ICT, Internet and Worker Productivity

Abstract

This article provides a brief overview of the role of information and communication technologies (ICT) as a driver of productivity. In particular, it focuses on the diffusion of computers and the Internet at the workplace and discusses the relationship with wages, the task composition of occupations and labour productivity at the firm level.

Information and communication technologies (ICT) enhance productivity and growth, as shown by various studies at the macroeconomic and microeconomic levels (see for instance Draca et al. 2007 or Kretschmer 2012, for a comprehensive overview). As so-called general purpose technologies (Bresnahan and Trajtenberg 1995), they diffuse throughout the whole economy and enable innovation in adopting firms and sectors (see for example Brynjolfsson and Saunders 2010), leading to higher productivity.

From a technological perspective, we can differentiate three basic stages of ICT: personal computers, the Internet, and more recently, mobile Internet.

When personal computers started to diffuse to workplaces, economists became interested in analysing whether the use of computers makes workers more productive. There are basically two approaches to measuring worker productivity. One approach takes an individual perspective. It is based on the concept of wage functions and assumes that wages reflect individual productivity. The other approach takes a firm-level perspective. It builds on production functions and analyses the relationship between labour productivity and firms’ input factors, labour, non-ICT capital and ICT capital, as well as other firm characteristics. Some studies go beyond these main approaches by taking into account job-related tasks, project-based information or regional aspects.

Worker Productivity at the Individual Level

Computer Use and Wages

In his seminal paper, Krueger (1993) analyses whether workers who use a computer at work earn higher wages than workers not using a computer at work. The following kind of wage equation is estimated:

$$ \ln {W}_i={\beta}_0+{\beta}_1\;{\mathrm{Computer}}_i+{\beta}_2{X}_i+{u}_i $$

where W_i is hourly wage for employee i, Computer_i is a dummy variable taking the value one if employee i uses a computer at work and the value zero otherwise, and X_i represents a vector of employee characteristics, such as education, age and gender. Krueger uses data from the Current Population Survey (CPS) collected in the USA in 1984 and 1989 and from the High School and Beyond Survey for the years 1980, 1982, 1984, 1986. The findings reveal that the wage rate of computer users is about 10–15% higher than that of the non-computer users. In 1984, about 25% of the employees used computers at work, whereas the number has increased to 37% in 1989. Krueger points out that it is not clear from the data and from his analysis whether using computers makes employees more productive and therefore means that they earn higher wages, or whether there are unobserved individual characteristics that correlate with computer use and wages. For instance, high-skilled employees may have abilities that make them earn higher wages and increase the probability of using computers. Owing to the fact that Krueger has only cross-sectional data he cannot control for individual unobserved heterogeneity using fixed effects regression.

DiNardo and Pischke (1997) replied provocatively to Krueger’s paper by asking: ‘Have pencils changed the wage structure too?’. They replicate Krueger’s estimations for the USA but extend the analysis to a further cross-section (1993) of the Current Population Survey. Furthermore, they use German employee-level data from the Qualification and Career Survey for the time periods 1979, 1985/86 and 1991/92 and compare the results to those found for the USA. In addition to looking at computers at work, the authors also consider the effect of other working tools such as calculators, telephones, writing tools like pens or pencils, and sitting on the job. They call these tools ‘white-collar tools’, since they are more probably used by white-collar than by blue-collar workers. The results for Germany, with respect to the wage premium for computer use, confirm the results found for the USA. However, similar wage differentials are found for pens and pencils, calculators, telephones and working while sitting. The authors draw the conclusion that the wage differential found for computer use cannot reflect true returns to computer use or skills, since otherwise no similar effects would have been found for the other white-collar tools. Similar results are found by Borghans and Ter Weel (2004) for British employee data collected in 1997. Computer use only at the advanced level is related to wage premiums, whereas mathematics and writing skills show significant wage premiums.

The study by Entorf et al. (1999) has the advantage over previous studies that it relies on panel linked employer–employee data for 1991 to 1993. This allows individual fixed effects that control for unobserved heterogeneity across employees to be taken into account. Moreover, the authors can observe what happens if an employee starts using a computer at work. The wage differential observed in the cross-sectional French data is more or less the same as in the USA and lies between 15 and 20%. Panel regressions, however, show that this wage differential decreases to only up to 2%, a result that confirms evidence found before for the 1980s (Entorf and Kramarz 1997). Moreover, employees were already better paid before they started using computers. This result implies that firms allocate computers to selected workers and these workers seem to have unobserved skills that are complementary with computer use. According to the French dataset, this seems to hold particularly for low-skilled workers. Wage premium estimates are summarised in Table 1.

