Are ICT displacing workers in the short run? Evidence from seven European countries

Accepted Manuscript

Are ICT Displacing Workers in the Short Run? Evidence from Seven European Countries Smaranda Pantea , Anna Sabadash , Federico Biagi PII: DOI: Reference:

S0167-6245(16)30161-5 10.1016/j.infoecopol.2017.03.002 IEPOL 772

To appear in:

Information Economics and Policy

Received date: Accepted date:

2 December 2016 16 March 2017

Please cite this article as: Smaranda Pantea , Anna Sabadash , Federico Biagi , Are ICT Displacing Workers in the Short Run? Evidence from Seven European Countries , Information Economics and Policy (2017), doi: 10.1016/j.infoecopol.2017.03.002

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT Highlights

 We study the short run substitution effect of ICT use on firms’ employment.  We use highly accurate quantitative measures of ICT use within firms.

CR IP T

 We use a longitudinal dataset containing internationally comparable firm level data for seven European countries, covering manufacturing and services sectors.  We find no evidence that ICT substitutes labour in the short run.

AN US

 The insignificant effect of ICT is very robust across ICT measures, countries and

AC

CE

PT

ED

M

sectors.

1

ACCEPTED MANUSCRIPT

Are ICT Displacing Workers? Evidence from Seven European Countries Smaranda Pantea (corresponding author) DG Internal Market, Industry, Entrepreneurship and SMEs European Commission BREY, Avenue D'Auderghem 45

CR IP T

Brussel, 1045 Belgium And Ministry of Public Finance, Romania

AN US

Email: [email protected]

Anna Sabadash Eurostat European Commission

M

BECH D4/720

ED

Kirchberg, 2920 Luxembourg

CE

Federico Biagi

PT

Email: [email protected]

Institute for Prospective Technological Studies (IPTS)

AC

European Commission

Edificio Expo, C/Inca Garcilaso, 3 Seville, 41092 Spain and University of Padua, Italy Email: [email protected]

2

ACCEPTED MANUSCRIPT

CR IP T

Abstract This paper examines the short run labour substitution effects of using ICT at firm-level in the manufacturing and services sectors in seven European countries, during the period 20072010. The data come from a unique dataset provided by the ESSLait Project on Linking Microdata, which contains internationally comparable data based on the production statistics linked at firm level with the novel ICT usage indicators. We adopt a standard conditional labour demand model and control for unobservable time-invariant firm-specific effects. The results show that ICT use has a statistically insignificant labour substitution effect and this effect is robust across countries, sectors and measures of ICT use. Our findings suggest that increased use of ICT within firms does not reduce the numbers of workers they employ.

1.

AN US

JEL codes: J23, J24, O33, L86 Keywords: Labour Demand, Technology Change, ICT, Employment, Productivity.

Introduction

Adoption of information and communication technologies (ICT) by European firms and their integration in enterprise processes are widely regarded as essential for the modernisation of EU industry and for regaining competiveness in international markets. This is reflected in policies such as Industry 4.0, Smart Industry, Free Flow of Data and The European Cloud,

M

which support the uptake of ICT with a view to improve the efficiency and flexibility of production processes and quality of the final products. However, there are serious concerns

ED

among policy-makers about the possible negative effects of these technologies on employment.

PT

In economic literature, two main effects of ICT adoption and use on employment are considered: the substitution effect and the compensation effect. On the one hand, ICT can

CE

substitute workers either directly (e.g. robots taking up tasks previously performed by workers) or indirectly (e.g. through increased labour productivity, which leads to less labour

AC

being needed to produce a unit of output). These effects are not limited to low-skilled workers: medium and high-skilled workers are affected as well1 (Autor et al., 2003; Acemoglu and Autor, 2011; Brynjolfsson and McAfee, 2011 and 2014). On the other hand, under favourable conditions2, a compensation effect may result from an increase in the demand for labour due to the ICT-driven gains in efficiency (and hence lower prices) or to the ICT-enabled product innovations, both of which lead to higher demand for firms’

1

In fact, there is a suggestion that ICT use might lead to de-skilling of workers, irrespective of their skills levels (Beaudry et al., 2014). 2 For a discussion of these conditions, see Vivarelli (2007).

