimensions.40 But the results also suggest that changes in political geography
are driving Democratic gains among inventors: after controlling for geography, inventorship actually
pulls in the opposite direction and would alone suggest that the Democratic Party has been losing, not
gaining, ground with this constituency.
Similar findings emerge from analyzing changes in inventor donor ideology arising from changes in
the population of Democratic inventor donors. Figure 8shows the results when the dependent variable
is donor ideology. It suggests that inventors in early election cycles were slightly more conservative than
their peers and slowly became more liberal than their peers over time, though the effect is not precisely
estimated and is not significantly different from zero until 2002. That trend, however, appears to have
reversed around 2006 and by 2014, inventors were only slightly more liberal than their peers (differing
37Between 1982 and 2014, the share of inventor donations going to the top 1 percent of Congressional candidates grew from
about 6 to about 20 percent for both parties.
38More information on this process is available in Section A.1 of the Appendix.
39The results include all inventors that could be matched, as described in Section A.1 of the Appendix, but are robust to
focusing only on first-time donors.
40Broockman, Ferenstein, and Malhotra (2019).
Business and Politics 75
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
−3
−2
−1
0
1
2
3
1980
1990
2000
2010
Inventors Have Become Less Likely to Donate to Democrats Than Their Peers
Figure 7. This figure shows the point estimates and 95 percent confidence intervals from regressing a binary variable indicating
whether the donor contributed to a Democratic candidate or committee on a binary variable indicating whether the donor is an
inventor, after matching inventors with non-inventors who have the same imputed gender, place of work, and place of
residence. These logistic regressions are run for each matched data set within each election cycle from 1980 through 2014. The
vertical axis reflects the estimated difference in the logged odds of donating to a Democratic candidate or committee between
inventors and non-inventors, with negative numbers implying less than even (50-50) odds.
−0.3
0.0
0.3
0.6
1980
1990
2000
2010
Inventors Have Not Become Significantly More Liberal Than Their Peers
Figure 8. This figure shows the point estimates and 95 percent confidence intervals from regressing ideology scores on a
binary variable indicating whether the donor is an inventor, after matching inventors with non-inventors who have the same
imputed gender, place of work, and place of residence. These linear regressions are run for each matched data set within each
election cycle from 1980 through 2014. The vertical axis reflects an estimated difference in mean ideology scores between
inventors and non-inventors.
76 Nicholas Short
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
only by 0.13 points on the common factor ideology scale in 2012).41 Accordingly, after controlling for
geography and other factors, differences between inventors and non-inventors explain only about
11 percent of the total change in average ideology scores among Democratic donors (of about 1.15 points).
Though individual characteristics, like inventorship, do not seem to explain increasing liberalism among
Democratic donors, it is still possible that we are confounding geographic with firm-level or sectoral
behaviors. To explore this question, I first expanded the inventor donor database, for the 1992 and 2012
election cycles, to include campaign contribution data on all workers at firms that produce intellectual
property. Specifically, as described in Section A.1 of the Appendix, I used patent data to identify all companies
that produced at least five patents42 in the five years prior to each election cycle. I then merged these firm
names with the DIME database to gather contribution and ideology data on all employees for these firms
(inventors and non-inventors alike) in each election cycle. Last, I linked new firm names (for those firms
which did not have inventor donors) to Capital IQ firm identifiers and 4-digit SIC codes.
With this dataset, I performed a variance decomposition on the subsets of knowledge economy
workers who contribute to Democrats and who contribute to Republicans. The motivation and
technical details for this analysis are described in Section A.2 of the Appendix.
The results suggest that Democratic knowledge economy workers are becoming more polarized
primarily by virtue of the place they live rather than the place they work, though residual variation in
ideology scores within districts and firms remains an important contributor as well. Figure 9illustrates
the main findings. It reveals that, for both Democratic and Republican knowledge economy workers,
the variance in the average ideology scores between firms did not materially change between 1992
Democratic Donors Republican Donors
0.0
0.2
0.4
0.6
0.0
0.2
0.4
0.6
Residual
Firm
District
Estimated Standard Deviation
Cycle 1992 2012
Political Geography Explains Increasing Polarization in the Knowledge Economy
Figure 9. This figure shows the empirical standard deviation in the distribution of average ideology scores across
Congressional Districts (top line) and across organizations (middle line) as well as the residual deviation within districts and
organizations (bottom line). The estimates are produced by fitting the model described in Section A.2 of the Appendix. The
estimates are reported for two different election cycles: 1992 (black points and 95 percent confidence intervals) and 2012 (gray
points and confidence intervals). The estimates are also reported from fitting the model to two different data sets: Democratic
donors (left panel) and Republican donors (right panel).
41Note that this turning point, in the 2006 election cycle, matches the point in time when inventors began developing a lower
propensity to donate to Democratic candidates and committees, shown in Figure 7.
42For patents assigned to multiple companies, each company received a proportional share.
Business and Politics 77
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
and 2012. This means that differences between firms cannot explain increasing polarization among
knowledge economy workers.
In contrast, for Democratic knowledge economy workers, the estimated variance in the average
ideology scores between Congressional Districts plummeted by about 84 percent and, as of 2012, was
close to zero (the point estimate is 0.058). In other words, Democratic knowledge economy workers
have come to increasingly reside in homogeneous liberal enclaves, so that there is almost no variation
left in the average ideology scores across the districts in which these workers reside.
A significant decline in the residual variance, by about 40 percent between 1992 and 2012, also
suggests that polarization among Democratic knowledge economy workers increased within firms and
districts as well. But the amount of ideological variation remaining within firms and districts is still
relatively large (comparable in size to the variance between firms). The most salient and surprising
result is the virtual dissipation of any meaningful variation between districts.
The results are similar when the variance decomposition is run using Congressional Districts and
4-digit SIC codes instead of firms. Whether the alternative source is hypothesized to be place of work or
industrial affiliation, increasing geographic polarization emerges as the more plausible source of
increasing polarization among knowledge economy workers who contribute to Democrats.
Strategic donation patterns
Do inventors exhibit strategic behavior that might allow their donations to sustain a broad political
coalition? I approached this question in two ways. First, I analyzed patterns of local and out of district
giving among inventor donors in the 2012 presidential election and the 2014 midterm elections
(the most recent elections in the data set). Second, the regression results from Section Changes in
Inventor Partisanship and Ideology Arise from Geographic Trendsrevealed that inventor donors since
2006 have become less likely to donate to Democrats, but additional analyses43 revealed that they have
not become more likely to donate to Republicans but have instead become more likely to donate to
relatively centrist PACs of unknown partisan affiliation (PAC-UPAs).44 I therefore audited these
contributions for the inventor donors in the matched data set for the 1992 and 2012 election cycles to
better understand this behavioral shift.
Closer inspection of the top recipients in 2012 and 2014 suggests that inventor donations remain
concentrated partly because inventors (from both parties) give mostly to local candidates and partly
because Democratic donors send much of their more expressive or strategic donations to candidates
who either reside in states that are leading knowledge economy development or who have publicly
promoted the knowledge economy. On the whole, local giving among inventors has declined, but still
made up a majority of donations in the 2014 election cycle (down from 71.6 percent to 51.8 percent of
all inventor donations from 1982 to 2014). Among Democratic recipients, in the 2012 presidential
election, three of the biggest recipients of inventor donations behind presidential candidate Barack
Obama were Senate candidate Elizabeth Warren from Massachusetts, Senator Maria Cantwell of
Washington, and Congresswoman Nancy Pelosi of California. Pelosi raised almost all (98.6 percent)
of those donations from local donors, but Warren and Cantwell both raised higher shares from out of
state donors (47.2 and 33.5 percent, respectively).
Similarly, in the 2014 midterm elections, Senators Ed Markey of Massachusetts and Kay Hagan of
North Carolina were two of the top four recipients and both drew significant shares of inventor
donations from out of state donors (47.9 and 37.5 percent, respectively). But the top recipient in that
cycle was Senator Cory Booker of New Jersey, who has taken prominent positions on the knowledge
economy45 and who received 81.8 of his donations from out of state inventors; the fourth largest
43These results are not shown but are available upon request.
