Skilled labor shortages and ICT services adoption in the United States

(1)

Faculty Economics and Business

Master’s Thesis

Skilled labor shortages and ICT services adoption in

the United States

An analysis of the New Digital Economy productivity paradox

By Jerrel King

MSc International Economics & Business

Specialisation: Globalization, Growth and Development

Academic year: 2016-2017

Student number: s3133281

Primary supervisor: dr. Bart van Ark

(2)

ABSTRACT

The productivity paradox associated with the New Digital Economy of the 21st century is a complex phenomenon. While there is a new digital revolution noticeable characterized by a shift from ICT assets to ICT services purchases, U.S. productivity growth still lacks behind. Specifically, U.S. productivity growth in the period 2007-2014 significantly slowed down compared to 1998-2006. Most of this slowdown is attributed to industries that most

intensively use ICT in their production processes. Several explanations have been sought for the occurrence of such paradoxes during technological revolutions. This thesis explores the role of skilled labor shortages as a major disabling factor in the adoption process of the newly implemented ICT services. Using unique data on labor shortages, it is found that an

(3)

technologies. For instance, today’s digitized U.S. economy is significantly different from that in the past century. The introduction of Information and Communications Technology (ICT) and the Digital Revolution in the later half of the 20th century, contributed to an overall change in the economies of advanced nations. Large-scale investments in ICT assets, such as computer hardware (PCs), software and telecommunications, changed U.S. industries as these assets spread throughout the economy and became the new normal. Several scholars have attributed the revival of economy-wide U.S productivity growth in the late 20th century to these extensive investments in ICT assets. There is accumulated evidence showing that industries that made the largest investments in ICT in the 1980s and early 1990s, showed larger productivity gains post-1995 (Stiroh, 2002). Most (case) literature suggests that one of the enabling factors of IT productiveness has been the supply of skilled labor. It is often shown that the successful transition toward an IT-based work environment entails

complementary organizational investments (i.e. practices) in addition to investments in IT technology. Corrado et al. (2014) find that labor productivity growth in high ICT-intensive industries is higher in countries experiencing larger increases in intangible organizational investments. Specifically, organizational investments related to human capital are expected to play a key role as the transition to new (IT) technology involves a protracted learning process and places increasing demands on skill, education and know-how of the workforce.

Therefore, in this thesis it is argued that the technological shift toward new IT technology requires sufficient (skilled) labor.

With that being said, the U.S. economy is undergoing a new digital transformation, away from the Old Digital Economy in the 20th century toward the New Digital Economy1.

1_{The exact definition of the New Digital Economy used throughout this thesis is: digitization and digital}

transformation driven by a combination of mobile technology; ubiquitous access to internet; and the shift toward storage, analysis, and development of new applications in the cloud.

Based on Bart van Ark, Abdul Erumban, Carol Corrado, and Gad Levanon, Navigating the New Digital

Economy: Driving Digital Growth and Productivity from Installation to Deployment. The Conference Board,

(5)

This New Digital era is driven by the shift from investments in ICT assets –as was prevalent in the late 20th_{century– toward purchases of ICT services, such as software applications,}

mobile and cloud computing, and information and data processing services (big data). Van Ark et al. (2016) note that the shift toward ICT services will improve utilization of data capabilities, change business processes and create new business models. Furthermore, ICT services will affect organizations mainly by altering the way business create value in the form of cheaper, higher quality, well-tailored goods and services. This transformation is

increasingly noticeable in people’s daily lives, as benefits of digitization span far beyond mere computational capabilities. Nevertheless, these technological developments do not seem to translate into widespread productivity gains. Specifically, average U.S. productivity

growth slowed down between 2007 and 2014 (compared to 1998-2006). More strikingly, over half of this productivity slowdown is caused by most intensive ICT-using sectors (van Ark et al., 2016). These are industries that are not involved in the actual production of ICT, but are a key factor in fostering widespread productivity gains through their intensive use of ICT assets and ICT services in their production processes. Thus, especially these industries should attract skilled workers and invest in human capital in order to leverage opportunities arising from ICT services and to spur productivity growth. Similar to van Ark et al. (2016), figure 1 below shows the contribution of four industry groups to U.S. productivity growth in 1998-2006 and 2007-20142_.

Evidently, U.S. industries that witnessed the largest increase in ICT assets and ICT services intensity (i.e. most intensive ICT-using industries) were the largest contributor to the U.S productivity slowdown in the period 2007-2014. This phenomenon –known as the productivity paradox– is not an unprecedented phenomenon as it has been witnessed during prior technological transformations. Examples go back as far as the “Electrical Age” in the early 1900s and the early stages of the “Information Age” in the 1970s. Similar to the New Digital Economy, these periods were characterized by rapid technological developments without immediate widespread productivity gains. Several scholars have attempted to provide explanations for the occurrence of such paradoxes in the development process of general-purpose technologies (GPTs). Overall, it can be stated that the adoption of new technologies, and the translation of these technologies into productivity gains, involve lengthy, volatile trials. Productivity gains occurring during the infant stages of technological developments

2_{van Ark et al. (2016) distinguish most intensive and least intensive ICT-using industries based on ICT intensity}

(6)

will be minimal and restricted to a selected amount of sectors. Specifically, in this thesis it is argued that intensive ICT services-using industries are held back from becoming more productive due to shortages of skilled workers. Accumulated evidence suggests that –on a firm level– the productivity paradox phenomenon is indeed associated with the protracted learning process for economic agents, and organizational investments within businesses. It is not until some time that GPTs are widely adopted and fully accessible, inducing widespread aggregated productivity gains.

Figure 1. Industry contributions to U.S. productivity growth

Note: most intensive ICT-using industries represent the top half of industries with the fastest growth rates in ICT intensity. Least intensive ICT-using industries represent the bottom half of industries in terms of ICT intensity growth rates. Here, ICT intensity is defined as ICT investments plus purchases of ICT services, as a percentage of “synthetic output” (value added plus intermediate use of ICT services).

