Immigrants to the U.S. play a particularly prominent rolein science and engineering commu-

nities. In the 2000 Census of Populations, immigrantsconstituted 25% and 48% of the U.S.

workforce employed in science and engineering occupationswith bachelor’s and doctorate ed-

ucations, respectively. Moreover, immigrants account formost of the recent growth in U.S.

scientists and engineers. This paper analyzes the impactthese individuals have had on the

global operations of U.S. firms by addressing three mainquestions. First, to what extent do

U.S. based innovators of a particular ethnicity enhancethe competitiveness of U.S. multinational

firms in countries associated with that ethnicity?Second, how do these immigrants in‡uence

the global distribution of the multinational’s researchand development (R&D) and patenting

e¤orts? Finally, are U.S. multinationals that employinnovators of a particular ethnicity less

dependent on joint venture partners when forming new affiliatesin countries associated with

that ethnicity?

High-skilled immigrants are likely to have severalattributes that could help U.S. multina-

tionals capitalize on foreign opportunities. Beyondlanguage skills, well-educated immigrants

typically possess specialized knowledge about how toconduct business in their home countries.

They are likely to have a strong understanding ofcustomer behavior there and to have insights

about what kinds of products would succeed. Furthermore,high-skilled immigrants often also

have relationships and are part of networks that canfacilitate foreign market access. In order

to study these e¤ects of skilled immigrants, it isparticularly useful to work with data that links

individuals of particular ethnicities to specific firrms.

Such data are drawn from a variety of sources. In orderto characterize the immigrant science

and engineering workforce of …rms, the analysis uses ameasure based on one type of their output,

namely patents. More specifically, the analysis usesdetailed filings from the U.S. Patent and

Trademark O¢ ce for all patents granted from 1975-2008.These filings include the names of the

inventors of each patent, their employer, and theirlocation. In order to the measure the degree to

which innovative activity is performed by individualsfrom each of nine ethnic groups, procedures

1

that make use of commercial databases of ethnic namesassign probable ethnicities to innovators.

For example, innovators with the surnames Ming or Yu areassigned a high probability of being

of Chinese ethnicity, while innovators with the surnamesAgrawal or Banerjee are assigned a

high probability of being of Indian ethnicity.

In order to conduct tests of the relation between ethnicinnovation and multinational firm

activity, the analysis links data on inventors to data onthe activities of U.S. multinational …rms

captured in the 1982, 1989, 1994, 1999, and 2004 Surveyof U.S. Direct Investment Abroad con-

ducted by the Bureau of Economic Analysis (BEA). Thesedata include measures of the activity

of each of the foreign subsidiaries of multinationalswith a U.S. parent, including measures of

assets, sales, employment, and employment compensation.The BEA data also contain informa-

tion on where multinationals perform R&D and measuresof the ownership structure of foreign

a¢ liates.

Tests that analyze panel data of parent-ethnicityobservations reveal that increases in the

share of innovation performed by individuals of a certainethnicity are associated with increases

in the share of multinational a¢ liate activity in theirnative countries. These tests include

parent-ethnicity …xed e¤ects so that responses aremeasured o¤ of time series variation in the

role played by innovators of a speci…c ethnicity at aspeci…c …rm, and they include a …xed e¤ect

for each ethnicity-year to control for trends in thegrowth of distinct ethnicities. The results of

these tests are particularly pronounced for …rms that arelikely to place high value on ethnic

innovators in the sense that these …rms are beginning toperform innovative activity in the home

countries of the innovators.

The results also do not seem to merely capture thepossibility that decisions to employ in-

novators of a certain ethnicity and to expand incountries associated with that ethnicity are

jointly determined. Measures of the share of ethnicinnovation re‡ect shares in the years pre-

ceding the measures of a¢ liate activity. Furthermore,results hold in speci…cations that use a

measure of the predicted extent of ethnic innovation thatis computed based on a …rm’s initial

level of ethnic innovation across U.S. cities and thesubsequent growth in ethnic innovation by

2

city. This approach is similar to the supply-pushimmigration framework of Card (2001). Taken

together, the results on the relation between the shareof innovation performed by an ethnicity

and the share of multinational …rm activity in the homecountries of that ethnicity indicate that

immigration enhances the competitiveness of U.S.multinationals. The knowledge and cultural

sensitivities of these innovators thus appear to bevaluable in helping multinationals unlock key

factors to succeeding in these markets.

The data allow for exploration of where U.S. firmsconduct R&D and of the extent to which

U.S. based innovators team up with foreign innovators togenerate patents. Linear probability

specifications that control for parent-ethnicity andethnicity-year fixed e¤ects illustrate that …rms

with more patents generated by innovators of a particularethnicity are more likely to conduct

R&D in the countries associated with that ethnicity.Similar speci…cations also reveal that …rms

with more patents generated by U.S. based innovators of aparticular ethnicity are more likely

to collaborate with innovators based in countriesassociated with that ethnicity when generating

patents. Thus, the paper shows that ethnic innovatorsfacilitate the disintegration of innovative

activity within multinational …rms across countries.

