OFFSHORING, SKILLS, AND EMPLOYMENT: EVIDENCE FROM INFORMATION TECHNOLOGY WORKERS, 1995-2006 Work in Progress: Comments Welcome

This study uses new employment micro data to evaluate how offshoring affects the employment of US-based IT workers. Our estimates suggest that workers providing services that do not require personal delivery are being displaced at firms that are offshoring. Offshoring firms appear to be retaining some of these workers and promoting them into new roles. However, workers that are displaced are less likely to be hired by firms that are offshoring, resulting in a net flow of these workers from firms that are offshoring to firms that are not. We discuss implications for workers, policy makers, and managers. The falling communication costs associated with information technologies have triggered important changes to the geographic organization of work. Among the most widely discussed of these has been the substitution of overseas labor for domestic labor, commonly referred to as “offshoring”. The term has been used by some to describe the movement of work to foreign affiliates by multinational firms, and by others to describe the outsourcing of work to third party vendors located overseas. In this study, we focus on the former definition, in keeping with the existing literature on offshoring and employment. Thus, we focus on how the expansion of affiliate employment abroad affects the employment of workers in the US, especially in skilled service industries. Our interest in this area is motivated by the recent upswing in the popularity of services offshoring, and accompanying concerns among the public, academic researchers, and policy makers about the future of American services jobs. While manufacturing jobs have long been subject to global competition, recent waves of services offshoring appear to be threatening the stability of high-wage, high-skill service jobs that were once considered safe from global competition. For instance, a number of recent press reports have linked offshoring to job displacement among US-based information technology workers and financial service workers (Amiti & Wei, 2005). Although current levels of offshoring appear to be small, recent analyses suggest that we may have so far only seen the tip of an “offshoring iceberg” (Blinder, 2006). Indeed, some of the more pessimistic estimates produced by recent studies suggest a substantial fraction of the US workforce will eventually feel job pressure from offshoring, either through displacement or increased wage competition (Blinder, 2007; Jensen & Kletzer, 2005). A potential employment shift of this magnitude may call for a reassessment of a wide array of government policies directed at trade, social insurance, worker training, regional development, and corporate tax policy. Evaluation of the potential benefits of these types of programs requires clear evidence on how offshoring impacts workers. However, there is currently disagreement not only about what jobs are at risk of being offshored, but also about whether offshoring ultimately leads to an expansion or contraction of domestic employment. For example, even if offshoring leads to the direct displacement of some workers, trade theorists argue that cheaper offshore labor inputs allow firms to expand production in other areas, ultimately creating new, better jobs to replace the ones that are lost (Baghwati, Panagariya, & Srinivasan, 2004). Furthermore, studies using skill-based categorizations of workers have generated a wide range of estimates of how many jobs are at risk of being offshored, making it difficult to understand the potential impact of services offshoring. For instance, recent estimates have ranged from essentially zero net job losses (Global Insight, 2004) to as much as 40% of US jobs at risk of being offshored (Blinder, 2007; Jensen & Kletzer, 2005). Thus, the implementation of effective policy measures requires developing a better understanding of the extent to which offshoring is displacing jobs, which workers are at risk of displacement, how offshoring-related displacement affects workers who are displaced, and how other workers are indirectly affected by offshoring. Despite the need for empirical guidance in this area, the evidence relating offshoring to employment is slim. This situation exists primarily because of an absence of nationally representative data describing the offshoring of services. Unlike manufacturing, where industry level import figures are collected, there is no definitive source of data on service imports by category. Therefore, changes in levels of service imports cannot be tied to changes in occupational demand. Most existing research in this area, primarily focused on skill-biased change and wage inequality, has related the offshoring of manufacturing to wages and employment at the industry level. However, services offshoring appears to be interacting with skills in different ways than earlier generations of technical change. Jobs that would traditionally be considered “high skill” such as programmers or financial analysis are readily offshored while less skilled service professions such as sales support have proven relatively safe. Thus, findings from manufacturing may not transfer to service occupations. Even with better access to data, however, there are still considerable challenges in producing accurate estimates of the employment effects of offshoring using industry data. Total labor offshoring is small compared to normal levels of job churn so any offshoring-related changes may be hard to detect. Some displaced workers may move within industry, making their displacements invisible in industry data. Finally, the actual decisions to offshore occur at the firm level and there may be substantial heterogeneity among firms within the same industry. Perhaps because of these difficulties, aggregate-level (industry or country) studies have produced conflicting findings about whether offshoring leads to a contraction or expansion of employment 1 Indeed, a report issued by the Government Accountability Office in 2004 was entitled “Current Government Data Provide Limited Insight Into Offshoring of Services” (Government Accountability Office, 2004) (Amiti & Wei, 