In the current stage of the New Digital Economy, much has to be learned about the use of new products, processes and business applications. As has become clear from existing literature on technological development processes, the enabling forces driving GPTs from inception to economy-wide productivity gains are interrelated and complex. There is still much to learn about this process. Nevertheless, a wider set of enabling factors can be

established, involving education, skills and the availability of skilled labor. This thesis argues that there need to be skilled workers in order to adequately leverage new opportunities arising from ICT services, and to foster widespread adoption and productivity growth. Therefore, the

0.00% 0.50% 1.00% 1.50% 2.00% 2.50% 3.00% 3.50% 1998-2006 2007-2014

Computer and related services

Telecom services ICT manufacturing "least intensive ICT-using industries"

(7)

emphasis of this thesis will be on U.S labor market characteristics in ICT (services) intensive industries and its role in the U.S. productivity slowdown. Specifically, this thesis addresses the issue whether or not (skilled) labor shortages in intensive ICT (services)-using industries are a contributor to the overall productivity slowdown in the New Digital Economy. Using a unique combined dataset on labor shortages and IT metrics, this thesis follows an unique approach to the productivity paradox as existing literature does not explicitly explore the role of skilled labor shortages in technological development processes. The next section provides a more in-depth analysis of the factors involved in the productivity paradox, elaborating on the existing literature regarding the development process of GPTs, and the use of ICT within organizations. In particular, this section discusses complementary organizational investments (in human capital) and the role of (skilled) labor as enabling factors in ICT productiveness. Furthermore, these theories will be linked to a discussion on labor shortages. Section III describes how the dataset underlying this thesis has been constructed, and describes the methodology used. In section IV, the results and implications are provided. Finally, section V provides the overall conclusion.

II. Literature review

There is a wide variety of business and academic literature regarding the use and

development process (or productivity paradox) of ICT. Most of this literature suggests that in order to stimulate ICT productivity, businesses need their organizational structures –and thus workers– to match technological capabilities. Thus, additional complementary organizational investments (i.e. practices) need to be made besides investments in ICT technology itself. Whereas this involves complex interrelated factors, a larger set of enabling factors involving human capital can be distinguished. The underlying notion here is that ICT (services)

productiveness requires sufficient investments in ICT and investments in human capital (i.e. skilled workers to implement the technologies). To allow for a clear overview on this subject, this section first discusses literature on the development process of GPTs (specifically ICT) on a macro-level. Subsequently, these theories are aligned with the literature on the adoption process of ICT and ICT productiveness within organizations. Finally, the role of labor shortages will be discussed.

(8)

adoption of new technologies and the translation of these technologies into productivity gains involve lengthy, volatile, trials leading to diffusion lags. Hence, productivity paradoxes in the development process of GPTs are not uncommon (rather expected). Additionally, other explanations for the (IT) productivity paradox have been suggested, such as the

mismanagement of IT, mismeasurement issues, and lags due to learning and adjustment. In accordance, Perez (2005) elaborates on the interrelated technological, economic and

institutional changes involved with technological developments. She notes that each

technological surge goes through a so-called installation and deployment phase. The former entails the infant stages of technological revolutions, characterized by large interest in the new technology as early applications emerge. This phase involves cultural adaptation by economic agents and a vast learning process about the production and use of new products. Productivity gains occurring during this stage will be minimal and restricted to a selected amount of sectors, while a large portion of industries will remain unaffected. It is not until the deployment phase that GPTs are widely adopted and fully accessible, inducing widespread aggregate productivity gains. This deployment phase is characterized by technological maturity. As GPTs mature, returns on investment will eventually have maximized and a new cycle will begin once a new GPT emerges.

In contrast, Brynjolfsson (2005) provides a deeper firm-level analysis of the

development process associated with ICT. His findings indicate that ICT productiveness is conditional on complementary organizational investments. He describes these

complementarities in seven practices, including employee empowerment, performance-based incentives, recruiting the right people and investing in human capital. It should be noted that the underlying notion here is that productivity provided by ICT is a function of the quality of people who are using it. It is therefore imperative that businesses are able to attract

(9)

organizational investments (practices) can be viewed as intangible unmeasured inputs. Obviously, these complementary practices, such as recruiting and training appropriate workers, and other organizational investments, take time to fully materialize. Therefore, it is not until some time until these IT investments will be able to disproportionally affect

measures of productivity.

As mentioned previously, Corrado et al. (2014) find that labor productivity growth in high ICT-intensive industries is higher in countries experiencing larger increases in intangible investments such as those discussed earlier. This line of reasoning is consistent with the characterizations of the installation phase and the productivity paradox as previously discussed. Furthermore, there are examples of other GPTs showing similar patterns of

complementary innovations spurring productivity (David, 1990; Brynjolfsson and Hitt, 2000; Milgrom and Roberts, 1992).

While most of the discussed literature does not pertain to ICT services specifically, similar patterns across GPTs suggest that complementary practices are also highly relevant in this case. Van Ark et al. (2016) note that a Business Application Research Center (BARC) survey identifies multiple barriers to the implementation of digital services, such as IT capabilities, analytical skills of the workforce, and companies’ organizational adaptability. In addition, they find that knowledge based assets (i.e. organizational investments) and ICT services are complementary in the U.S. This is indeed consistent with literature regarding other GPTs. Furthermore, Katz and Goldin (2008) show that most technological changes over the course of history have placed increasing demands on skill, education and know-how of the workforce. They state that new and complex technologies (GPTs) have a history of changing the workplace in such a way that rewards educated individuals.

(10)

to be fully explored.

Haskel and Martin (1993) serve as a good departure point regarding labor shortages. They examine the effect of skilled labor shortages on labor productivity. They argue that skilled labor shortages reduce productivity through two channels. First, skilled labor shortages increase hiring costs per skilled worker, leading firms to substitute to less productive unskilled workers. Second, shortages put workers in a stronger bargaining position. By estimating a Cobb-Douglas production function and controlling for labor and product market effects, Haskel and Martin find that increases in skill shortages reduced productivity by 0.7% per annum. Furthermore, skill shortages have a larger impact in

industries that require more skilled labor. They estimate (variants of) the following equation:

(y − l) = α₁(k − l) + a₂SKSHORT + a₃UNSKSHORT + 𝛼₄(PSKILL x SHORT)_𝑡−1+

a₅CONTROLS + FE_i + time dummies (1)

, where (y − l) is gross output per unit of labor (logarithms), (k − l) is capital per unit of labor (logarithms), SKSHORT and UNSKSHORT denoted skilled and unskilled shortages respectively, (PSKILL x SHORT) an interaction term between shortages and skills required in an industry3, and FE_𝑖 are industry-specific fixed effects. In their analysis, the skill

shortages variables represent the percentage of firms replying yes to the question whether they expect output to be constrained by a shortage of skilled labor.

