Printer Friendly

Firms' expectations about the impact of AI and robotics: evidence from a survey.

I. INTRODUCTION

This study presents new evidence on the impact of artificial intelligence (AI), robotics, and big data, sometimes referred to as the "fourth industrial revolution." Specifically, we collect original survey data from more than 3,000 Japanese firms operating in both the manufacturing and service sectors to investigate their views about the impacts that these new technologies may have on future business and employment. The focus of this study is on the complementarity between these new technologies and workforce skills.

Since the 1990s, productivity growth has accelerated in the United States owing to the "IT revolution." However, some recent studies have noted that the productivity effects of traditional types of IT had already been exhausted by the mid-2000s (e.g., Bergeaud, Cette, and Lecat 2016; Fernald 2015). However, development and diffusion of the next generation of general-purpose technologies (GPTs), including AI and robotics, may substantially affect the future economy and society.

Under these circumstances, the Japanese government has begun efforts to develop and diffuse robotics and AI technologies. The Robot Revolution Initiative Council was established in 2014 and published a report in 2015 titled "New Robot Strategy," which includes a 5-year action plan to actualize the robot revolution. The Artificial Intelligence Research Center was established in the National Institute of Advanced Industrial Science and Technology in 2015. The purpose of this center is to promote research by industry and research institutions of basic AI technologies and their applications to solving real-world problems. The Japan Revitalization Strategy 2015, which is the core growth strategy of the Japanese government, seeks to modify industrial and employment structures through the utilization of the Internet of things (IoT), big data, and AI.

In spite of enthusiastic interest in the AI revolution in policy circles, there is a lag in economic research on AI and robotics. Currently, there are two types of studies on this subject: (1) theoretical arguments from the viewpoint of economic growth theory and (2) inferences from past innovations voiced mainly by labor economists. In the field of growth theory, the "singularity" hypothesis has been discussed widely. According to a standard endogenous growth model, if AI and robots increasingly were to replace labor, then capital share would rise, and the economic growth rate would accelerate. If AI and robots were to replace labor completely, growth rates would explode (Fernald and Jones 2014). However, using economic growth theory and available data, Nordhaus (2015) conducts several calculations on whether we are rapidly approaching singularity and concludes that we will not reach this point in the near future.

In the field of labor economics, substitution of human labor by AI and robots is hotly discussed. Although this discussion is a natural extension of numerous studies on the relationship between information and communications technology (ICT) and labor, the estimation by Frey and Osborne (2013) on the number of jobs at risk of replacement by future computerization, including advances in machine learning and mobile robotics, has attracted the attention of media and policy practitioners. According to Frey and Osborne (2013), roughly 47% of total U.S. employment is at risk from computerization. However, according to Autor (2015), automation and technological progress have not led to the obsolescence of human labor. In fact, automation and labor are highly complementary, particularly for employees that have adaptable, creative, and solutions-oriented skills. Based on historical lessons learned since the industrial revolution, Mokyr, Vickers, and Ziebarth (2015) argue that computers and robots will create new products and services and that these innovations will create currently unimaginable new occupations.

However, these arguments are inferences from past experiences and are not based on actual data or information regarding AI and robotics. (1) The purpose of this study is to present new empirical findings on this issue using data from an original survey of Japanese firms. Special attention is paid to the complementarity betveen human resources skills and Al-related technologies. There are numerous studies on the complementarity/substitutability of ICT and the skills of workers. Earlier studies have produced evidence based on the skill-biased nature of ICT, which indicates that skilled labor and ICT are complementary (e.g., Autor, Katz, and Krueger 1998; Bresnahan, Brynjolfsson, and Hitt 2002; Doms, Dunne, and Troske 1997; Krueger 1993). More recent studies (e.g., Autor, Katz, and Kearney 2006, 2008; Goos and Manning 2007; Goos, Manning, and Salomons 2014; Van Reenen 2011) have indicated that ICT substitutes routine tasks conducted by mid-skilled employees, which results in the "polarization" of the labor market. However, the relationship between Al-related technologies and human skills has not yet been studied explicitly.

The results of this study can be summarized as follows. First, a number of Japanese firms have intentions to use big data for their businesses and have positive expectations about the impacts of AI and robotics on their business. Second, we observe complementarity between Alrelated technologies and the skill level of firm employees, similar to the findings of earlier studies on the complementarity between ICT and highly skilled workers. Importantly, in the case of Al-related technologies, the complementarity is more prominent for employees with postgraduate education. Third, firms that operate in global markets tend to respond positively to the impact of Al-related technologies.

The remainder of this article is organized as follows. Section 11 explains the survey data used in this study and the method of analysis. Section III reports the descriptive findings on firms' views on the use of big data as well as the expected impacts of AI and robotics. Thereafter, it presents regression results on the relationship between various firm characteristics and firms' views on the new technologies. Section IV summarizes the study's conclusions and presents the implications of the research.