3

ACCEPTED MANUSCRIPT products. Which of the two effects prevails is subject to intense debate, but is ultimately an empirical issue. Finally, ICT could affect the skill/occupation composition of employment (Falk and Biagi, 2016), though it may not have a systematic effect on overall employment (Aghion et. al., 2013). Among these effects, the substitution effect is the focus of most debate as it may lead to major disruption and uncertainty in labour markets and it is conventionally believed to cause job destruction in the short term. Moreover, the jobs eventually created through the

CR IP T

compensation effect may require different skills than those displaced by ICT. Therefore, the magnitude of the substitution effects of ICT is important for policies which aim to support the take up of these technologies. This is particularly relevant for periods of low economic growth, such as the recent economic crisis and the current fragile recovery, when the compensation effect is likely to play a limited role.

AN US

Empirical evidence on the effect of ICT at firm level is mixed and inconclusive (Chennells and Van Reenen, 2002; Vivarelli, 2014). Studies examining the effect of ICT on employment by means of accurate firm-level measures of ICT use within firms are scarce, mainly due to lack of data (Falk, 2001, and Bloom et al., 2011, are exceptions). Most studies focused on

M

ICT-related technology change, such as process and organisational innovations, report mixed, often insignificant, effects on employment, despite theoretical suggestions that they should

ED

lead to employment destruction at least in the short run (Pianta, 2005; Vivarelli, 2014). Moreover, these studies often differ in their country, time and sector coverage, with most focusing only on the manufacturing sector. This variety in the results to certain extent reflects

PT

the ambiguous predictions of theory and the variety of methodological approaches (Vivarelli, 2014). However, when it comes to ICT, it mainly reflects the lack of accurate indicators of

CE

ICT use and internationally-comparable firm-level data. In this paper, we address the existing gap in the empirical literature by exploiting the unique

AC

linked micro-data that relates ICT variables to indicators of firms’ economic characteristics on a comparable basis across industry groups and countries. We examine the short run substitution effect of ICT use on firms’ employment separately for firms in manufacturing and services sectors in seven European countries (Finland, France, the Netherlands, Norway, Sweden, Poland, and the United Kingdom) during the period 2007-2010. To identify the substitution effect of ICT, we adopt a standard conditional labour demand model and control for unobservable time-invariant firm-specific effects which may be correlated with ICT use and firm labour demand.

4

ACCEPTED MANUSCRIPT Our results show no indication of significant ICT-induced labour substitution effects. This result is remarkably robust across countries, sectors and measures of ICT use. Our findings lend support to the current EU policies of promoting the take up of these technologies. The paper makes several major contributions to the literature. First, it is one of the very few studies that examine employment effect of ICT by means of highly accurate quantitative measures of ICT use within firms. On top of capturing aspects of ICT use that remain unexploited in the literature, these variables have an advantage of measuring both ICT use

CR IP T

and ICT diffusion within and across firms, without reaching full saturation common for the swiftly developing technologies. Second, it uses a unique dataset that contains internationally-comparable firm level data over the period 2007-2010 and examines how the relationship between ICT use and employment varies across countries and sectors. To our knowledge, this is the most representative multi-country micro-data to-date. Third, the paper

AN US

focuses explicitly on the substitution effect of ICT, which is highly policy relevant in the short to medium term and in periods characterised by low economic growth. The paper is organized as follows. Section 2 reviews the related literature. Section 3 describes the empirical specification, while Section 4 describes the data used in our paper. Section 5

Related Literature

ED

2.

M

discusses the results of the estimation. Section 6 offers conclusions.

Theoretical literature on the effect of ICT on firm labour demand suggests that several drivers are at work and that the overall effect is ambiguous. First, ICT may directly substitute

PT

workers whose tasks become automated. Second, by increasing firm productivity (as documented by a large body of literature, recently reviewed by Van Reenen et al., 2010, and

CE

Biagi, 2013), ICT enable firms to produce the same level of output with fewer inputs, including less labour. These two possible effects are often referred to in the literature as the

AC

"substitution effect" of ICT. Third, ICT-driven productivity improvements may lead to price decreases and higher demand for the firms' products, which, under certain circumstances3, would induce an outward shift of the (unconditional) labour demand. This effect is often referred to in the literature as the "compensation effect"4. The combined result of the