Analysis of new a¢ liates reveals that U.S.multinationals are able to own larger shares of new

entities in countries that are home to …rms’ethnicinnovators. Linear probability speci…cations

that include parent-year …xed e¤ects indicate that higherlevels of patenting activity by inventors

of a particular ethnicity are associated with higherpropensities to form new a¢ liates as wholly

owned or majority owned entities. Previous work indicatesthat one motivation for the use

of joint ventures is to gain access to a local partnerwho can provide information about local

demand and customs.1The …ndings in this paper suggest that the input of ethnic innovators

makes the input of local partners less valuable andlowers entry barriers to foreign countries.

These …ndings contribute to several literatures byillustrating the role …rms play in linking

immigration, foreign direct investment (FDI), andknowledge di¤usion. A signi…cant body of

research documents the e¤ects of immigration on otherforms of international economic inter-

1

See, for example, Balakrishnan and Koza (1993) and Desai,Foley, and Hines (2004).

3

action.2 Ethnic networks have been shown to playimportant roles in promoting international

trade, investment, and cross-border …nancing activity,with recent work particularly emphasiz-

ing the role of educated or skilled immigrants.3 Much of this work uses aggregated data and

cross-sectional techniques, so the panel analysis of …rm-leveldata in this paper complements it

and identi…es key mechanisms in these linkages.

Recent work also considers the possibility that socialand ethnic ties facilitate transfers of

technology.4 Individuals who are geographically mobileappear to play a signi…cant role in these

kinds of transfers.5Because this paper’s …ndings illustrate a mechanism by which knowledge is

transferred globally, it also adds to research on therole multinational …rms play in the interna-

tional di¤usion of knowledge.6Finally, the results inform a growing body of work that analyzes

…rm decisions about whether to locate innovative activityin a single place or in multiple loca-

tions.7

The remainder of this paper is organized as follows.Section 2 provides details about the

data. Section 3 includes three parts; the …rst describesthe analysis of how U.S. based ethnic

innovation shapes the share of a multinational’s activityin countries associated with that eth-

nicity. The second part describes the analysis of theextent to which ethnic innovators facilitate

the disintegration of innovative activity across borders.The third part presents the examination

of whether …rms that employ innovators of a certainethnicity are less likely to use joint ventures

when they form new a¢ liates in countries associated withthat ethnicity. Section 4 concludes.

2

Rauch (2001) reviews papers on the economic impact ofethnic networks, and Saxenian, Motoyama, and Quan

(2002) provide survey evidence on the cross-borderlinkages of science and engineering immigrants in particular.

3

Papers in this literature include Saxenian (2002, 2006),Arora and Gambardella (2005), Buch, Kleinert, and

Toubal (2006), Kugler and Rapoport (2007, 2011),Bhattacharya and Groznik (2008), Docquier and Lodigiani

(2010), Huang, Jin, and Qian (2010), Iriyama, Li, andMadhavan (2010), Hernandez (2011), and Javorcik et al.

(2011). Related work on trade includes Gould (1994), Headand Ries (1998), Rauch (1999), Rauch and Trindade

(2002), Kerr (2009), and Rangan and Sengul (2009).Clemens (2009) and Docquier and Rapoport (2011) provide

broader reviews.

4

Examples of this work include Agrawal, Cockburn, andMcHale (2006), MacGarvie (2006), Oettl and Agrawal

(2008), Kerr (2008), Papageorgiou and Spilimbergo (2008),and Agrawal et al. (2011).

5

For evidence of this point, see Almeida and Kogut (1999),Rosenkopf and Almeida (2003), Nanda and Khanna

(2010), Choudhury (2010), and Hovhannisyan and Keller(2010).

6

Papers on this topic include Keller (2004), Veugelers andCassiman (2004), Singh (2004, 2005, 2007), Mac-

Garvie (2005), Branstetter (2006), Alcacer and Chung(2007), and Nachum, Zaheer, and Gross (2008).

7

Recent work on this topic includes Zhao (2006), Singh(2008), Alcacer and Zhao (2011), and Zhao and Islam

(2011).

4

2

Data

This section describes the data employed, starting withthe ethnic patenting data developed for

U.S. multinational …rms. The second part describes theBEA data on the foreign operations of

these …rms and the merger of the two data sources.

2.1

Data on Ethnic Innovators

Measures of the ethnicity of innovators employed at U.S.multinational …rms are created on the

basis of data on each patent granted by the United StatesPatent and Trademark O¢ ce between

January 1975 and May 2008. Hall, Ja¤e, and Trajtenberg(2001) provide extensive details about

these data, and Griliches (1990) surveys the use ofpatents as economic indicators of technological

advancement. Each patent lists at least one and oftenseveral inventors and includes information

on the location and employer of each inventor. These dataare extensive, containing over eight

million inventors and four million granted patents duringthe sample period. Much of the analysis

below considers the impact of U.S. based innovators, andinventors are classi…ed as being based

in the U.S. if they are located in a U.S. city. Althoughthe data are selected using a screen

related to the date of patent grants, the date of patentapplications is used to identify the timing

of innovative activity.

