The Causal Effects of Education on Technology Adoption: Evidence ...

43 downloads 2172 Views 160KB Size Report
Oct 28, 2008 - between the level of education and the rate of technology adoption, .... from two different workplaces, then the person will be counted as two ...
The Causal Effects of Education on Technology Adoption: Evidence from the Canadian Workplace and Employee Survey ∗

W. Craig Riddell [email protected] University of British Columbia

Xueda Song [email protected] York University

October 28, 2008

Preliminary: Not for Citation



We thank both the British Columbia Interuniversity Research Data Centre and Toronto Region-Statistics Canada Research Data Centre for access to the data. In particular, we thank Yves Decady and Lee Grenon at Statistics Canada for excellent statistical and data support.

Abstract Adoption of innovations by firms and workers is an important part of the process of technological change. Many prior studies find that highly educated workers tend to adopt new technologies faster than those with less education. Such a positive correlation between the level of education and the rate of technology adoption, however, does not necessarily reflect the true causal effect of education on technology adoption. Relying on data from the Workplace and Employee Survey (WES) (1999-2004), this study assesses the causal effects of education on technology adoption by using instrumental variables for schooling derived from Canadian compulsory school attendance laws. WES is an employer-employee linked panel data set, which provides rich information on firms and workers, including not only information on computer use, but also general information on technology adoption. We find that education increases the probability of using computer in the job. We also find that employees with more education possess longer work experiences in using computer, and are more likely to experience upgrade in computer-controlled or computer-assisted technology and experience upgrade in technological device than those with less education. Findings from this study not only shed light on the causal relationships between education and technology adoption, but also contribute to the growing literature on the private and social benefits of education.

2

1

Introduction

In a “knowledge-based economy” such as the modern OECD (Organization for Economic Co-Operation and Development) economies, information, technology, and learning have been playing an increasingly important role, and the creation and diffusion of new and more advanced knowledge and technologies are now recognized as a major drive for productivity and economic growth. The adoption of new technologies by firms and workers constitutes an important part of the process of technological diffusion and advancement. Thus, investigation of both firm-specific and employee-specific characteristics that influence decisions to adopt new technologies has long been recognized as an important area of research. The purpose of this study is to assess the causal effects of workers’ educational attainment on their adoption of new technologies, using an employer-employee linked panel dataset in Canada, the Workplace and Employee Survey (WES) (1999-2004). This study also reveals other determinants of technology adoption within a firm and illuminates the mechanism of technology diffusion. A large body of prior research has shown that highly educated workers tend to adopt new technologies faster than those with less education (Lleras-Muney & Lichtenberg, 2002; Welch, 1970; Weinberg, 2004; Wozniak, 1987). Generally a new technology is associated with uncertain returns and fixed costs of adoption. How quickly producers and employees can adapt to a changing set of production possibilities partly depends on their human capital and their knowledge of the new technology. Wozniak (1987) finds that education and information reduce adoption costs and uncertainty, and thereby raise the probability of early adoption. Weinberg (2004) also finds a complementarity between existing human capital and computer adoption. Technology adoption can also be viewed as a reallocation decision made in response to disequilibria, which allows firms to take advantage of the opportunities provided by the introduction of innovative inputs. Given that highly educated workers have a comparative advantage in dealing with disequilibria and in implementing new technology (Bartel & Litchenberg, 1987; Khaldi, 1975; Nelson & Phelps, 1966; Shultz, 1964, 1975; Welch, 1970, 1973), firms with more highly educated workers are more likely to be early adopters of new technologies.

3

One thing to note is that how education affects technology adoption may depend on the type of technology to be adopted. Dunne and Troske (2004) find that the relationship between technology adoption and workforce skills varies across different types of technologies. The adoption of technologies related to engineering and design tasks, for instance, is more strongly associated with the skills of the workforce than is the adoption of technologies related to production activities. They find little correlation between the adoption or use of technologies and changes in workforce skills at the plant level. However, they do find that plants adopting technologies related to engineering and design tasks grow faster. The positive correlations between the level of education and technology adoption found by previous studies, however, is likely to be confounded by the endogeneity of education, and thus does not necessarily reflect the true causal effects of education on technology adoption. In particular, positive associations between education and technology adoption could arise because of unobserved factors that are correlated with both measures. For example, individuals with higher innate ability and stronger motivation are likely to be early adopters of new technologies at the workplace and also likely to acquire more schooling. Therefore, positive correlations between technology adoption and education based on ordinary least squares (OLS) estimates tend to overestimate the effects of education on technology adoption and fail to reveal the true causal link between the two. 1 To our knowledge, there has been no prior research that has established a causal link between education and technology adoption largely due to the endogeneity of education problem. To overcome this problem, we make use of historical changes in compulsory schooling laws in Canada as natural experiments to create instrumental variables for education in our study, which allows us to draw valid causal inference about the effects of education on technology adoption. In addition, our study extends previous research in this area by making use of the rich information provided by WES data. As a joint decision made by the firm and the worker, technology adoption is affected by the characteristics of both the workplace and

1

The influential reviews conducted by Card (1999, 2001) and Griliches (1977) discuss the endogeneity of education and the potential biases of OLS estimates in the context of estimating the return to schooling.

4

the workers. Most previous studies on the determinants of technology adoption, however, focus on either employer characteristics or employee characteristics, but not on both, mainly due to the data limitation (Hannan & McDowell, 1984; Levin, Levin, & Meisel, 1987; Stoneman & Kwon, 1996; Weinberg, 2004). 2 WES, as an employer-employee linked dataset, enables us to take into account the characteristics of both employers and employees that affect technology adoption. Taking advantage of the panel data nature of WES data, our estimation further control for unobserved heterogeneity in both the workers and the workplace, and is thus likely to produce more accurate estimates than prior research. Another advantage of WES data is that they provide rich information on technology use and recent adoptions of new technology at both the employee and employer levels. We are thus able to create and analyze a variety of measures of technology adoption, while prior studies typically focus on a very limited set of measures of technology adoption, such as the use of computers. Based on data from WES (1999-2004), we find that education increases the probability of using computer in the job. We also find that employees with more education possess longer work experiences in using computer, and are more likely to experience upgrade in computer-controlled or computer-assisted technology and experience upgrade in technological device than those with less education. Given the general consistency in findings across alternative measures of technology adoption, our results provide empirical support for the hypothesis that there exists a causal link between education and technology adoption. Findings from this study not only shed light on the causal relationships between education and individuals’ technology adoption, but also contribute to the growing

2

Hannan and McDowell (1984) examine the determinants of adoption of automatic teller machines by banks and find that larger banks and banks operating in more concentrated local banking markets register a higher conditional probability of adopting automatic teller machines. Levin, Levin, and Meisel (1987) investigate the role of market structure in the time path of adoptions of a new technology (optical scanners) in the U.S. food store industry. They find that during the early stage leading firms with larger store size who are not members of chains and who operate in less concentrated markets with higher incomes and wage rates tend to adopt the scanners sooner. Stoneman and Kwon (1996) examine the relationship between firm profitability and technology adoption. They find that non-adopters experience reduced profits as other firms adopt new technologies and that the gross profit gains to adopters of new technology are related to firm and industry characteristics, the number of other users of new technologies, and the cost of acquisition.