3

See Vivarelli (2007) for discussions on the factors that may affect this effect. Other "compensating" effects might arise due to the fact that i) employment will be increasing in the industries creating the new capital goods (the “machines”) that substitute for workers; ii) there could be positive indirect effects from ICT use and diffusion that originate in partial (i.e. industry level) and in general equilibrium (such as an increase in aggregate consumption and investment). 4

5

ACCEPTED MANUSCRIPT substitution and compensation effects is unclear5. The empirical literature on this topic is characterised by considerable heterogeneity in terms of the aggregation level of the analysis, effects studied (total, substitution or compensation effects), technology measures, empirical methods used and country and sector coverage. The employment effects of ICT are closely related to the employment effects of process and organisational innovations. These innovations are important drivers of productivity growth and they have negative substitution effects (at least in the short run) and positive

CR IP T

compensation effects on labour demand (Pianta, 2005; Vivarelli, 2014). The link between ICT investment and process and organisational innovations is complex, with ICT often being an enabler of these innovations. In particular, it has been shown that organizational innovation and ICT adoption tend to complement each other in determining productivity improvements and that ICT are a driver of (product and) process innovation (Brynjolfsson et

AN US

al., 2002; Brynjolfsson and Hitt, 2003; Polder et al., 2010).

From an empirical point of view, the employment impact of ICT is difficult to estimate: due to their multifaceted and intangible nature, technological change in general, and ICT diffusion in particular, are difficult to capture in data. A major challenge is faced by cross-

M

country studies where international comparability often comes at the expenses of the precision in ICT proxies. The vast majority of studies addresses the effect of ICT on

ED

employment at the macro and sector level, and, depending on the proxy used for ICT, different employment effects surface (this heterogeneity of results can be attributed in part to the fact that different economic and institutional mechanisms are not fully accounted for in

PT

the regressions6).

Firm-level analysis has significant measurement advantages for examining the effects of ICT

CE

use as it allows to eliminate many biases resulting from aggregation. However, it is often difficult to find good quality data representative of national economies, let alone

AC

multinational regions.

There are only a few studies that have explored firm-level data on ICT use (for the United States, see, for example, Brynjolfsson et al., 2002; Brynjolfsson and Hitt, 2003; Lichtenberg,

5

In addition to these direct effects, ICT diffusion may generate indirect employment effects in firms other than those using ICT, such as their competitors (a likely negative effect due to business stealing) and suppliers (a likely positive effect, through increased demand), with potential additional general equilibrium macroeconomic effects through increased consumption and investment. 6 A good example of this ambiguity is Severgnini (2009), who contrasts a set of specifications, commonly employed in studies of labour demand, and applies them to data within and across European and non-European countries. Depending on the choice of ICT variable, the effect on employment appears to be highly dispersed and rarely significant.

6

ACCEPTED MANUSCRIPT 1993; for Europe see van Leeuwen, 2008; Polder et al., 2010; Bartelsman, 2014), most of which document the positive effects of ICT on firm productivity, but also the complex links between the two. Papers that examine the effect of ICT on firm-level employment are even scarcer. Falk (2001), using data on German firms for the period 1995-1997, finds that ICT has an insignificant substitution effect on employment but a significant indirect effect that works through organisational change. Bloom et al. (2011) using firm-level data from 12 European

CR IP T

countries, find that ICT has a positive long-term effect on employment growth.

At the same time, due to the availability of micro-data, there is a large body of empirical literature on the employment effects of process and organisational innovation7.

Particularly relevant are the studies that estimate the substitution effect of process and organisational innovations. Harrison et al. (2014) find negative effects for manufacturing

AN US

firms in Germany and the UK, insignificant effects in Spain and France and insignificant effects for firms in services sectors in all four countries. Hall et al. (2008) and Giliodori and Stucchi (2012) find mixed, but mostly insignificant effects for manufacturing firms in Italy and Spain. Evangelista and Vezzani (2011) find a negative effect for firms in manufacturing

M

and service sectors in six European countries, while Falk (2014) finds insignificant results for firms in manufacturing and service sectors in Austria. Dachs and Peters (2014) find negative

ED

effects for firms in manufacturing and services sectors in 16 European countries. Overall, studies that focus on employment substitution effects tend to find negative, but often insignificant effects.

