?: The Impact of Internet Use on the Earnings of U.S. Workers
With the emergence of the Internet as a popular means of communication and information retrieval in the mid-1990s, policy makers and scholars became concerned about the “digital divide” — the emerging gulf between people with access to the Internet and those without. The literature on the digital divide has grown in size and sophistication: Whereas early work simply documented and tracked intergroup differences, more recent research attempts to explain such differences statistically, and has also explored digital inequality within the online population in extent and types of use, autonomy of use, and the effectiveness with which desired information can be retrieved (DiMaggio et al. 2004).
Much of this work is motivated by faith that access to the Internet and the ability to use it effectively is an important form of human capital that influences labor-market success. An early study of the digital divide warned that “the consequences to American society” of racial inequality in Internet access “are expected to be severe” and noted that “the Internet may provide for equal opportunity…but only for those with access” (Hoffman and Novak 1998: 390). A more recent paper makes a similar point: “The ‘Digital Divide’ may have serious economic consequences for disadvantaged minority groups as information technology skills become increasingly important in the labor market” (Fairlie 2004). Many policy makers share this faith. For example, the Statement of Findings for Illinois’s 2000 “Eliminate the Digital Divide” Act noted the existence of a “digital divide” and asserted as settled fact that citizens who have mastered and have access to “the tools of the new digital technology” had “benefited in the form of improved employment possibilities and a higher standard of life,” whereas those without access to and mastery of the technology “are increasingly constrained to marginal employment and a standard of living near the poverty level” (Illinois General Assembly 2000, Section I-5).
But although we have learned a lot about the nature and causes of inequality in access to and use of the Internet, we know surprisingly little about such inequality’s effects on individual mobility. To be sure, there are other reasons to worry about the digital divide: Internet use is becoming necessary for certain kinds of social and political participation and for access to some private markets and government services (Fountain 2001). Ultimately, however, the expectation that people without Internet access are disadvantaged in their pursuit of good jobs and adequate incomes is a central basis for concern about the digital divide, and therefore an important topic for research.
The digital divide is also significant for students of social stratification as an example of what many believe to be the increasingly important influence of technological access and knowhow on social inequality in an era in which rapid technological change has become the norm. Charles Tilly (2005: 118, 120), for example, asks “To what extent and how does unequal control over the production and distribution of knowledge generate or sustain” inequality? He contends that control over information, science, and “media for storage and transmission of capital, information and scientific-technical knowledge” are “newly prominent bundles of value-producing resources” that have displaced ownership of the material means of production as primary bases of intergroup inequality.
Limitations of Existing Research on the Effects of Technology Use on Earnings Research on organizations suggests that command of new technologies increases the power and centrality to the labor process of those who possess it. For example, Barley (1986) reported that the introduction of CT scanners in hospital radiology labs enhanced the status and autonomy of technicians trained to use them in their relations with senior radiologists, to whom the new methods were unfamiliar. Kapitzke (2000) found similar dynamics when computers were introduced into public schools.
Such studies have not determined whether such increments in power are converted into higher earnings, however. Indeed, sociologists who study inequality have rarely asked whether variation in access to or command of new technologies influences individual life chances. Economists have addressed this question more thoroughly and have found positive impacts of computer use on earnings (Krueger 1993). Very little economic research has addressed Internet use, however. Moreover, most economic studies of effects of technology use on earnings have exhibited two shortcomings. First, they usually have employed cross-sectional data. Second, they have assumed technology use influences income through a single mechanism – i.e. , that any nonspurious effects of technology use on income reflect increases in human capital and productivity.
Some economists have called for employing longitudinal data and using other means to counteract effects of reciprocity bias (DiNardo and Pischke 1997; Card and DiNardo 2002) inherent in (but not limited to) cross-sectional designs. The obvious problem is reciprocity bias: workers may adopt a new technology because they are better paid (and can therefore afford it) rather than being paid better because they use the technology. Cross-sectional studies are also vulnerable to three kinds of selectivity bias. First, employers may choose their highest-quality workers to implement new technologies. Thus earnings advantages that appear to be caused by the use of new technology may instead reflect unmeasured variation in human capital (Entorf and Kramarz 1997). Second, successful firms with slack resources may adopt new technologies sooner than their less successful competitors and pay their employees higher wages (Domes, Dunne and Troske 1997). Third, firms with skilled (and highly paid) workers can more easily implement technological changes requiring an educated work force than those with less well-trained employees (Acemoglu 2002), producing additional opportunities for spurious correlation between technology use and earnings.
The second problem with existing research is that economists have restricted their hypothesis-testing to a single mechanism: technology use increases human capital, which in turn boosts productivity, which in turn leads to higher wages. From a sociological perspective, this view is unnecessarily narrow: Earnings may be determined not only by productivity (correctly appraised) but also by efforts of groups or networks of workers to monopolize access to certain skills (monopolistic closure [Weber 1978: 336]), to use social ties to receive disproportionate access to desirable jobs (opportunity hoarding [Tilly 1998]), or to employ culturally embedded status cues to signal virtue and ability (cultural capital [Bourdieu 1986]). (Economists refer to such devices as “rent-seeking” but regard them as less central and ubiquitous features of labor markets than do most sociologists.)