5

literature on the private and social benefits of education. To the extent that education increases the probability of technology adoption, the private and social benefits of education may be understated by standard outcome measures (e.g., earnings). In particular, the WES data allow us to identify the impacts of an individual employee’s education on employer and co-worker outcomes. In contrast, most studies of the nonwage effects of education focus on outcomes such as criminal activity, health and civic participation, impacts that are experienced by the individual receiving the additional education (Riddell, 2007). Further, this study provides empirical evidence that supports education as an effective means to enhance technology adoption and diffusion, an important channel for technological advancement and economic growth. The remainder of the paper is organized as follows. We describe the data for this study in Section 2, and explain the econometric models used in our data analyses in Section 3. In Section 4, we report and discuss empirical results about the effects of education on alternative measures of technology adoption. We conclude the paper in Section 5.

2

Data

2.1 The Micro Dataset: Workplace and Employee Survey 3 The data used in this study include all panels of the annual Canadian Workplace and Employee Survey (WES) from 1999 to 2004, the latest release. WES is designed to explore a broad range of issues relating to employers and their employees. The survey is unique in that employers and employees are linked at the micro data level; employees are selected from within sampled workplaces. Thus, information from both the supply and demand sides of the labour market is available to enrich studies on either side of the market. The survey aims to shed light on the relationships among competitiveness, innovation, technology use and human resource management on the employer side and technology use, training, job stability and earnings on the employee side. The target population for the employer component is defined as all business locations operating in Canada that have paid employees in March in the survey year, with

3

The description of the Workplace and Employee Survey in this subsection was mainly taken from the Statistics Canada website.

6

the following exceptions: employers in Yukon, Nunavut and Northwest Territories; and employers operating in crop production and animal production; fishing, hunting and trapping; private households, religious organizations and public administration. The target population for the employee component is all employees working or on paid leave in March in the survey year in the selected workplaces who receive a Canada Customs and Revenue Agency T-4 Supplementary form. If a person receives a T-4 slip from two different workplaces, then the person will be counted as two employees on the WES frame. The WES draws its sample from the Business Register maintained by the Business Register Division of Statistics Canada and from lists of employees provided by the surveyed employers. The Business Register is a list of all businesses in Canada and is updated each month using data from various surveys, business profiling and administrative data. The survey population is the collection of all units for which the survey can realistically provide information. The survey population may differ from the target population due to operational difficulties in identifying all the units that belong to the target population. The initial sample of 6,322 workplaces selected in 1999 is followed over time and is supplemented at two-year intervals with a sample of births selected from units added to the Business Register since the last survey occasion. Stratification of units remains constant over the life of the initial panel (set at eight years). Whenever possible, the same sampling fractions are applied to all panels. Sometimes the sampling fractions are adjusted to offset stratum erosion, or to compensate for upswings or downswings in the economy. For 2001, they were revised slightly upward. This resulted in a birth panel of 1,792 workplaces. For 2003 this resulted in a birth panel of 2,334 workplaces. In 2004, the sample size of employers was 6,159. The employee data and the workplace data are collected separately. The frame of the employee component of WES is based on lists of employees made available by the selected workplaces. A maximum of 24 employees are sampled using a probability mechanism. In workplaces with fewer than four employees, all employees are selected. Employees are followed for two years only, due to the difficulty of integrating new employers into the location sample as workers change companies. As such, fresh samples

7

of employees are drawn on every second survey occasion (i.e. first, third, fifth). Therefore, data at the employee level are essentially two-year panel data although the data at the workplace level are six-year panel data from 1999 to 2004. The sample size of employees was 23,540 in 1999 and was 16,804 in 2004. 2.2 Measures of Technology Adoption WES provides rich information on the use of technology and recent adoptions of new technology at both the employee and employer level. The information/variables on technology use at the employee level are divided into four categories: use of computer; use of computer-controlled or computer-assisted technology; use of other machine or technical device; and change in technological complexity. Our study focuses on technology adoption at the employee level. The four measures of employee-level technology adoption used in our study and the relevant survey questions are listed below. •

Use of computers: Do you use a computer in your job? Please exclude sales terminal, scanners, machine monitors, etc. By a computer we mean a microcomputer, minicomputer, personal computer, mainframe computer or laptop that can be programmed to perform a variety of operations.



Number of years using a computer: Considering all jobs you have held, how many years have you used a computer in a work environment?



Upgrade or change in computer-controlled or computer-assisted technology: If you use a computer-controlled or computer-assisted technology in the course of your normal duties, such as industrial robots and retail scanning systems, has there been any upgrade or change in that technology in the past twelve months?



Upgrade or change in technological device: If you use any other machine or technological device for at least one hour a day in the course of your normal duties, including cash registers, sales terminals, typewriters, vehicles and industrial machinery, has there been any upgrade or change in that technology in the past twelve months?

2.3 Educational Attainment and Compulsory Schooling Laws in Canada “Educational attainment” is a key variable in our empirical analyses. In the WES survey, employees who are surveyed for the first time, i.e., in odd survey years, are first asked about the highest grade of elementary or high school (secondary school) that they have completed. Employees will then report whether they have graduated from high school

8

(secondary school) and whether they have received any other education. Conditional on having received other education beyond elementary or high school, employees will further report the type of the additional education received, which may include trade or vocational diploma or certificate; some college, CEGEP, 4 institute of technology or nursing school; completed college, CEGEP, institute of technology or nursing school; some university; teachers' college; university certificate or diploma below the bachelor level; bachelor or undergraduate degree or teachers' college (e.g. B.A., B.Sc., B.A.Sc., or four-year B.Ed.); university certificate or diploma above the bachelor level; Master's degree (M.A., M.Sc., M.Ed., MBA, MPA or equivalent); earned doctorate; degree in medicine, dentistry, veterinary medicine, law, optometry or theology (M.D., D.D.S., D.M.D., D.V.M., LL.B., O.D., or M.DIV.) or one-year B.Ed. after another bachelor's degree; industry certified training or certification courses; or other. We assign to each employee a total number of years of schooling based on the level of schooling reported. Because WES follows each sample of employees for two consecutive years, WES only asks about any additional education obtained by the employees during the past 12 months in even-year surveys. Conditional on having received additional education during the past 12 months, WES will further ask about the type of education received using the same categories of education as in the odd-year surveys. Based on the assumption that the number of years of schooling increases by one if an employee received additional education during the past 12 months, we created a variable for the total number of years of schooling for all employees in all survey years. One methodological challenge to our research is that the positive correlations between education and technology adoption that we expect to observe are likely to be confounded by the endogeneity of education, and thus do not necessarily reflect the true causal effects of education on technology adoption. To address the endogeneity of education problem, we use changes in compulsory schooling laws over time to instrument for schooling. Changes in these laws have been shown to have significant effects on educational attainment, and have been a commonly-used instrument for 4

CEGEPs are post-secondary education institutions exclusive to the province of Quebec in Canada. CEGEP is a French acronym for “Collège d'enseignement général et professionnel,” meaning "College of General and Professional Education." The purpose of CEGEPs is to make post-secondary education more accessible in Quebec, as well as to provide proper academic preparation for university. Students in Quebec who want to continue a post-secondary education must attend a CEGEP prior to a Quebec university.