II. SURVEY DESIGN AND ANALYSIS METHOD

The data used in this study originate from the Survey of Corporate Management and Economic Policy (SCMEP) conducted by the Research Institute of Economy, Trade, and Industry. The survey questionnaire was mailed to and collected from a variety of public and private Japanese firms operating in both the manufacturing and service industries between October and December 2015. The survey is designed to be linked to the Basic Survey of Japanese Business Structure and Activities (BSJBSA. conducted by the Ministry of Economy, Trade and Industry). The BSJBSA, an annual survey started in 1991, accumulates representative statistics for all Japanese firms with 50 or more regular employees engaged in mining, manufacturing, electricity and gas, wholesale, retail, and several service industries to provide a comprehensive picture of Japanese firms. Since the BSJBSA is one of the "fundamental statistical surveys" designated as such by the Statistics Act, firms have obligations to report back and the response rate of the BSJBSA is higher than 85%. Approximately 30,000 firms are surveyed every year. We randomly chose 15,000 firms from the registered list of the BSJBSA in 2014, excluding firms classified in mining and utilities, and sent the questionnaire of the SCMEP.

A total of 3,438 firms responded to the SCMEP (response rate of 22.9%). The industrywise breakdown of firms is as follows: manufacturing 1,647 (48.1%), ICT 199 (5.8%), wholesale 639 (18.6%), retail 403 (11.8%), and services 395 (11.5%). (2) In order to check the representativeness of sample firms, Panel A of Table I presents the numbers and percentages of firms that responded to the SCMEP, and the entire BSJBSA firms by industry. The percentages of the entire sample of BSJBSA firms in 2014 by industry (excluding firms classified in mining and utilities) are as follows: manufacturing 43.5%, ICT 8.7%, wholesale 19.1%, retail 11.9%, and services 14%. The industry distribution of the SCMEP sample is not much different from the BSJBSA firms overall.

Furthermore, in order to verify the representativeness of the sample, panels B-D of Table 1 report the means and standard deviations of the firm size, firm age, and the ratio of nonstandard employees by industry for the SCMEP and BSJBSA firms. (3) The table indicates that the mean size of the sample firms covered by the SCMEP is smaller, and the mean ratio of nonstandard employees is higher than the figures for all the BSJBSA firms, but that the quantitative differences are small. The means and standard deviations of firm age of the SCMEP sample are similar to those observed from all the BSJBSA firms.

It should be mentioned, however, that (1) small firms of less than 50 regular employees and (2) firms that have no establishment classified in mining, manufacturing, electricity and gas, wholesale, retail, and the designated service industries--for example, firms engaged only in industries such as construction, transport, banking and finance, or healthcare--are not covered by the BSJBSA. (4) Consequently, the results should be interpreted as representing almost all medium and large Japanese firms active in industries covered by the BSJBSA.

The questionnaire of the SCMEP is wide-ranging, but in this study, we focus on three questions related to big data, AI, and robotics, as well as firm characteristics available from the survey, such as industry, firm size (total number of employees), and employee composition.

The specific wording of the three multiple-choice questions regarding Al-related technologies should be described in detail. The first is a simple query regarding the use of big data: "What does your firm think about big data?" The four possible responses are (1) "already using for business," (2) "intend to use in future business," (3) "not related to our business," and (4) "have no idea." We should note that although AI is not a prerequisite for using big data, AI and big data are complementary in business applications. That is, the availability of big data will enhance the use of AI on one hand, and the progress of AI technologies will accelerate the accumulation of big data on the other hand.

The second question enquires about the possible impact of AI and robotics on business: "What does your firm think about the impact of the development and diffusion of AI and robotics on the future business of your firm?" The five possible responses are (1) "significantly positive impact," (2) "positive impact," (3) "neither positive nor negative," (4) "negative impact," and (5) "significantly negative impact."

The third question is about the possible impact of AI and robotics on employment: "What does your firm think about the impact of the development and diffusion of AI and robotics on the future employment of your firm?" The four possible responses are (1) "increase in the number of employees," (2) "decrease in the number of employees," (3) "no impact on the number of employees," and (4) "have no idea."

In this study, we first report simple tabulation results of the responses to each question and then cross-tabulate the responses with the education level of the workforce. Next, we run ordered probit models to analyze the firm characteristics associated with the respondents' views regarding big data, AI, and robotics. In order to check the robustness of the ordered probit estimation results, we run multinomial probit regressions. The firm characteristics used as explanatory variables include industry (manufacturing, ICT, wholesale, retail, services, and other industries), firm size (log number of employees), geographic market area of the firms' products/services (city, prefecture, Japan, Asia, and the world), and the existence of labor unions. Industry and geographic market area are dummy variables, where manufacturing industry and city are used as reference categories. Under the rapid development and diffusion of Al-related technologies worldwide, we expect that firms engaged in global markets tend to have positive views about the business application of Al-related technologies.