ICT, Internet and Worker Productivity, Table 1 Overview of wage premium estimates

The Task-Based Approach

In order to obtain deeper insights into the unobserved characteristics that are complementary with computer use, Autor et al. (2003) suggested a so-called task-based approach. This approach assumes that work consists of a series of routine and non-routine tasks. While manual and cognitive routine tasks can be performed and thus substituted by a computer, non-routine tasks cannot. Analytical and interactive non-routine cognitive tasks are, by contrast, supported (i.e. complemented) by computers. For instance, doing research or advising customers are non-routine cognitive tasks that can be better performed using a computer. By contrast, bookkeeping or controlling machines can be performed by computers (see Spitz-Oener 2006, p. 243, for a classification of tasks). Autor et al. (2003) and Spitz-Oener (2006) have shown for the USA and Germany, respectively, that the diffusion of computers goes hand in hand with a shift in the content of work from manual and cognitive routine tasks towards non-routine cognitive tasks. This shift implies an increase in the demand for skilled employees (in line with the hypothesis that technological change is skill-biased), leading to increased wages for these skills. Another, more direct, channel for how computers affect wages is that employees become increasingly productive when complementing their tasks with computer use.

This latter aspect corresponds to the complementarities between computer use and organisational change found at the firm level by Bresnahan et al. (2002) (see next section). For the empirical analysis of the task-based approach, task compositions within occupations are calculated for each employee i (see Spitz-Oener 2006, p. 242 for the following definition):

$$ {\mathrm{Task}}_{ijt}=\frac{\mathrm{number}\kern0.34em \mathrm{of}\kern0.34em \mathrm{activities}\kern0.34em \mathrm{in}\kern0.34em \mathrm{category}\;j\;\mathrm{performed}\kern0.22em \mathrm{by}\kern0.22em i\kern0.22em \mathrm{at}\kern0.22em \mathrm{time}\kern0.22em t}{\mathrm{total}\kern0.34em \mathrm{number}\kern0.34em \mathrm{of}\kern0.34em \mathrm{activities}\kern0.34em \mathrm{in}\kern0.34em \mathrm{category}\kern0.22em j\kern0.22em \mathrm{at}\kern0.22em \mathrm{time}\kern0.220em t} $$

where j represents the tasks, i.e. j = 1 (nonroutine analytic tasks), j = 2 (nonroutine interactive tasks), j = 3 (routine cognitive tasks), j = 4 (routine manual tasks) and j = 5 (nonroutine manual tasks), and t reflects the cross-section for which data is available. According to the example given by Spitz-Oener, if employee i performs two out of four analytical activities, his or her analytical task measure is 50. Based on these task measures the change in the shares of tasks within occupations over time can be calculated, showing which of the tasks have become more or less important.

Spitz-Oener (2008) extends her previous analysis conducted in 2006 to take up the issue raised by DiNardo and Pischke (1997), i.e. that there is also an effect of pencil use on wages. She shows for West German employee data, again from the Qualification and Career Survey in 1998/99, that wage premiums are observed for employees with skills that are complementary with computer use. In contrast with the study by DiNardo and Pischke (1997), no similar effects are found for the use of pencils. This result underpins what has been found before: computer use has shifted the task composition of occupations towards analytical and interactive tasks and away from routine cognitive and manual tasks. While computers complement the first, they tend to substitute for the latter.

Worker Productivity at the Firm Level ICT and Labour Productivity

Taking a firm-level perspective, the relationship between labour productivity and ICT can be captured by a production function approach. Output Q is related to the input factors of labour, non-ICT capital and ICT capital. Although sometimes materials explicitly are taken into account, we do not consider them here. If we assume a Cobb-Douglas form of the production function, the function for firm i looks as follows:

$$ {Q}_i=A{L}_i^{\beta_1}{K}_i^{\beta_2}{C}_i^{\beta_3} $$

where Q is output, L is labour, K is capital, C is ICT capital and A represents a technology or efficiency parameter. The parameters β₁, β₂ and β₃ represent the output elasticities of the respective input factors. Taking logarithms, setting lnA = β₀ and adding an error term u_i leads to the following equation:

$$ \ln {Q}_i={\beta}_0+{\beta}_1\ln {L}_i+{\beta}_2\ln {K}_i+{\beta}_3\ln {C}_i+{u}_i $$

Subtracting lnL from both sides results in

$$ \ln \left(\frac{Q_i}{L_i}\right)={\beta}_0+\left({\beta}_1-1\right)\ln {L}_i+{\beta}_2\ln {K}_i+{\beta}_3\ln {C}_i+{u}_i $$

where $\ln \left(\frac{Q_i}{L_i}\right)$ represents labour productivity, i.e. output per worker. This equation can be estimated by econometric methods using firm-level data. If panel data are available, an index t is added. Panel data usually allow taking account of firm-specific fixed effects and thus unobserved heterogeneity across firms. Depending on the available data, labour productivity is measured by sales per employee, sales per hour worked, value added per employee or value added per hour worked. ICT capital often is not observable in firm-level data sets. In this case, it may be approximated by ICT investment or by the percentage of employees working with computers. If panel data is available and information about ICT investment, then ICT capital stocks can be calculated according to the so-called perpetual inventory method (see for example Bloom et al. 2012, or Hempell 2005). Some studies, instead of using measures of ICT capital, analyse the relationship of labour productivity with specific ICT applications such as B2B e-commerce that are measured by dummy variables (see for example Bertschek et al. 2006).