9

education (see, for example, Acemoglu & Angrist, 2000; Lochner & Moretti, 2004; Milligan, Moretti & Oreopoulos, 2004; and Oreopoulos, 2003, 2006a). Using the compulsory schooling laws data compiled by Oreopoulos (2003, 2006a), we first created three indicator variables to indicate whether the youngest school leaving age is 14, 15, or 16, and then created another three indicator variables to indicate whether the oldest school entry age is 6, 7, or 8. The linkage between the WES data and data on compulsory schooling laws is established based on the current province of residence of each individual and the year when the individual turned 14 for matching school leaving age or 6 for matching school entry age. 5 Schmidt (1996) finds that the effects of compulsory schooling laws in the U.S. were largest when matched to individuals at age 14. Acemoglu and Angrist (2000), Lleras-Muney and Lichtenberg (2002), Schmidt (1996), and Goldin and Katz (2003) adopt the same procedure in their studies based on the U.S. data, and Oreopoulos (2003, 2006a) adopts the same procedure when analyzing Canadian data. Because the WES does not report the birthplace of the respondents or the province of residence when the respondents turned 6 or 14, we need to rely on the respondents’ current province of residence at the time of the survey in linking WES data with data on compulsory schooling laws. Although this might introduce some measurement error into our instrumental variables, previous studies suggest that such measurement error is not likely to seriously bias the instrumental variable estimates (Milligan, Moretti & Oreopoulos, 2004). Relying on the instrumental variables thus created, we estimate the causal effects of high-school graduation and years of schooling on individuals’ technology adoption. Identification of the causal effects of education is based on changes over time in the youngest school leaving age and oldest school entry age in a given province as well as variations in compulsory schooling laws across provinces. The identifying assumption is that conditional on province of residence, cohort of birth, and survey year, the timing of the changes in compulsory schooling laws within each province is orthogonal to

5

Individuals having moved across provinces before age 6 were mismatched for both the school leaving age and school entry age, while individuals having moved across provinces between age 6 and age 14 were mismatched for school leaving age. Because changes in compulsory schooling laws were unlikely to cause people to move across provinces, this should not cause significant bias in our estimates for the full sample.

10

unobserved characteristics that affect schooling choices, such as ability and family background. Ideally, we would like to estimate a general model where the effect of education on technology adoption varies across years of schooling. This is not empirically feasible, however, because the instruments we use are limited in both the range of schooling years affected and the amount of actual variation. In fact, it is not possible to use two-stage least squares (2SLS) to estimate a model of technology adoption that is linear in years of schooling with a separate “sheepskin” effect of high-school graduation. Therefore, for the instrumental variable analyses, we use years of schooling or a dummy for high-school graduation as the main independent variable. 2.4 Sample Selection and Descriptive Statistics We restrict our analytic sample to those aged 20-65. In the analyses of each measure of technology adoption, we drop from the sample employees who have missing values on the dependent variable. Because WES does not report the specific provinces within the Atlantic region, we could not link the compulsory schooling laws, specified by province, with WES data for employees from the Atlantic region, and thus drop these employees from the sample. The resulting sample size is 99,871 for analyzing computer use and computer use experience, 12,823 for analyzing upgrade in technology, and 23,154 for analyzing upgrade in technological device. As shown by the sample descriptive statistics in Table 1, the average number of years of schooling completed by employees in our sample is between 13.6 and 14.6, and 80 to 92% of our sample had graduated from high school, depending on the outcome measure. The mean of each of the measures of technology adoption can be found in Tables 2 to 7. Specifically, 67.8% of employees use a computer in the job. Considering all jobs held, employees use a computer in a work environment for an average of 7.8 years. Conditional on using a computer-controlled or computer-assisted technology in the course of normal duties, 38.5% of the employees have experienced upgrade or change in that technology in the past 12 months. Conditional on using any other machine or technological device for at least one hour a day in the course of normal duties, 23.7% of the employees have experienced upgrade or change in that technology in the past 12 months.

11

3

Econometric Model

In order to study the effects of education on technology adoption, we estimate three different models. Firstly, we analyze the data in its cross-section dimension, i.e., the pooled cross-sectional data for employees linked with the workplace data from 1999 to 2004. Secondly, we analyze the three two-year panel data sets, i.e., 1999-2000 panel, 2001-2002 panel, and 2003-2004 panel, both separately and in pooled form. Finally, we conduct the instrumental variables estimation on the pooled cross-sectional data to quantify the causal effects of education on technology adoption. For each model, we analyze the four measures of technology adoption separately. Technology adoption is a joint decision between employers and employees, who interact with each other in the workplace. Any outcome variable, for both the employers and the employees, is an equilibrium resulting from the interactions. To be consistent with the actual data generating process, our analyses of technology adoption decision take into account both the employer side and the employee side. Given that our aim is to determine whether education has any causal role in the technology adoption process, rather than to determine the magnitude of its effect relative to the effects of other factors, we restrict our analytic model to a simple, reduced form specification. For the crosssection analyses, the specification of the estimation equation is as follows: Tech _ Adopt ij = α + X ij β + Y j γ + ε ij ,

(1)

where Tech _ Adopt ij is one of the four measures of technology adoption for the ith worker in the jth workplace; α is a constant; X ij is a vector of explanatory variables for the ith worker in the jth workplace, including education, age, work experience, gender, marital status, union status, industry, occupation, province, and survey year dummies; β is a vector of coefficients for variables in X ij ; Y j is a vector of workplace characteristics, including the three dummy variables for the size of the workplace; γ is a vector of coefficients for variables in Y j ; and ε ij is a randomly distributed error term. When estimating Equation (1) with the ordinary least squares (OLS), we use the employee survey final weights provided by the Workplace and Employee Survey (WES). The calculations of t-statistics and all inferences are based on robust standard errors. Because workplaces are the primary sampling units from which a random sample of

12

employees are further selected, the estimation procedure need to adjust the calculation of the standard errors to take into account of the potential correlations among employees within the same workplace. In all our estimations, therefore, we correct for the clustering by workplaces. The advantage of analyzing the pooled cross-sectional data is the large analytic sample size, which is particularly important for conducting instrumental variables estimation. By using the pooled cross-sectional data, however, we fail to take into account the panel data feature of the WES, which is essentially a two-year panel data set, with the same sample of employees surveyed for two consecutive years. Therefore, we also estimate the following model with the three two-year panel data sets separately: Tech _ Adopt ijt = α + X ijt β + Y jt γ + u ij + ε ijt ,