PT

In conclusion, while the theoretical literature predicts a negative substitution effect, the empirical evidence often finds insignificant results. The vast majority of the existing evidence

CE

is based on technology variables conceptually related to ICT, but only two studies (Falk, 2001 and Bloom et al., 2011) so far have used direct measures of ICT. This paper aims to fill

AC

this gap in the empirical literature by testing the employment substitution effect of several innovative direct indicators of ICT use, consistently defined and measured along a unique panel of seven European countries over four years.

3.

7

Empirical Specification

Chennells and Van Reenen (2002) and Vivarelli (2014) provide reviews of this literature.

7

ACCEPTED MANUSCRIPT We adopt a standard conditional labour demand framework and we assume that profitmaximising firms choose production inputs that minimise production costs, conditional on a given level of output and on factor prices, which the firm takes as given. We use the log linear model of the conditional labour demand, which can be derived from a CES production function with two inputs, capital and labour (for the derivation, see Hammermesh, 1993). There are two advantages of using this model. First, the coefficients can be interpreted as constant-output labour demand elasticities (Hamermesh, 1993). Second,

CR IP T

it enables comparability with previous studies, as this model has been commonly used to study the effects of technology changes on labour demand (Van Reenen, 1997; Zimmermann, 2009; Lachenmaier and Rottmann, 2007; 2011; Meriküll, 2010; Falk, 2014).

We assume capital to be a quasi-fixed input - following Berman et al. (1994) - and we include the capital stock instead of the user's cost of capital in the labour demand equation.

AN US

This substitution is reasonable for yearly variations, as it is difficult for firms to adjust the capital stock over such periods even if the price of capital changes significantly. This is in line with a standard neo-classical approach that conventionally assumes that physical capital is quasi-fixed in the short term, and with most empirical papers on employment effects of

M

technological change (Berman et al., 1994; Van Reenan, 1997). In addition, assuming that capital is quasi-fixed and using the capital stock instead of the user's cost of capital, allows us

ED

to avoid possible problems related to the measurement of the user's cost of capital, for which there is no reliable data at firm level.

To isolate the substitution effect of ICT, we estimate the constant-output effect of ICT on

PT

labour demand. We cannot separately estimate the compensation effect of ICT, but by controlling for output we control for all changes in labour demand due to changes in output,

CE

including those due to the compensation effects of ICT. Thus, the coefficient of the ICT variable captures only the substitution effect caused by the use of ICT.

AC

The relationship between ICT use and employment cannot be identified directly from the variation in intensity of ICT use and employment across firms because both variables could be correlated with unobserved firm characteristics. Not taking into account these characteristics could lead to omitted variable bias. Potentially relevant but unobserved in our data firm' characteristics affecting labour demand (and productivity) are: management ability, offshoring/outsourcing, use of other technologies, propensity to innovate and management and organisation characteristics including recruitment strategies. To the extent that these characteristics are time invariant over short periods of time, they can be eliminated by time

8

ACCEPTED MANUSCRIPT differentiating the conditional labour demand. Therefore, we estimate the labour demand equation in first differences. The conditional labour demand equation in first differences takes the following form: ΔlnLijt = δwΔlnWijt + δkΔlnKijt + δyΔlnYijt + δictΔICTijt + δxΔXijt + αt + αj +υijt

(1)

Δ denotes the difference between year t and year t-1, i represents the firm and j the industry in which firm i operates. L denotes employment, K real capital stocks, W real average wages and Y real output.

CR IP T

The detailed definitions of all variables used in the econometric analysis are reported in Table 1. The definitions of employment and the computation of the capital stocks slightly differ across countries. These differences, however, do not pose problems for the estimation of Eq.1, since the regressions are run separately for each country and these measures are consistent within countries over time. Given the log linear form of Eq. 1, the coefficients of

AN US

wages, capital and output can be interpreted as short-term elasticities. Other things being equal, we expect an increase in labour costs (real wages) to reduce labour demand, and an increase in output to increase it. As for capital, which enters the model as a quasi-fixed production input, we would expect it to have a positive effect on labour demand.