In this thesis a similar analysis is applied. With the transition towards the New Digital Economy in mind, it is examined how shortages of skilled workers across intensive ICT (services)-using industries has affected the U.S. productivity slowdown of the past decade. However, contrary to Haskel and Martin (1993) data on labor shortages are not based on survey data but represent actual (projected) demand-supply gaps (i.e. the difference between job entrants and exits).

3_{Haskel and Martin created their skill measure based on the 1984 Workplace Industrial Relations Survey. The}

(11)

III. Methodology

A. Data description and sources

The core empirical analysis in this thesis relies on a combination of two separate datasets: longitudinal data on ICT investments at industry-level and cross-sectional data on labor shortage indices at industry-level. The first dataset consists of estimates in the Industry Economic Accounts of the Bureau of Economic Analysis (BEA). This detailed dataset provides longitudinal data on 75 industries and their investments in ICT and ICT services purchases for the years 1987 through 2014. Industries follow a 3-digit North American Industry Classification System (NAICS). The second dataset contains data on labor shortage risks (denoted in percentile indexes) at the occupational and industrial-level, projected for the following decade (2014-2024). Specifically, in this dataset from The Conference Board industry labor shortage indexes are constructed as a weighted average of multiple normalized components. These key components are: labor shortage risks due to insufficient supply (i.e. demand-supply gap); labor shortage risks due to educational and skill requirements; and labor shortage risks due to limits on how, when and where suitable candidates can be obtained. In this thesis, the main focus is on (a combination of) the first two components. This thesis specifically focuses on the first two components. The first component represents the

difference between entrants and exiting workers in a certain industry (demand-supply gap). The higher the demand-supply gap, the higher the labor shortages risk4_{. Figure 2 below}

illustrates the relation between the demand-supply gap for industries and the change in productivity from the relatively fast growth period (1998-2006) to the slowdown period (2007-2014).

The second component allows a distinction of labor shortages based on education and skill-level (i.e. skilled labor shortages), by means of adding measures of labor quality to the indexes. These measures include schooling, typical required prior work experience, skills, training, and etcetera. The higher the labor quality measures, the higher the risk of labor shortages given that it will be more difficult for the economy to supply these workers (Levanon et al., 2016). It should be noted that these variables have been chosen and constructed in such a way that they reflect risks in labor shortages5.

4_{For estimation purposes the actual demand-supply gap denoted in numbers of people is used instead of the}

percentile index

(12)

Figure 2. Labor shortages and productivity change (R2 _{= .1361)}

Unlike the industries in the dataset on ICT investments, the industries in this particular dataset follow a less detailed, yet similar, BEA classification. These two datasets are merged by means of aligning the industries across the two datasets. This entails adjusting and

aggregating the 3-digit NAICS industries in the more detailed dataset on ICT investments, to match the BEA industries of the labor shortages dataset. This method has been chosen to preserve the dataset containing labor shortages indexes. In the end, these modifications result in a data sample of 64 BEA industries for the years 1998 through 20146. Furthermore,

WORDKLEMS data on U.S. factor inputs are used. These industries are assumed to represent the U.S. economy for the years considered.

B. Methods and estimation

Throughout this thesis four major industry groups are distinguished: ICT using industries,

6_{Due to the restricted availability of ICT services, data used in this thesis concerns only investment levels in the}

(13)

non-ICT industries, ICT producing industries, and ICT services industries. Similar to van Ark et al. (2003; 2014), the first two groups are distinguished based on their intensity of ICT use. In this thesis three ICT intensity measures are considered: ICT intensity in terms of ICT assets usage, ICT intensity in terms of ICT services usage, and ICT intensity based on a combination of both ICT assets and ICT services. By ranking industries based on one of these measures of ICT intensity, the top half of industries can be classified as “most intensive” ICT-using industries (in terms of either ICT assets, ICT services or a combination of ICT assets and ICT services) and the bottom half as “least intensive” ICT-using industries. The third group of ICT producing industries includes producers of computer and electronic products as specified under the North American Industry Classification System (NAICS 334). Finally, computer and related services industries and telecom services industries represent the group of ICT services industries. The former includes data processing, information services, computer systems design and related services. The latter includes broadcasting and publishing (including software).

As mentioned before, the New Digital Economy is driven by a shift from investments in ICT assets (including PCs, ubiquitous access to Internet) to purchases of ICT services (mobile and cloud computing). Hence, it is imperative that the measure of ICT intensity that is used to distinguish between ICT-using and non-ICT industries, considers the intermediate use of ICT services. ICT measures used are based on van Ark et al. (2016). Specifically, ICT intensity is defined as the share of ICT (assets) investments plus ICT services purchases as a percentage of “synthetic output”, where “synthetic output” is defined as industry level value added plus intermediate use of ICT services. These ratios are calculated not as a share of GDP (value added) but as a share of synthetic output because one needs to consider the intermediate inputs from ICT services. ICT services purchases are expenditures whereas other factor inputs (labor; (ICT)-capital) are considered investments (and therefore included in value added). Furthermore, the analysis in this thesis also considers two additional ICT intensity measures, where each component is examined individually: investments in ICT assets (%synthetic output); and ICT services purchases (%synthetic output). These two ICT intensity measures allow a clear distinction between the two different technologies of the Old Digital Economy and New Digital Economy respectively. Thus, separate industry groups can be created based on these three measures of ICT intensity. As mentioned before, in this thesis four major industry groups are considered: most intensive ICT-using industries, least

(14)

composition of most intensive ICT using industries and least intensive ICT-using industries will differ depending on the measure of ICT intensity that is considered. Three different segmentations are defined regarding these industries: most intensive ICT-using industries in terms of ICT assets, most intensive ICT using industries in terms of ICT services purchases, and most intensive ICT-using industries in terms of investments in ICT assets and ICT services purchases combined. A full description of the industries in each industry group, based on each of the three considered ICT intensity measures, is provided in appendix A. Moreover, these industry characteristics are modeled through dummy variables. Specifically, these dummy variables equal 1 when the ICT intensity measure considered exceeds a certain threshold such that the most intensive ICT-using industries represent the top half of

industries.