(2)

the specification of which is similar to that of Equation (1), but without survey year dummies. Equation (2) also differs from Equation (1) in that it includes an index for time period, t. Because this is a panel data model, there is also an individual fixed effect, u ij . Because the key variable we are interested in is an employee’s educational attainment, which is time-constant, neither the first-difference estimator nor the fixedeffects estimator work for our model. Instead, we use the random effects estimator to estimate Equation (2). As in the cross-section analyses based on Equation (1), we correct for the clustering by workplaces in all estimations. Assuming that the marginal effect of each explanatory variable on technology adoption is the same across three panels, we pool together the three two-year panel data sets and estimate Equation (2) augmented by dummy variables for each of the three panels. The advantage of analyzing pooled panel datasets is the increased sample size, which leads to improved precision of instrumental variables estimates. Finally, we make use of historical changes in compulsory schooling laws in Canada as natural experiments to create instrumental variables for education in order to assess the causal effects of education on technology adoption. Specifically we conduct the two-stage least squares (2SLS) estimation on Equation (1) using the pooled crosssectional data. In order to obtain accurate estimates with 2SLS, it is essential that the sample size is large enough. We correct for clustering by workplaces as well as clustering by province of residence and year of birth in all estimations. The calculations of t13

statistics and all inferences are based on robust standard errors. All estimations are weighted by the final employee weights.

4

Empirical Results

This section presents empirical evidence on the causal effects of education on the four measures of technology adoption based on data from the Workplace and Employee Survey (WES) (1999-2004). 4.1 The Effects of Education on Computer Use in the Job In this subsection, we present and discuss our findings on the effects of education on a commonly-used measure of technology adoption--the probability of using a computer in the job. We begin by reporting the OLS estimates, and then present the random effects estimates and the instrumental variable estimates. 4.1.1 OLS Estimates Since the distribution of years of schooling is concentrated around 12 to 13 years, with a very small percentage of individuals having 8 years of schooling or less, we used 8 years of schooling or less as the base category and regressed the probability of using a computer in the job on the complete set of years of schooling dummies. 6 The regression also controls for survey year, province/territory, a quartic in age, five experience groups (years of experience 1-9, 10-19, 20-29, 30-39, and 40 or above), gender, marital status, union coverage, 13 dummy variables for industry, 5 dummy variables for occupation, and size of the firm. Based on the OLS coefficient estimates on the complete set of schooling dummies, Figure 1 displays how education affects the probability of using a computer in the job at each schooling level. It shows a steady increase in the probability of using a computer as schooling increases from 10 to 17 years. The partial relationship between education and the probability of using a computer in the job is approximately linear beyond 10 years of schooling. In addition to the above analyses where schooling was represented as a set of dummy variables, we also conducted analyses where the main independent variable is a

6

Because the percentage of the sample reporting more than 18 years of schooling was only 0.6%, we used the dummy variable for 19 years of schooling to cover all of those having more than 18 years of schooling.

14

dummy for high-school graduation or the number of years of schooling. OLS results from these additional analyses are presented in Column (1) in Tables 2 and 3 respectively. As shown in Table 2, graduating from high school is associated with a 16.2% increase in the probability of using a computer in the job. Table 3 reveals that an additional year of schooling is associated with a 3.7% increase in the probability of using a computer in the job. The estimation results not reported in the tables further indicate that females, married people, people not covered by unions, and older people are more likely to use a computer in the job than are others. People with 30 to 39 years of work experiences are most likely to use a computer among all experience groups. Conditional on all the other explanatory variables included in our analyses, employees from the following industries are most likely to use a computer in the job: finance and insurance; real estate, rental and leasing operations; business services; and information and cultural industries. Employees from the following industries, on the other hand, have the lowest probability of using a computer in the job: forestry, mining, oil, and gas extraction; education and health services; labour intensive tertiary manufacturing; primary product manufacturing; secondary product manufacturing; retail trade and consumer services; and particularly construction. The probability of using a computer is located somewhere in between for the other industries: capital-intensive tertiary manufacturing; transportation, warehousing, and wholesale; and communication. As for the impacts of occupation on the probability of

computer

use

in

the

job,

we

find

that

managers,

professionals,

and

clerical/administrative workers are most likely to use a computer in the job, whereas production workers are least likely to use a computer. Technical/trades workers and marketing/sales workers are somewhere in between. Consistent with findings from Wozniak (1987), we also find that the larger the size of the firms, the more likely that employees use a computer in the job. 4.1.2 Random Effects Estimates OLS estimates using the pooled cross-sectional data provide empirical evidence on the positive impacts of education on computer use in a straightforward way. Such analyses, however, fail to take into account the panel data feature of the WES data, which are essentially two-year panel data at employee level, with each sample of employees

15

interviewed for two consecutive years. Therefore, we pool together the three two-year panels, i.e., 1999-2000 panel, 2001-2002 panel, and 2003-2004 panel, and conduct the empirical analyses using panel data models with dummy variables for each of the three panels included in all regressions. We also conduct the same analyses for each two-year panel separately. Because the key explanatory variable we are interested in–employee’s educational attainment—is time-constant, neither the first-difference estimator nor the fixed effects estimator is appropriate for our model. Instead, we use the random effects estimator to estimate the effects of education on computer use. The estimation results are presented in Column (3) in Tables 2 and 3, which indicate slightly stronger effects of education on computer use in the job than the OLS estimates. Random effects estimate in Column (3) of Table 2 shows that graduating from high school is associated with a 21.1% increase in the probability of using a computer in the job. Table 3 reveals that an additional year of schooling is associated with a 4.3% increase in the probability of using a computer. Figure 2 illustrates the regression-adjusted probability of computer use in the job by years of schooling. It shows that conditional on all the other explanatory variables, the probability of using a computer increases steadily as schooling increases from 9 to 18 years. Due to space limitations, we do not report the estimation results based on each of the three two-year panel datasets, which are very similar to the results based on the pooled panel data. 4.1.3 Instrumental Variable Estimates Both the OLS and the random effects estimates just presented are consistent with the hypothesis that education increases the probability of using a computer in the job. These estimates, however, may reflect the effects of unobserved individual characteristics that influence both the probability of computer use and schooling choices. Therefore, the positive correlations between the probability of computer use and education as shown by the OLS estimates and the random effects estimates may be biased estimates of the effects of education on computer use and fail to reveal the true causal link between the two. It is also possible that the OLS estimates and the random effects estimates could underestimate the effect of education on computer use, for instance, due to the existence of measurement error in educational attainment.