2005; Desai, Foley, & Hines, 2005; Riker & Brainard, 1997). Thus, there may be considerable advantage to finding novel empirical approaches that can be used to estimate these changes at the worker and firm level. In this study, we use new sources of microeconomic data to address how services offshoring is affecting the employment of US workers. Our framework treats the decision of individual firms to offshore as exogenous, and empirically examines the relationship between firm level offshoring and worker outcomes. Our principal model relates individual worker-level employment outcomes (displacement, promotion, hiring) to individual worker characteristics, such as skills and prior job market behavior, and to various employer characteristics, including the extent of firm offshoring. We introduce two new data sources that address limitations in prior offshoring research. First, we utilize an employee-level data set on the overseas employment of information technology workers to create detailed measures of services offshoring activity. By aggregating these data to the firm level, we can estimate the extent to which different firms are offshoring information technology services work. These data, however, are not sufficiently comprehensive to track employment outcomes in the broader US labor market. Therefore, we utilize a second data set, a very large (>10 million workers), nationally representative sample of US based workers, which includes information about employee characteristics and job histories. These data enable us to examine worker flows among firms and measure the employment outcomes of individual workers. Moreover, these data provide information about workers’ occupational categories and job titles which can be linked to data on the skill composition of different occupations (through the O*Net database). Together, these two data sets, combined with other public sources of firm-level data, enable us to connect firm-level offshoring decisions to the employment outcomes of US IT workers. We restrict our analyses to the employment of IT workers because they represent the class of workers most affected by offshoring to date, and are well represented in our offshoring data set. However, lessons learned from these workers are also likely to apply to other skilled service workers, such as financial analysts, who are beginning to face significant offshore competition (Deloitte, 2007). A key part of our analysis focuses on how the relationship between services offshoring and US employment is affected by employee skills. To explore the specific role of skills, we map job titles to a skills index that categorizes jobs according to how important it is for the services performed by the worker to be personally delivered (Blinder, 2007). Our findings suggest a division in how the employment prospects of IT workers are affected by offshoring. Offshoring increases displacement among IT workers, but only among those who provide services that do not require personal delivery. For workers that are not displaced, it increases the frequency of promotion, consistent with arguments that by lowering the costs of certain skills, offshoring increases demand for other higher-value jobs (Baily & Lawrence, 2004; Kierkagaard, 2004). We also find that displaced workers are not being hired into other firms that offshore, but instead, appear to be moving to firms that are not yet aggressively offshoring. Thus, our analysis suggests three key findings: 1) that the adverse effects of offshoring are experienced by workers that perform tasks that can be electronically delivered, 2) that workers that are retained by firms may actually benefit from offshoring, and 3) the adverse impact of offshoring, thus far, has been tempered by a shift of vulnerable workers from firms that offshore to those that have not. In the next section, we describe the theoretical perspectives that we use to guide our analysis, our hypotheses, and the limitations of existing empirical work. Section 3 describes the data and methods we use to investigate the effects of offshoring on IT workers’ employment outcomes. Section 4 describes our results. The final section includes a discussion of our findings, and addresses their implications for workers, policy makers, and managers. THEORETICAL BACKGROUND AND HYPOTHESES Firms and the Use of Offshore Labor As part of a multi-stage process, firms choose both whether to offshore, and what work should be sent offshore (Moxon, 1975). The decision to offshore depends on the perceived costs and benefits. Although labor cost savings can often be realized by hiring offshore employees, there are likely to be significant offsetting costs associated with establishing new offshore operations, such as new capital investments and the costs of transferring knowledge to new workers. Furthermore, the availability of skilled labor pools, tariffs, materials shipping costs, political risk and other factors may shape managers’ thinking about whether or not to send work offshore (Moxon, 1975). Additional operational costs may be incurred by introducing geographic and cultural divisions among employees. For instance, having workers spread across different continents can make it difficult for employees to build trust with one another, an important factor in the execution of projects where tacit communication is important for project success (Jarvenpaa & Leidner, 1999). Agency problems can also be magnified because remote employees are costly to monitor (Roth & O’Donnell, 1996). Additionally, tasks that are amenable to offshoring are often interwoven with other tasks that must be kept in a particular location, making it difficult to offshore specific tasks without disrupting existing processes. Thus, firms seeking to move work offshore face difficult operational and organizational decisions when determining what work can and should be moved offshore, and are likely to establish offshore operations only if they can recoup costs associated with the move. Recent technological innovations, however, have significantly altered incentives for firms to establish offshore operations. For example, the adoption of information technologies has dramatically decreased the costs of shipping