Finally, this thesis is centered around two separate time periods: 1998-2006 and 2007-2014. The first period can be viewed as a period of (relatively) fast productivity growth. The latter period on the other hand is the period in which average U.S. productivity growth rates slowed down (see figure 1 for a comparison of the two periods). The change in productivity over these two periods adequately represents the slowdown in average U.S. labor

productivity over time. In support of the hypothesis of this thesis, it is expected that labor shortages will negatively affect the change in average U.S. labor productivity over time (while controlling for ICT asset and ICT service use). The change in labor productivity over these periods, and the effect of (skilled) labor shortages will be examined using (variants of) the following equation:

∆ (Y L)_i = β0+ β1∆ ( K L)_i+ β2∆ ( C L)_i + β3∆(S)i + β4( Y L)_{i,1998−2006}+ β5ICTi+

β₆Shortage_i+ β₇EDUCSKILL_i + β₈ICT_i x Shortage_i (2)

, where ∆ (Y

L)_i is the change in labor productivity (value added per hours worked) for industry

i over the periods considered, ln [(Y

L)_{𝑖,2007−2014}] − ln [( Y

L)_{𝑖,1998−2006}]; ∆ ( K

L)_i is the change

in non-IT capital/hours worked for industry i, ln [(K

L)_{𝑖,2007−2014}] − ln [( K

L)_{𝑖,1998−2006}]; ∆ ( C L)_i

(15)

investments in ICT assets), ln [(C

L)_{𝑖,2007−2014}] − ln [( C

L)_{𝑖,1998−2006}]; ∆(S)i is the change in

ICT services purchases for industry i, ln(S𝑖,2007−2014) − ln(S𝑖,1998−2006); ( Y

L)_{i,1998−2006}is

the productivity level of industry i for the period 1997-2006 (accounting for

convergence/divergence); ICT_i is a dummy variable accounting for most intensive ICT-using industries (in terms of ICT services; ICT services and ICT assets combined); SHORTAGE_𝑖 is the projected labor shortage in industry i (i.e. the demand-supply gap).

Van Ark et al. (2016) calculate the shortage gap as the difference between

Labor demand = (employment growth + 0.65 ∗ replacement growth) and

Labor supply = (entrants growth + reentrants growth)

,where employment growth is the projected employment growth (2014-2024)7_{, replacement}

growth is the growth in jobs that need to be replaced; EDUCSKILL is the education and skill component of the labor shortages index (percentile ranking) and accounts for labor shortages related to industry education and skill requirements; and finally ICT_i x Shortage_i is a cross term between the dummy variable accounting for most intensive ICT-using industries and labor shortages (demand-supply gap).

To assess whether skilled labor shortages negatively affect productivity growth the following hypothesis is tested:

H₀: β₇ = 0 H₁: β7 < 0

The expectation is that labor shortages related to education and skills are detrimental to U.S. productivity. Not only is it a reasonable expectation that labor shortages related to an

insufficient supply of workers (i.e. the demand-supply gap) negatively affects the U.S. productivity (that is β₆< 0)8_{, labor shortages related to education and skills in particular are}

7_{For estimation purposes it is assumed that the projected labor shortages also can be extended to the earlier}

periods considered in this thesis.

8_{A positive relation, however, is also a reasonable expectation for certain most intensive ICT-using industries,}

(16)

expected to be relevant in light of the ongoing digital transformations. In particular, this hypothesis needs to be tested across different industry groups to determine whether this is particularly the case for most intensive ICT services-using industries.

IV. Results

A. Productivity analysis

As noted in the introduction, van Ark et al. (2016) find that the slowdown in US productivity growth in the 21st century coincided with the largest productivity growth slowdown occurring in the most intensive ICT-using industries. Particularly, decline in productivity growth during 2007-2014 was largest in these industries compared to the 1998-2006 period. First, it can be shown that these productivity paradox also hold when measures of ICT intensity based on growth rates are used instead of absolute levels (see figure 1). In this particular case, the largest decline in productivity growth during 2007-2014 (compared to 1998-2006) can be attributed to most intensive ICT-using industries in terms of a combination of ICT services and ICT assets (i.e. upper half of industries with the largest average increases in ICT investments and ICT services purchases). An interesting finding is that this particular result does not hold when solely examining the role of the New Digital Economy. Specifically, when distinguishing most intensive ICT-using industries and least intensive ICT-using industries based on ICT services intensity, it can be shown that the slowdown in US

(17)

slowdown could also mean that the shift toward buying ICT services actually helps businesses become more productive as the correlation analysis by van Ark et al (2016) suggests. Nevertheless, these industries still witness a slowdown in productivity. This suggests that the productivity paradox is still present and that –apparently– ICT services purchases at this current stage are insufficient to spur aggregate productivity growth. Therefore, it can be argued that further examination is necessary to assess whether labor shortages play a role in this phenomenon.

Figure 3. Industry group contributions to U.S. productivity growth

Note: “ICT services-using industries” refers to the top half of industries with the highest share of ICT services purchases/”synthetic output”.

Third, these findings could mean that a combination of both explanations stated above is likely to be the case. Specifically, the Old Digital Economy has matured and maximized its returns. As mentioned in the introduction, prior productivity growth gains have been realized through ICT asset investments. These gains are realized until the point of technological maturity. On the other hand, the New Digital Economy as characterized by ICT services is in its installation phase and productivity gains are minimal and restricted to a selected amount of industries (of which the explanation will be sought in industry labor shortages).