It is important to note that the SCMEP collects relatively rich information on the characteristics of the firms' employees. Specifically, it surveys employees' education, using the ratio of employees who graduated from university or higher and the ratio of employees holding postgraduate degrees as the subset; average age; female ratio; the ratio of nonstandard workers; and the existence of labor union. We analyze the association between these employee characteristics and the firms' views on Al-related technologies. We hypothesize that firms with many employees who have complementary skills with Al-related technologies would have positive views about the impacts of the new technologies. Conversely, firms with many low-skilled employees would generate negative views about the impacts of the development and diffusion of Al-related technologies on their businesses and employment.

This study does not seek to uncover a causal relationship; the purpose is simply to present new evidence from cross-sectional survey data. Major variables and their summary statistics by industry are shown in Table 2. (5) For example, the sample means of the shares of university graduates and postgraduate degree holders are 37.8% and 2.4%, and the standard deviations are 27.1% and 5.9%, respectively, indicating that skill levels are well dispersed among sample firms. By industry, ICT firms show the largest share of highly educated employees.

III. RESULTS

A. Overview

The share of firms currently using big data for their business is notably small (3%), but 18.1% of firms have intention to use big data in their future business (reported in Table S1, Supporting Information). However, a relatively large number of firms (39.5%) responded "have no idea," reflecting that business applications of big data have not yet been well understood. By industry, positive responses to using big data (the sum of the shares of firms "already using for business" and "intend to use in future business") are the highest for the ICT industry (50.5%) followed by services (27.7%), retail (26.9%), and manufacturing (20.7%). It is interesting to observe that firms in the nonmanufacturing sector generally have positive views about the use of big data, which is similar to the well-known fact that nonmanufacturing "IT-using industries" reap the benefits of the IT revolution (Oliner, Sichel, and Stiroh 2007; Pilat, Lee, and van Ark 2002; Stiroh 2002). The mean size of firms showing a positive view (816 employees) is larger than that of firms that responded "not related to our business" (253 employees), and the difference in firm size is statistically significant at the 1% level.

Responses regarding the impact of the development and diffusion of AI and robotics on future business (reported in Table S2) are as follows. Positive responses (27.5%)--the sum of "significantly positive impact" (3.9%) and "positive impact" (23.6%)--are far larger than negative responses (1.3%)--the sum of "negative impact" (1%) and "significantly negative impact" (0.3%)--although more than 70% of firms do not have a clear outlook (they responded as "neither positive nor negative"). By industry, firms operating in the ICT industry have the most positive views on the impacts of AI and robotics (42.3%) followed by manufacturing firms (32.5%). However, many firms in non-manufacturing industries also exhibit positive responses. The mean size of firms expressing positive views (607 employees) is larger than that of firms expressing negative views (298 employees), and the difference in firm size is statistically significant at the 1% level.

The expectation about the impact of AI and robotics on employment (reported in Table S3) is generally negative: 21.8% of firms responded that the development and diffusion of new technologies will decrease the number of their employees, and the share of firms expecting positive effects on their employment is notably small (3.7%). However, 28.6% of firms expect no impact of AI and robotics on employment and 45.8% of firms responded "have no idea." By industry, with the exception of the ICT industry, the number of firms expecting a negative employment effect is larger than the number expecting a positive employment effect. However, we should interpret the result carefully because, as mentioned in the introduction, innovative technologies, such as AI and robotics, may create new employment opportunities that are currently unimaginable, and technology-intensive emerging firms may create many new occupations. When cross-tabulating the results of the two questions, firms concerned about the negative impact of AI and robotics on their businesses tend to have negative views about employment effects.

Lessons learned from the IT revolution suggest that firms with relatively low-skilled employees are likely to be affected negatively by the fourth industrial revolution, and those with highly skilled employees are expected to reap the benefits of the revolution. To detect this technology-skill complementarity, we compare the relationship between firms' views on new technologies and the education levels of their employees. Firms that responded they are already using big data for business and those that intend to use big data in future business have highly educated employees (panel A, Table 3). The ratios of university graduates and postgraduate degree holders of these firms are higher than the ratios of firms responding that big data is unrelated to their business, and the differences of both are statistically significant at the 1% level.

Similar relationships can be observed regarding firms' views about the impact of AI and robotics on business (panel B. Table 3). Firms that expected positive outcomes on their businesses have higher ratios of university graduates and employees with postgraduate degrees than firms that responded with "neither positive nor negative" (the reference category), and the differences are both statistically significant. Conversely, the education levels of the employees are lower among firms that anticipated negative impacts on their businesses than the reference category, although the differences are statistically insignificant at the conventional level. This technology-skill complementarity is confirmed after controlling for firm size and industry as well as the other covariates by the ordered probit estimations reported in the next subsection.