The immigration status of inventors is not listed onpatents, but it is possible to determine

their probable ethnicity through their names. Thematching approach exploits the fact that

people with particular …rst names and surnames are likelyto be of a certain ethnicity and makes

use of two databases of ethnic names. The …rst wasdeveloped by the Melissa Data Corporation

for use in direct-mail advertisements and the second byLSDI, also for marketing purposes. The

process a¤ords the distinction of nine ethnicities:Anglo-Saxon, Chinese, European, Hispanic,

Indian, Japanese, Korean, Russian, and Vietnamese. Whenthere is more than one inventor

associated with a patent, each individual is given anethnicity assignment and then these are

averaged. The name match rate is 99%. Kerr (2007, 2010)provides details on the matching

process, lists frequent ethnic names, and providesmultiple descriptive statistics and quality

5

assurance exercises.

Table 1 displays the share of U.S. based innovationperformed by ethnic innovators working

at public companies over the time periods that areanalyzed in more detail in Section 3. The

Anglo-Saxon ethnic share declines from 81% of U.S.domestic patents for public …rms in the

1975-1982 period to 68% in the 2000-2004 period. Thisdeclining share is primarily due to the

growth in innovation among Chinese and Indianethnicities, which increase from under 3% to

10% and 7%, respectively. The data also indicate thatethnic inventors are more concentrated

in high-tech industries than in other industries and thatthis gap has widened substantially

over the past three decades. Furthermore, while ethnicinnovation was particularly prevalent

in pharmaceuticals and chemicals industries in the 1970s,ethnic contributions to innovation in

computers and electronics industries were particularlyprevalent in the 2000s.

The tests below exploit variation within …rms in theshare of innovation performed by in-

ventors of a certain ethnicity and control forethnicity-year …xed e¤ects. Therefore, the tests

depend on there being variation in evolution of ethnicinnovation across …rms. Figure 1, which

is constructed from the patent database, illustrates thatsuch heterogeneity exists among seven

large U.S. …rms that report earning foreign income inCompustat.8 Each line plots the share

of U.S. based innovation that is attributed to Chineseand Indian innovators for one of seven

large …rms. As indicated, there is substantial variationin the levels and changes of the share of

innovation performed by Chinese and Indian inventorsacross …rms.

The analysis described below uses data on ethnicinnovation aggregated to the …rm-ethnicity-

year level. The analysis calls for measures of ethnicinnovation that precede the measures of the

outcomes of interest. Therefore, levels and shares ofinnovation performed by each ethnicity for

each …rm are calculated for each time period listed inTable 1. The years associated with each

period relate to the timing of patent applications. Onaverage, slightly more than 50 patents

per …rm and time period are used to calculate theserelative ethnic contributions.

8

In order to protect the con…dentiality of the BEA data,to which the patent data are linked, the names of

these …rms are not identi…ed.

6

2.2

Data on U.S. Multinational Firm Activity

Data on the activities of U.S. multinational …rms aredrawn from the Survey of U.S. Direct

Investment Abroad conducted by the Bureau of EconomicAnalysis. U.S. direct investment

abroad is de…ned as the direct or indirect ownership orcontrol by a single U.S. legal entity of at

least 10% of the voting securities of an incorporatedforeign business enterprise or the equivalent

interest in an unincorporated foreign businessenterprise. A U.S. multinational …rm includes the

U.S. legal entity that has made the direct investment,called the U.S. parent, and at least one

foreign business enterprise, called a foreign a¢ liate.9 The sample includes records drawn from

the 1982, 1989, 1994, 1999, and 2004 benchmark surveys.These surveys capture …nancial and

operating data for each foreign a¢ liate of each U.S.multinational, so it is possible to create a

panel of data on the assets, sales, employment, andemployment compensation for each …rm in

each country. The BEA data also include information onthe parent’s ownership share of each

a¢ liate, as well as the amount a¢ liates spend onR&D.

A number of steps were taken to link the data on U.S.multinationals with the data on ethnic

innovators. Data on the CUSIPs of employers of ethnicinnovators were taken from the NBER

Patent Citations Data File and have been manually updatedto assign patents to subsidiaries

of major corporations and to account for major mergersand acquisitions.10 These CUSIPs

were matched with Employment Identi…cation Numbers (EINs)from Compustat. The BEA

data include EINs, and an automated merge was performedon the basis of these. Automated

matches were manually con…rmed and augmented with avisual comparison of …rm names. One

notable consequence of this process is that the matchedsample only includes publicly listed

…rms because CUSIPs are used as the starting point.

Much of the analysis below also aggregates the data onU.S. multinational …rm activity to the

…rm-ethnicity-year level. This requires relatingethnicities to countries. There is a one-to-one

mapping of ethnicity and country for …ve cases. Chinese,European, and Hispanic ethnicities each

9

As a result of con…dentiality assurances and penaltiesfor noncompliance, BEA believes that survey coverage

is close to complete and levels of accuracy are high.Mataloni (1995) and Mataloni and Yorgason (2002) provide

further details on these FDI data.

10

Debbie Strumsky and Bill Lincoln performed portions ofthis update.

7

relate to more than one country. Chinese economiesinclude Mainland China, Hong Kong, Macao,

Singapore, and Taiwan. European economies includeAustria, Belgium, Denmark, Finland,

France, Germany, Italy, Luxembourg, Netherlands, Norway,Poland, Sweden, and Switzerland.