0.00% 0.50% 1.00% 1.50% 2.00% 2.50% 3.00% 3.50% 1998-2006 2007-2014

Computer and related services

Telecom services ICT manfacturing "least intensive ICT services-using industries"

(18)

B. Regression analyses Having illustrated the contributions of the different industry groups to U.S. productivity growth in the two periods considered, it is now imperative to assess the role of labor

shortage. Table 1 provides results for (five variants of) the regression analysis as specified in equation (2). In each specification the dependent variable represents the productivity change over the two periods considered (the period of relatively fast productivity growth, 1997-2006, and the slowdown period 2007-2014).

Column (1) shows the relation between labor shortages (due to insufficient supply of workers) and productivity change, while controlling for factor inputs, ICT services usage, the (average) productivity level for the period 1998-2006, and most intensive ICT-using

industries (in terms of services and assets and services combined). First, it is important to assess the effects of factor inputs and intermediate usage of ICT services on productivity growth. As shown in column (1), growth in ICT-capital/labor positively affects productivity change over the two periods. Specifically, a 1% increase leads to an increase in productivity change of .38%. On the other hand, ICT-capital/labor negatively affects productivity change (column (1)) (or does not show any significant relation at all (column (2)). A 1% increase in ICT-capital/labor growth decreases productivity growth by .015%. This suggests that while increases in non-ICT capital foster productivity growth, this cannot be said about ICT-capital. Furthermore, ICT services purchases show a positive sign in relation to productivity growth. However, this relation is insignificant, as productivity growth over the period 1998-2014 seems to remain unaffected by changes in ICT services purchases. Overall, these variables show the expected sign as it suggests that the returns on ICT capital have been exhausted, whereas ICT services are yet to translate into aggregate productivity growth9_{. This scenario is}

consistent with the employment (deployment) phase of the New (Old) Digital Economy as discussed in the literature review. Furthermore, average productivity levels in the period 1998-2006 cannot explain productivity growth, suggesting that the productivity change over that period follows a random walk.

9_{As stated prior, van Ark et al. (2016) show that (for a specific period) there is a positive correlation between}

(19)

Table 1. Effect of labor shortages on U.S. productivity change 1998-2014 Dependent variable: Δ(Y/L) (1) Δ(Y/L) (2) Δ(Y/L) (3) Δ(Y/L) (4) Δ(Y/L) (5) Δ(Y/L) (6) Δ(Y/L) (7) Constant Δ(K/L) Non-ICT capital/labor Δ(C/L) ICT capital/labor .080 (.110) .380*** (.058) -.015* (.009) .138 (.111) .349*** (.064) -.013 (.009) .065 (.089) .352*** (.061) -.018** (.008) .101 (.116) .372*** (.059) -.017* (.009) .130 (.099) .365*** (.054) -.021** (.008) .206 (.125) .326*** (.067) -.020** (.009) .119 (.099) .336*** (.063) -.021*** (.008) Δ(S)

ICT services purchases (.035) .016 _(.035)-.001 (.036) .043 (.035) .017 (.035) .049 _(.036)-.001 (.037) .050 Productivity 1998-2006 ICT-using industries (assets, services) ICT-using industries (services) Labor shortage (demand-supply gap) EDUCSKILL (shortage index) Labor shortage (demand-supply gap) x ICT-using (assets, services) Labor shortage (demand-supply gap) x ICT-using (services) .016 (.023) -.051 (.034) -.026*** (.010) .004 (.022) -.027 (.028) -.059*** (.015) .055*** (.021) .016 (.020) -.034 (.027) -.046*** (.014) .036* (.021) .012 (.024) -.053 (.036) -.026** (.010) .006 (.013) .004 (.022) -.071** (.031) -.030*** (.010) .021* (.012) -.009 (.024) -.023 (.029) -.061*** (.016) .022* (.013) .062*** (.022) .006 (.021) -.046 (.028) -.045*** (.014) .015 (.012) .034 (.022) 𝑅2_{=.5855 𝑅}2_{=.6370 𝑅}2_{= .6075} _𝑅2_{=.5764 𝑅}2_{=.5908 𝑅}2=.6393 _𝑅2_{= .6033}

(20)

With regards to labor shortages (the variable of interest), we observe that industry labor shortages (i.e. demand and supply gap) negatively affect productivity growth. Specifically, a 1%-point increase in labor shortages negatively affects productivity growth by .026%-points. This finding supports part of the hypothesis by indicating that a lack of workers, irrespective of skill or industry, indeed negatively affects productivity growth.

Column (2) adds an interaction effect of the most intensive ICT-using industry dummy and labor shortages to the specification. In this specification, it is found that –again– an increase in labor shortages negatively affects productivity growth. What’s more, the results show that the effect of labor shortages is conditional on industries being heavy users of ICT (in terms of a combination of ICT assets and ICT services). Labor shortages

negatively affect productivity growth rates less in most intensive ICT-using industries. Specifically, a 1%-point increase in labor shortages in least intensive ICT industries

decreases productivity growth with .059%, whereas a 1%-point increase in labor shortages in most intensive ICT-using industries decreases productivity with .004%-point. This suggests that most intensive ICT-using industries are more capable of coping with labor shortages compared to least intensive ICT-industries. A possible explanation for this finding is that there is an automation effect associated with becoming more ICT intensive. This automation effect is generally associated with ICT assets (PCs, hardware etc.). Therefore, it is not surprising to obtain this finding for most intensive ICT-using industries in terms of both ICT assets and ICT services use. However, the results in column (3) suggest that this is possibly also the case when examining most intensive ICT-services using industries separately.