information-based work to overseas locations, and has been empirically linked to offshoring (Abramovsky & Griffith, 2006). Information technologies have also lowered the costs of monitoring and coordinating remote employees, reducing agency problems faced by firms that have a geographically dispersed workforce (Argyres, 1999). Computer networks can also enable the facilitation of communications among diverse project teams, making it easier to coordinate far flung activities (Montoya-Weiss, Massey, & Song, 2001). Furthermore, the business process reengineering efforts that have been associated with the integration of computers into the workplace have forced managers to consider how business processes can be reorganized to take advantage of offshore labor costs (Dossani & Kenney, 2003; Hammer & Champy, 1993). Indeed, in modern, globally competitive organizations, managers are charged with understanding how work can be most effectively reorganized to maximize the value jointly derived from domestic labor, information technologies and offshore labor. Thus, while offshoring is not new, press reports suggest that recent technological, organizational, and managerial changes have dramatically accelerated the movement of service processes and information services offshore. For the purposes of our analysis, it is critical that a significant number of firms engage in offshoring and that the determinants of offshoring vary by firm in a way that is independent of the identity of the individual workers they employ. The incentives to offshore may vary among firms according to the skill distribution within the firm. Firms that employ a large number of high-wage computer programmers, for example, may have greater incentives to offshore than other firms because of the availability of foreign, skilled labor and potential cost savings. The fixed costs of offshoring may also differ among firms. For example, multinational firms that already have overseas operations in developing countries may have capacity that can be used to offshore additional work. Managerial preferences and the perceived risks associated with offshoring may also influence the decision to offshore. Firms that work with sensitive data may be more reluctant to send work to countries where intellectual property laws are less well defined. However, in this study, we do not attempt to model the outsourcing decision at the firm-level directly, instead using the resulting variation across firms to identify the impact of offshoring on individual workers. The Employment of Workers in the U.S. The expansion of offshore employment has raised concerns that US firms are using offshore workers to substitute for US workers. In particular, much of the interest in this area has centered on the fact that services offshoring appears to be affecting high-skill jobs, which appear to be at equal or greater risk of being offshored than the low-skill jobs that have traditionally been affected by technological change. Theoretically, workers are at risk of having their job offshored if the costs of doing so are outweighed by the potential benefits to the firm. Costminimizing employers will choose to offshore skills for which the gains are positive, where gains can be modeled as the difference between local wages and the combination of overseas wages and overseas remote coordination costs. The wages paid by the firm to local labor are location and skill specific, overseas wages are skill specific, and the costs of remote coordination are skill specific. For example, transactions costs associated with moving management jobs overseas may be higher because of the extensive tacit communication required for such jobs. When the costs of remote coordination are sufficiently high, the net gains from moving a job offshore will be negative. Therefore, holding constant the state of technology, offshoring should affect some jobs but not others. Because offshoring is likely to increase the risk of displacement among some workers, while having no effect on the displacement rate of others, we hypothesize: H1. IT workers are more likely to be displaced from their jobs at firms that offshore. An emerging literature concerned about the possible extent of offshoring focuses on categorizing occupations according to the risk that they face from offshoring. Researchers have suggested that services that can be delivered from afar, such as call-center services and computer programming, can be easily offshored, while services that require personal interaction, such as waitressing, nursing, or hairdressing, are difficult to send overseas. Bardhan and Kroll use a number of occupational attributes, such as the need for face-to-face interaction and the degree of information content, to determine which jobs are vulnerable to being offshored. They find that about 11 million jobs may potentially be offshored (Bardhan & Kroll, 2003). Jensen and Kletzer examine the geographic distribution of activities, reasoning that services that are currently being provided over state boundaries can be provided just as easily over international boundaries. Using these methods, they estimate that about 38% of total US employment falls into the tradable service category (Jensen & Kletzer, 2005). Blinder uses a subjectively derived index based on job task and content indicators from the O*Net database, and characterizes the offshorability of skills depending on whether the skill must be personally delivered (Blinder, 2007). Using this index, he finds that 30-40% of US jobs are vulnerable to being offshored. In this study, we use the classification advanced by Blinder, categorizing workers according to whether the services they provide must be personally delivered. Examples of some IT related occupations and their assigned indices are shown in TABLE 1. Skills at the upper end of the spectrum, closer to 100, are more likely to be offshored, while those at the lower end of the spectrum, closer to 25, are unlikely to be offshored. One difficulty with the use of this index is that it is an ordinal scale, ranking the relative vulnerability of different jobs to offshore competition. Blinder draws the line between offshorable and non-offshorable workers at an index value of 50. However, he