44As described in Section A.1 of the Appendix, PAC-UPAs are committees that either do not have a partisan designation in the
underlying DIME data, do not have an ideological score or have a middlingideological score (greater than 0.5 and less
than 0.5) which makes it difficult to impute a partisan tendency based on donation patterns, and do not have the text strings
Republicanor Democratin their name.
45Techonomy (2015).
78 Nicholas Short
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
recipient was Senator Gary Peters of Michigan who has sought to promote technological innovation in
his home state, especially within the auto industry,46 and who received 73.1 of his donations from out of
state inventors. As these examples suggest, there is some opportunity to cultivate inventor support for
candidates in non-leading states, but few have successfully capitalized on those opportunities.
The dominant tendency is instead for large shares of Democratic inventor donations to flow to local
candidates, candidates in regions that are leading the knowledge economy transition, or to knowledge
economy advocates in relatively safe seats.47
Similarly, since 2006, inventors have become less likely to donate to Democrats and more likely to
donate to PAC-UPAs. An audit of these donations revealed that the vast majority of these donations are
going to PACs representing the donors employer, and that the share of PAC-UPA donations flowing to
employer PACs increased from about 83 to about 89 percent between 1992 and 2012 (see Table 2).
The data cannot tell us whether this reflects underlying donor preferences, as employees may feel
obligated to contribute to their employers PAC for many reasons.48 Irrespective of the behavioral
motivation, the outcome does not suggest that inventors are behaving in a way to strategically benefit
Democrats. On the contrary, they appear to be more frequently contributing to and contributing larger
shares to employer PACs with weaker attachments to the Democratic Party.
5 Conclusion
As a group, inventors, or those who produce the new technologies that drive the knowledge economy,
have come to favor Democrats when donating to political campaigns, and those who contribute to
Democrats have also become more liberal since the 1992 election cycle. Similarly, from 1982 to 2004,
relatively large shares of single-party donors switched their party affiliation from Republican to
Democrat. These findings generally support the hypothesis that the Democratic Partys rhetoric and
policy commitments on knowledge economy formation have allowed it to reap electoral rewards.
But closer inspection of the data suggests that these rewards may be highly unequal and qualified, in
ways that call into question the knowledge economys viability as a dominant platform of economic
development. Just as innovation has become increasingly concentrated in geographic space, so have
donations from inventors become more concentrated, with roughly 60 percent of donations to each
party coming from only 148 zip codes in the 2012 election cycle. Also, larger shares of these donations
increasingly flow to presidential candidates and a small number of Congressional candidates who either
live in states that are leading knowledge economy development or who have publicly supported the
knowledge economy. In a majoritarian system with single-member districts, the concentration of
knowledge economy development to a few regions seems to be generating large political payoffs for
some candidates and more muted results for many others.
Table 2. The PAC-UPAs Inventors Favor Are Employer Corporate PACs
PAC Type Share (%) in 1992 Share (%) in 2012
Corporate PACs 83.11 88.65
Trade and Prof. Assn PACs 16.02 9.02
Union PACs 0.00 0.03
Unknown affiliation 0.87 2.30
Note: This table shows the results of sub-categorizing the PAC-UPAs that received donations from inventors in the matched data
set as PACs representing trade and professional associations, corporations and their employees, or unions, with the rema inder
being categorized as PACs of no discernible affiliation. The share of donations to each type of PAC-UPA are shown for the 1992
election (column 2) and the 2012 election (column 3).
46Detroit Economic Club (2018).
47Both Booker and Peters both won their 2014 elections with about 55 percent of the statewide vote.
48Hertel-Fernandez (2018).
Business and Politics 79
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
Of course, the campaign finance system has developed institutions for distributing money to more
competitive races, but it is not yet clear whether it does so in ways that create concrete electoral
incentives for the many Democratic candidates outside of known knowledge economy hubs to take up
or maintain the cause of knowledge economy development. If as yet unobserved voting behavior among
inventors follows their donation behavior, institutions for redistributing donations may offer little
recourse, as inventor votes will remain concentrated, nevertheless. Moreover, even if the Democratic
Party finds it useful to draw resources from affluent knowledge economy hubs to fund competitive races
around the country, the evidence suggests that the parties are somewhat evenly matched when it comes
to leveraging urban donor networks49 and that this form of monetary support is not likely influence
member policy.50
More broadly, these results suggest that American political institutions create at least two major
perils for those who seek to mobilize the force of the government to undertake bold new programs of
economic development. First, in a setting where the two main political parties have staked out divergent
philosophies on macroeconomic management, those aspects of the nations governing institutions that
protect the rights of political minorities may tend to give neoliberal strategies a higher likelihood of
becoming law.51 But the market-oriented reforms that those strategies rely upon and the unequal (and
limited) response of state governments in a federalist system may also tend to exacerbate pre-existing
regional inequalities.
Then, in the absence of a more robust effort by the federal government to equalize regional patterns
of economic development, winner-take-all elections in single-member districts may cause economic
agglomeration52 to turn into political agglomeration,53 where geographic concentration in economic
gains leads to concentration in electoral payoffs and relatively few political candidates perceive a benefit
from supporting or broadening the policies driving the economic transition. This in turn may impede
the formation of the kinds of wide cross-regional coalitions, at the electoral level, that would be needed
to overcome the imperatives of divided government to assert the federal governments hand more
forcefully. Opportunities to broaden these policies to bring benefits to more regions may then be
confined to relatively rare moments when the Democratic Party controls the government. The Biden
administrations success with the Inflation Reduction Act, the CHIPS and Science Act, and the
Infrastructure Investment and Jobs Act from 2021 to 2022 suggests that Democratic Party leaders are
aware that its previous approach to knowledge economy development has produced unequal economic
gains that have turned into significant political liabilities.
Political scientists have previously documented that the nations governing institutions tend to
create gridlock and are resistant to policy change.54 And some have shown how the fact of
agglomeration has created geographic cleavages (the ruralurban divide) that tend to impede the
Democratic Partys ability to translate aggregate economic growth into a governing majority.55
These institutional constraints may, however, be linked in a way that scholars have yet to appreciate and
which deserves further exploration.
References
Barnes, Lucy. 2022. Public Investment in the Knowledge Economy.InThe American Political Economy: Politics, Markets, and
Power, edited by Jacob S. Hacker, Alexander Hertel Fernandez, Paul Pierson and Kathleen Thelen. New York: Cambridge
University Press. Chap. 12.
Barth, Erling et al. 2016. Its Where You Work: Increases in the Dispersion of Earnings across Establishments and Individuals in
the United States.Journal of Labor Economics 34 (2): S67S97.
49Gimpel, Lee, and Kaminski (2006).
50Canes-Wrone, Brady, and Cogan (2002); Canes-Wrone and Miller (2022).
51Short (2022).
52Moretti (2013).
53Rodden (2019).
54See, e.g., Hacker et al. (2022).
55Rodden (2019).
80 Nicholas Short
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
Bonica, Adam. 2016. Database on Ideology, Money in Politics, and Elections: Public version 2.0 [Computer file]. Stanford, CA:
Stanford University Libraries. https://data.stanford.edu/dime.
Bramlett, Brittany H., James G. Gimpel, and Frances E. Lee. 2011. The Political Ecology of Opinion in Big-Donor
Neighborhoods.Political Behavior 33: 565600.
Broockman, David E., Gregory Ferenstein, and Neil Malhotra. 2019. Predisposition and the Political Behavior of American
Economic Elites: Evidence from Technology Entrepreneurs.American Journal of Political Science 63 (1): 212233.
Canes-Wrone, Brandice, David W. Brady, and John F. Cogan. 2002. Out of Step, Out of Office: Electoral Accountability and
House MembersVoting.American Political Science Review 96 (1): 127140.
Canes-Wrone, Brandice, and Kenneth M. Miller. 2022. Out-of-District Donors and Representation in the US House.Legislative
Studies Quarterly 47 (2): 361395.
Denwood, Matthew J. 2016. runjags: An R Package Providing Interface Utilities, Model Templates, Parallel Computing Methods
and Additional Distributions for MCMC Models in JAGS.Journal of Statistical Software 71 (9): 125.
Detroit Economic Club 2018. Michigan Should Lead the Next Frontier of Innovation.Recorded interview. https://www.eco
nclub.org/meeting/senator-peters/.