Column (3) provides a similar specification where most intensive ICT-using

(21)

could still be highly related to ICT assets as prevalent in the Old Digital Economy. Nevertheless, little evidence is found for the latter explanation10_{. The specifications in}

columns (4) and (5) add the education shortages and accountsfor both the level of labor shortages (demand-supply gap) and the education and skills labor shortage (percentile) index (i.e. labor shortages related to education and skills). The specification in column (4) shows factor inputs and ICT service usage have to expected sign. Furthermore, labor shortages seem to negatively affect productivity growth over the period 1997-2014. Specifically, a 1%-point increase in labor shortages (negatively affects productivity growth by 2.6%-points. Moreover, labor shortage risks related to education and skills do not seems to affect productivity growth over the periods considered. In the specification of column (5) however, different results are found. We find that labor shortages related to education and skills positively affect

productivity change over the periods considered. Specifically, a 1%-point increase in the percentile ranking of the labor shortage education and skills index increases productivity growth with .021%. It should be noted that this effect is not highly significant (significance at 10%). On one hand, the sign is not surprising given that the education and skills index is a measure of labor quality. Expectedly, industries with higher labor quality are expected to have higher productivity growth. On the other hand, this sign may very well be negative in the case of labor shortages due to education and skill requirements negatively affecting productivity change. Furthermore, most intensive ICT services-using industries have on average .069% lower productivity growth compared to least intensive ICT services-using industries. Again, we find that labor shortages (insufficient supply irrespective of skills and education) negatively affect productivity growth. A 1%-point increase in labor shortages is associated with a .03% decrease in productivity growth.

Columns (6) and (7) test the full specification. Again, labor shortages (irrespective of skill) negatively affect productivity change in both specifications. The only significant difference between the two pertains to education shortages and the conditional effect of labor shortages. Specifically, in column (6) it is found that for least intensive ICT-using industries (in terms of a combination of ICT assets and ICT services) labor shortages negatively affect productivity change, while this is positive for most intensive ICT-using industries. In column (7) these results do not seem to hold when solely examining for most intensive ICT

10_{A preliminary correlation exercise shows that the correlation between industries belonging to most intensive}

(22)

using industries. With regards to education shortages, column (6) shows that there is a positive effect on productivity change. In column (7) there is no significant effect.

The results above show that labor shortages do play a role in determining productivity change over the periods considered. However, education shortages show a positive effect on

productivity change rather than a negative effect. To compare how these findings change across different industry groupings and to assess whether these findings hold for most

intensive ICT services-using industries in particular, table 2 provides a similar analysis based on the different industry groupings as mentioned before. In this table each column pair can be compared directly as each pair is based on the same measure of ICT intensity. For instance, columns (1) and (2) present results for most intensive ICT service-using industries and least intensive ICT services-using industries respectively. The results indicate that non-IT

capital/labor positively affects productivity growth over the period considered, while IT capital/labor seems to have either no significant effect (for most intensive ICT services-using industries) or a negative effect (for least intensive ICT-services using industries) on

productivity growth. As mentioned before, these effects are consistent with the deployment phase of the Old Digital Economy in which the benefits of ICT-capital investments are expected to be exhausted. What’s more, we find that ICT services purchases positively affect productivity growth for most intensive ICT services-using industries. Specifically, a 1%-point increase in ICT services purchases increases productivity growth with .096%. As shown in column (2), this result does not hold for least intensive ICT services-using industries, suggesting that sufficient ICT services purchases need to be made in order to foster productivity growth. This confirms the preliminary correlation findings by van Ark et. al (2016), who suggest that indeed ICT services purchases help businesses come more productive. These results are consistent with the explanations provided for figure 3 at the top of this section. Figure 3 showed that while the productivity slowdown was still present across ICT services intensive industries, the slowdown was not driven by most intensive ICT

(23)

play a key role. As found in columns (1) and (2), labor shortages in most intensive ICT services-using industries show a negative sign whereas this sign is positive for least intensive ICT services-using industries. Nevertheless, both variables are insignificantly different from zero. Whereas most of the discussed literature suggests that ICT (services) productiveness is a function of investments in technology and additional organizational practices (i.e. the people that use it), the latter cannot be found in the results.

Table 2: labor shortages in ICT-using industries

Dependent variable: Δln(Y/L) (1) Δln(Y/L) (2) Δln(Y/L) (3) Δln(Y/L) (4) Δln(Y/L) (5) Δln(Y/L) (6) Sample Constant Δln(K/L) Non-ICT capital/labour Δln(C/L) ICT capital/labour ICT-using (services) 11.497* (5.570) .222** (.080) .038 (.035) Non-ICT (services) -.4.815** (1.812) .531*** (.064) -.038*** (.009) ICT-using (assets, services) 1.756 (2.582) .125 (.091) .010 (.027) Non-ICT (assets, services) -4.236* (2.043) .528*** (.063) -.028*** (.008) ICT-using (assets) 2.738 (3.329) .092 (.086) .017 (.022) Non-ICT (assets) -1.900 (2.357) .504*** (.059) -.035*** (.008) Δln(S)

ICT services purchases (.051) .096* -.005 _(.041) _(.059)-.090 (.021) .016 -.128* _(.069) (.024) .019 Productivity 1998-2006 Shortage (demand-supply gap) EDUCSKILL (shortage index) SKILL Shortage x SKILL .127** (.061) -3.309 (2.629) .176* (.086) -1.065* (.518) .304 (.025) -.031 (.021) .617 (.453) -.046 (.030) .450** (.161) -.059 (.042) .003 (.045) .317 (.754) .045 (.041) -.129 (.226) -.028 (.070) -.000 (.016) -.435 (.587) -.038 (.030) .387** (.180) .036 (.053) .016 (.026) 1.474* (.796) .066 (.054) -.218 (.299) -.135* (.073) -.001 (.024) .987 (.967) -.018 (.029) .174 (.212) -.095 (.089) 𝑅2_{=.2881 𝑅}2=.8724 _𝑅2_{=.0269 𝑅}2=.8467 _𝑅2_=.2598 _𝑅2=..8341

Estimation method: Pooled OLS

Notes: Robust standard errors are provided in parentheses. All (continuous) variables are expressed in natural logarithms.

(24)

Labor shortages in most intensive (and least intensive) ICT services-using industries do not seem to affect the change in the U.S. productivity based on these results in columns (1) and (2). A possible explanation for this could depend on the data. Evidently, the labor shortages data underlying this thesis is static and does not allow one to track labor shortages over time (specifically for the periods considered). Nevertheless, when examining over the whole sample, one does find that labor shortages play a role in affecting the productivity slowdown (as was shown in table 1 above).