16

To address the endogeneity of education problem, we use changes in compulsory schooling laws over time to instrument for schooling, as explained in Section 2.3. Column (2) in Tables 2 and 3 present the two-stage least squares (2SLS) estimates of the impact of education on computer use in the job based on the pooled cross-sectional data with specifications identical to those used to obtain the OLS estimates. Column (2) in the upper panels of Tables 2 and 3 reports the first-stage results, i.e., coefficient estimates for the effects of different school leaving ages and school entry ages on educational attainment. The base categories are those with school leaving age of 14 and those with school entry age of 8. The first-stage results indicate that, in general, the more stringent the compulsory schooling legislation, the higher the probability of high-school graduation and the more years of schooling completed. Table 2 shows, for example, individuals who lived in provinces requiring the youngest school leaving age to be 15 when they were 14 were 8.3% more likely to have completed high school by the time of the survey compared with individuals living in provinces requiring the youngest school leaving age to be 14 when they were 14 (the excluded category). Individuals who lived in provinces requiring the oldest school entry age to be six when they were six were 4.5% more likely to have completed high school by the time of the survey compared with individuals who lived in provinces requiring the oldest school entry age to be eight when they were six (the excluded category). In addition, results in Table 3 indicate that the average years of schooling completed was 0.58 years higher with a youngest school leaving age of 15 compared with a youngest school leaving age of 14, and 0.53 years higher with an oldest school entry age of 7 compared with an oldest school entry age of 8. These results are similar to those obtained in Oreopoulos (2006a). To assess the adequacy of the instrumental variables, we performed F-tests for exclusion of instruments in the first-stage regression. As shown in Column (2) in the upper panels of Tables 2 and 3, the F-statistic for exclusion of instruments is 3.97 with high-school graduation as the endogenous variable and 8.23 with years of schooling as the endogenous variable, which suggests significant positive correlations between the instruments and schooling.

17

Column (2) in the bottom panels of Tables 2 and 3 presents instrumental variable estimates of the effects of high-school graduation and years of schooling on the probability of using a computer in the job. The IV coefficient for high-school graduation is 0.32 when the four indicator variables for school leaving and entry ages are used as the instruments but is not significant. The IV coefficient for years of schooling is 0.06 and is significant at the .05 level. Therefore, graduating from high school increases the probability of using a computer in the job by 32%. An additional year of schooling increases this probability by 6%. The IV estimates are consistently higher than both the OLS and the random effects estimates, although the differences are not always statistically significant. We discuss the implications of the differences between the IV and OLS estimates or random effects estimates later in the paper. 4.2 The Effects of Education on Computer Use Experience In this subsection, we focus on the empirical results on the effects of education on computer use experience, the analyses of which are very similar to those on computer use as explained in detail in Section 4.1. 4.2.1 OLS Estimates Column (1) in Tables 4 and 5 present OLS estimates of the effects of high-school graduation and years of schooling respectively on the total number of years of experience in using a computer in a work environment. The regressions also control for survey year, province/territory, a quartic in age, five experience groups (years of experience 1-9, 1019, 20-29, 30-39, and 40 or above), gender, marital status, union coverage, 13 dummy variables for industry, 5 dummy variables for occupation, and size of the firm. As shown in Table 4, conditional on all other explanatory variables, high-school graduates tend to have three years more experience in using computer than those without a high-school degree. Table 5 reveals that an additional year of schooling is associated with a 0.6 years increase in computer use experience. The estimation results not reported in the tables further indicate that females, married people, people not covered by unions, and people with longer work experience tend to possess more experience in using computer in a work environment than are others. Conditional on all other explanatory variables, employees from the following industries have the longest computer use experience in a work environment: finance and insurance;

18

real estate, rental and leasing operations; business services; and information and cultural industries. Employees from the following industries, on the other hand, have the shortest computer use experience in a work environment: labour intensive tertiary manufacturing; and construction. The computer use experience is located somewhere in between for other industries: forestry, mining, oil, and gas extraction; primary product manufacturing; secondary product manufacturing; capital-intensive tertiary manufacturing; transportation, warehousing, and wholesale; communication; retail trade and consumer services; and education and health services. As for the impacts of occupation on computer use experience, we find that managers, professionals, and clerical/administrative workers possess the most experience in using a computer in a work environment while production workers have the least computer use experience. Technical/trades workers and marketing/sales workers are somewhere in between. Overall the pattern of computer use experience across different industries and occupations is similar to the pattern of the probability of computer use and computer use time. Moreover, we find that the larger the size of the firms, the more years of experience that employees have in using a computer in a work environment. 4.2.2 Random Effects Estimates The random effects estimates based on the pooled panel data are similar to the OLS estimates of the effects of education on computer use experience. As shown in Column (3) of Table 4, graduating from high school is associated with a 3.2 years increase in computer use experience. Table 5 reveals that an additional year of schooling is associated with a 0.6 years increase in computer use experience. Estimation results based on each of the three two-year panel datasets, which are not reported here, are very similar to the results based on the pooled panel data. 4.2.3 Instrumental Variable Estimates We employ 2SLS estimator to identify the causal effects of high-school graduation and years of schooling on computer use experience. The first-stage results are exactly the same as those from the analysis of the probability of computer use reported in Section 4.1.3 due to the identical analytic samples. As shown in Column (2) in the bottom panels of Tables 4 and 5, the IV coefficient is 10.7 for high-school graduation (p < .05) and 1.7 for years of schooling (p

19

< .01). Therefore, graduating from high school increases computer use experience by 10.7 years while an additional year of schooling increases computer use experience by 1.7 years. Employees with more education are not only more likely to use computer in the job, but also start to use computer in a work environment earlier. 4.3 The Effects of Education on Upgrade in Technology Besides measures of technology adoption based on the use of computers, upgrades or changes in technology used at workplaces also reflect technology adoption within a firm. In this subsection, we present the empirical results on the effects of an employee’s educational attainment on his/her probability of experiencing upgrade or change in computer-controlled or computer-assisted technology during the past 12 months conditional on having used such technology in the course of his/her normal duties. Because the sample size is relatively small, we were not able to obtain precise instrumental variable estimates. Thus empirical results presented here include just OLS and random effects estimates. 4.3.1 OLS Estimates Column (1) in Table 6 presents OLS estimates of the effects of high-school graduation and years of schooling respectively on the probability of experiencing upgrade or change in computer-controlled or computer-assisted technology during the past 12 months. The regressions also control for survey year, province/territory, a quartic in age, five experience groups (years of experience 1-9, 10-19, 20-29, 30-39, and 40 or above), gender, marital status, union coverage, 13 dummy variables for industry, 5 dummy variables for occupation, and size of the firm. Conditional on all other explanatory variables, graduating from high school is associated with a 5.7% increase in the probability of experiencing upgrade or change in computer-controlled or computerassisted technology during the past 12 months, and an additional year of schooling is associated with a 1.1% increase in this probability. Estimation results not reported here, however, show that employees with more education are actually less likely to use computer-controlled or computer-assisted technology than those with less education. 4.3.2 Random Effects Estimates The random effects estimates based on the pooled panel data are very similar to the OLS estimates of the effects of education on the probability of experiencing upgrade or change