M

The main variable of interest is ICT, which denotes the intensity of ICT use within the firm. Three ICT use indicators are used in separate regression sets: i) the share of broadband

ED

internet-enabled workers; ii) the share of mobile internet-enabled workers; iii) the share of sales due to e-commerce activities. The first two measures are closely related to the number of computers available per worker (i.e. the measure used by Bloom et al., 2011). The third

PT

measure reflects the development stage of the firm's e-commerce activity. All three can be interpreted as refined measures of ICT infrastructures and of their use at the firm level. While

CE

each indicator has its specific aspects, we regard them as general indicators of ICT use within firms8 and we expect an increase in them to be positively related to labour productivity, in

AC

line with existing evidence9, and to reduce labour demand for a given level of output. Since Eq. 1 controls for real output, the coefficient of ICT would capture the labour substitution effects of ICT. Xijt represents other firm characteristics that may influence labour demand, such as a firm's age, its size, and multinational and export status. There is comprehensive evidence that small 8

These measures do not reflect automation and robotisation of production. However, the Brynjolfsson and McAfee (2011 and 2014) hypothesis indicates that the ICT effect is not limited to the effects of automation and robotisation of production workers, but that it affects a large number of occupations with different skill levels. 9 Hagsten and Sabadash (2014) found that the share of broadband internet-enabled workers and the share of mobile internet-enabled workers are associated with higher productivity in six European countries. Liu et al. (2014) found e-commerce activities have a positive effect on productivity for firms in Taiwan.

9

ACCEPTED MANUSCRIPT and large firms and young and old ones experience different employment dynamics. Multinationals and exporting firms are exposed to greater competition in international markets and, hence, they are likely to be more productive than domestic firms, which ceteris paribus is likely to reduce labour demand. Sector (αj) and year (αt) fixed effects control for unobservable factors that can influence labour demand within specific sectors or common macroeconomic shocks. We assume that they also capture any effects of prices of intermediate goods or energy. Due to the data

CR IP T

construction, sector dummies are based on two-digit NACE rev.1.1 classification of economic activities.

Eq (1) is estimated using OLS separately for manufacturing and services for each country, which allows the parameters of the labour demand to differ across countries and sectors.

The period studied covers the economic crisis, whose impacts at country and sector level are

AN US

captured by the year fixed effects (we allow them to vary by sector, i.e. manufacturing vs. services, and by country).

Conceptually, yearly differences are in line with the focus of the study on the short term substitution effect of ICT. If ICT affects employment similarly to process innovations, we

M

can refer to the literature suggesting that substitution effects of process innovation show up fairly soon after their introduction (Evangelista and Vezzani, 2011; Dachs and Peters, 2014;

ED

Peters et. al., 2014). On the other hand, the compensation effect (which is not the focus of this study) is generally thought to take longer time to materialise, since it mostly works through changes in the demand for firms' products and through general equilibrium

PT

adjustments. A related concern is that the substitution effect could be difficult to identify if employment adjustments are delayed due to labour market institutions or high costs of hiring

CE

and firing. Table 4 suggests that, over the period 2006-2010 and in all countries and sectors covered, employment displayed large variation, so that the inertia in the employment variable

AC

is not likely to be an important concern for the estimation. In addition to the short term substitution effect, ICT can also have a long term substitution effect on labour demand. We cannot examine this effect due to the short time dimension of the panel and to the large sample attrition. Imposing a two year lag structure on Eq.1 would lead to an important loss of observations (close to 30%). Therefore, we only look at the short term effect of ICT. 4

Data and Summary Statistics

10

ACCEPTED MANUSCRIPT The data come from the most informative datasets on ICT use in European firms. All estimations were carried out using the Distributed Microdata Approach (DMD) within the Esslait project. This method executes the same estimations from a centrally-managed platform on different harmonised national firm-level datasets kept at data repositories in different countries10. The use of the same code and the harmonisation of the national level datasets ensure that the results are comparable across countries. The data we used are widely regarded as state-of-art in linked micro data (Iancu et al. 2013). It is used extensively for

CR IP T

research (Lorenzani and Varga, 2014; Hagsten, 2015; Hagsten and Sabadash, 2016) and as a base for policy initiative such as Digital Single Market strategy (European Commission, 2014), which suggest that it is widely regarded as reliable.

While the DMD11 approach offers a solution to the challenge of generating firm-level and multi-country comparable analyses, it imposes restrictions on the number of specifications

AN US

and type of empirical methods that could be used and does not allow to pool data across countries.