Hispanic economies include Argentina, Belize, Brazil,Chile, Colombia, Costa Rica, Dominican

Republic, Ecuador, El Salvador, Guatemala, Honduras,Mexico, Nicaragua, Panama, Paraguay,

Peru, Philippines, Portugal, Spain, Uruguay, andVenezuela.

The …nal sample has several limitations, but it o¤ersbroad coverage of U.S. multinational ac-

tivity nonetheless. The …rms only include publicly listedentities that have been granted patents

during the sample period and have a foreign a¢ liate.Anglo-Saxon innovators and multinational

activity in Anglo-Saxon countries are removed from thesample because such innovators are less

likely to be recent immigrants and to have distinctiveties to countries associated with their

ethnicity. Although it is not possible to identify ethnicnames associated with many countries

like Thailand or Saudi Arabia, the aggregated data cover45 foreign countries. The …nal sample

includes 641 …rms which account for more than two-thirdsof aggregate foreign a¢ liate sales

in each of the locations associated with non-Anglo Saxonethnicities in each benchmark year.

Furthermore, these shares are higher in industries thatintensively employ patenting. Table 2

presents descriptive statistics for the variables used inthe analysis below.

3

Empirical Tests and Results

This section describes the empirical tests and presentsthe results. It contains three subsections.

The …rst presents analyses of the relation between theshare of innovation performed by a partic-

ular ethnicity and the share of multinational a¢ liateactivity that occurs in countries associated

with that ethnicity. The second, which includes twoparts, explores the association between

ethnic innovation and the amount and location ofinnovative activity that U.S. multinationals

perform outside of the U.S. The third describes tests ofwhether U.S. multinationals own larger

shares of a¢ liates in countries that are home to …rms’ethnicinnovators.

8

3.1

Ethnic Innovation and Shares of Multinational A¢ liateActivity

One of the questions this paper seeks to address iswhether U.S. based innovators of a particular

ethnicity enhance the competitiveness of U.S.multinational …rms in countries associated with

that ethnicity. Several tests shed light on this questionby examining the relationship between

the share of innovation performed in the U.S. by acertain ethnicity and the subsequent share of

a¢ liate activity that occurs in the countries of originof those inventors. The basic estimating

equation takes the following form:

M N E%fet =

fe

+

et

+

EI %fet +

f et:

(1)

The observations employed in this test relate to aparticular …rm for a particular ethnicity in

a particular year. M N E%f etis a measure of the share of …rm f ’s foreign activity that occurs

in countries associated with ethnicity e in benchmarksurvey year t: Four measures of this

share are calculated using data on foreign a¢ liateassets, sales, employment, and employment

compensation. EI %f etmeasures the share of U.S. based innovation performed by individuals of

ethnicity e in the period leading up to benchmark surveyyear t: These periods span seven years

for the 1982 and 1989 benchmark years and …ve years forthe 1994, 1999, and 2004 benchmark

years.

fe

and

et

are vectors of …rm-ethnicity and ethnicity-year …xed e¤ects.Standard errors

are clustered by ethnicity-year.

Several features of this speci…cation are noteworthy. The…rm-ethnicity …xed e¤ects remove

time invariant di¤erences in the extent to which …rmsinvest in countries associated with a par-

ticular ethnicity and employ innovators of a particularethnicity. The

parameter is therefore

identi…ed o¤ of changes in these …rm characteristics overthe sample period. A potential con-

cern is that there appear to be secular trends in theshares of innovation performed by certain

ethnicities, as indicated in Table 1, and these mightcoincide with secular trends in the growth

of a¢ liate activity. Including ethnicity-year …xed e¤ectsaddresses this concern. Finally, …rm

speci…c changes in the scale of activity could generatecoincident changes in the levels of ethnic

innovation and multinational a¢ liate activity. Measuringthe extent of ethnic innovation and

9

the location of multinational a¢ liate activity usingshares, as opposed to levels, addresses this

concern.

Table 3 presents results of tests using speci…cation (1).The dependent variable in the …rst

column is the share of a¢ liate assets in countriesassociated with a particular ethnicity. The

0.1008 coe¢ cient in column 1 is statistically signi…cantand implies that a one standard deviation

increase in the share of innovation by individuals of aparticular ethnicity is associated with a 2.3

percentage point increase in the share of multinational a¢liate activity in the native countries of

the innovators. Consistent results are obtained for othermeasures of the distribution of a¢ liate

activity that are computed using data on sales,employment, and employment compensation, as

indicated in columns 2-4. Because the estimates thatappear in columns 3 and 4 are of a similar

magnitude, the results suggest that changes in the shareof ethnic innovation are not associated

with changes in the wage structures of foreignoperations.

These basic results are robust to a variety of checks.They do not depend on the inclusion

of any particular ethnicity; the results hold droppingeach of the ethnicities. They also do not

appear to be a consequence of activity in particularindustries where patenting is especially

prevalent. Removing …rms that are primarily engaged inthe production of pharmaceuticals or

other chemicals; audio, video and communicationequipment; or computer and o¢ ce equipment

does not overturn the results. The measured relationshipsalso do not seem to be driven by the

recent rapid growth in innovative activity by individualsof Chinese or Indian ethnicity; removing

observations related to the 2004 benchmark survey doesnot a¤ect the results.