20

in computer-controlled or computer-assisted technology during the past 12 months. As shown in Column (2) of Table 6, graduating from high school is associated with a 6.2% increase in the probability of experiencing upgrade or change in computer-controlled or computer-assisted technology during the past 12 months. An additional year of schooling is associated with a 1.2% increase in this probability Random effects estimates not reported here also indicate that highly educated employees are less likely to use computer-controlled or computer-assisted technology. Estimation results based on each of the three two-year panel datasets are very similar to the results based on the pooled panel data. 4.4 The Effects of Education on Upgrade in Technological Device This subsection presents the OLS and random effects estimates of the effects of an employee’s educational attainment on his/her probability of experiencing upgrade or change in any other machine or technological device during the past 12 months, conditional on having used such device for at least one hour a day in the course of his/her normal duties. Because the sample size is relatively small, we were not able to obtain precise instrumental variable estimates for this measure of technology adoption. 4.4.1 OLS Estimates Column (1) in Table 7 present OLS estimates of the effects of high-school graduation and years of schooling respectively on the probability of experiencing upgrade or change in any other machine or technological device during the past 12 months. The regressions also control for survey year, province/territory, a quartic in age, five experience groups (years of experience 1-9, 10-19, 20-29, 30-39, and 40 or above), gender, marital status, union coverage, 13 dummy variables for industry, 5 dummy variables for occupation, and size of the firm. Conditional on all other explanatory variables, graduating from high school is associated with a 3.6% increase and an additional year of schooling is associated with a 1.2% increase in the probability of experiencing upgrade or change in any other machine or technological device during the past 12 months. Estimation results not reported here, however, show that employees with more education are actually less likely to use other machine or technological device than those with less education. 4.4.2 Random Effects Estimates

21

The random effects estimates based on the pooled panel data are very similar to the OLS estimates of the effects of education on the probability of experiencing upgrade or change in any other machine or technological device during the past 12 months. As shown in Column (2) of Table 7, graduating from high school is associated with a 3.4% increase, and an additional year of schooling is associated with a 1% increase in the probability of experiencing upgrade or change in any other machine or technological device during the past 12 months. Random effects estimates not reported here also indicate that highly educated employees are less likely to use other machine or technological device. Estimation results based on each of the three two-year panel datasets are very similar to the results based on the pooled panel data. 4.5 Why Are IV Estimates Consistently Higher than OLS Estimates? The empirical results in this study based on the WES data indicate that the IV estimates are consistently higher than the OLS estimates. There are several potential explanations for such a difference. According to Griliches (1977) and Card (1999), the existence of measurement error in educational attainment, in the classical sense, could cause the OLS estimates to underestimate the effects of education on outcome variables, such as earnings. Kane, Rouse, and Staiger (1999) discusses how OLS estimates and IV estimates relying on “natural experiments” in particular may both be biased estimates of the returns to schooling due to measurement error in educational attainment. An alternative explanation is that the OLS and IV estimates measure different things in the presence of heterogeneity in the net benefits of additional education across individuals. When OLS is applied to a sample representative of the overall population, it yields an estimate of the average treatment effect (ATE), which shows the expected benefits of education experienced by an individual chosen at random from the population. IV estimates, on the other hand, represent the local average treatment effect (LATE) for the subset of the population who are actually affected by the instruments (Card, 2001). In estimations using changes in compulsory attendance laws as the IV, for example, the IV estimates indicate the benefits of education in terms of technology adoption for the subset of the population who acquired more schooling than they otherwise would have chosen to acquire as a consequence of the changes in the laws. For this subset of the population, therefore, the payoff to incremental investments in

22

education may exceed the average return in the population. Policy interventions that result in additional schooling being acquired by individuals from disadvantaged backgrounds, or those who face other barriers to acquiring human capital, may yield a substantial return in the form of earlier or faster technology adoption, in addition to contributing to equity objectives. Recent studies by Carneiro, Heckman, and Vytlacil (2003) and Oreopoulos (2006b), however, question the ATE vs. LATE interpretation of the difference between OLS and IV estimates in the compulsory schooling laws context. Apparently further research is needed in this area.

5

Conclusions

In an environment characterized by rapid technological changes as is the case with the current economy in North America, studies on whether and how education promotes technology adoption and subsequently technology diffusion have become especially relevant. Relying on data from the Workplace and Employee Survey (1999-2004), this study assesses the causal effects of education on technology adoption by using instrumental variables for schooling derived from Canadian compulsory school attendance laws. We find that Employees with more education are more likely to use computer in the job. Specifically, graduating from high school increases the probability of using a computer in the job by 32% and an additional year of schooling increases such a probability by 6%. We also find that employees with more education possess longer work experiences in using a computer. High-school graduates, for instance, have 10.7 more years of experience in using a computer at work than those without a high-school degree. Moreover, employees with more education are more likely to experience upgrade in computer-controlled or computer-assisted technology and to experience upgrade in technological device than those with less education. Given the consistency of our findings, we conclude that the estimated effects of education on technology adoption cannot be easily explained away by unobserved factors that are correlated with both variables. Findings from this study not only shed light on the causal relationships between education and individuals’ technology adoption, but also contribute to the growing literature on the private and social benefits of education.

23

Further, this study provides empirical evidence that supports education as an effective means to enhance technology adoption and diffusion, an important channel for technological advancement and economic growth.

References Acemoglu, Daron, Joshua Angrist. 2000. “How Large Are Human Capital Externalities? Evidence from Compulsory Schooling Laws.” NBER Macroeconomics Annual 2000 :9-59. Bartel, Ann P., and Frank R. Lichtenberg. 1987. “The Comparative Advantage of Educated Workers in Implementing New Technology.” The Review of Economics and Statistics 69:1-11. Card, David. 1999. “The Causal Effect of Education on Earnings.” Pp. 1801-1863 in Handbook of Labor Economics, edited by O. Ashenfelter and D. Card: Elsevier Science B.V. Card, David. 2001. “Estimating the Return to Schooling: Progress on Some Persistent Econometric Problems.” Econometrica 69:1127-1160. Carneiro, Pedro, James Heckman, and Edward Vytlacil. 2003. “Understanding What Instrumental Variables Estimate: Estimating Marginal and Average Returns to Education.” Working Paper, University of Chicago. Dunne, Timothy and Kenneth Troske. 2004. “Technology Adoption and Workforce Skill in U.S. Manufacturing Plants.” IZA Discussion Paper No. 1427. Goldin, Claudia, and Lawrence F. Katz. 2003. “Mass Secondary Schooling and the State: The Role of State Compulsion in the High School Movement.” NBER Working Paper #10075. Griliches, Zvi. 1977. “Estimating the Returns to Schooling: Some Econometric Problems.” Econometrica 45:1-22 Hannan, Timothy H., and John M. McDowell. 1984. “The Determinants of Technology Adoption: the Case of the Banking Firm.” Rand Journal of Economics 15:328-335. Kane, Thomas J., Cecilia Elena Rouse, and Douglas Staiger. 1999. “Estimating Returns to Schooling when Schooling is Misreported.” NBER Working Paper #7235.