is interested in potentially offshorable jobs, rather than jobs that are already being offshored, so many jobs that he deems potentially offshorable will not yet be affected by offshoring. Furthermore, because our chosen industry, information technology workers, is primarily an information services industry, most variation occurs among the degree of worker offshorability, not whether or not a worker is potentially offshorable at all. Therefore, in this study, we divide workers into those who have an offshorability index of greater than or equal to 90, and those that have an index less than 90. Programmers and graphic designers fall into the first category, while sales personnel, managers, and software engineers tend to fall into the second category. We choose a value of 90 because press reports suggest that to date, programmers, analysts, and computer operators have been the occupations that have been affected by offshoring. ------------------------------Insert Table 1 about here ------------------------------Classification of jobs according to the risk of being offshored allows us to formulate hypotheses relating skills to offshoring and employment. All other things being equal, firms will achieve the largest gains by offshoring the production of services that do not require personal delivery. Therefore, we hypothesize: H2. IT workers with skills that do not require personal delivery are more likely to be displaced from their jobs at firms that offshore. Job displacement may not imply a separation between worker and firm. Firms may choose to move displaced workers into new positions. For instance, firms may retain some highperforming programmers who are familiar with the firm’s systems, and move them into project management positions if their skills are no longer needed in old positions. Therefore, we hypothesize: H3. Conditional on remaining with the firm, IT workers with skills that do not require personal delivery are more likely to be promoted at firms that offshore. Furthermore, a reduction in the demand for these skills will reduce the rate of hiring of these skills among firms that offshore. We therefore hypothesize: H4. US-based IT workers with skills that do not require personal delivery are less likely to be hired at firms that offshore. Employment predictions based solely on direct job loss estimates, however, ignore the indirect impacts of offshoring on employment. Economic theory suggests that as the price of easily offshored skills falls, firms will invest in greater quantities of these skills, and complementarities between these and other skills will raise demand for non-offshorable skills. For example, as firms hire a greater number of lower wage computer programmers abroad, more managers, sales agents, and designers will be required by firms. If these positions are easier to fill locally, then offshoring will create new domestic jobs. Indeed, many trade theorists feel that ultimately, offshoring will ultimately create new, higher-value jobs in the economy to replace jobs lost to offshoring (Baghwati et al, 2004; Baily & Lawrence, 2004; Kirkegaard, 2004). If offshoring increases demand for skills that are difficult to offshore beyond what is available in the firm, firms may choose to fill this demand by hiring these skills from external labor markets. The net impact on hiring, however, will be determined by which skills firms ultimately find profitable to offshore. While managerial skills may be relatively more difficult to offshore, firms may still offshore these positions if it generates cost efficiencies. And while firm-specific knowledge may buffer some existing workers from displacement, the hiring of new managers may be immediately affected if the labor savings outweigh the transaction costs associated with remote coordination. Therefore, if firms find it profitable to offshore workers with skills that require personal delivery, the hiring of these types of workers may also be reduced along with the hiring of workers with skills that do not require personal delivery. Thus, the effects of offshoring on overall hiring and employment are ultimately empirical questions. Limitations of Existing Empirical Work Despite the attention paid to offshoring by the media, evidence on how offshoring affects employment has been limited because of difficulties in obtaining reliable, fine-grained data describing services trade (Government Accountability Office, 2004). The predominant approach in the literature has been to relate imports of goods and services by firms or industries to changes in employment, at the country, industry, or firm level. Feenstra and Hanson used this approach for manufacturing industries in the 1990’s, and found that imports of materials were linked to an increase in the demand for non-production workers relative to production workers (Feenstra & Hanson, 1996, 1999). Studies conducted at the country level found that foreign direct investment is associated with an increase in domestic employment (Amiti & Wei, 2005; Desai, Foley, & Hines, 2005). The studies closest to ours use firm-level survey data available from the Bureau of Economic Analysis (BEA). Studies based on this data have found weak evidence that the expansion of offshore employment substitutes for domestic jobs (Brainard & Riker, 2001; Riker & Brainard, 1997), and that these effects are moderated by whether offshore workers are employed in high-wage or low-wage countries (Harrison, McMillan, & Null, 2007). Some authors, interested in how globalization affects the wages of low-skill workers, have focused on how offshoring affects the relative demand for skilled labor (Falk & Koebel, 2002; Hijzen, Gorg, & Hine, 2005; Strauss-Kahn, 2003; Wood, 1998). However, most of these studies focus on offshore manufacturing. Although these studies make important contributions to our understanding of the relationship between offshoring and employment, they have a number of significant limitations. As discussed in the introduction, offshoring is small compared to the normal movement of labor among firms, and movements among firms in the same industry may be lost entirely using industry data. Second, these studies cannot separately identify the dynamics of workers