Using this approach, in the context of three consecutive Eurostat-funded projects12, it was possible to link and harmonise several firm-level national data sources: the Survey on ICT

M

Usage and e-Commerce in Enterprises (also known as the e-Commerce or ICT usage survey), the Community Innovation Survey and the Production Survey.

ED

In our study, we used two linked surveys: the Production Survey (PS), which provides data on the economic performance of firms, and the e-Commerce Survey (EC), which provides data on ICT use. The PS compiles a set of variables describing the economic characteristics

PT

of firms that are drawn largely from the Business Register (BR) and the Structural Business Survey. It is the source of information on employment, real wages, real capital and real

CE

output, firms’ age and size, and exporting and multinational status. The EC survey is used for information on ICT usage, as captured by the three indicators:

AC

broadband internet-enabled employees, mobile internet-enabled employees and the share of e-sales. To our knowledge, there are no studies that use these variables to explore the impact of ICT use on changes in employment at the firm level. On top of capturing aspects of ICT

10

Because access to firm-level data is restricted in most countries, it is not possible to collect information on firms in a unique multi-country dataset. On the Distributed Microdata Approach see Bartelsman and Barnes (2001), Bartelsman (2004), and Bartelsman et al. (2013). 11 Bartelsman (2004); Eurostat (2008, 2012, 2013). 12 Eurostat ICT Impacts (Eurostat 2008), ESSnet on Linking of Microdata on ICT Usage (ESSLimit, Eurostat 2012) and ESSNet on Linking of Microdata to analyse ICT Impact (ESSLait, Eurostat 2013).

11

ACCEPTED MANUSCRIPT use that remain unexploited in the literature, these variables have an advantage of measuring both ICT use and ICT diffusion within and across firms without reaching full saturation13. The linked datasets cover firms in seven European countries14 (Finland, France, the Netherlands, Norway, Sweden, Poland, and the United Kingdom), in manufacturing and service sectors15, for four consecutive years, 2007-2010. When administrative data sources are linked to sample surveys, as in our case, the whole population of firms is not represented in the linked data. Moreover, the overlap between two

CR IP T

primary source datasets differs across countries, as a result of the difference in the underlying national sampling strategies. The coverage of the PS sample is about 50% of employment of firms in the BR in the Netherlands and in the UK and close to 100% in the other five countries studied, while the share of employment covered in the linked data is considerably lower, from 19.5% in the Netherlands to 78.1% in the UK (Table 2). To address the above

AN US

sampling bias issues, a weighting strategy was applied to ensure the representativeness of linked data. This method inflated the sample, so that the weighted sum of firms in the sample equals the number of firms in the population.

The linking of data16 from PS and EC leads to observation losses also because the EC covers

M

only firms with 10 or more employees17. As a result, the linked datasets represent a smaller share of firms compared to the share of employment (as Table 3 shows, coverage in terms of

ED

number of firms ranges from 0.3% in France to 18% in Poland), which leads to a slight bias towards larger firms in data merged across surveys or over time. However, Fazio et al. (2007) and Hagsten et al. (2013) show that this bias does not distort results from marginal analyses.

PT

Due to a certain amount of exit and entry by firms over time and due to the rotating nature of the surveys, only a smaller subset of firms appears in the sample every year. To reduce this

CE

bias we kept the panel unbalanced. Table 4 reports the mean scores of ICT use indicators and firm employment for each country

AC

and sector in 2007 and 2010 (the first and the last year in our sample) and the change over this period. In most countries and sectors, the shares of employees using mobile internet and the share of employees with access to broadband increased significantly between 2007 and 13

In the EC data, a full or nearly full saturation is reached, for example, by the dummy variable that indicates whether a firm uses internet. In other words, in developed economies, virtually all businesses use internet, which makes this variable ineffectual in discriminating between firms in a marginal analysis. 14 Only five countries provide data on the share of broadband enabled employees. 15 In the service sector, data does not cover public administration and defense, compulsory social security, education, health and social work and other community, social and personal service activities. 16 When linking data, we kept only those firms for which information on all variables of interest was available. 17 This reflects the sampling strategy aimed at covering as much of value added and employment in a given country as possible.