We observe a similar pattern from the responses to the impact of AI and robotics on employment (panel C, Table 3). The ratios of highly educated employees are higher among firms that anticipate a positive impact and lower among firms that anticipate a negative impact compared to the reference firms, which responded that there was "no impact," and the differences are statistically significant.

To summarize, all of these results suggest complementarity between new Al-related technologies and the skills level of employees.

B. Estimation Results

In this subsection, we report ordered probit estimation results on the relationships between various firm characteristics and views about Al-related technologies. As explained in Section II, the explanatory variables are industry, firm size, the spatial market area of the firms' products/services. the existence of labor unions, and employee composition (education, age, gender, and type of employment). The reference categories of the dummy variables are "city" for the geographic area of the market and "manufacturing" for industry.

The regression result regarding the use of big data is shown in column (1) of Table 4. In this ordered probit estimation, firms that responded "have no idea" are removed from the sample. The dependent variable is the use of big data: "already using for business" = 3, "intend to use in future business" = 2, and "not related to our business" = 1. Accordingly, positive coefficients mean the characteristics are associated with a positive response on the use of big data. The result indicates that the larger the firm size is, the higher is the education level of the employees; in addition, the lower the average age of the employees is, the more positive firms are about using big data. (6) It is noteworthy that the coefficient for the ratio of postgraduate education is far larger than that of university graduates, suggesting that the threshold of the complementary skills for utilizing big data is relatively high.

The coefficient for female ratio is positive and significant at the 5% level. We conjecture that it is important for firms serving wide ranges of consumers to collect detailed information on the needs of consumers and that such firms may be active in employing female employees and utilizing customer data. The coefficient for the world market is positive and highly significant, confirming that globalized firms are positive about using big data after accounting for the other firm characteristics. Interest in the development and application of big data is not limited to Japan, and fierce international competition is expected in this new frontier. Thus, it is natural that firms selling their products/services globally are positive about using big data in their businesses. The coefficients for the ICT industry are positive and highly significant, but the coefficients for other service industries are generally insignificant, with the exception of a significantly negative coefficient for the wholesale industry.

An advantage of employing the ordered probit model is its full utilization of information about the order of multiple-choice questions. However, in this specification, we have to disregard firms that responded as "have no idea." In order to include this category and to confirm the robustness of the result, we conduct a multinomial probit estimation by including "have no idea" as a choice. Since the estimated coefficients from the multinomial probit regression indicate the odds relative to the base category, it is preferable to choose a category as the base for interpreting the results easily. Accordingly, we choose firms that responded as "not related to our business" as the neutral base category.

The multinomial probit regression result is presented in Table S4. The result is consistent with that obtained from the ordered probit estimation. The estimated coefficients of the explanatory variables of interest to us, such as firm size, education dummies, and mean age for the choices "already using for business" and "intend to use in future business," exhibit the expected signs (columns (1) and (2)). In addition, although some coefficients lose statistical significance, sizes of absolute values of the coefficients significant in the ordered probit model are generally larger for the choice "already using for business" than for "intend to use in future business." On the other hand, most of the coefficients for the response "have no idea" are insignificant (column (3)), indicating that disregarding this category does not fundamentally affect the ordered probit estimation result.

Next, we report the estimation results for the questionnaire regarding the impact of AI and robotics on business. In this estimation, the order of the five response categories is reversed and used as the dependent variable. For example, "5" and "4" are assigned for the response of "significantly positive impact," and "positive impact," respectively. Thus, the positive coefficient can be interpreted as positive views about the effects of AI and robotics on business. According to the estimation result (column (2) of Table 4), the larger the firm is, the higher is the ratio of employees with postgraduate education; in addition, the lower is the average age of the employees, the more positive views the firms have about the impacts of AI and robotics on their businesses. (7) However, although the coefficients for the ratio of university graduates are positive, they are statistically insignificant, suggesting that the threshold of the complementary skill necessary to take advantage of the development and diffusion of AI and robotics is relatively high.

The coefficients for the dummies for geographic market area monotonically increase as market areas widen, even after controlling for the other firm characteristics. This result confirms the complementarity between globalization and the use of new technologies. The dummies for nonmanufacturing industries are generally negative and significant, with the exception of the ICT industry. After accounting for the other firm characteristics, manufacturing firms have a more positive expectation about the impact of AI and robotics on their future business than firms operating in the service sector.