The …ndings in Table 3 suggest that innovation byindividuals of a particular ethnicity en-

hances the competitiveness of U.S. multinationals incountries associated with that ethnicity. If

this interpretation is correct, one would expect U.S.based ethnic innovation to have particularly

large e¤ects when …rms are also beginning to engage ininnovative activity in countries associ-

ated with an ethnicity. U.S. based ethnic innovatorscould play a valuable role in facilitating

cooperation between innovators working in di¤erentlocations and in identifying products and

services that could be developed further abroad to meetlocal demands. In order to identify such

10

situations, it is possible to use the patent datadescribed above to isolate …rm-ethnicities for

which: 1) the …rms had previously applied for patents forinnovations of U.S. based inventors

and 2) subsequently applied for patents for innovationsinvolving inventors located in countries

of a particular ethnicity. This sample is labeled thesample of new foreign innovators.

Table 4 presents the results of running speci…cation (1)on two subsamples, the sample of new

foreign innovators and other observations. The top panelpresents results for the new foreign

innovator sample and the bottom panel for otherobservations. The 0.2155 coe¢ cient on the

Ethnic Share of U.S. Patents in the top panel isstatistically signi…cant and much larger than

the 0.0551 insigni…cant coe¢ cient on this variable inthe bottom panel. A similar pattern holds

across the panels for the speci…cations in columns 2-4.The results therefore indicate that the

association between U.S. based ethnic innovation andmultinational a¢ liate activity are more

pronounced in situations where U.S. based ethnicinnovations are arguably more valuable to the

…rms they work for.

An additional and perhaps more fundamental concern thatcan be raised about the results

in Table 3 is that they may re‡ect omitted variable biasor reverse causality. In particular, …rms

might jointly make decisions about the use of ethnicinnovators and about where to expand

internationally. Alternatively, conducting FDI abroad maylead to identi…cation of promising

scientists and engineers that are then brought to theU.S. to work. It is therefore desirable

to create an alternative measure of ethnic innovationthat is more likely to exhibit exogenous

variation.

One such measure can be computed using the patent dataand is based on the initial distri-

bution of ethnic innovation across U.S. cities for speci…c…rms and the subsequent local growth

of ethnic innovation. This framework is based on thesupply-push work of Card (2001), which

has also been applied in the immigration and patentingcontext by Hunt and Gauthier-Loiselle

(2010) and Kerr and Lincoln (2010). The identi…cationbuilds o¤ the fact that immigrants of

di¤erent ethnicities tend to agglomerate in certaincities and the fact that rates of immigration

to the U.S. have di¤ered across ethnicities. For example,many Chinese immigrants settle in

11

San Francisco, while many Hispanic immigrants settle inMiami. The immigration of Chinese

scientists and engineers to the U.S. is therefore morelikely to in‡uence …rms in San Francisco

than …rms in Miami.

More speci…cally, the Predicted Ethnic Share of U.S.Patents is computed by …rst calculating

ExpEIfet as follows:

ExpEIfet =  EIf;cet0                   +                 :         (2)

c

The …rst term in the expression following the summationcaptures the initial distribution of

ethnic innovation for a …rm. It is the count of patentsapplied for by …rm f in which the

inventor is based in city c and is of ethnicity e at timet0, which is the …rst benchmark year the

…rm appears in the data. The analysis considers 281cities de…ned as Metropolitan Statistical

Areas, and Kerr (2010) lists major cities and theirinventor shares.

The terms within the parentheses measure growth inpatenting activity for …rms other than

…rm f . Taking this approach increases the likelihoodthat this measure of ethnic innovation

is exogenous. For cities in which a single …rm isresponsible for a large share of patenting

activity, growth in local patenting by ethnicity forother …rms can exhibit irregular properties.

Therefore, the terms in parentheses calculate growthrates using a weighted average of city

speci…c and national growth in ethnic patenting for other…rms. The two weights are captured

by

EI  f;cet0

EIcet0

and

EIf;cet0

EIcet0

.

These two weights sum to one, and the …rst is the shareof the initial

patent counts attributable to …rms other than …rm f ,while the second is the share attributable

to …rm f .

EI

EI

f;cet

f;cet0

is the local growth in patent applications …led by …rmsother than …rm f for

patents in which the inventor is based in city c and isof ethnicity e in period t relative to t0.

EI

EI

f;et

f;et0

is a similar measure of growth, but it is measured acrossall cities and is not city speci…c.

As such, city speci…c growth gets more weight when a …rmis responsible for a smaller share of

total innovative activity in the city.

The Predicted Ethnic Share of U.S. Patents is equal toExpEIf et scaled by the total number

of patents …rm f is expected to apply for in year t, andit is thus a predicted share for each

ethnicity in each period. The predicted share for anindividual ethnicity in …rm f increases over

X

EI f;cet0 EI f;cet   EIf;cet0 EI f;et

EIcet0   EI f;cet0   EIcet0  EI f;et0

the sample period if the initially observed ethnicinnovation of the …rm occurred in cities that

subsequently experienced strong in‡ows of researchers ofthat ethnicity. The spatial distribution

of each …rm is held …xed at its initial level to avoidcapturing …rms expanding into new cities to

take advantage of di¤erential growth in innovation.