24

Khaldi, Nabil. 1975. “Education and Allocative Efficiency in U.S. Agriculture.” American Journal of Agricultural Economics 57:650-657. Levin, Sharon G., Stanford L. Levin, and John B. Meisel. 1987. “A Dynamic Analysis of the Adoption of a New Technology: The Case of Optical Scanners.” The Review of Economics and Statistics 69:12-17. Lleras-Muney, Adriana, and Frank Lichtenberg. 2002. “The Effect of Education on Medical Technology Adoption: Are the More Educated More Likely to Use New Drugs?” NBER Working Paper 9185. Lochner, Lance, Enrico Moretti. 2004. “The Effect of Education on Crime: Evidence from Prison Inmates, Arrests and Self-Reports.” American Economic Review 94:155189. Milligan, Kevin, Enrico Moretti and Philip Oreopoulos. 2004. “Does education improve citizenship: Evidence from the U.S. and U.K.” Journal of Public Economics 88 (August): 1667-1695. Nelson, Richard, and Edmund P. Phelps. 1966. “Investments in Humans, Technological Diffusion, and Economic Growth.” American Economic Review 56:69-75. Oreopoulos, Philip. 2003. “Do Dropouts Drop Out Too Soon? International Evidence from Changes in School leaving Laws.” NBER Working Paper No. 10155. Oreopoulos, Philip. 2006a. “The compelling effects of compulsory schooling: Evidence from Canada” Canadian Journal of Economics 39 (February) 22-52. Oreopoulos, Philip. 2006b. “Average Treatment Effects of Education when Compulsory School Laws Really Matter.” American Economic Review 96: 152–175. Riddell, W. Craig. 2007. “The Impact of Education on Economic and Social Outcomes: An Overview of Recent Advances in Economics” in Fulfilling Potential, Creating Success: Perspectives on Human Capital Development edited by Garnett Picot, Ron Saunders and Arthur Sweetman. Montreal: McGill-Queen’s University Press, pp. 55100. Schmidt, Stefanie. 1996. “School Quality, Compulsory Education Laws, and the Growth of American High School Attendance, 1915-1935.” Ph.D. dissertation, MIT. Schultz, Theodore. 1964. “Transforming Traditional Agriculture.” New Haven: Yale University Press.

25

Schultz, Theodore. 1975. “The Value of the Ability to Deal With Disequilibria.” Journal of Economic Literature :827-846. Stock, James, and Motohiro Yogo. 2003. “Testing for Weak Instruments in Linear IV Regression.” Working paper, Harvard University. Stoneman, Paul, and Myung Joong Kwon. 1996. “Technology Adoption and Firm Profitability.” The Economic Journal 106:952-962. Weinberg, Bruce A. 2004. “Experience and Technology Adoption.” Ohio State University, Working Paper. Welch, Finis. 1970. “Education in Production.” Journal of Political Economy 78:35-59. Welch, Finis. 1973. “Education, Information, and Efficiency.” NBER Working Paper #1. Wozniak, Gregory D. 1987. “Human Capital, Information, and the Early Adoption of New Technology.” Journal of Human Resources 22:101-112.

26

Figure 1 OLS Regression-Adjusted Probability of Computer Use in the Job by Years of Schooling Data source: Workplace and Employee Survey (1999-2004) Number of observations: 99,871 OLS Regression-Adjusted Probability of Com puter Use in the Job by Years of Schooling

Probability of Computer Use

0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 9

10

11

12

13

14

15

16

17

18

19

Years of Schooling

Note: Regression-adjusted probabilities of computer use in the job were obtained by OLS on pooled cross-sectional data conditioning on survey year, province/territory, a quartic in age, five experience groups (years of experience 1-9, 10-19, 20-29, 30-39, and 40 or above), gender, marital status, union coverage, 13 dummy variables for industry, 5 dummy variables for occupation, and size of the firm. The graph displays the coefficient estimates on the complete set of schooling dummies with those with eight years of schooling or less as the base category.

27

Figure 2 Random Effects Regression-Adjusted Probability of Computer Use in the Job by Years of Schooling Data source: Workplace and Employee Survey (1999-2004) Number of observations: 99,871 Random Effects Regression-Adjusted Probability of Com puter Use in the Job by Years of Schooling

Probability of Computer Use

0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 9

10

11

12

13

14

15

16

17

18

19

Years of Schooling

Note: Regression-adjusted probabilities of computer use in the job were obtained by random effects estimation on panel data conditioning on survey year, province/territory, a quartic in age, five experience groups (years of experience 1-9, 10-19, 20-29, 30-39, and 40 or above), gender, marital status, union coverage, 13 dummy variables for industry, 5 dummy variables for occupation, and size of the firm. The graph displays the coefficient estimates on the complete set of schooling dummies with those with eight years of schooling or less as the base category.

28

Table 1 Descriptive Statistics for Analyses Based on Workplace and Employee Survey (1999-2004)

Variable Years of schooling High-school graduation Married Male Age Years of work experience Union Firm size: 1-19 Firm size: 20-99 Firm size: 100-499 Firm size: 500 or more School leaving age = 14 School leaving age = 15 School leaving age = 16 School entry age = 6 School entry age = 7 School entry age = 8 Number of observations

Analyses on computer use/computer use experience 13.940 (2.352) 0.842 (0.365) 0.599 (0.490) 0.562 (0.496) 41.362 (10.296) 18.635 (10.523) 0.315 (0.465) 0.220 (0.414) 0.331 (0.470) 0.286 (0.452) 0.164 (0.370) 0.039 (0.194) 0.388 (0.487) 0.573 (0.495) 0.577 (0.494) 0.379 (0.485) 0.043 (0.204) 99,871

Analyses on upgrade in technology 13.917 (2.123) 0.851 (0.356) 0.585 (0.493) 0.694 (0.461) 39.694 (10.151) 17.973 (10.335) 0.404 (0.491) 0.160 (0.367) 0.307 (0.461) 0.343 (0.475) 0.190 (0.393) 0.022 (0.147) 0.351 (0.477) 0.627 (0.484) 0.577 (0.494) 0.389 (0.487) 0.033 (0.179) 12,823

Note: Standard deviations are reported in parentheses.