Finally, the estimation results for the impact on employment are reported in column (3) of Table 4. The dependent variable is the impact of AI and robotics on employment: "increase in the number of employees" = 3, "no impact on the number of employees" = 2, and "decrease in the number of employees" = 1. Firms that responded "have no idea" are dropped from the sample. Again, a positive coefficient means a less negative view of the impact of Al-related technologies on employment. The coefficients for the ratio of employees with postgraduate education and for the ratio of university graduates are both positive and the coefficients for the ratio of university graduates are significant at the 5% level, indicating that firms with less educated employees have more negative views on the effects of AI and robotics on their employment. The coefficients for the ratio of female employees and the ratio of nonstandard employees are negative and statistically significant, suggesting that these types of workers at the current skills level may be affected adversely by the development and diffusion of AI and robotics. Different from the previous two regression results, we do not observe a systematic relationship among the coefficients for the geographic market area. The coefficients for industry dummies are all positive and highly significant, indicating that manufacturing firms used as the reference category are more concerned about disemployment effects from the diffusion of AI and robotics.

To check the robustness of the results, we run the multinomial probit estimation by including "have no idea" as a choice (reported in Table S5). Since we use the response "no impact on the number of employees" as the base category, the result indicates the odds relative to this category. According to the estimation results, since the number of firms responding as "increase in the number of employees" is very small (3.7%), some of the coefficients for this choice exhibit unexpected signs (column (1)). For example, the coefficients for wholesale and retail dummies are negative and significant, which are inconsistent with the ordered probit regression results. However, for the response "decrease in the number of employees," the signs of the coefficients are generally consistent with the ordered probit estimation results (column (2)), with an exception in the coefficient of firm size. In particular, the estimated coefficient of the ratio of employees with university or higher education is negative and significant and the coefficient of the ratio of nonstandard employees is positive and significant. Finally, the coefficients for the response "have no idea" are generally insignificant (column (3)), but the coefficient for the ratio of employees with university or higher education is negative and significant, suggesting that disregarding this category affects the ordered probit estimation results to some extent. However, the inference that firms with less skilled employees tend to have negative views on the impact of AI and robotics on their employment holds.

The estimation results presented above using cross-sectional data cannot be interpreted as causality in an econometric sense. For example, the positive association between education of employees and views about new technologies might be the result of firms' plans to use AI and robotics and their propensity to hire highly educated young employees. The observed relationships should be interpreted as an indication of complementarity or bidirectional causality.

IV. CONCLUSION

This study investigates the views of firms about AI, robotics, and big data, as well as firms' expectations about the impacts of these new technologies on future business and employment. The analysis utilizes original survey data of more than 3,000 Japanese firms operating in both the manufacturing and service sectors. Many speculative arguments have arisen regarding the economic and social impacts of the fourth industrial revolution, but quantitative evidence on this issue has rarely been presented.

The results of this study indicate the following. First, both manufacturing and non-manufacturing firms are generally positive about using big data and expect favorable impacts of AI and robotics on their business. Since improving productivity performance is imperative to enhance the potential growth rate of advanced economies, including Japan, there are high expectations about the diffusion and application of Al-related technologies. Second, we observe complementarity between AI and the skill levels of firms' employees. Although this is similar to the findings of earlier studies on the complementarity between ICT and high skills, it is noteworthy that there is strong complementarity found at the relatively higher end of the skills distribution. This finding suggests that it is important to upgrade human capital, such as increasing the number of employees with postgraduate education, in order to reap the benefits of the new technologies. (8) Third, firms operating in global markets reported a positive outlook about the impact of Al-related technologies, suggesting that further globalization of economic activity and the development and diffusion of new technologies will proceed hand in hand.

Although this study presents novel findings on the possible impact of Al-related technologies, the analysis is limited to simple calculations from a cross-sectional survey data and it depends on firms' subjective assessments. Further research on this important topic is necessary.
ABBREVIATIONS

AI:       Artificial Intelligence
BSJBSA:   Basic Survey of Japanese Business Structure and Activities
GPT:      General-Purpose Technology
ICT:      Information and Communications Technology
IoT:      Internet of Things
SCMEP:    Survey of Corporate Management and Economic Policy


doi: 10.1111/ecin.12412

REFERENCES

Autor, D. H. "Why Are There Still So Many Jobs? The History and Future of Workplace Automation." Journal of Economic Perspectives, 29. 2015. 3-30.

Autor. D. H., L. F. Katz, and A. B. Krueger. "Computing Inequality: Have Computers Changed the Labor Market?" Quarterly Journal of Economics, 113, 1998, 1169-213.

Autor, D. H., L. F. Katz, and M. S. Kearney. "The Polarization of the U.S. Labor Market." American Economic Review, 96, 2006, 189-94.

--. "Trends in U.S. Wage Inequality: Revising the Revisionists." Review of Economics and Statistics, 90. 2008. 300-23.

Bergeaud. A., G. Cette, and R. Lecat. "Productivity Trends in Advanced Countries between 1890 and 2012." Review of Income and Wealth. 62, 2016, 420-44.