There is meanwhile a large number of firm-level studies analysing the role of ICT for labour productivity. Draca et al. (2007) provide a comprehensive overview of the studies published between 1996 and 2005 and summarise the main results. One main finding of most studies is that labour productivity is positively and significantly related with ICT. The average of the estimated coefficients of ICT is about 5% to 6% and has increased over time (see Kretschmer 2012). This relationship, however, might be heterogeneous with respect to firms and industries, i.e. some firms or industries are more successful in employing ICT than others.

This firm-specific heterogeneity in reaping the potential of ICT might be due to differences in complementary investment in organisational capital and human capital across firms, an argument put forward for instance in Bresnahan et al. (2002) and underpinned in several further studies. The relationship between productivity and ICT is stronger if investment in ICT is supported by investment in organisational capital (for the particular role of organisational capital see for example Black and Lynch 2001, for the USA; Bertschek and Kaiser 2004, for Germany). Since ICT lowers the cost of communication, employees can communicate and exchange information more efficiently. Thus, working in teams and with a low number of hierarchies becomes more feasible and organisational structures may become more decentralised and flexible. Moreover, communicating and coordinating with customers and suppliers become easier and costs may decrease. Investment in human capital is considered to be complementary with ICT since the implementation of a new ICT system or application in a firm often requires that firms train their employees in order to be able to work with these new technologies or applications. Recent evidence on the relationship between ICT, organisational capital and human capital and its productivity-enhancing effect is presented by Bloom et al. (2012) who find that US multinationals located in Europe obtain higher productivity effects from using ICT than their non-US counterparts due to better people management practices. Bartel et al. (2007) analyse specific ICT applications in valve-producing plants, and Aral et al. (2012) present econometric evidence on a three-way complementarity between firms’ adoption of software for human capital management, performance pay and firms’ practice of human resource analytics (including worker monitoring, performance feedback, the integration of workforce support data, and talent management). Hall et al. (2012) consider investment in ICT and in research and development (R&D) as potential sources of innovation which in turn may enhance labour productivity. They use four cross sections of Italian manufacturing firms covering the period 1995–2006. The econometric results show that R&D and ICT contribute directly to labour productivity but also indirectly through enabling innovation.

One big issue empirically working economists are faced with is endogeneity. It is a priori not evident whether investment in ICT increases labour productivity or whether productivity growth implies more investment in ICT. Depending on the available datasets, the studies are more or less able to tackle this issue.

Looking at the Internet as a specific ICT, there is not so much empirical work yet as exists for ICT in general. The following section will summarise some of the empirical results.

Internet and Labour Productivity

The Internet started to diffuse to workplaces later than computers. If a firm connects to the Internet, it does not necessarily mean that all employees have Internet access, but this might rather be restricted to a certain group of persons such as the chief executive officers or the administration staff. Also, Internet use as well as computer use varies considerably between manufacturing and services industries.

Figure 1 shows the diffusion of computers and Internet access in German firms. The percentage of employees using a computer at least once per week at work has increased from 46% in 2002 to 63% in 2011. In the same period, the percentage of employees with connection to the Internet has increased from 29% in 2003 to 54% in 2011.

ICT, Internet and Worker Productivity, Fig. 1

Percentage of employees using computers and Internet in German firms, 2002–2011. In 2008, NACE code classification has changed

What does the Internet add to only having a computer? It allows access to information and connects employees with each other, facilitating the search and exchange of information. Moreover, Internet technologies or web-based applications such as wikis or collaboration platforms facilitate processing of information, documentation and cooperation.