Enamorado, Ted, Benjamin Field, and Kosuke Imai. 2019. Using a Probabilistic Model to Assist Merging Large-Scale
Administrative Records.The American Political Science Review 113 (2): 353371.
Geismer, Lily. 2015. Dont Blame Us: Suburban Liberals and the Transformation of the Democratic Party. Princeton & Oxford:
Princeton University Press.
Gelman, Andrew, and Jennifer Hill. 2007. Data Analysis Using Regression and Multilevel/Hierarchical Models. New York:
Cambridge University Press.
Gimpel, James G., Frances E. Lee, and Joshua Kaminski. 2006. The Political Geography of Campaign Contributions in American
Politics.The Journal of Politics 68 (3): 626639.
Gimpel, James G., Frances E. Lee, and Shanna Pearson-Merkowitz. 2008. The Check Is in the Mail: Interdistrict Funding Flows in
Congressional Elections.American Journal of Political Science 52 (2): 373394.
Gingrich, Jane, and Ben Ansell. 2022. Concentration and Commodification: The Political Economy of Postindustrialism in
America and Beyond.In The American Political Economy: Politics, Markets, and Power, edited by Jacob S. Hacker, Alexander
Hertel-Fernandez, Paul Pierson and Kathleen Thelen. New York: Cambridge University Press. Chap. 13.
Grenzke, Janet. 1988. Comparing Contributions to U.S. House Members from outside Their District.Legislative Studies
Quarterly 13 (1): 83103.
Hacker, Jacob S. et al. 2022. Introduction: The American Political Economy: A Framework and Agenda for Research.In The
American Political Economy: Politics, Markets, and Power, edited by Jacob S. Hacker, Alexander Hertel-Fernandez, Paul
Pierson and Kathleen Thelen. Working Paper. New York: Cambridge University Press.
Haskel, Jonathan, and Stian Westlake. 2018. Capitalism Without Capital: The Rise of the Intangible Economy. Princeton & Oxford:
Princeton University Press.
Hertel-Fernandez, Alexander. 2018. Politics at Work: How Companies Turn Their Workers Into Lobbyists. New York: Oxford
University Press.
Iversen, Torben, and David Soskice. 2019. Democracy and Prosperity: Reinventing Capitalism Through a Turbulent Century.
Princeton and Oxford: Princeton University Press.
Mettler, Suzanne. 2011. The Submerged State: How Invisible Government Policies Undermine American Democracy. Chicago and
London: The University of Chicago Press.
Moretti, Enrico. 2013. The New Geography of Jobs. Boston and New York: Mariner Books of Houghton Mifflin Harcourt.
Pearlstein, Joanna. 2018. Its True: Tech Workers Overwhelmingly Support Democrats in 2018.InWired.
Rodden, Jonathan. 2019. Why Cities Lose: The Deep Roots of the Urban-Rural Political Divide. Basic Books.
Schickler, Eric. 2016. Racial Realignment: The Transformation of American Liberalism, 1932-1965. Princeton & Oxford: Princeton
University Press.
Schlozman, Daniel. 2015. When Movements Anchor Parties: Electoral Realignments in American History. Princeton & Oxford:
Princeton University Press.
Schwartz, Herman Mark 2022. MoPatents, MoProblems: Corporate Strategy, Structure, and Profitability in Americas Political
Economy.InThe American Political Economy: Politics, Markets, and Power, edited by Jacob S. Hacker, Alexander Hertel-
Fernandez, Paul Pierson and Kathleen Thelen. New York: Cambridge University Press.
Scola, Nancy. 2014. CaliforniasPolitical ATMis now located closer to San Francisco than L.A.In The Washington Post.
Short, Nicholas. 2022. The Politics of the American Knowledge Economy.Studies in American Political Development.
DOI: 10.1017/S0898588X21000134.
Song, Jae et al. 2019. Firming Up Inequality.The Quarterly Journal of Economics 134(1), 150.
Soskice, David. 2022. The United States as Radical Innovation Driver: The Politics of Declining Dominance?In The American
Political Economy: Politics, Markets, and Power, edited by Jacob S. Hacker, Alexander Hertel-Fernandez, Paul Pierson and
Kathleen Thelen. New York: Cambridge University Press. Chap. 11.
Techonomy 2015. Booker, Fischer, and Parker: Technology, Innovation and American Progress. Videotaped interview.
Accessed November 06, 2015. https://techonomy.com/video/booker-fischer-parker-technology-innovation-and-american-pro
gress/.
Business and Politics 81
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
A Appendix
A.1 Constructing the Inventor-Donor Database
As indicated in the main article, the process for creating the inventor-donor data set involved three main steps: (1) identify all
inventors (first and last name, firm, and city and state of residence) listed on US patents that were applied for on or after 1 January
1979 and who resided within the United States using research datasets provided by the US Patent and Trademark Office;
(2) identify the subset of these US inventors that also appear in the DIME database using fastLink and acquire data on their
contribution history and imputed ideology; and (3) match the self-reported employer names from the DIME database to
organizations in Capital IQ to generate unique identifiers for these organizations plus other information, like SIC codes, where
available.
To implement the first step, I used the research datasets published by the PTO on the Patentsview website to build a database
containing the first and last name, city and state, and organizational assignee (a firm, a university, a government agency, etc.) for
all inventors who applied for a US patent on or after 1 January 1979, who listed an address in the US in their correspondence with
the PTO (i.e., were American residents at the time of the patent application), and who assigned their patent to some organizational
entity. Assignees are usually employers; by law, the inventors named on a patent must be people, but ownership of the patent
routinely passes to that persons employer by virtue of the employment contract. If that does not happen, ownership passes to the
inventors (there is no assignee). Because employer is an essential field for matching with DIME data, I exclude instances where
ownership passes to the inventors and keep only instances where patent ownership passes to some organization.
To implement the second step, I gathered the same information (name, city and state, and employer) from the DIME database.
Using fastLink, I then identified those American inventors who also contributed to a political campaign at some point from 1979
through 2014 (the 1980-2014 cycles). I completed the matching in three steps. First, I stratified the patent and donor data by both
election cycle and state. The algorithm would therefore only find a match if an inventor both applied for a patent and made a
campaign contribution in the same election cycle (an election year and the prior year). These matches are the strongest because
the invention and donation occur close in time. Second, I stratified the remaining data (after purging matches from the first step)
by state and repeated the matching for inventors in all states except California, New York, and Texas. These results introduce the
possibility of more error because the acts of invention and donation are not close in time. But it captures instances where, for
example, an inventor at Microsoft who lives in Washington and stays in Washington applies for a patent in, say, 1991 but does not
donate to a campaign until, say, 2008. Third, and finally, for the remaining data in California, New York, and Texas, I stratified by
both state and the first letter of the inventors last name. Without this further stratification for these three large states, probabilistic
matching was not computationally feasible.
The administrators of both the Patentsview and the DIME data sets have run their own disambiguation algorithms to generate
unique identifiers for inventors (in Patentsview) and donors (in DIME). To ensure a higher quality of matching, I kept only those
high probability matches where both datasets agreed that the match identified a unique individual. In other words, I abandoned
instances where a single DIME identifier was matched to more than one Patentsview identifier and vice versa. This produced a
dataset of 30,603 American inventors who contributed to a political campaign from 1979 through 2014.
Once inventor-donors are matched in this fashion, it is possible to use the unique identifiers in both data set to construct an
invention record, containing data on all patents applied for by these inventor-donors from 1979-2019, and a donor record,
containing data on all campaign contributions made by these inventor-donors from 1979-2014. Below, I focus exclusively on
analyzing the donor record of American inventor-donors.56 I also confine the donor record to campaign contributions made in all
federal elections from the 1980 cycle through the 2014 cycle. The donor and recipient party coding in the DIME database appear
to be a mix of FEC codes and legacy Voteview codes. In the analysis below, I re-coded the recipient types as Democratic candidates
and committees, Republican candidates and committees, and political actions committees of unknown partisan affiliation (PAC-
UPAs) and ignored contributions to other partisan entities (which were not substantial in any time period). PAC-UPAs are
committees that either do not have a partisan designation in the underlying DIME data, do not have an ideological score or have a
middlingideological score (greater than 0.5 and less than 0.5) which makes it difficult to impute a partisan tendency based on
donation patterns, and do not have the text strings Republicanor Democratin their name.