Bresnahan, T. F., E. Brynjolfsson, and L. M. Hitt. "Information Technology, Workplace Organization, and the Demand for Skilled Labor: Firm-level Evidence." Quarterly Journal of Economics, 117, 2002, 339-76.

Doms, M., T. Dunne, and K. R. Troske. "Workers. Wages, and Technology." Quarterly Journal of Economics, 112. 1997, 253-90.

Fernald. J. G. "Productivity and Potential Output Before, During. and After the Great Recession," in NBER Macroeconomics Annual 2014, edited by J. A. Parker and M. Woodford. Chicago: University of Chicago Press, 2015, 1-51.

Fernald, J. G., and C. I. Jones. "The Future of US Economic Growth." American Economic Review, 104, 2014, 44-49.

Frey, C. B., and M. A. Osborne. "The Future of Employment: How Susceptible Are Jobs to Computerisation?" Mimeo. Oxford Martin School. 2013.

Goos, M., and A. Manning. "Lousy and Lovely Jobs: The Rising Polarization of Work in Britain." Review of Economics and Statistics, 89. 2007. 118-33.

Goos. M., A. Manning, and A. Salomons. "Explaining Job Polarization: Routine-Biased Technological Change and Offshoring." American Economic Review, 104. 2014, 2509-26.

Graetz, G., and G. Michaels. "Robots at Work." CEPR Discussion Paper No. 10477. 2015.

Krueger. A. B. "How Computers Have Changed the Wage Structure: Evidence from Microdata, 1984-1989." Quarterly Journal of Economics. 108, 1993, 33-60.

Lechevalier, S., J. Nishimura, and C. Storz. "Diversity in Patterns of Industry Evolution: How an Intrapreneurial Regime Contributed to the Emergence of the Service Robot Industry." Research Policy, 43, 2014, 1716-29.

Mokyr, J., C. Vickers, and N. L. Ziebarth. "The History of Technological Anxiety and the Future of Economic Growth: Is This Time Different?" Journal of Economic Perspectives, 29. 2015, 31-50.

Morikawa, M. "Postgraduate Education and Labor Market Outcomes: An Empirical Analysis Using Micro Data from Japan." Industrial Relations, 54, 2015. 499-520.

Nordhaus. W. D. "Are We Approaching an Economic Singularity? Information Technology and the Future of Economic Growth." NBER Working Paper No. 21547, 2015.

Oliner. S. D., D. E. Sichel, and K. J. Stiroh. "Explaining a Productive Decade." Brookings Papers on Economic Activity, I, 2007. 81-152.

Pilat, D., F. Lee, and B. vanArk. "Production and Use of ICT: A Sectoral Perspective on Productivity Growth in the OECD Area." OECD Economic Studies, 35. 2002. 47-78.

Stiroh. K.J. "Information Technology and the U.S. Productivity Revival: What Do the Industry Data Say?" American Economic Review, 92. 2002. 1559-76.

Van Reenen. J. "Wage Inequality. Technology and Trade: 21st Century Evidence." Labour Economics, 18. 2011. 730-41.

SUPPORTING INFORMATION

Additional Supporting Information may be found in the online version of this article:

Table S1. Use of big data

Table S2. Impact of AI and robotics on business

Table S3. Impact of AI and robotics on employment

Table S4. Multinomial probit estimation result: use of big data

Table S5. Multinomial probit estimation result: impact of AI and robotics on employment

MASAYUKI MORIKAWA, I thank the editor (Charles Mason), anonymous reviewers, Shota Araki. Masahisa Fujita, Arata Ito. Keisuke Kondo, Yoko Konishi, Atsushi Nakajima, Hiroshi Ohashi. Akihiko Tamura. Willem Thorbecke. Isamu Yamauchi, and the seminar participants at RIETI for their helpful comments and suggestions. I am grateful to the Ministry of Economy, Trade and Industry for providing the microdata from the Basic Survey of Japanese Business Structure and Activities employed in this study. Any errors are my own. This research is supported by the JSPS Grants-in-Aid for Scientific Research (B26285063). Morikawa: Research Institute of Economy. Trade and Industry (RIETI), Tokyo, 100-8901, Japan. Phone 3-3501-1362, Fax 3-3501-8391. E-mail morikawamasayuki@rieti.go.jp

(1.) Exceptions include international comparison of AIrelated patent applications (Lechevalier, Nishimura, and Storz 2014), and empirical research on the economic impacts of industrial robots (Graetz and Michaels 2015).

(2.) One hundred and forty-four firms (4.2%) in the SCMEP and 844 firms (2.8%) in the BSJBSA are classified as "other industries." Industry classifications of the remaining 11 firms are unknown. The percentages are calculated by excluding "unknown" firms from the denominator.

(3.) Other employee characteristics--education, age, and female ratio--and information on whether firms are unionized or not. are unavailable in the BSJBSA.

(4.) The designated service industries include eating and drinking services, goods rental and leasing, scientific research, professional and technical services, and amusement and recreation services.