For the case of New Zealand, Grimes et al. (2012) find based on a firm-level cross section collected in 2006 that firms using broadband Internet have a 7 to 10% higher labour productivity. By contrast, for the early phase of broadband diffusion in Germany, 2000 to 2002, Bertschek et al. (2011) find positive and significant effects of broadband on firms’ innovation activity but not on labour productivity. Polder et al. (2010) analyse the role of ICT and R&D for innovation success. Their estimations are based on three data waves of Dutch firms. ICT is measured as investment in ICT per employee. Additionally, they use a measure for Internet use (the percentage of employees having access to broadband Internet), and include e-commerce as a specific ICT application. The results show that broadband Internet is particularly important for service firms, where broadband is positively related to product and process innovation as well as to organisational innovation. By contrast, in the manufacturing sector, broadband is significant only for product and organisational innovation. For process innovation, it is e-commerce that plays a significant role. These results support the hypothesis of complementarity between ICT and innovation or organisational change. Moreover, they show that it also depends on what firms concretely do with their ICT or with their Internet access to enable innovation. This latter issue is taken up in a recent paper by Colombo et al. (2012). The authors show for a sample of small Italian firms that it is not the broadband connection itself that makes firms more productive. It depends rather on the kind of application as well as on complementary organisational and strategic changes whether or not firms profit with respect to their productivity.

Internet and Wages: A Regional Perspective

Studies looking at Internet and wages are still scarce. Forman et al. (2011) take a regional perspective. Their initial hypothesis is that Internet lowers the cost for economic engagement also in geographically isolated regions. Thus, Internet should have effects on the performance of firms and employees also in regions whose performance was comparably low before the diffusion of the Internet. The study does not look at broadband Internet itself but at business investment in advanced Internet technologies. These comprise investment in enterprise resource planning (ERP), customer service, education, extranet, publications, purchasing and technical support. The time span of the analysis is from 1995 to 2000, a time period when Internet had just started to diffuse more broadly and when there was still a lot of variation in the use of broadband Internet or Internet-based applications with respect to firms, individuals and regions. The authors use data from different sources on firms with more than 100 employees as well as county-level data.

The estimations show that although advanced Internet applications diffused widely in the USA from 1995 to 2000, the economic benefits in terms of wage growth were concentrated in a few well-performing counties only. More precisely, only 6% of US counties profited from investment in Internet technologies in terms of wage growth. This wage growth amounted to 28% from 1995 to 2000, whereas the average growth over all counties was 20%. These counties, however, had a better performance already before 1995, i.e. they were characterised by relatively high income, large population, high skills and high IT intensity. The results of the study thus do not support the initial hypothesis that the Internet contributes to economic regional inclusion, but rather imply that the Internet aggravates regional wage inequality.

Including Social Network Data

In order to analyse the relationship between multitasking, knowledge networks and productivity, the approach by Aral et al. (2012) goes beyond the firm level and the individual level. The authors focus on only one firm, a mid-size executive recruiting firm. They use detailed data on employees’ characteristics, on their project output and team membership for projects, and on email messages sent and received by these employees, i.e. on the workers’ digital network. A recruiting firm offers services, and for services, measuring output, input and productivity is harder than in manufacturing firms. The authors of the study have accounting records for all projects covering the period 2001 to 2005, including the number of projects completed and the revenue generated by individual recruiters. They measure output as the number of projects completed per month, i.e. the number of days a recruiter works on the project per month divided by the total number of days for which the project runs. Completing a project means that the recruiter has found an appropriate candidate for the client and the candidate has signed a contract. Output is set into relation with the heterogeneity of multitasking measured as the number of projects recruiters work on per month and the heterogeneity of recruiters’ contacts resulting from the work on prior projects as well as from the number of email contacts.

There are several interesting findings from this analysis: Recruiters’ output is increasing with multitasking, but only up to a certain threshold. A further increase of multitasking then implies diminishing rates of return. Although heterogeneity of contacts is negatively related with output, it complements task heterogeneity. Having access to heterogeneous information via email makes multitasking recruiters more productive. This result again supports the hypothesis of complementarity between workplace organisation and IT as well as the complementarity between specific tasks and IT.

Current Technological Trends

While computers allow for digitisation, and the Internet for connectedness, the mobile Internet, which has started to diffuse only recently, additionally offers the possibility to work at any time from any place. For example, in Germany, on average 25% of employees have broadband access via a mobile device such as a smartphone or a tablet (Statistisches Bundesamt 2011, p. 21). This development is supported by so-called cloud computing – the concentration of computing capacity, data and software in data centres that employees can connect to from anywhere. There are so far no empirical econometric studies based on large-scale data analysing whether mobile Internet adds to worker productivity additionally to computers and Internet. We can imagine what might happen if working environments get more and more flexible and independent from time and space. On the one hand, this technological opportunity, by decreasing information and communication costs, supports further decentralisation of work as suggested in Bresnahan et al. (2002). On the other hand, these flexible working environments require a high degree of self-reliance and coordination. Very probably, new studies will soon give insights into the net effects of these new technological advances.

See Also

• Internet, Economics of the

• Production Functions

• Social Networks, Economic Relevance of