The DIME dataset does not have disambiguated firm or organizational identifiers, and it is problematic to use those provided
in the Patentsview dataset for several reasons. I therefore implemented my own name matching between the self-reported
employer listed in the donation record of American inventor-donors and the organizations in Standard & Poors Capital IQ
database. To execute this third step, I first excluded instances where the DIME employer was missing or appeared to be conflated
with occupation or employment status (CEO, engineer, retired, etc.). I then ranked the remaining employer names in descending
order by the number of inventor donations (not the dollar amount) associated with that employer. I fed all of these names into
Capital IQs proprietary lookup algorithm to generate a suggested match and then audited the matches in two steps. First, because
the top 2,212 of these names account for roughly 74.6 percent of all inventor donations across all election cycles, I manually
audited the proposed matches, leading to 2,050 valid matches. For the remaining results, I implemented a relatively soft constraint
on name similarity: that the DIME employer name and Capital IQ organization name had a Jaro-Winkler distance less than or
56In the Introduction, I appealed to the invention record to identify the technological domain (based on patent data) in which
each inventor-donor predominately works.
82 Nicholas Short
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
equal to 0.15, which produced another 21,204 matches.57 Together, these 23,254 self-reported employer names were linked to
14,735 unique organizations with Capital IQ identifiers.
The matching analysis implemented in Section Construction of the Datasetutilizes the subset of inventor-donor data where
the DIME employer was linked to a Capital IQ organization through one of these 23,254 matches. Table A1 presents some
summary statistics about this subset of the inventor-donor data for each election cycle. The second column shows the total
number of donors in the DIME database in thousands. The third column shows the percentage of all donors that are inventor-
donors, which varies over time between 0.5 and 2.4 percent of all donors. The fourth column shows the percentage of all donors
that are linked to Capital IQ organizations. It reveals that the link between DIME employers and Capital IQ organizations is
weakest in the 1980s, which is to be expected given that the Capital IQ database has the best coverage from the mid-1990s to the
present. The fifth column shows the share of all donors that are linked to Capital IQ organizations that are inventor-donors, which
essentially defines the pool of inventor-donors eligible for matching. It shows that the linking to Capital IQ organizations slightly
reduces the share of inventors compared to all donors (column three) in the 1980s, that there is no relative loss in the 1990s, and
that the linking slightly reduces the relative share of non-inventors from 2000 to 2014. But it does not do so dramatically in any
election cycle. The sixth column shows the number of inventor-donors that were matched to non-inventor donors by
organization, Congressional District, and imputed gender, and the seventh columns shows the matching success rate, which is
number of matched inventor-donors as a share of inventor-donors linked to Capital IQ organizations. It shows that matching
succeeds in 20-31 percent of cases in the 1980s, 3544 percent of cases in the 1990s, and in 5667 percent of cases from 2002
to 2014.
Table A1. Summary of the inventor-donor dataset
Cycle
Total donors
(Thousands)
Inventor share
(%)
CIQ linked share
(%)
Linked inventor
Share (%)
Matched
inventors
Matched share
(%)
1980 225.1 0.6 4.5 0.7 29 39.7
1982 101.4 2.2 4.6 1.1 14 26.4
1984 152.9 2.0 4.0 1.4 24 27.9
1986 155.9 2.6 5.8 1.4 44 34.6
1988 247.6 2.0 5.9 1.2 63 37.3
1990 287.8 2.2 7.7 1.3 131 44.3
1992 451.1 1.7 6.9 1.7 254 47.8
1994 428.7 2.1 8.3 1.9 284 42.6
1996 595.8 1.8 8.2 1.9 379 40.5
1998 487.2 2.7 9.1 2.5 456 41.6
2000 777.2 2.0 9.2 2.3 698 43.1
2002 894.2 1.9 13.2 1.8 1,231 56.7
2004 1,693.3 1.1 12.9 1.8 2,484 61.5
2006 1,357.0 1.6 16.0 2.1 2,669 59.9
2008 2,603.7 0.9 13.4 1.9 4,241 63.8
2010 1,689.6 1.5 15.1 2.3 3,676 62.2
2012 3,310.9 0.9 12.9 2.0 5,821 67.1
2014 2,433.0 1.3 12.5 2.2 4,269 65.1
Note: This table presents basic summary statistics about the inventor-dono r data set and the subset of that data linked to Capital IQ organizations
used for the matching analysis in Section Construction of the Dataset. For each election cycle (column 1), it shows the total number of donors in
the DIME data (column 2), the share of total donors that are inventors (column 3), the share of total donors and that are linked to Capital IQ
organizations (column 4), and the share of donors linked to Capital IQ organizations that are also inventors (column 5). The last two columns
show the number of inventor-donors matched to non-inventor donors by firm, gender, and Congressional District (column 6) and the matching
success rate as a share of inventor-donors linked to Capital IQ organizations (column 7).
57That cutoff was chosen because, after auditing small samples, I observed that the proposed matches below this cutoff
generated very few potentially false matches, while matches with a Jaro-Winkler distance between 0.15 and 0.2 had about
30 percent potentially false matches.
Business and Politics 83
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
The ANOVA analysis implemented in Section Analysisis slightly different. Here, the goal is to understand whether
polarization is increasing among knowledge economy workers who contribute to Democrats even if that organizations inventors
do not donate. For this exercise, carried out only in the 1992 and 2012 election cycles, I supplemented the data set with data on
non-inventor donors at known IP producers. Specifically, I used patent data to first identify all IP producers (any organization
that was issued a patent) from 1987-1991 and from 2007-2011 (the five years prior to each relevant election year). I then linked
self-reported DIME employers to these IP producers and, for those employers not already matched above, I linked the IP producer
names to Capital IQ firm names. This allowed me to link DIME employers to an additional 887 Capital IQ organizations in 1992
and an additional 7,200 Capital IQ organizations in 2012. These organizations produced IP in the years leading up to the election
cycle and had employees who donated in federal elections but did not have inventor-donors who made contributions.
Table ?? characterizes the data set used in the ANOVA analysis. For each election cycle, column 2 shows the total number of
donors in the DIME data and column 3 shows the share of those donors (inventors and non-inventor employees at IP producers)
that are linked to Capital IQ organizations. Columns four through six show the number of organizations, Congressional Districts,
and industries (4-digit SIC codes) that are represented in this data set. As shown, the data set covers virtually all Congressional
Districts in each election cycle,58 and captures data on donors from 957 organizations in 339 industries in 1992 and 6,038
organizations in 714 industries in 2012.
A.2 ANOVA Analysis Motivation and Details
This variance decomposition described in the paper was predicated upon and modeled after similar analyses conducted in
prominent studies of rising wage inequality.59 In those studies, the question was whether increasing wage inequalityreflected by
an increasing variance in the overall wage distribution over timewas best explained by changes between firms, with the average
wages of some superior firms pulling away from the average wages of their competitors, or within firms, with executive pay (for
example) pulling away from pay for administrative staff across many firms.
Here, the phenomenon to explain is not increasing variance in the wage distribution over time but decreasing variance in the
ideology distribution over time among knowledge economy workers that give to Democratic candidates and committees.
To determine whether geographic or firm-level shifts are driving the declining ideological variance among Democratic inventors,
I implemented a Bayesian form of ANOVA decomposition for each subset of knowledge economy workers in the 1992 and 2012
election cycles using the runjags library in R.60 Specifically, I fit the following non-nested hierarchical model to each data set in
each election cycle:
yiNji bki
;σ2
y

ajN0;σ2
a

bjN0;σ2
b

Here yirepresents the ideology score for donor iresiding in Congressional District jiand working at organization ki. The
estimated standard deviations, σa,σb,σycan be interpreted as point estimates of the variation in the average ideology across
districts, the average ideology across organizations, and the residual variation within districts and organizations, respectively.
Following Gelman and Hill, I report finite population empirical standard deviations since there is no super-population of
Congressional Districts beyond those observed in the data, though this choice does not impact the results.61
A Bayesian form of ANOVA is preferable, here, because the goal is not to test whether the batches of coefficients for
Congressional Districts and organizations, aji and bki
, are statistically significant sources of variation in ideology among
Democratic knowledge economy workers. Both variables are highly significant in this respect in both election cycles. The goal is
rather to precisely estimate (and efficiently compute) the amount of observed variation between the batch of district effects and
organization effects in each period, and the residual variation within both districts and organizations, and determine which
plausibly explains the overall decline in the variance of ideology scores among Democratic knowledge economy workers.