with different skill sets, which is critical since not all skills are equally affected, and the value of some skills may actually be enhanced by offshoring. While some studies have focused on the demand for skilled labor, studies focused on education-based skill measures may not be particularly relevant. For example, Blinder reports that the correlation between the offshorability index and educational attainment associated with each occupation in his data set .08, which indicates 1) that offshorability is not closely correlated with education, and 2) to the extent that they are correlated, increased education is associated with more not less, risk of being offshored (2007). Thus, traditional measures of education and skill may not be sufficient to identify the effects of services offshoring. If offshoring is leading to the displacement of workers with some skills while increasing the demand for workers with other skills, these dynamics will be difficult to separate in aggregate employment data. Our data, describing services offshoring activity at the firm-level and service worker outcomes at the individual level, allow us to overcome many of these limitations. Because both offshoring and employment decisions are made at the firm-level, having data describing economic activity at these levels makes it easier to frame testable hypotheses relating offshoring to worker outcomes. Firm-level identifiers also allow us to control for other firm-level factors that may bias our estimates. Furthermore, the human capital information available in the worker data provides significant advantages. A challenge in testing the effect of particular skills on displacement is that workers endowed with these skills may also be rich in other attributes that affect employment, such as experience. If these attributes are omitted, correlations between skills and displacement may erroneously reflect the effects of other attributes. Similarly, measures of offshorability based on the importance of communication or personal proximity may reflect managerial experience, which has been empirically linked with displacement in technology intensive firms (Cappelli, 1991; Osterman, 1986). It may be necessary, therefore, to control for these characteristics in order to identify whether workers with certain skills have truly suffered disproportionate levels of displacement. Some preliminary evidence related to these hypotheses can be found in the administrative occupational employment data, published annually by the Bureau of Labor Statistics (BLS). TABLE 2 shows the offshorability of different computer occupations alongside employment shifts from 2001 to 2006. Employment in categories with a high offshoring index has fallen significantly, while employment in job categories that are associated either with hands-on analysis or personal interaction, such as database administrators, and network analysts, has risen. While this is consistent with our hypotheses, evidence from aggregate employment data is not sufficient to identify the effects of offshoring on employment because other trends relating to new technologies, organizational changes, and worker attributes may also be influencing aggregate employment numbers. Furthermore, because the aggregate employment data does not enable us to track individuals over time, we cannot observe whether, for example, computer programmers are being displaced from their jobs, or instead, whether retiring programmers are being replaced by workers in new job categories. It is important to test, therefore, whether the connection between job displacement and offshoring persists after controlling for other employer and individual characteristics. ------------------------------Insert Table 2 about here ------------------------------METHODS Data In this study, we use data from unique sources, electronic resume databases. Employee resumes contain a wealth of information about employees, including positions that they have held, as well as information on employer names, job titles, dates of entry and exit, and additional information on education, skills, and experience (an example of some typical entries in our larger database appear in TABLE 3). In large samples, resume data can be used to generate matched employer-employee data, where firm names can be matched to other public firm databases and job titles can be matched to other public databases on jobs and skills. Although employeremployee data has been used before (e.g. the Census Bureau LEHD) our dataset is unique in that it has occupational category and employee job titles which do not appear in any prior dataset. We use data of this type, albeit from two different sources, to construct measures of offshoring activity and firm-level labor flows. ------------------------------Insert Table 3 about here ------------------------------To measure the movements of workers among US firms, we take advantage of micro data on the employment histories of a nationally representative sample of US-based workers. The data 2 One disadvantage of our data is that we do not have comprehensive pay data which limits our ability to examine the effect of offshoring on wages. Thus, our focus will be on labor quantity and the skill distribution of labor quantity rather than the impact of offshoring on wages. come through a research partnership with a leading online job search site, and include complete resume information for about ten million workers, including information on job history, employer name, dates of employment, and job title for all prior spells of employment. In addition, we have a number of human capital variables for each worker, including education and experience. Using these data, we can observe when employees join and leave firms, and through firm-level aggregation, can observe the flow of employees into and out of the firm as well as the characteristics of these employees. While other data exist that describe worker flows, this is the first data set of which we are aware that includes skill and occupational data, important for the questions we address in this study. From these data, we extracted about 500,000 workers who appeared in the data set between 1995 and 2006 and identified themselves as information