Table 5 presents the results of tests that make use ofthis alternative measure of ethnic in-

novation. As in the previous two tables, the speci…cationspresented include …rm-ethnicity and

ethnicity-year …xed e¤ects, and standard errors areclustered by ethnicity-year. It is notewor-

thy that the …xed e¤ects absorb the impact of di¤erencesin the initial distribution of ethnic

innovation for a …rm as well as the aggregate immigrationtrends of di¤erent ethnicities. The

identi…cation therefore comes from di¤erences in theextent to which …rms were exposed to

di¤erent growth in ethnic innovation across U.S. cities.The speci…cation in the …rst column

provides evidence that the Predicted Ethnic Share of U.S.Patents is positively correlated with

the Ethnic Share of U.S. Patents. The 0.1917 coe¢ cienton the Predicted Ethnic Share of U.S.

Patents indicates that the two measures are closelyrelated, but it is less than one, implying that

factors besides growth in ethnic innovation across citiesin‡uence how inventor compositions

evolve in large …rms.

The dependent variables in the next four columns are thesame ones considered in Table 3.

The coe¢ cients on Predicted Ethnic Share of U.S. Patentsare positive in each of these speci…-

cations, and in three of the four speci…cations they havea similar or larger magnitude than the

coe¢ cients on the Ethnic Share of U.S. Patents reportedin Table 3. The coe¢ cients are statis-

tically signi…cant in columns 3 and 4. The …ndingsindicate that changes in ethnic innovation

related to plausibly exogenous changes in the growth ofethnic innovation across U.S. cities are

associated with changes in the distribution of U.S.multinational a¢ liate sales and employment.

Therefore, these results alleviate some concerns aboutthe potential endogeneity of the Ethnic

Share of U.S. Patents in Table 3. The tests are notperfect, as a forward looking manager might

have located the …rm’s initial inventive facilities toattract innovators of a particular ethnicity

in anticipation of foreign expansion, for example.Nevertheless, this approach does show the

13

robustness of Table 3’s results to several endogeneityconcerns.

3.2

Ethnic Innovation and the Disintegration of InnovativeActivity

Two pieces of analysis shed light on the role of ethnicinnovators in breaking up innovative

activities across borders. The …rst piece examines a¢liate R&D activity, and the second piece

considers the patenting of foreign innovations.

3.2.1

A¢ liate R&D Activity

Although U.S. multinationals perform a large share oftheir R&D within the U.S., this share

has been shrinking. According to the aggregate publishedBEA data, majority owned foreign

a¢ liates performed 6.4% of U.S. multinational R&D in1982, but this ratio was 13.6% in 2004.

This globalization of R&D activities has receivedconsiderable recent attention in the academic

literature.11 While early foreign R&D e¤orts focusedon re…ning products so they were suitable

for foreign markets and on accessing foreigntechnologies, recent e¤orts also attempt to tap

into the large supply of foreign scientists and engineersregardless of their knowledge of speci…c

foreign technologies.12U.S. based ethnic innovators could be especially valuable in facilitating

the disintegration of inventive activity acrosscountries.

Linear probability speci…cations shed light on thispossibility, and these take the following

form:

R&Dfet =

fe

+

et

+

ln(EIfet) +

f et:

(3)

R&Dfet is a dummy variable equal to one if …rmf conducts R&D in countries of ethnicity e in

benchmark year t. Like speci…cation (1), this speci…cationincludes …rm-ethnicity and ethnicity-

year …xed e¤ects. Because the dependent variable does notmeasure the share of R&D performed

in countries of a particular ethnicity but insteadcaptures the extensive margin of R&D activity,

the measure of ethnic innovation is not measured as ashare either. ln(EIf et) is the log of the

count of the number of patents a …rm applies for in theperiod before the benchmark year for

11

12

See, for example, Dalton et al. (1999), Freeman (2006),Zhao (2006), and Puga and Tre‡er (2010).

Studies of these issues include Niosi (1999), vonZedtwitz and Gassmann (2002), Thursby and Thursby

(2006), and National Science Foundation (2010).

14

which the inventor is of ethnicity e. One concern thatcould be raised about this approach is

that ln(EIfet) might re‡ect something about theoverall scale of parent activity. Growing …rms

might increase employment of ethnic innovators and bemore likely to conduct R&D abroad. To

address this possibility, tests include the log of parentR&D expenditures and the log of parent

sales.

Table 6 presents the results. The 0.0192 coe¢ cient incolumn 1 implies that a one standard

deviation increase in the log of ethnic U.S. patents isassociated with a 4.2 percentage point

increase in the likelihood of conducting R&D incountries associated with that ethnicity. This

e¤ect is sizeable given that the mean likelihood that a …rmconducts R&D in countries associated

with a particular ethnicity is 48%, implying a relativeincrease of 9%. The speci…cation in column

2 adds the log of parent R&D as a control. Itattracts a positive and signi…cant coe¢ cient,

implying that …rms that conduct more R&D in the U.S.are more likely to conduct R&D abroad.