58There is a 436th district because the at-large district for the District of Columbia is included.
59Barth et al. (2016); Song et al. (2019).
60Denwood (2016).
61Gelman and Hill (2007, Ch. 22).
84 Nicholas Short
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
Table A2. Regression results for ideology modelfull matched datase
Dependent variable: Common factor ideology score
1980 1982 1984 1986 1988 1990
inventor 0.143 (0.200) 0.009 (0.263) 0.091 (0.235) 0.010 (0.134) 0.072 (0.119) 0.052 (0.096)
Constant 0.531*** (0.127) 0.783*** (0.186) 0.424** (0.162) 0.533*** (0.091) 0.606*** (0.082) 0.525*** (0.062)
Observations 35 20 38 76 92 216
R20.015 0.0001 0.004 0.0001 0.004 0.001
Adjusted R20.015 0.055 0.024 0.013 0.007 0.003
Residual Std. error 0.580 (df =33) 0.587 (df =18) 0.724 (df =36) 0.581 (df =74) 0.571 (df =90) 0.695 (df =214)
FStatistic 0.508 (df =1; 33) 0.001 (df =1; 18) 0.150 (df =1; 36) 0.005 (df =1; 74) 0.369 (df =1; 90) 0.293 (df =1; 214)
1992 1994 1996 1998 2000 2002
inventor 0.062 0.020 0.023 0.063 0.074 0.121**
(0.080) (0.073) (0.070) (0.072) (0.062) (0.050)
Constant 0.394*** (0.052) 0.441*** (0.048) 0.401*** (0.047) 0.297*** (0.049) 0.256*** (0.042) 0.298*** (0.031)
Observations 433 479 656 694 1,095 1,825
R20.001 0.0002 0.0002 0.001 0.001 0.003
Adjusted R20.001 0.002 0.001 0.0004 0.0004 0.003
Residual Std. error 0.817 (df =431) 0.792 (df =477) 0.888 (df =654) 0.948 (df =692) 1.017 (df =1093) 1.037 (df =1823)
FStatistic 0.614 (df =1; 431) 0.076 (df =1; 477) 0.108 (df =1; 654) 0.751 (df =1; 692) 1.431 (df =1; 1093) 5.900** (df =1; 1823)
2004 2006 2008 2010 2012 2014
inventor 0.161*** (0.037) 0.176*** (0.038) 0.157*** (0.029) 0.142*** (0.033) 0.097*** (0.025) 0.082*** (0.029)
Constant 0.106*** (0.024) 0.103*** (0.024) 0.268*** (0.019) 0.248*** (0.021) 0.492*** (0.016) 0.494*** (0.018)
A.3 Regression Output
Business and Politics 85
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
Table A2. (Continued )
2004 2006 2008 2010 2012 2014
Observations 3,945 3,818 6,938 5,305 9,321 6,158
R20.005 0.006 0.004 0.004 0.002 0.001
Adjusted R20.004 0.005 0.004 0.003 0.002 0.001
Residual Std. error 1.155 (df =3943) 1.138 (df =3816) 1.188 (df =6936) 1.159 (df =5303) 1.184 (df =9319) 1.112 (df =6156)
FStatistic 18.667*** (df =1; 3943) 21.838*** (df =1; 3816) 29.627*** (df =1; 6936) 19.100*** (df =1; 5303) 15.267*** (df =1; 9319) 7.971*** (df =1; 6156)
Note: This table shows the results of regressing common factor (CF) ideology scores on a binary variable indicating whether the donor is an inventor in the matched inventor data set (as described in Sections PriorWork and
Hypothesesand Construction of the Dataset) for each election cycle from 1980-2014.
*p<0.1; **p<0.05; ***p<0.01.
86 Nicholas Short
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
Table A3. Regression output for ideology modelswitchers
Dependent variable: Common factor ideology score
1982 1984 1986 1988 1990 1992
inventor 0.277 (0.269) 0.147 (0.268) 0.135 (0.234) 0.052 (0.240) 0.276 (0.234) 0.062 (0.197)
Constant 0.830** (0.190) 0.607** (0.190) 0.828*** (0.165) 0.691*** (0.163) 0.563*** (0.152) 0.271** (0.126)
Observations 6 6 8 13 38 75
R20.210 0.069 0.053 0.004 0.037 0.001
Adjusted R20.012 0.163 0.105 0.086 0.011 0.012
Residual Std. error 0.329 (df =4) 0.329 (df =4) 0.331 (df =6) 0.431 (df =11) 0.712 (df =36) 0.838 (df =73)
FStatistic 1.060 (df =1; 4) 0.299 (df =1; 4) 0.333 (df =1; 6) 0.047 (df =1; 11) 1.398 (df =1; 36) 0.098 (df =1; 73)
1994 1996 1998 2000 2002 2004
inventor 0.115 (0.212) 0.118 (0.229) 0.083 (0.177) 0.011 (0.162) 0.352** (0.151) 0.183* (0.102)
Constant 0.260* (0.139) 0.353** (0.157) 0.167 (0.122) 0.317*** (0.111) 0.427*** (0.091) 0.197*** (0.067)
Observations 65 91 130 151 214 500
R20.005 0.003 0.002 0.00003 0.025 0.006
Adjusted R20.011 0.008 0.006 0.007 0.020 0.004
Residual Std. error 0.846 (df =63) 1.089 (df =89) 1.010 (df =128) 0.994 (df =149) 1.059 (df =212) 1.133 (df =498)
FStatistic 0.297 (df =1; 63) 0.266 (df =1; 89) 0.220 (df =1; 128) 0.005 (df =1; 149) 5.428** (df =1; 212) 3.174* (df =1; 498)
2006 2008 2010 2012 2014
inventor 0.160 (0.109) 0.177** (0.090) 0.268*** (0.093) 0.127* (0.072) 0.060 (0.084)
Constant 0.146** (0.069) 0.341*** (0.059) 0.235*** (0.057) 0.578*** (0.046) 0.640*** (0.052)
Observations 429 738 670 1,071 668
R20.005 0.005 0.012 0.003 0.001
Adjusted R20.003 0.004 0.011 0.002 0.001
Residual Std. error 1.109 (df =427) 1.204 (df =736) 1.166 (df =668) 1.152 (df =1069) 1.058 (df =666)
FStatistic 2.154 (df =1; 427) 3.910** (df =1; 736) 8.247*** (df =1; 668) 3.124* (df =1; 1069) 0.508 (df =1; 666)
Note: This table shows the results of regressing common factor (CF) ideology scores on a binary variable indicating whether the donor is an inventor among switchers (as described in Sections PriorWork and Hypotheses
and Construction of the Dataset) for each election cycle from 1980-2014.
*p<0.1; **p<0.05; ***p<0.01.
Business and Politics 87
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
Table A4. Regression results for democratic donor modelfull matched dataset
Dependent variable: Democratic donor
1980 1982 1984 1986 1988 1990
inventor 0.000 (0.725) 0.000 (1.491) 0.214 (0.654) 0.000 (0.514) 0.421 (0.462) 0.281 (0.284)
Constant 1.163** (0.512) 2.197** (1.054) 0.619 (0.469) 1.131*** (0.364) 1.240*** (0.342) 0.828*** (0.194)
Observations 42 20 40 82 98 250
Log likelihood 23.053 6.502 26.409 45.554 56.276 146.793
Akaike Inf. Crit. 50.105 17.003 56.818 95.108 116.553 297.586
1992 1994 1996 1998 2000 2002
inventor 0.190 (0.195) 0.137 (0.185) 0.080 (0.163) 0.050 (0.158) 0.000 (0.124) 0.010 (0.100)
Constant 0.760*** (0.135) 0.688*** (0.129) 0.775*** (0.115) 0.787*** (0.112) 0.631*** (0.088) 1.223*** (0.071)
Observations 502 538 704 748 1,152 2,252
Log likelihood 305.668 336.862 433.918 461.522 743.858 1,209.641
Akaike Inf. Crit. 615.336 677.723 871.835 927.044 1,491.716 2,423.281
2004 2006 2008 2010 2012 2014
inventor 0.102* (0.060) 0.067 (0.062) 0.187*** (0.045) 0.319*** (0.051) 0.391*** (0.039) 0.607*** (0.047)
Constant 0.274*** (0.042) 0.578*** (0.043) 0.014 (0.032) 0.186*** (0.036) 0.420*** (0.028) 0.283*** (0.033)
Observations 4,574 4,596 7,910 6,394 10,736 7,472
Log likelihood 3,109.524 2,980.247 5,467.983 4,319.290 7,325.466 5,093.655
Akaike Inf. Crit. 6,223.047 5,964.493 10,939.970 8,642.579 14,654.930 10,191.310
Note: This table shows the results of regressing a binary variable indicating whether the donor contributed to a Democratic candidate or committee on a binary variable indicating whether the donor is an inventor in the full
matched dataset (as described in Sections PriorWork and Hypothesesand Construction of the Dataset) for each election cycle from 1980-2014.