technology workers. Summary statistics for these workers are shown in TABLE 4. From 1995 to 2006, we have slightly less than 1,900,000 information technology worker-year observations. As expected for information technology workers, the population is highly educated, with the largest group of workers having at least a four year college degree. The average job tenure for a worker in our sample is about four years. In column (2), we include comparable statistics, where available, from the information technology workers in the 2006 Current Population Survey (CPS), a sample of randomly drawn workers from the larger US population. The education distribution is generally similar to the distribution in our data. The average job tenure in the CPS sample is slightly lower than the average job tenure in our sample, reflecting the tendency of the workers in our sample to be “job-hoppers”. ------------------------------Insert Table 4 about here ------------------------------Indeed, one shortcoming of this type of data source is that it may be associated with sampling issues. For example, on a job-search site, more technologically savvy workers from knowledge-intensive firms may be more likely to participate than others. However, because we focus on information technology workers in this study, these problems should not significantly influence our sample. In addition, the particular service that provided the data also has a significant presence in the offline recruitment advertising market making the worker population more likely to be broadly representative than a typical online job site. Furthermore, we control for education, time in the labor market, managerial experience, and a variety of other human capital variables to adjust for differences among workers. A second concern is that “jobhoppers” may be more likely to post their resumes on these types of sites than workers looking for long term employment. However, we are interested primarily in worker transitions, and to the extent that job-hoppers are responsible for a disproportionately large fraction of worker transitions, this should not adversely affect the reliability our results. To ensure that our estimates on skills are not reflecting the intrinsic job-hopping tendencies of our workers, we include controls in our analyses for an employee’s job-hopping propensity, measured as the average job tenure at all other jobs that the employee has held. Our measures of firm offshoring come from a similar, but separate data source that focuses on career networking rather than job search. The offshoring data were obtained in late 2006 from a leading online networking service through which about ten million service workers post employment information, including primary industry affiliation and geographic location, as well as information for each professional position that they have held, employer name, job title, years spent at the firm, and for public companies, a ticker symbol. This source contains data similar to the type described above, but it differs in that the participants are seeking to network rather than find new jobs. Furthermore, while the first data set contains only US workers, the second is rich in international workers, especially in the information technology domain. From the full set of site participants, we obtained a random sample of about one million workers. Between 1995 and 2006, these data contains about 22,000 offshore IT worker-year observations. The availability of data on offshore IT workers and their employers allows us to develop measures of how many overseas IT workers are being employed by different US companies. As with the employee flow data, the nature of the offshoring data source raises some sampling considerations. Within the bounds of sampling error, the networking data would accurately reflect the distribution of offshore IT employment if the workers that participated in the network were a random sample of all offshore IT workers. However, if these sites are more popular in some firms than in others, or more popular with some groups of workers than others, than the offshoring measures that we construct from these data will include error. There is currently little theoretical guidance on how online network participation might differ among firms. However, it is reasonable to expect that biases that exist among workers at international firms in the sample would also exist among US based employees that participate on the site. Therefore, we compare firm-level counts of the US based IT workers in the networking data with measures of US firm level IS employment from the job search data described above. While, selection issues in the job search data may influence worker counts, these sources of error should be uncorrelated with sources of error in the networking data. The correlation coefficient between the two sets of measures is 0.57, and a Spearman test firmly rejects the hypothesis that the two measures are independent (p<0.00). We believe therefore, that the IT worker counts from the networking measures are a valid measure of the number of IT workers employed by firms, both 3 Further details of the data provider cannot be disclosed as the firm has chosen to remain anonymous. domestically and abroad. Finally, it is important to note that any error included in our offshoring measures will serve only to understate our regression estimates of the effects of offshoring. A potentially more serious problem is that the available evidence on social network sites suggests that there is significant variation in the usage of different networks across countries, so our measures of the number of offshore IT workers may be confounded with firms’ geographic choices. For example, if a firm sends much of its offshore work to a country in which the network service is not widely used, than our measures may understate the number of IT workers offshored by that firm. However, we can bound some of these effects because we have information on the countries in which these offshored IT workers are employed. The countries with the largest numbers of offshore IT workers in our sample, shown in TABLE 5, are led by France, Germany, and India. While most of the media attention has been focused on