(5.) We removed a small number of firms whose reported number of standard employees was larger than the total number of employees as well as firms with extremely low figures of average age of their employees as outliers.

(6.) We conduct an estimation using firm age (years since establishment) as an additional explanatory variable by linking the survey data with microdata of the BSJBSA. The coefficient for firm age is statistically insignificant, and the size and significance level of the coefficient for the age of employees are unaffected (not reported in the table). In other words, firms employing young workers have positive intentions to use big data, irrespective of firm age.

(7.) The estimation result using firm age as an additional explanatory variable is similar to the result for the use of big data. The coefficient for firm age is statistically insignificant and the size and significance level of the coefficient for the mean age of employees remain highly significant.

(8.) Morikawa (2015), an empirical study on the return to postgraduate education in Japan, indicates that the rate of return to postgraduate education exceeds 10%.
TABLE 1

Comparison of the SCMEP and BSJBSA Firms by Industry

                              (1) All      (2) Manufacturing

Variables                  SCMEP   BSJBSA   SCMEP   BSJBSA

A. Number         Number   3,427   30,031   1,647   13,053
of firms          %        100.0   100.0    48.1     43.5

B. Number of      Mean      380     482      360     404
regular           SD       2,082   1,939    2,741   1,618
employees

C. Firm age       Mean     44.8     43.9    46.7     49.0
                  SD       20.7     21.0    21.5     20.3

D. Nonstandard    Mean     0.243   0.205    0.215   0.146
ratio (%)         SD       0.248   0.246    0.210   0.181

                               (3) ICT      (4) Wholesale

Variables                  SCMEP   BSJBSA   SCMEP   BSJBSA

A. Number         Number    199    2.612     639    5,741
of firms          %         5.8     8.7     18.7     19.1

B. Number of      Mean      314     382      181     261
regular           SD        937    1.162     306     614
employees

C. Firm age       Mean     36.7     29.3    47.7     48.3
                  SD       19.6     16.3    19.7     20.7

D. Nonstandard    Mean     0.120   0.086    0.174   0.157
ratio (%)         SD       0.158   0.147    0.196   0.186

                             (5) Retail      (6) Services

Variables                  SCMEP   BSJBSA   SCMEP   BSJBSA

A. Number         Number    403    3,570     395    4,211
of firms          %        11.8     11.9    11.5     14.0

B. Number of      Mean      744     880      522     768
regular           SD       1.707   3,211    1.566   2,857
employees

C. Firm age       Mean     42.9     40.5    39.3     33.9
                  SD       19.0     20.7    19.5     17.2

D. Nonstandard    Mean     0.407   0.383    0.366   0.349
ratio (%)         SD       0.311   0.308    0.302   0.315

Notes: The figures are calculated from the SCMEP in 2015 and the BSJBSA
in 2014. The BSJBSA firms in this table exclude firms classified
in mining and utilities.

TABLE 2

Variables and Summary Statistics by Industry

                                         (1) All

Variables                        Mean     SD     No. of obs.

Regular employees                 380    2.082      3,145
Ratio of university or higher    0.378   0.271      2,996
Ratio of postgraduates           0.024   0.059      2,847
Mean age of employees            40.62   4.34       3,159
Female ratio                     0.300   0.204      3,145
Nonstandard ratio                0.243   0.248      3,145
Labor union dummy                0.318   0.466      3,281

                                      (2) Manufacturing

Variables                        Mean     SD     No. of obs.

Regular employees                 360    2,741      1,518
Ratio of university or higher    0.277   0.208      1.461
Ratio of postgraduates           0.026   0.057      1,406
Mean age of employees            40.98   3.89       1,525
Female ratio                     0.259   0.182      1.518
Nonstandard ratio                0.215   0.210      1.518
Labor union dummy                0.403   0.491      1,582

                                           (3) ICT

Variables                        Mean     SD     No. of obs.

Regular employees                 314     937        184
Ratio of university or higher    0.672   0.230       167
Ratio of postgraduates           0.054   0.067       164
Mean age of employees            37.84   3.70        184
Female ratio                     0.229   0.119       184
Nonstandard ratio                0.120   0.158       184
Labor union dummy                0.209   0.408       191

                                       (4) Wholesale

Variables                        Mean     SD     No. of obs.

Regular employees                 181     306        595
Ratio of university or higher    0.513   0.274       573
Ratio of postgraduates           0.016   0.053       537
Mean age of employees            41.18   3.79        598
Female ratio                     0.296   0.152       595
Nonstandard ratio                0.174   0.196       595
Labor union dummy                0.174   0.379       615

                                        (5) Retail

Variables                        Mean     SD     No. of obs.