*p<0.1; **p<0.05; ***p<0.01.
88 Nicholas Short
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
Table A5. Regression results for democratic donor modelswitchers
Dependent variable: Democratic donor
1982 1984 1986 1988 1990 1992
Inventor 0.000 (106,969.800) 1.386 (1.732) 0.000 (1.633) 0.000 (1.183) 0.613 (0.646) 0.000 (0.450)
Constant 24.566 (75,639.060) 0.693 (1.225) 1.099 (1.155) 0.916 (0.837) 0.368 (0.434) 0.659** (0.318)
Observations 6 6 8 14 44 88
Log Likelihood 0.000 3.819 4.499 8.376 27.775 56.464
Akaike Inf. Crit. 4.000 11.638 12.997 20.752 59.549 116.928
1994 1996 1998 2000 2002 2004
Inventor 0.117 (0.483) 0.480 (0.441) 0.000 (0.354) 0.161 (0.328) 0.000 (0.296) 0.155 (0.168)
Constant 0.496 (0.339) 0.990*** (0.325) 0.565** (0.250) 0.619*** (0.234) 1.315*** (0.209) 0.119 (0.118)
Observations 74 96 138 160 274 574
Log Likelihood 48.527 59.791 90.354 105.205 141.431 394.705
Akaike Inf. Crit. 101.054 123.582 184.708 214.410 286.863 793.409
2006 2008 2010 2012 2014
Inventor 0.017 (0.185) 0.142 (0.138) 0.411*** (0.142) 0.563*** (0.114) 0.727*** (0.142)
Constant 0.595*** (0.131) 0.047 (0.097) 0.239** (0.098) 0.535*** (0.082) 0.414*** (0.101)
Observations 512 844 840 1,278 824
Log Likelihood 332.589 584.424 558.162 863.701 557.539
Akaike Inf. Crit. 669.178 1,172.847 1,120.324 1,731.402 1,119.078
Note: This table shows the results of regressing a binary variable indicating whether the donor contributed to a Democratic candidate or committee on a binary variable indicating whether the donor is an inventor among
switchers (as described in Sections 2 and 3) for each election cycle from 1980-2014.
*p<0.1; **p<0.05; ***p<0.01.
Business and Politics 89
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
Table A6. Regression results for democratic share modelfull matched dataset
Dependent variable: Democratic share of donations
1980 1982 1984 1986 1988 1990
inventor 2.111 (5.332) 0.000*** (0.000) 17.198*** (6.028) 4.461 (3.401) 3.747 (3.816) 1.470 (2.427)
dem_donor 60.096*** (6.259) 100.000*** (0.000) 76.415*** (6.225) 76.309*** (3.960) 71.244*** (4.307) 72.724*** (2.705)
Constant 1.056 (4.054) 0.000*** (0.000) 8.255* (4.782) 2.231 (2.591) 1.742 (2.841) 0.760 (1.900)
Observations 42 20 40 82 97 247
R20.703 1.000 0.808 0.825 0.745 0.749
Adjusted R20.688 1.000 0.798 0.821 0.739 0.747
Residual Std. error 17.276 (df =39) 0.000 (df =17) 19.036 (df =37) 15.398 (df =79) 18.696 (df =94) 19.037 (df =244)
FStatistic 46.173***
(df =2; 39)
38,330,146,607,691,530,952,077,391,953,920.000*** (df =2; 17) 77.803***
(df =2; 37)
186.563***
(df =2; 79)
137.025***
(df =2; 94)
363.990***
(df =2; 244)
1992 1994 1996 1998 2000 2002
inventor 0.125 (1.665) 3.114* (1.591) 0.510 (1.425) 0.106 (1.316) 2.545** (1.081) 1.097 (0.706)
dem_donor 74.837*** (1.812) 76.332*** (1.703) 75.435*** (1.542) 77.164*** (1.424) 80.722*** (1.132) 74.284*** (0.839)
Constant 0.064 (1.313) 1.578 (1.258) 0.258 (1.121) 0.053 (1.033) 1.281 (0.865) 0.547 (0.535)
Observations 496 535 698 745 1,135 2,236
R20.776 0.792 0.775 0.798 0.818 0.778
Adjusted R20.775 0.791 0.774 0.798 0.818 0.778
Residual Std. error 18.517 (df =493) 18.388 (df =532) 18.826 (df =695) 17.957 (df =742) 18.213 (df =1132) 16.699 (df =2233)
FStatistic 854.673***
(df =2; 493)
1,009.887*** (df =2; 532) 1,197.789***
(df =2; 695)
1,468.023***
(df =2; 742)
2,547.524***
(df =2; 1132)
3,917.237***
(df =2; 2233)
2004 2006 2008 2010 2012 2014
inventor 1.233** (0.490) 0.216 (0.527) 0.175 (0.412) ** 1.123 (0.481) 0.468 (0.326) 0.658 (0.412)
dem_donor 89.208*** (0.496) 81.852*** (0.551) 86.351*** (0.412) 78.483*** (0.487) 90.435*** (0.329) 89.994*** (0.412)
(Continued)
90 Nicholas Short
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
Table A6. (Continued )
2004 2006 2008 2010 2012 2014
Constant 0.630 (0.408) 0.110 (0.424) 0.092 (0.358) 0.599 (0.406) 0.260 (0.304) 0.379 (0.372)
Observations 4,539 4,536 7,823 6,309 10,681 7,431
R20.877 0.830 0.849 0.805 0.878 0.868
Adjusted R20.877 0.830 0.849 0.805 0.877 0.868
Residual Std.
error
16.488 (df =4536) 17.759 (df =4533) 18.185 (df =7820) 19.028 (df =6306) 16.791 (df =10678) 17.554 (df =7428)
FStatistic 16,226.480*** (df =2;
4536)
11,053.980*** (df =2;
4533)
22,025.750*** (df =2;
7820)
13,050.360*** (df =2;
6306)
38,245.970*** (df =2;
10678)
24,467.720*** (df =2;
7428)
Note: This table shows the results of regressing the share of donations given to Democratic candidates and committees on a binary variable indicating whether the donor is an inventor (inventor) and a binary variable
indicating whether the donor contributed to a Democratic candidate or committee (dem_donor). Conditional on being a Democratic donor, it captures the difference between inventors and non-inventors in the share of total
contributions made to Democratic candidates and committees. The regressions were run in the full matched dataset (as described in Sections PriorWork and Hypothesesand Construction of the Dataset) for each
election cycle from 1980-2014.
*p<0.1; **p<0.05; ***p<0.01.