The coe¢ cient on the log of ethnic U.S. patents becomessomewhat smaller in this speci…cation,

but it remains statistically signi…cant. The speci…cationin column 3 also adds the log of parent

sales, and the coe¢ cient on the log of ethnic U.S.patents remains signi…cant in this test as well.

Thus, the results in Table 6 indicate that U.S. basedinnovation by inventors of a certain

ethnicity facilitate R&D activity in countriesassociated with that ethnicity. Further evidence

of the manner in which ethnic innovators support thedisintegration of inventive activity across

borders comes from the analysis of patent data.

3.2.2

Patenting Foreign Inventions

If ethnic innovators promote meaningful foreign R&D,this activity should result in patents that

list inventors located outside of the U.S. Speci…cationsthat take the following form consider this

possibility:

F oreignP atentf et =

fe

+

et

+

ln(EIfet) +

f et:

(4)

15

F oreignP atentf et is a dummy equalto one if …rm f applies for at least one patent in which at

least one inventor is based in a country associated withethnicity e in the period that precedes

benchmark year t.13Other variables are de…ned as in speci…cation (3). To account for potential

scale e¤ects, some tests further control for a …rm’stotal patent applications from the U.S., ex-

cluding the focal ethnicity. This latter variableincludes Anglo-Saxon contributions that comprise

the majority of U.S. multinational innovation. The sampleemployed in this test di¤ers from

the samples used elsewhere. This sample is not restrictedto …rm-ethnicity observations where a

foreign a¢ liate exists in the BEA data. Thus, the patentsample includes public U.S. …rms that

never conduct foreign operations, and it includesethnicities within …rms where domestic ethnic

invention occurs but where foreign a¢ liate activitynever occurs.

Results of running this speci…cation appear in Table 7.In column 1, the log of ethnic U.S.

patents has a positive and signi…cant coe¢ cient,implying that changes in innovation by inventors

of a certain ethnicity are correlated with changes in theextent of innovative activity in countries

associated with that ethnicity. The coe¢ cient on the logof ethnic U.S. patents is smaller and

only marginally signi…cant when the speci…cationconditions on the log of …rm patents, as in

column 2. A one standard deviation increase in the log ofethnic U.S. patents is associated with

a 0.6% increase in the likelihood of a foreign patent inthe region, a 14% relative increase from

the baseline probability.

The speci…cations in the next two columns attempt topinpoint more directly the mechanism

by which ethnic inventors facilitate the disintegrationof innovative activity across borders by

analyzing two distinct types of patenting. The dependentvariable used in column 3 is a dummy

equal to one if …rm f applies for at least onecollaborative patent in which at least one inventor

is based in a country associated with ethnicity e andanother listed inventor is located in the

U.S. The dependent variable used in column 4 is de…ned ina similar way, but it is equal to one

only if there are no U.S. based co-inventors for apatent.

13

One concern that could be raised about this analysis isthat …rms are not required to patent foreign innovations

in the U.S. The inclusion of parent …rm-ethnicity andethnicity-year …xed e¤ects alleviates this concern as the

…xed e¤ects control for any systematic di¤erences inpatenting propensities on either of these dimensions.

16

If U.S. based ethnic innovators support innovativeactivity abroad, one would expect to see a

larger coe¢ cient on the log of ethnic U.S. patents inthe speci…cation for collaborative patenting

presented in column 3 than in the speci…cation fornon-collaborative patents presented in column

4. The results indicate that this is the case. A onestandard deviation increase in the log of

ethnic U.S. patents is associated with a 0.8% increase inthe likelihood of a collaborative foreign

patent in the region, a 28% relative increase from thebaseline probability. This evidence on

collaborative patenting highlights one way that U.S.based innovative workers can spur changes

in foreign activity. Thus, the results of analysis offoreign patenting are consistent with the

…ndings on a¢ liate R&D activity. They suggest thatethnic inventors in the U.S. promote

innovation activity abroad. Furthermore, they indicatethat such foreign innovative activity

continues to require support of U.S. personnel.

3.3

Ethnic Innovations and A¢ liate Ownership Structure

If innovators of a certain ethnicity facilitate theexpansion of U.S. multinationals and innovative

activity in countries associated with that ethnicity,they might also be associated with distinct

ownership choices for new a¢ liates in those countries.Prior work on ownership structure high-

lights the role local partners play in providing U.S.multinationals valuable market information.

This bene…t of shared ownership is weighed against highercoordination costs when multinational

…rms make ownership choices. Innovators of a particularethnicity might make local partners

less valuable and allow multinationals to own largershares of new a¢ liates in countries associ-

ated with that ethnicity. Ethnic innovators can oftenprovide both codi…ed and tacit knowledge

that can substitute for the local expertise typicallyacquired through partnerships with local

…rms abroad. These informational advantages are oftencited as a key advantage that diasporas

confer.