Moving to the labor shortages related to education and skills, column (1) shows a positive effect on productivity growth from most intensive ICT services-using industries (although not highly significant). The more difficult it is for the U.S. economy to provide appropriate workers due to education and skill requirements, the larger the positive

productivity change in most intensive ICT services-using industries. For least intensive ICT services-using industries this sign is actually negative. Nevertheless, this effect seems to be insignificant. Given that this variable is represents labor quality, this positive effect is not unexpected. What’s more, it is found that the higher the skills required in most intensive ICT services-using industries, the lower productivity growth. This explains part of the hypothesis. Given that technological developments place increasing demands on skills, education, and know-how this suggests that inefficient use of ICT services technology by workers may play a role. Finally, similar to Haskel and Martin (1993) an interaction effect is also considered to determine whether labor shortages are dependent on the average level of skills required in industries (i.e. skilled shortages). The results in columns (1) and (2) show that this interaction effect is not significant.

(25)

Surprisingly, we do not find any significant variables when examining the sample of most intensive ICT-using industries. For least intensive ICT-using industries, it is found that non-ICT capital/labor has a positive effect on productivity growth, whereas non-ICT capital/labor (ICT assets) has a negative effect on productivity growth. Furthermore, a 1%-point increase in ICT services purchases is associated with a .128% decrease in productivity change. Nevertheless, neither ICT-using industry group considered shows any relation regarding labor shortages.

Finally, Columns (5) and (6) provide the results for most intensive and least intensive ICT asset-using industries. We find that for most intensive ICT assets-using industries, ICT services purchases negatively affect productivity growth. Apparently, intensive ICT asset-users are not fostering ICT services productiveness as was found for most intensive ICT services-using industries. This indicates that most-intensive ICT asset-using industries are not necessarily also ICT services intensive (as stated previously). On the other hand, (ICT) capital/labor does not seem to affect productivity growth. This suggests that the returns on investments in the Old Digital Economy (as characterized by ICT assets (capital)) have indeed been maximized. Turning to the variable of interest, labor shortages positively affect productivity change only in most intensive ICT asset-using industries. There is no significant relationship for least intensive ICT asset-using industries. This could be explained by the fact that ICT assets are generally associated with an automation effect. Therefore most intensive ICT asset-using industries are expected to be able to deal with labor shortages. What’s more, it is found that for most intensive ICT asset-using industries the effect of labor shortages is dependent on the average skill level of industries (as indicated by the skill-shortage

interaction term in column 5). Whereas labor shortages in these industries increase

(26)

V. Conclusion

In this thesis it was argued that skilled labor is necessary to foster ICT services

productiveness. Various literature suggests that indeed IT productiveness is a function of the people that use it. Furthermore, with ICT services being a relatively new-implemented technology, it is assumed that the successful adoption of ICT services in the production processes involves a learning process requiring appropriate workers. The results show that – while taking into account the switch from the Old Digital Economy to the New Digital Economy– labor shortages do play a role in the U.S. productivity slowdown of the 21st

century. Irrespective of the average skill level of workers, it can be shown that simply an insufficient supply of workers in certain occupations has been detrimental to U.S.

productivity change. What’s more, it is found that these labor shortages are more of a constraint for less intensive ICT-using industries than for the more intensive ICT-using industries (in terms of ICT assets and ICT services combined). This suggests that there is an automation effect in play, as more intensive ICT-using industries seem to be more capable of coping with shortages. When examining labor shortages across specific industry groups, these findings do not hold. Specifically, there is no hard evidence of labor shortages (that is, an insufficient supply of workers) being a major productivity constraint in any of the industry groups considered in this thesis. Turning to education shortages, there does not seem to be much of an impact when examining specific industry groups. Only for most intensive ICT services-using industries it is found that education shortages affect productivity change. Furthermore, this effect is positive rather than the expected negative sign. Thus, based on these findings the hypothesis that skilled labor shortages in most intensive ICT services-using industries are driving the U.S. productivity slowdown cannot be accepted. Another

interesting finding is that this might be related to the fact that ICT services purchases do not seem to positively affect productivity change when considering the whole sample. Only when considering most intensive ICT services-using industries does one find a significant positive effect. This suggests that the ICT services adoption process might be at too much of a

(27)

driving the shift toward the New Digital Economy (i.e. most intensive ICT services-using industries). Whereas these findings reject the hypothesis that skilled labor shortages in most intensive ICT services-using industries are driving the U.S. productivity slowdown, there is still some evidence that the learning curve involved with new ICT services technology is a deterring factor to U.S. productivity growth. For instance, it is found that the higher the average skill level required in most intensive ICT services-using industries, the more negative the U.S. productivity change in these industries. This could suggest that workers in these industries are not capable of effectively leveraging their know-how and skills into

(28)

References

Brynjolfsson, E., (2005). “Seven Pillars of Productivity”, Optimize

Brynjolfsson, E., Hit, L.M. (2000). “Beyond Computation: Information Technology, Organizational Transformation and Business Performance”, Journal of Economic Perspectives, Vol. 14, No.4, pp. 23-48

Brynjolfsson, E., Hitt, L.M. (2003). “Computing Productivity: Firm-level Evidence”, The Review of Economics and Statistics, Vol. 85, No. 4, pp. 793-808

Corrado, C., Haskel J., Jona-Lasinio, C. (2014). “Knowledge Spillovers, ICT and

Productivity Growth”, Economics Program Working Paper Series No. 14-02, The Conference Board, May

David, P. (1990). “The Dynamo and the Computer: An Historical Perspective on the Modern Productivity Paradox”, The American Economic Review, Vol. 80, No. 2, pp. 355-361 Goldin, C., Katz, L. (2008). “The Race Between Education and Technology”. Belknap Press

for Harvard University Press.

Haskel, J., Martin, C. (1993). “Do Skill Shortages Reduce Productivity? Theory and

Evidence from the United Kingdom”. The Economic Journal, Vol. 103, No. 417, pp. 386-394

Levanon, G., Erumban, A. (2016). What Looming Labor Shortages Mean for Your Busines”, The Conference Board

Milgrom, P., Roberts, J. (1992). Economics, Organization and Management. New York: Prentice-Hall

Perez, C. (2005). Technological revolutions and financial capital. 1st ed. Cheltenham: E. Elgar.