Business and Politics 91
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
Table A7. Regression results for democratic share modelswitchers
Dependent variable: Democratic share of donations
1982 1984 1986 1988 1990 1992
inventor 0.000 (0.000) 34.375 (23.868) 8.655 (6.704) 0.851 (14.308) 10.747* (6.194) 2.530 (4.524)
dem_donor 65.625* (23.868) 32.690*** (7.741) 55.655*** (15.836) 77.635*** (6.534) 68.339*** (4.733)
Constant 0.000 (0.000) 11.458 (17.790) 4.327 (5.120) 0.426 (11.083) 5.930 (5.092) 1.288 (3.635)
Observations 6 6 8 14 44 85
R20.724 0.796 0.529 0.789 0.718
Adjusted R20.540 0.714 0.443 0.779 0.711
Residual Std. error 0.000 (df =4) 27.560 (df =3) 9.481 (df =5) 26.767 (df =11) 20.330 (df =41) 20.854 (df =82)
FStatistic 3.934 (df =2; 3) 9.749** (df =2; 5) 6.178** (df =2; 11) 76.655*** (df =2; 41) 104.454*** (df =2; 82)
1994 1996 1998 2000 2002 2004
inventor 6.086 (5.103) 0.497 (4.071) 0.917 (3.486) 4.270 (2.878) 1.507 (1.922) 0.630 (1.393)
dem_donor 71.844*** (5.300) 73.674*** (4.329) 75.154*** (3.615) 83.835*** (2.978) 78.590*** (2.349) 89.904*** (1.399)
Constant 3.108 (4.127) 0.221 (3.058) 0.458 (2.801) 2.093 (2.297) 0.750 (1.445) 0.326 (1.188)
Observations 74 94 136 159 273 568
R20.724 0.764 0.765 0.836 0.806 0.880
Adjusted R20.716 0.758 0.761 0.833 0.804 0.879
Residual Std. error 21.938 (df =71) 19.569 (df =91) 20.329 (df =133) 18.132 (df =156) 15.876 (df =270) 16.588 (df =565)
FStatistic 93.122*** (df =2; 71) 146.983*** (df =2; 91) 216.093*** (df =2; 133) 396.318*** (df =2; 156) 559.941*** (df =2; 270) 2,068.380*** (df =2; 565)
2006 2008 2010 2012 2014
inventor 1.178 (1.655) 0.461 (1.236) 1.235 (1.441) 1.141 (0.971) 0.272 (1.269)
dem_donor 80.267*** (1.725) 87.459*** (1.236) 73.059*** (1.472) 90.399*** (0.980) 89.812*** (1.270)
Constant 0.595 (1.329) 0.238 (1.082) 0.666 (1.208) 0.666 (0.923) 0.161 (1.169)
Observations 504 835 826 1,270 821
(Continued)
92 Nicholas Short
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
Table A7. (Continued )
2006 2008 2010 2012 2014
R20.812 0.858 0.751 0.872 0.863
Adjusted R20.811 0.857 0.750 0.872 0.863
Residual Std. error 18.574 (df =501) 17.849 (df =832) 20.605 (df =823) 17.131 (df =1267) 17.892 (df =818)
FStatistic 1,083.590*** (df =2; 501) 2,505.006*** (df =2; 832) 1,240.425*** (df =2; 823) 4,330.088*** (df =2; 1267) 2,586.726*** (df =2; 818)
Note: This table shows the results of regressing the share of donations given to Democratic candidates and committees on a binary variable indicating whether the donor is an inventor (inventor) and a binary variable
indicating whether the donor contributed to a Democratic candidate or committee (dem_donor). Conditional on being a Democratic donor, it captures the difference between inventors and non-inventors in the share of total
contributions made to Democratic candidates and committees. The regressions were run among switchers (as described in Sections Prior Work and Hypothesesand Construction of the Dataset) for each election cycle
from 1980-2014.
*p<0.1; **p<0.05; ***p<0.01.
Business and Politics 93
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
Table A8. Regression results for republican share modelfull matched dataset
Dependent variable: Republican share of donations
1980 1982 1984 1986 1988 1990
inventor 13.827 (13.953) 1.860 (1.908) 14.857 (10.916) 3.173 (8.169) 0.255 (8.491) 10.928** (5.407)
dem_donor 25.069 (16.380) 98.967*** (3.180) 61.767*** (11.274) 57.887*** (9.512) 48.375*** (9.585) 26.626*** (6.026)
Constant 66.812*** (10.609) 99.897*** (1.386) 77.868*** (8.660) 77.491*** (6.225) 65.930*** (6.321) 47.405*** (4.233)
Observations 42 20 40 82 97 247
R20.079 0.983 0.457 0.320 0.215 0.085
Adjusted R20.031 0.981 0.428 0.303 0.198 0.077
Residual Std. error 45.212 (df =39) 4.266 (df =17) 34.474 (df =37) 36.989 (df =79) 41.606 (df =94) 42.414 (df =244)
F Statistic 1.662 (df =2; 39) 484.748*** (df =2; 17) 15.590*** (df =2; 37) 18.594*** (df =2; 79) 12.879*** (df =2; 94) 11.304*** (df =2; 244)
1992 1994 1996 1998 2000 2002
inventor 8.362** (3.747) 6.047* (3.664) 5.206* (3.056) 4.510 (3.067) 3.796* (2.218) 5.871*** (1.682)
dem_donor 25.045*** (4.079) 34.805*** (3.922) 50.822*** (3.305) 43.115*** (3.319) 57.927*** (2.321) 15.798*** (1.999)
Constant 43.715*** (2.956) 52.469*** (2.897) 70.686*** (2.403) 57.661*** (2.408) 71.569*** (1.773) 30.415*** (1.273)
Observations 496 535 698 745 1,135 2,236
R20.078 0.132 0.255 0.187 0.356 0.032
Adjusted R20.074 0.128 0.253 0.185 0.355 0.032
Residual Std. Error 41.689 (df =493) 42.352 (df =532) 40.360 (df =695) 41.852 (df =742) 37.354 (df =1132) 39.776 (df =2233)
FStatistic 20.775*** (df =2; 493) 40.359*** (df =2; 532) 119.217*** (df =2; 695) 85.212*** (df =2; 742) 312.792*** (df =2; 1132) 37.397*** (df =2; 2233)
2004 2006 2008 2010 2012 2014
inventor 8.046*** (1.096) 5.464*** (1.022) 8.656*** (0.808) 6.382*** (0.852) 8.772*** (0.638) 5.376*** (0.725)
dem_donor 32.021*** (1.109) 17.324*** (1.067) 38.636*** (0.809) 22.130*** (0.863) 40.166*** (0.643) 23.783*** (0.725)
Constant 41.680*** (0.913) 26.300*** (0.821) 46.707*** (0.704) 29.204*** (0.720) 46.665*** (0.595) 28.585*** (0.656)
Observations 4,539 4,536 7,823 6,309 10,681 7,431
(Continued)
94 Nicholas Short
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
Table A8. (Continued )
2004 2006 2008 2010 2012 2014
R20.162 0.060 0.231 0.098 0.271 0.127
Adjusted R20.162 0.060 0.231 0.098 0.271 0.127
Residual Std.
error
36.905 (df =4536) 34.404 (df =4533) 35.701 (df =7820) 33.751 (df =6306) 32.819 (df =10678) 30.907 (df =7428)
FStatistic 438.471*** (df =2;
4536)
144.684*** (df =2;
4533)
1,177.555*** (df =2;
7820)
343.496*** (df =2;
6306)
1,983.195*** (df =2;
10678)
540.543*** (df =2;
7428)
Note: This table shows the results of regressing the share of donations given to Republican candidates or committees on a binary variable indicating whether the donor is an inventor (inventor) and a binary variable
indicating whether the donor contributed to a Democratic candidate or committee (dem_donor). Conditional on being a Democratic donor, it captures the difference between inventors and non-inventors in the share of
contributions made to Republican candidates and committees. The regressions were run in the full matched dataset (as described in Sections Prior Work and Hypothesesand Construction of the Dataset) for each
election cycle from 1980-2014.
*p<0.1; **p<0.05; ***p<0.01.