Because of their preoccupation with earnings increases caused by workplace productivity enhancement, economists’ empirical efforts have focused almost exclusively on examining the impact on earnings of current technology use in the workplace . By contrast, we believe that an exclusive focus on the human-capital/productivity-enhancement mechanism produces three kinds of mischief. First, it leads one to neglect two other mechanisms by which workers may gain earnings advantages: social-capital/information-hoarding, i.e., the use of technology to gain privileged access to information about desirable jobs; and cultural-capital/signaling, i.e. the use of technology to signal positive qualities that the worker may or may not possess. Second, an exclusive emphasis on human-capital/productivity-enhancement leads analysts to rely exclusively on measures of technology use – current use at work – for which the potential for endogeneity related to employer decisions is greatest; and to neglect measures of technology use that are less likely to be affected by employers (for example, prior use or use outside the workplace), and which may affect earnings independently.1 Third, the focus on current Internet use neglects research indicating that experience leads to more effective use, which suggests that returns to current users should be higher for those with more accumulated experience (Eastin and LaRose 2000; Hargittai 2003). $$$$1000 For Free
Assessing confidently the impact of Internet use on earnings, then, requires that we:
(1) Go beyond cross-sectional analyses to examine the influence of technology use on earnings change over time;
(2) Control for as many individual differences that may be associated with both earnings and technology use as possible, including occupation and industry characteristics; and
(3) Distinguish between types of Internet use and include independent measures of Internet use at home and in the past, as well as measures of current Internet use on the job.2
We take the following steps to accomplish these goals:
1. Panel data. We exploit a fortuitous feature of the Current Population Survey (CPS) to produce a panel with two measures of both Internet use and earnings. The CPS has conducted periodic surveys of respondents’ use of communications technologies, as well as taking multiple measures of respondents’ incomes. CPS empanels respondents for a span of 16 months. Two of their periodic surveys of communications-technology use, in 2000 and 2001, captured several thousand employed respondents toward the beginning and end of their periods of empanelment. Thus it was possible to explore the impact of Internet use on earnings change over a thirteenmonth interval. To our knowledge this is the first study to exploit this feature to study the overtime effects of Internet use on earnings.
2. Controls for other factors affecting income. Including lagged wages in a wage-determination model helps to correct for selectivity bias, but other factors may influence both technology use and the rate at which wages rise. It is therefore important to include a variety of additional controls and to employ additional means of correcting for possible selectivity bias. The CPS sample’s large size enables us to explore differences in the effects of Internet use associated with industry and occupation and job-specific skill requirements, as well as educational attainment, union membership, gender, race and Hispanic ethnicity, marital status, age, and place and region of residence. We also employ propensity-score matching to address sample selection bias based on observable characteristics of Internet users and non-users; and change-score models to address selectivity on unobserved characteristics the effects of which are not incorporated in the lagged term.
3. Distinguishing among types of Internet use. Almost all economic accounts posit that technology-linked wage gains reflect enhanced productivity due to the use of the new technology at work. By contrast, we argue that Internet use may also contribute to earnings by enhancing access to labor-market information and by serving as a signal of status and/or competence. We use measures of Internet use from the 2000 and 2001 CPS Internet modules to divide our sample into groups of non-users, consistent users, adopters (2000 non-users who were users in 2001), and disadopters (Internet users in 2000 but not 2001). We also use the CPS to compare the impact on earnings, respectively, of Internet use at work and at home. The latter is less likely to be a product of firm-level decisions than Internet use at work, and therefore less likely to be a function of unmeasured employer characteristics that influence both workplace technology and wages. We believe that ours is the first earnings study to use separate measures of technology use at work and at home, and separate measures of Internet use at two points in time.
The CPS data offer substantial purchase on the relationship between Internet use and earnings for U.S. workers at the turn of the 21st century. We first look at the relative earnings gains of consistent Internet users, new adopters, and dis adopters (compared to never-users) between 2000 and 2001. Next we explore the effects of Internet use at home as compared to Internet use in the workplace. Finally, after testing several model specifications to examine the models’ robustness to differing assumptions, we evaluate the hypothesis that gains result from computer use per se rather than from Internet use. But first we discuss in more detail the mechanisms – human-capital/productivity-enhancement, social-capital/information-hoarding, and culturalcapital/signaling – that might lead us to expect, and enable us to explain, an association between Internet use and wages.
Explaining The Relationship Between Internet Use And Earnings
Why might we expect to find positive empirical associations between Internet use and earnings (and, more generally, between technology use and socioeconomic achievement)? Whereas most work in economics has focused on mechanisms that link technology use to worker productivity and thence to earnings (summarized below under the heading of “human capital/productivity enhancement”), we describe additional mechanisms that link technology use, respectively, to better labor-market information and social networks (“social capital/information-hoarding”) and to the worker’s ability to establish a positive face (Goffman 1955) before potential and actual employers (“cultural capital/signaling”). Skill Online We begin by anticipating an objection from Internet-savvy academic readers to our focus on longterm use and home use. Even if new employees have not used the Internet at home or in a previous job, can they not pick up necessary skills quickly? Finding information and communicating with other people on-line, after all, is not rocket science. This objection underestimates the strangeness of cyberspace to neophytes, the difficulty of mastering online search and communication skills for workers without previous experience, and the range of competencies that Internet use entails. New users must (1) understand graphic conventions prevalent in web design (for example, the difference between a list and a drop-down menu) and learn the cues that make it easy for experienced users to tell one from the other; (2) acquire a mental map of the Internet as a “space” across which one can “navigate,” and master the instrumentalities (hyperlinks, URLs, search engines) through which one can do so; (3) learn the basics of on-line search ( e.g., generating queries that are neither too broad nor too narrow, using Boolean operators to refine a search) (4) acquire information about the uses and reputations of major websites; (5) develop skill in distinguishing between trustworthy online information sources and amateurish or misleading sites; and (6) master the pragmatics of online communicative competence (e.g. knowing when it is appropriate to contact a stranger or participate in an online forum, the appropriate formality of address, appropriate message length and contents, use of abbreviations and emoticons, and so on) (Warschauer 2003; Van Dijk 2005, ch. 5).