(29)

APPENDIX A: LIST OF INCLUDED INDUSTRIES, 1996-2006 (and 2007-2014) NAICS 211 213 22 322 323 325 326 333 337 339 42 44,45 481 482 483 485 486 487,488,492 111,112 113-115 212 23 311,312 313,314 315,316 321 324 327 331 332 335 336 336 334 511 515,517 518,519 5415

ICT using industries (assets and

services)

Oil and gas extraction

Support activities for mining Utilities

Paper products

Printing and related support activities

Chemical products Plastics and rubber products

Machinery Furniture and related products Miscellaneous manufacturing Wholesale trade Other Retail Air transportation Rail transportation Water transportation

Transit and passenger transportation

Pipeline transportation Other transportation and support

activities

Non-ICT industries

Farms Forestry, fishing, and related activities Mining, except oil and gas Construction Food and beverage and tobacco products

Textile mills and textile product mills Apparel and leather and allied

products Wood products

Petroleum and coal products Nonmetallic mineral products Primary metals Fabricated metal products Electrical equipment, appliances, and components

Other transportation equipment Motor vehicles, bodies and trailers,

and parts

ICT producing industries

Computer and electronic products

ICT services industries

Publishing industries, except internet, includes software

Broadcast and telecommunications Data processing, internet publishing, and other information services Computer systems design and related services 512 521,522 523,525 524 532,533 54 5411 55 561 61 622,623 44,45 44,45 44,45 484 493 531 562 621 622,623 624 711,712 713 721 722 81

Motion picture and sound recording industries Federal Reserve banks, credit intermediation and related activities Funds, trusts, securities, commodity contracts, and other financial vehicles Insurance carriers and related activities Rental and leasing services and lessors of intangible assets Miscellaneous professional,

scientific, and technical services Legal services

Management of companies and enterprises

Administrative and support activities

Educational services Hospitals

Motor Vehicle and Parts Dealers

Food and Beverage Stores General Merchandise Stores

Warehousing and storage Truck transportation

Housing and other real estate Waste management and remediation services Ambulatory health care services

Nursing and residential care facilities Social assistance

Performing arts, spectator sports, museums and related activities Amusements, gambling, and recreation industries

(30)

NAICS 211 213 22 311,312 313,314 315,316 321 322 323 326 327 332 333 337 339 44,45 481 482 512 111,112 113-115 212 23 324 325 331 335 336 336 42 44,45 44,45 44,45 483 334 511 515,517 518,519 5415

ICT using industries (services)

Oil and gas extraction

Food and beverage and tobacco

products

products Wood products Paper products

Printing and related support activities Plastics and rubber products

Nonmetallic mineral products Fabricated metal products Machinery Furniture and related products Miscellaneous manufacturing

Other Retail Air transportation

Rail transportation Motion picture and sound recording industries

Farms Forestry, fishing, and related

activities Mining, except oil and gas Construction Petroleum and coal products

Chemical products Primary metals

Electrical equipment, appliances, and components Other transportation equipment Motor vehicles, bodies and trailers,

and parts

Wholesale trade Motor Vehicle and Parts Dealers Food and Beverage Stores General Merchandise Stores Water transportation

Broadcast and telecommunications Data processing, internet publishing, and other information services Computer systems design and related services 521,522 523,525 524 532,533 54 5411 55 561 562 61 622,623 722 484 485 486 487,488,492 493 531 621 622,623 624 711,712 713 721 81

Federal Reserve banks, credit intermediation and related activities Funds, trusts, securities, commodity contracts, and other financial vehicles Insurance carriers and related activities Rental and leasing services and lessors of intangible assets Miscellaneous professional,

scientific, and technical services Legal services

Management of companies and enterprises

Administrative and support activities Waste management and remediation services

Educational services Hospitals

Food services and drinking places

Warehousing and storage Transit and passenger transportation

activities Truck transportation

Housing and other real estate Ambulatory health care services

Nursing and residential care facilities Social assistance

Performing arts, spectator sports, museums and related activities Amusements, gambling, and recreation industries

(31)

NAICS 213 22 323 326 333 335 339 42 44,45 44,45 44,45 44,45 481 482 483 485 486 487,488,492 111,112 113-115 211 212 23 311,312 313,314 315,316 321 322 324 325 327 331 332 336 334 511 515,517 518,519 5415

ICT using industries (assets)

Printing and related support activities Chemical products

Machinery Electrical equipment, appliances, and components

Miscellaneous manufacturing Wholesale trade

Other Retail Motor Vehicle and Parts Dealers Food and Beverage Stores

General Merchandise Stores Air transportation

Rail transportation Water transportation

Transit and passenger transportation

activities

Farms Forestry, fishing, and related activities Oil and gas extraction Mining, except oil and gas Construction Food and beverage and tobacco products

products Wood products

Paper products Petroleum and coal products

Plastics and rubber products Nonmetallic mineral products Primary metals Fabricated metal products Other transportation equipment

Broadcast and telecommunications Data processing, internet publishing, and other information services Computer systems design and related services 493 512 521,522 523,525 524 532,533 54 55 561 61 622,623 336 337 484 531 5411 562 621 622,623 624 711,712 713 721 722 81 Truck transportation Motion picture and sound recording industries Federal Reserve banks, credit intermediation and related activities Funds, trusts, securities, commodity contracts, and other financial vehicles Insurance carriers and related activities Rental and leasing services and lessors of intangible assets Miscellaneous professional, scientific, and technical services Management of companies and enterprises

Administrative and support activities Educational services

Nursing and residential care facilities

Motor vehicles, bodies and trailers, and parts

Furniture and related products Warehousing and storage

Housing and other real estate Legal services Waste management and remediation services Ambulatory health care services Hospitals

Social assistance Performing arts, spectator sports,

museums and related activities Amusements, gambling, and recreation industries

Accommodation Food services and drinking places Other services, except government

Skilled labor shortages and ICT services adoption in the United States

Faculty Economics and Business

Master’s Thesis