Business and Politics 95
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
Table A9. Regression results for republican share modelswitchers
Dependent variable: Republican share of donations
1982 1984 1986 1988 1990 1992
inventor 6.200 (6.200) 34.375 (23.868) 8.655 (6.704) 9.030 (9.706) 2.085 (13.424) 6.097 (9.017)
dem_donor 65.625* (23.868) 32.690*** (7.741) 7.152 (10.743) 31.366** (14.160) 17.557* (9.433)
Constant 100.000*** (4.384) 88.542** (17.790) 95.672*** (5.120) 15.835* (7.518) 46.525*** (11.036) 39.344*** (7.244)
Observations 6 6 8 14 44 85
R20.200 0.724 0.796 0.106 0.107 0.045
Adjusted R20.000 0.540 0.714 0.056 0.064 0.022
Residual Std. error 7.593 (df =4) 27.560 (df =3) 9.481 (df =5) 18.158 (df =11) 44.059 (df =41) 41.560 (df =82)
FStatistic 1.000 (df =1; 4) 3.934 (df =2; 3) 9.749** (df =2; 5) 0.654 (df =2; 11) 2.467* (df =2; 41) 1.950 (df =2; 82)
1994 1996 1998 2000 2002 2004
inventor 0.398 (9.552) 4.317 (8.448) 5.015 (7.002) 8.402 (5.704) 7.198 (4.583) 11.568*** (2.996)
dem_donor 25.945** (9.922) 50.221*** (8.983) 45.000*** (7.261) 58.817*** (5.903) 14.351** (5.602) 31.140*** (3.008)
Constant 39.796*** (7.725) 73.091*** (6.346) 56.971*** (5.625) 75.166*** (4.553) 27.308*** (3.445) 42.865*** (2.555)
Observations 74 94 136 159 273 568
R20.088 0.265 0.226 0.397 0.032 0.174
Adjusted R20.062 0.249 0.215 0.389 0.025 0.171
Residual Std. error 41.070 (df =71) 40.608 (df =91) 40.830 (df =133) 35.940 (df =156) 37.864 (df =270) 35.679 (df =565)
FStatistic 3.426** (df =2; 71) 16.394*** (df =2; 91) 19.460*** (df =2; 133) 51.280*** (df =2; 156) 4.522** (df =2; 270) 59.572*** (df =2; 565)
2006 2008 2010 2012 2014
inventor 4.967* (2.858) 10.250*** (2.437) 6.530*** (2.187) 10.029*** (1.752) 5.054*** (1.949)
dem_donor 13.294*** (2.978) 40.337*** (2.437) 12.672*** (2.234) 30.970*** (1.767) 18.618*** (1.950)
Constant 22.665*** (2.295) 48.537*** (2.133) 21.473*** (1.833) 37.711*** (1.666) 22.526*** (1.794)
Observations 504 835 826 1,270 821
(Continued)
96 Nicholas Short
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
Table A9. (Continued )
2006 2008 2010 2012 2014
R20.044 0.257 0.044 0.200 0.101
Adjusted R20.040 0.255 0.042 0.199 0.099
Residual Std. error 32.076 (df =501) 35.189 (df =832) 31.271 (df =823) 30.901 (df =1267) 27.468 (df =818)
FStatistic 11.418*** (df =2; 501) 143.584*** (df =2; 832) 19.055*** (df =2; 823) 158.775*** (df =2; 1267) 46.004*** (df =2; 818)
Note: This table shows the results of regressing the share of donations given to Republican candidates or committees on a binary variable indicating whether the donor is an inventor (inventor) and a binary variable
indicating whether the donor contributed to a Democratic candidate or committee (dem_donor). Conditional on being a Democratic donor, it captures the difference between inventors and non-inventors in the share of
contributions made to Republican candidates and committees. The regressions were run among switchers (as described in Sections Prior Work and Hypothesesand Construction of the Dataset) for each election cycle
from 1980-2014.
*p<0.1; **p<0.05; ***p<0.01.
Business and Politics 97
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
Table A10. Regression results for PAC-UPA share modelfull matched dataset
Dependent variable: PAC-UPA share of donations
1980 1982 1984 1986 1988 1990
inventor 15.939 (13.086) 1.860 (1.908) 2.340 (9.016) 1.288 (7.527) 4.003 (8.490) 12.398** (5.382)
dem_donor 35.027** (15.362) 1.033 (3.180) 14.647 (9.311) 18.422** (8.764) 22.869** (9.583) 46.098*** (5.998)
Constant 32.133*** (9.950) 0.103 (1.386) 13.877* (7.152) 20.278*** (5.735) 32.329*** (6.320) 51.835*** (4.213)
Observations 42 20 40 82 97 247
R20.146 0.058 0.063 0.053 0.058 0.215
Adjusted R20.103 0.052 0.013 0.029 0.038 0.208
Residual Std. error 42.402 (df =39) 4.266 (df =17) 28.472 (df =37) 34.079 (df =79) 41.598 (df =94) 42.218 (df =244)
FStatistic 3.341** (df =2; 39) 0.528 (df =2; 17) 1.253 (df =2; 37) 2.224 (df =2; 79) 2.875* (df =2; 94) 33.382*** (df =2; 244)
1992 1994 1996 1998 2000 2002
inventor 8.487** (3.674) 9.161** (3.601) 5.716** (2.898) 4.616 (3.040) 6.341*** (2.163) 6.968*** (1.690)
dem_donor 49.791*** (3.999) 41.527*** (3.855) 24.614*** (3.135) 34.049*** (3.290) 22.795*** (2.264) 58.487*** (2.009)
Constant 56.221*** (2.898) 45.953*** (2.847) 29.056*** (2.279) 42.285*** (2.387) 27.150*** (1.730) 69.038*** (1.279)
Observations 496 535 698 745 1,135 2,236
R20.249 0.189 0.087 0.129 0.089 0.279
Adjusted R20.246 0.186 0.084 0.127 0.087 0.278
Residual Std. error 40.869 (df =493) 41.631 (df =532) 38.280 (df =695) 41.487 (df =742) 36.436 (df =1132) 39.965 (df =2233)
FStatistic 81.597*** (df =2; 493) 62.071*** (df =2; 532) 33.067*** (df =2; 695) 54.895*** (df =2; 742) 55.084*** (df =2; 1132) 432.001*** (df =2; 2233)
2004 2006 2008 2010 2012 2014
inventor 9.278*** (1.078) 5.680*** (1.053) 8.480*** (0.840) 5.259*** (0.917) 9.240*** (0.663) 6.035*** (0.776)
dem_donor 57.187*** (1.091) 64.528*** (1.099) 47.715*** (0.841) 56.353*** (0.929) 50.269*** (0.668) 66.211*** (0.776)
Constant 57.690*** (0.898) 73.590*** (0.846) 53.385*** (0.732) 71.394*** (0.774) 53.076*** (0.618) 71.037*** (0.702)
Observations 4,539 4,536 7,823 6,309 10,681 7,431
(Continued)
98 Nicholas Short
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
Table A10. (Continued )
2004 2006 2008 2010 2012 2014
R20.386 0.435 0.302 0.376 0.364 0.509
Adjusted R20.385 0.435 0.302 0.375 0.364 0.509
Residual Std.
error
36.286 (df =4536) 35.445 (df =4533) 37.125 (df =7820) 36.303 (df =6306) 34.121 (df =10678) 33.079 (df =7428)
FStatistic 1,424.310*** (df =2;
4536)
1,744.568*** (df =2;
4533)
1,692.067*** (df =2;
7820)
1,896.364*** (df =2;
6306)
3,059.940*** (df =2;
10678)
3,854.735*** (df =2;
7428)
Note: This table shows the results of regressing the share of donations given to PAC-UPAs (PACs of unknown partisan affiliation) on a binary variable indicating whether the donor is an inventor (inventor) and a binary
variable indicating whether the donor contributed to a Democratic candidate or committee (dem_donor). Conditional on being a Democratic donor, it captures the difference between inventors and non-inventors in the
share of contributions made to PAC-UPAs. The regressions were run in the full matched dataset (as described in Sections Prior Work and Hypothesesand Construction of the Dataset) for each election cycle from
1980-2014.
*p<0.1; **p<0.05; ***p<0.01.
Business and Politics 99
https://doi.org/10.1017/bap.2023.25 Published online by Cambridge University Press
Table A11. Regression results for PAC-UPA share modelswitchers
Dependent variable: PAC-UPA share of donations
1982 1984 1986 1988 1990 1992
inventor 6.200 (6.200) 0.000 (0.000) 0.000 (0.000) 8.179 (16.104) 12.832 (12.959) 8.627 (8.571)
dem_donor 0.000 (0.000) 0.000 (0.000) 48.502** (17.823) 46.269*** (13.669) 50.781*** (8.967)
Constant 0.000 (4.384) 0.000 (0.000) 0.000 (0.000) 84.591*** (12.474) 47.545*** (10.654) 59.368*** (6.886)
Observations 6 6 8 14 44 85
R20.200 0.411 0.250 0.288
Adjusted R<