Regular employees                 744    1,707       365
Ratio of university or higher    0.367   0.261       343
Ratio of postgraduates           0.005   0.017       320
Mean age of employees            39.18   4.90        366
Female ratio                     0.430   0.271       365
Nonstandard ratio                0.407   0.311       365
Labor union dummy                0.396   0.490       381

                                        (6) Services

Variables                        Mean     SD     No. of obs.

Regular employees                 522    1,566       350
Ratio of university or higher    0.429   0.303       329
Ratio of postgraduates           0.031   0.078       308
Mean age of employees            40.52   5.68        351
Female ratio                     0.383   0.234       350
Nonstandard ratio                0.366   0.302       350
Labor union dummy                0.206   0.405       369

Note: The denominator to calculate the ratios of female and nonstandard
employees is the total number of regular employees,
including part-time workes.

TABLE 3

Firms' Responses on AI-Related Technologies and Education of Employees

                                    (1) Ratio of       (2) Ratio of
                                University or Higher   Postgraduates

A. Use of big data
1) Already using for business        51.7% ***           7.3% ***
2) Intend to use in future           46.4% ***           3.4% ***
  business
3) Not related to our                  35.6%               1.9%
  business (reference)

B. Impact on business
1) Significantly positive             41.8% **           4.6% ***
2) Positive                           39.3% **           3.6% ***
3) Neither positive nor                37.2%               1.8%
  negative (reference)
4) Negative                            35.7%               0.9%
5) Significantly negative              33.6%               0.6%

C. Impact on employment
1) Increase                          50.3% ***           4.3% ***
2) No impact (reference)               41.2%               2.6%
3) Decrease                          33.1% ***            2.1% *

***, **, and * indicate statistical significance relative
to the reference categories at the 1%, 5%, and 10% levels, respectively.

TABLE 4

Ordered Probit Estimation Results: Firm Characteristics and Responses
to AI-Related Technologies

Variables                (1)             (2)               (3)
                     Use of Big       Impact of       Impact of AI
                        Data       AI and Robotics   and Robotics on
                                     on Business       Employment

Ln employees         0.2195 ***      0.0783 ***          0.0290
                      (0.0356)        (0.0289)          (0.0369)
University or        0.6067 ***        0.0950           0.3446 **
higher                 0.1396          0.1135            0.1426
Postgraduate         1.6271 ***      2.1829 ***          0.3126
                       0.5323          0.4498            0.5646
Mean age             -0.0262 ***     -0.0166 ***         0.0078
                      (0.0081)        (0.0062)          (0.0076)
Female ratio          0.4447 **        -0.0611         -0.4261 **
                       -0.1976        (0.1528)          (0.1907)
Nonstandard ratio      -0.1162         0.0418          -0.6908 ***
                       -0.1724        (0.1292)          (0.1669)
Union dummy           -0.1267 *      -0.1146 **          0.0415
                       -0.0753        (0.0581)          (0.0728)
Market: prefecture     0.1799          0.2210           0.2939 *
                      (0.1810)        (0.1394)          (0.1641)
Market: Japan         0.3283 *        0.2479 *           0.1119
                      (0.1819)        (0.1397)          (0.1657)
Market: Asia           0.2259         0.3544 **          0.1946
                      (0.2190)        (0.1669)          (0.2030)
Market: world        0.6717 ***      0.6601 ***          0.2012
                      (0.2062)        (0.1588)          (0.1920)
ICT                   0.3299 **        0.0338          0.8717 ***
                      (0.1313)        (0.1133)          (0.1429)
Wholesale            -0.2060 **      -0.3725 ***       0.37904 ***
                      (0.0996)        (0.0779)          (0.0986)
Retail                 0.0915        -0.2296 **        0.6299 ***
                      (0.1288)        (0.1018)          (0.1272)
Services               -0.1618       -0.3747 ***       0.8788 ***
                      (0.1133)        (0.0928)          (0.1076)
Other industries       -0.0027       -0.2862 **         0.3674 *
                      (0.1715)        (0.1392)          (0.1756)
No. of obs.             1.621           2.643             1.466
Pseudo [R.sup.2]        0.090           0.046             0.072

Notes: Ordered probit estimation results with standard errors in
parentheses. Firms that responded "have no idea" are dropped
from the estimations of (1) and (3). The reference categories are
"city" for the spatial area of the market and "manufacturing" for
industry.

***, **, and * indicate statistical significance at the 1%, 5%,
and 10% levels, respectively.
COPYRIGHT 2017 Western Economic Association International
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2017 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Morikawa, Masayuki
Publication:Economic Inquiry
Article Type:Survey
Geographic Code:9JAPA
Date:Apr 1, 2017
Words:6934
Previous Article:College advising and gender.
Next Article:Sentiment bias and asset prices: evidence from sports betting markets and social media.
Topics:

Terms of use | Privacy policy | Copyright © 2019 Farlex, Inc. | Feedback | For webmasters