India, our data is consistent with IMF Balance of Payments statistics that detail the primary sources of business services imports, shown in Column (2). The absence of the United Kingdom and Netherlands from the IT worker data set is most likely because these countries provide business services (e.g., call center operations), but are not major information technology service providers. Indeed, as shown in (4), the Netherlands and the UK are two of the largest computer service importers in the world. Finally, the higher position of France than Germany in our data may have been influenced by the site popularity in each country, shown in Column (5). In aggregate, therefore, the data appears to provide information on offshore workers from most of the world’s largest software export centers. ------------------------------Insert Table 5 about here ------------------------------4 Available at http://www.comscore.com/press/release.asp?press=1555. Measures Displacement, Promotion, and Hiring. Our primary dependent variables measure whether an employee has been displaced from a firm, has been promoted within a firm, or has joined a firm. Measures of these events are extracted from individual worker employment histories. The data describe whether, in any given year, an employee has joined a new employer, has left an old employer, or has been promoted by the current employer. We code a given worker-year as 1 if the worker has joined, left, or been promoted in the firm in a particular year, and 0 otherwise. Combinations of these events can occur within the same year if, for instance, an employee leaves a firm and joins another within the same year. As with other work in this area, individual turnover can include both voluntary and involuntary separations, so our measures will include error to the extent that we are primarily interested in the effects of offshoring on involuntary separations. However, this should result in understatement of our estimates of the effects of offshoring on displacement. Promotions are measured as events where employees report shifting to new positions within the same firm. Therefore, our promotions measure may include error in cases where promotions actually reflect lateral moves or demotions. Offshoring. Our primary independent variable of interest, offshoring, is a firm-level construct measured as the logarithm of the number of offshore IT workers employed by the worker’s employer. To develop measures of offshore IT employment, we extract individuals from the networking data who are associated with computer services based industries, are located outside the US, and are employed by public US firms. US firms are defined as those that have headquarters located in the United States, where firms’ headquarter locations are taken from the 5 Specifically individuals who identify “Information Technology and Services”, “Computer Software”, “Internet”, “Computer Networking”, “Computer and Network Security”, “Computer Hardware”, “Telecommunications”, or “Semiconductors” as their primary industry affiliation. Compustat Database. We use the sample distribution of offshore IT workers across all firm-years to infer the population distribution of offshore IT employees across all firm-years. The data is used to estimate both differences in offshoring intensity across firms, and differences in offshoring intensity across time within the same firm. Skills Index. The theoretical discussion above suggests that employees providing services that need to be personally delivered are less vulnerable to offshoring. To construct measures of how vulnerable a worker’s job is to being offshored, we match workers’ job titles to standard occupational codes in the O*Net database by hand-mapping titles that appear at least fifty times in our data set against O*Net job titles. Then, to relate occupational codes to offshoring vulnerability, we use the offshorability index developed by Blinder (2007). As discussed above, Blinder uses the O*Net database in addition to subjective assessment to rank different occupational codes on a scale from 25 to 100 according to their vulnerability to offshoring trends, based on whether or not they can be electronically delivered with little or no degradation. TABLE 1 reports mappings between some common information technology related occupations and their skills measure. It is important to note that even within the information technology industry, there is considerable variation in the offshorability of jobs. Control Variables. We include a number of individual and firm-level variables to control for factors that may influence our estimates of offshoring and skills. Although the literature has provided support for a number of employee attributes that influence the employee turnover decision, our particular interests are on factors that contribute to employee displacement and simultaneously influence our offshoring estimates. Therefore, we focus on firm-level attributes that may be correlated with the employer’s offshoring decision. We also include individual-level 6 http://online.onetcenter.org human capital measures that may be correlated with our measure of skills, such as education and experience, to ensure that our estimates of the effects of skills on employment outcomes are not reflecting correlations with other worker attributes. However, we omit other employee level constructs, such as organizational commitment, employee motivation, number of dependents, job embeddedness and alternative job offers, which have been shown to influence the employee turnover decision but are unlikely to be correlated with the employer’s offshoring decision. Our choice of individual level control variables is based on a review of the literature on individual level determinants of job displacement and turnover (e.g. Cotton & Tuttle, 1986, review an extensive literature on employee turnover). At the individual level, a number of factors correlated with job classification may influence the probability of displacement. We include measures of education coded as one of the five categories shown in TABLE 4, as well as number of years of overall related experience, self-reported by workers in the employment data. We also include measures of job