Speci…cations that consider this possibility take thefollowing form:

OW Nayet =

ft

+

ln(EIfet) +

REST RICTyet +

aet:

(5)

The dependent variable measures the extent to which theparent owns the equity of a¢ liate a

17

located in country y associated with ethnicity e at timet. The analysis considers two measures

of ownership: a whole ownership dummy which is equal toone for a¢ liates that are wholly

owned by their parent and a majority ownership dummywhich is equal to one for a¢ liates

that are at least 50% owned by their parent. To isolatenew a¢ liates, the sample only includes

a¢ liates the …rst time they appear in the BEA benchmarksurveys, and a¢ liates that appeared

in the …rst survey in the sample, which occurred in 1982,are excluded. 80% of new a¢ liates are

wholly owned by their parents, and 92% of new a¢ liatesare majority owned. ln(EIfet) is the

log of the count of the number of patents the …rm appliesfor in the period before benchmark

year t for which the inventor is of ethnicity e. Severalcountries limit the ownership stake that

can be held by U.S. multinationals during the sampleperiod. Speci…cations include a measure

of these restrictions, REST RICT , to capture the impactof these restrictions and to compare

the relationship between restrictions and ownershipchoices with the relationship between ethnic

innovation and ownership choices. REST RICT is a dummybased on Shatz (2000), and it is

equal to one if both the acquisition and sectoral scoreare at least three in a particular country

and year. The speci…cation also includes parent-year …xede¤ects

14

The speci…cations are

linear probability models, and standard errors areclustered by parent-year.

The results of these speci…cations appear in Table 8. Thepositive and signi…cant coe¢ cient in

the …rst column implies that …rms that have moreinnovation performed in the U.S. by inventors

of a certain ethnicity are more likely to wholly, asopposed to partially, own new a¢ liates

in countries associated with that ethnicity. The secondcolumn also includes the ownership

restriction dummy, and it has a negative coe¢ cient,indicating that ownership restrictions limit

the use of whole ownership, as one might expect. Theresults in column 2 imply that a one

standard deviation decrease in ethnic innovation isassociated with a decrease in the use of whole

ownership that is about one half the size of the decreaseassociated with ownership restrictions.

The next two columns present a similar analysis where thedependent variable is a dummy

14

Previous speci…cations include parent-ethnicity andethnicity-year …xed e¤ects. There is not su¢ cient entry

within parent-ethnicities to identify e¤ects whenparent-ethnicity …xed e¤ects are included. If ethnicity-year …xed

e¤ects are included, there is little variation inownership restriction within ethnicity-years, yielding results that

do not allow for a comparison of the relationship betweenownership restrictions and ownership structure and

the relationship between ethnic innovation and ownershipstructure.

f t.

for the use of majority ownership. The results aresimilar, but, relative to ownership restrictions,

ethnic innovators appear to be more strongly associatedwith majority ownership decisions than

whole ownership decisions. Thus, ethnic innovators appearto allow U.S. multinationals to

serve countries without the assistance of a localpartner. Using ethnic innovators therefore

likely increases the ability of multinationals to enjoythe coordination bene…ts that come with

majority and whole ownership. These results support theview that high-skilled immigrants

possess knowledge and connections that aid …rms innavigating entry abroad.

4

Conclusion

This paper studies the e¤ects that immigrant scientistsand engineers have on the global activities

of the …rms that employ them. The analysis uses detaileddata on the names of inventors that

appear in patent applications to infer the ethnicity ofU.S. based innovators. This information

is used in conjunction with detailed data on a¢ liates ofU.S. multinationals.

Tests reveal that increases in the share of innovationperformed by inventors of a certain

ethnicity are associated with increases in the share of a¢liate activity in countries related to

that ethnicity. This result is stronger for …rms that aremore likely to value ethnic innovators;

more speci…cally, it is stronger when …rms are beginningto engage in innovative activity abroad,

and ethnic innovators could play a role in facilitatingcooperation between innovators working

in di¤erent locations and in identifying products andservices that could be developed further

to meet foreign demands. This result also holds in teststhat use a measure of ethnic innovation

that exhibits plausibly exogenous variation. This resultimplies that innovators of a particular

ethnicity enhance the competitiveness of U.S.multinational …rms in countries associated with

that ethnicity.

The data also illustrate that …rms with more innovativeactivity performed by inventors of

a certain ethnicity are more likely to conduct R&D incountries associated with that ethnicity.

Furthermore, they are more likely to collaborate withinventors located in such countries to

generate new patents. Recent literature points out that …rmsare increasingly breaking up

19

innovative activities across countries to perform di¤erentsteps in settings where they can be

performed most e¢ ciently. The …ndings in this papersuggest that ethnic innovators facilitate

this change in the manner in which innovation occurs.

Finally, tests show that U.S. multinational …rms relyless on joint venture partners when

forming new a¢ liates in countries that are home to the …rms’ethnicinnovators. Joint ventures

typically entail substantial coordination costs and aresubject to con‡icts over transfer pricing

issues and technology transfers. Ethnic innovators appearto make local partners less valuable

by providing insights about foreign markets that allowmultinationals to majority or wholly own

foreign a¢ liates.

Taken together, these results have implications forimmigration policies. Many debates about

immigration focus on the potentially deleterious impactof low wage immigrants on the domestic

workforce. However, this paper points out that immigrantswho are skilled enough to engage in

innovative activity generate bene…ts for …rms that areseeking to do business abroad. Immigrants

play a signi…cant role in science and engineeringcommunities in the U.S., so these kinds of e¤ects

deserve consideration.