29

Analyses on upgrade in technological device 13.591 (2.258) 0.803 (0.398) 0.587 (0.492) 0.624 (0.484) 40.388 (10.478) 18.320 (10.576) 0.353 (0.478) 0.244 (0.429) 0.346 (0.476) 0.274 (0.446) 0.136 (0.343) 0.030 (0.171) 0.325 (0.468) 0.645 (0.479) 0.539 (0.498) 0.419 (0.493) 0.041 (0.198) 23,154

Table 2 Estimates of the Effect of High-school Graduation on Computer Use in the Job Data source: Workplace and Employee Survey (1999-2004) Number of observations: 99,871 (1) (2) OLS IV First stage: dependent variable is high-school graduation School leaving age = 15 0.083** (0.027) School leaving age = 16 0.060* (0.029) School entry age = 6 0.045* (0.023) School entry age = 7 0.058* (0.027) F-statistic for exclusion 3.97 of instruments p-value

(3) Random Effects

0.003

Second stage: dependent variable is an indicator variable for whether the employee uses computer in the job Mean of dependent variable: 0.678 High-school graduation

0.162** (0.013)

0.321 (0.233)

0.211** (0.006)

Note: All regressions control province/territory, a quartic in age, five experience groups (years of experience 1-9, 10-19, 20-29, 30-39, and 40 or above), gender, marital status, union coverage, 13 dummy variables for industry, 5 dummy variables for occupation, and size of the firm. Estimations using the cross-sectional data further control the survey year while estimations using the panel data include dummy variables for each of the three panels. Robust standard errors are reported in parentheses. All estimations correct for clustering by workplaces while IV estimation further corrects for clustering by province of residence and year of birth. Estimations using the cross-sectional data are weighted by the final employee weights. *Significant at the 5% level. **Significant at the 1% level.

30

Table 3 Estimates of the Effect of Years of Schooling on Computer Use in the Job Data source: Workplace and Employee Survey (1999-2004) Number of observations: 99,871 (1) (2) OLS IV First stage: dependent variable is years of schooling School leaving age = 15 0.582** (0.154) School leaving age = 16 0.452** (0.166) School entry age = 6 0.169 (0.138) School entry age = 7 0.527** (0.165) F-statistic for exclusion 8.23 of instruments p-value

(3) Random Effects

0.000

Second stage: dependent variable is an indicator variable for whether the employee uses computer in the job Mean of dependent variable: 0.678 Years of schooling

0.037** (0.002)

0.058* (0.028)

0.043** (0.001)

Note: All regressions control province/territory, a quartic in age, five experience groups (years of experience 1-9, 10-19, 20-29, 30-39, and 40 or above), gender, marital status, union coverage, 13 dummy variables for industry, 5 dummy variables for occupation, and size of the firm. Estimations using the cross-sectional data further control the survey year while estimations using the panel data include dummy variables for each of the three panels. Robust standard errors are reported in parentheses. All estimations correct for clustering by workplaces while IV estimation further corrects for clustering by province of residence and year of birth. Estimations using the cross-sectional data are weighted by the final employee weights. *Significant at the 5% level. **Significant at the 1% level.

31

Table 4 Estimates of the Effect of High-school Graduation on Computer Use Experience Data source: Workplace and Employee Survey (1999-2004) Number of observations: 99,871 (1) (2) OLS IV First stage: dependent variable is high-school graduation School leaving age = 15 0.083** (0.027) School leaving age = 16 0.060* (0.029) School entry age = 6 0.045* (0.023) School entry age = 7 0.058* (0.027) F-statistic for exclusion 3.97 of instruments p-value

(3) Random Effects

0.003

Second stage: dependent variable is number of years using a computer at work Mean of dependent variable: 7.751 High-school graduation

2.991** (0.159)

10.686* (4.225)

3.168** (0.096)

Note: All regressions control province/territory, a quartic in age, five experience groups (years of experience 1-9, 10-19, 20-29, 30-39, and 40 or above), gender, marital status, union coverage, 13 dummy variables for industry, 5 dummy variables for occupation, and size of the firm. Estimations using the cross-sectional data further control the survey year while estimations using the panel data include dummy variables for each of the three panels. Robust standard errors are reported in parentheses. All estimations correct for clustering by workplaces while IV estimation further corrects for clustering by province of residence and year of birth. Estimations using the cross-sectional data are weighted by the final employee weights. *Significant at the 5% level. **Significant at the 1% level.

32

Table 5 Estimates of the Effect of Years of Schooling on Computer Use Experience Data source: Workplace and Employee Survey (1999-2004) Number of observations: 99,871 (1) (2) OLS IV First stage: dependent variable is years of schooling School leaving age = 15 0.582** (0.154) School leaving age = 16 0.452** (0.166) School entry age = 6 0.169 (0.138) School entry age = 7 0.527** (0.165) F-statistic for exclusion 8.23 of instruments p-value

(3) Random Effects

0.000

Second stage: dependent variable is number of years using a computer at work Mean of dependent variable: 7.751 Years of schooling

0.601** (0.030)

1.714** (0.482)

0.617** (0.016)

Note: All regressions control province/territory, a quartic in age, five experience groups (years of experience 1-9, 10-19, 20-29, 30-39, and 40 or above), gender, marital status, union coverage, 13 dummy variables for industry, 5 dummy variables for occupation, and size of the firm. Estimations using the cross-sectional data further control the survey year while estimations using the panel data include dummy variables for each of the three panels. Robust standard errors are reported in parentheses. All estimations correct for clustering by workplaces while IV estimation further corrects for clustering by province of residence and year of birth. Estimations using the cross-sectional data are weighted by the final employee weights. *Significant at the 5% level. **Significant at the 1% level.

33

Table 6 Estimates of the Effect of Education on Upgrade in Technology Data source: Workplace and Employee Survey (1999-2004) Number of observations: 12,823 (1) OLS

(2) Random Effects

Dependent variable: an indicator variable for whether there is upgrade or change in computer-controlled or -assisted technology in past twelve months Mean of dependent variable: 0.385 Main independent variable is high-school graduation High-school graduation

0.057* (0.027)

0.062** (0.012)

Main independent variable is years of schooling Years of schooling

0.011* (0.005)

0.012** (0.002)

Note: All regressions control province/territory, a quartic in age, five experience groups (years of experience 1-9, 10-19, 20-29, 30-39, and 40 or above), gender, marital status, union coverage, 13 dummy variables for industry, 5 dummy variables for occupation, and size of the firm. Estimations using the cross-sectional data further control the survey year while estimations using the panel data include dummy variables for each of the three panels. Robust standard errors are reported in parentheses. All estimations correct for clustering by workplaces. Estimation using the cross-sectional data is weighted by the final employee weights. *Significant at the 5% level. **Significant at the 1% level.

34

Table 7 Estimates of the Effect of Education on Upgrade in Technological Device Data source: Workplace and Employee Survey (1999-2004) Number of observations: 23,154 (1) OLS

(2) Random Effects

Dependent variable: an indicator variable for whether there is upgrade or change in any other machine or technological device in past twelve months Mean of dependent variable: 0.237 Main independent variable is high-school graduation High-school graduation

0.036* (0.016)

0.034** (0.007)

Main independent variable is years of schooling Years of schooling

0.012** (0.003)

0.010** (0.001)

Note: All regressions control province/territory, a quartic in age, five experience groups (years of experience 1-9, 10-19, 20-29, 30-39, and 40 or above), gender, marital status, union coverage, 13 dummy variables for industry, 5 dummy variables for occupation, and size of the firm. Estimations using the cross-sectional data further control the survey year while estimations using the panel data include dummy variables for each of the three panels. Robust standard errors are reported in parentheses. All estimations correct for clustering by workplaces. Estimation using the cross-sectional data is weighted by the final employee weights. *Significant at the 5% level. **Significant at the 1% level.

35