tenure, measured as the number of years that the employee has been with the firm. We include measures of managerial experience, a binary variable indicating whether or not the worker has had experience managing other workers. Finally, worker mobility, which can differ across occupations, has been shown to be strongly related to the employee’s previous job change frequency (Farber, 1994). Therefore, we include average job tenure into the model to control for an employee’s intrinsic tendency to job-hop. Average job tenure is computed by averaging the length of employment of all work spells for each worker, excluding the worker’s current job. The data also include some earnings information. For each worker, 2006 wages are reported. Therefore, we include measures of workers’ wages, converting hourly wages to an annual basis and including a dummy variable for whether the reported wages are annual or hourly wages. The inclusion of 2006 wages rather than matched year wages will add some error into our analysis, but is a measure of wage classification of the employee relative to other employees. Firm-level controls are obtained by matching the employer to external data sets, and were chosen through a review of the literature relating worker turnover to firm-level constructs (e.g., Huselid, 1995). Most measures are taken from the Compustat database. We included firm size, measured as the logarithm of the number of employees, because larger firms may be more likely to offshore and will have more developed internal labor markets. Both offshoring and the rate at which employees leave a firm may also be influenced by the health of the firm. Successful firms should experience relatively less frequent employee turnover than firms that are in financial distress. Therefore, we include measures of sales growth, measured as the year-on-year difference in sales, normalized by total sales. We also include measures of sales concentration, measured as the market-share based Herfindahl of the four digit SIC industry, capital intensity, measured as the log of the ratio of property, plant, and equipment (PPE) to employees, and R&D intensity, measured as R&D investment normalized by sales. We also include measures of industry union coverage taken from the Hirsch and Macpherson data set (Hirsch & Macpherson, 2003). Finally, to control for the fact that multinational firms will have more offshore workers to satisfy foreign markets, we included measures of foreign income. Descriptive Statistics and Analytic Approach TABLE 6 shows the means and standard deviations for all variables in our analysis. The number of worker-year observations is reduced to 66,633 after removing observations for which data on individual and firm-level attributes is missing, and after keeping only observations for which the job titles fell into our list of most common occupations and therefore could be mapped to a standard occupational code. The average worker in our sample is employed at a firm that has just over thirty-two IT workers employed offshore. The average skills index of the workers in our sample is just over 74 out of 100. ------------------------------Insert Table 6 about here ------------------------------TABLE 7 shows the correlations between our primary variables. Our primary independent variable, number of offshore workers, is positively correlated with firm size and foreign income, and negatively correlated with industry union coverage. However, these correlations are not strong enough to raise multicollinearity concerns. Only wages appear to be significantly correlated with the skill index, perhaps because management positions have lower offshorability values. FIGURE 1 illustrates how the occupational distribution in our data translates to the distribution across the skills index, where a higher number indicates a job that is more vulnerable to being offshored. The dispersion in the distribution indicates that the different job titles that appear in our data provide sufficient variation to link skill differences to differences in outcomes. ------------------------------Insert Table 7 about here ------------------------------------------------------------Insert Figure 1 about here ------------------------------7 The drop in sample size is due primarily to missing data on R&D investment and foreign income. SEC regulations mandate that foreign income must be reported by firms only if it comprises at least 10% of their total income. To ensure that sample selection problems caused by missing foreign income data did not bias our estimates, we ran comparison models setting foreign income to zero where it is unreported (for a similar approach, see Foley, Titman, & Twite, 2007). Although we report only the results from the raw data in this paper, both sets of analyses produced similar estimates. To relate job displacement to offshoring and worker skills, we embed employers’ offshoring intensity and control variables into models of individual worker displacement. To test the role of skills. We use bivariate probit equations in which the dependent variable is 1 if the worker is displaced. To examine the role of skills, we compare estimates across regressions performed on samples of workers, categorized by skill type. We use a similar approach to test how offshoring affects the rate of promotion, as well as the rate of new hires within a firm. Our approach is consistent with a methodological literature recommending the use of probit models to examine dichotomous worker employment outcome variables (Huselid & Day, 1991; Winship & Mare, 1984). Because time trends may influence both the strength of external labor markets as well as the costs of offshoring, we include year dummy variables to control for time trends in the data. We also include industry dummy variables to account for industry-specific economic shocks, because worker mobility and offshoring may vary systematically between industries. In some models, we include firm effects to eliminate the effects of unobserved firm heterogeneity. Finally, in all analyses, error terms will be correlated across firms. We account for this correlation by clustering errors on firm using the Huber-White robust (clustered) standard errors for panel data models.