12

ACCEPTED MANUSCRIPT 2010. The evolution of the e-sales share in total sales varied considerably across countries: it increased significantly in the Netherlands and the UK, but it remained constant or even decreased in other countries. Average firm employment decreased in most countries and sectors over the period studied, which may be attributed in part to the effect of the economic crisis of 2007/2008 on the labour dynamics. We run country-year correlation tests separately for manufacturing and services to assess the relationship between employment and ICT use indicators (Table 5). Pearson's correlation

CR IP T

shows insignificant coefficients for all ICT use indicators in both services and manufacturing. To summarize, while most countries experienced an increase in ICT use and a decrease in employment during the period of analysis, our correlation test shows no significant negative correlation between the two variables, whose behaviour could be driven by different factors, such as the economic crisis in the case of employment and global technology diffusion in

AN US

case of ICT.

The next section will examine whether increases in ICT use are associated with decreases in employment at firm level.

Estimation Results

M

5.

Table 6 to Table 8 show the results of the estimation of Eq. 1. Results are reported separately

ED

for each measure of intensity of ICT use, sector and country. Specifically, Table 6 shows results for the share of employees with access to broadband, Table 7 for the share of employees with access to mobile internet, and Table 8 for the share of e-sales in total sales. In

PT

each table, the upper-panel reports the results for manufacturing firms and the lower-panel reports the results for service firms. Each column reports the results for a different country18.

CE

The results show that coefficients of output, capital and wages are statistically significant, and have the expected signs and plausible magnitudes in almost all estimations. The wage

AC

growth rate has a statistically significant negative effect on labour demand in all countries and sectors studied. Real output and capital rates of growth have significant positive effects in almost all estimations19. Overall, the variation in the explanatory variables included explains a large part of the variation of firms' labour demand (between 44% and 86%20). The high R squared value is remarkable, given that all estimations are in first differences, and indicates that the model has a good overall fit. The coefficients on other control variables (not 18

The number of countries differs across measures of ICT-use intensity, as some measures are not available for all countries. 19 The only exception is the coefficient of capital in Finland. 20 Except Poland, where it explains a lower share of the variation.

13

ACCEPTED MANUSCRIPT reported) are mostly insignificant and where they are significant, they do not suggest any clear pattern. The results indicate that ICT use within firms has an insignificant effect on firms’ labour demand in almost all countries and sectors studied. The hypothesis that the effect of ICT use on firms' demand for labour is zero cannot be rejected at conventional levels of statistical significance. Moreover, the magnitude of the coefficients (essentially zero, in most cases) and their small standard errors suggest that their insignificance is not due to the lack of power to

CR IP T

precisely estimate these effects.

More specifically, the share of employees with access to broadband (Broadpct) has an insignificant effect on labour demand in all countries and sectors, while the share of mobile internet-enabled employees (Mobpct) has an insignificant effect in all country-sector pairs except manufacturing firms in Poland and Sweden and service firms in the UK. E-commerce

AN US

activities (esales) have an insignificant effect on labour demand in all countries and sectors, except in the service sector in Finland. The consistency of the insignificant effect of ICT across ICT measures, countries and sectors is remarkable.

These results indicate that an increase in the intensity of ICT use is not significantly

M

associated to within-firm reductions in employment, hence lending no support to the hypothesis that ICT substitutes labour in the short run.

ED

The coefficient on sales captures all the effects of a change in output on labour demand, including the possible compensations effect of ICT. This coefficient is always positive and statistically significant. This suggests that the insignificant substitution effect of ICT may

PT

coexist with a positive effect of ICT on employment that works through increased demand for firm products ("compensation effects").

CE

These results are consistent with the insignificant direct effect of ICT on employment growth for German firms found by Falk (2001). They are not directly comparable, but compatible

AC

with the positive long-term total (substitution + compensation) effect of ICT found by Bloom et al. (2011). They are also broadly in line with the results of firm-level studies on the labour substitution effects of process innovation, which found insignificant or very small job displacement effects (Hall et al., 2008; Evangelista and Vezzani, 2011; Giuliodori and Stucchi, 2012; Falk, 2014; Harrison et al., 2014). Despite consistency with previous empirical results, the results may be seen as contradicting the prediction of the more pessimistic theoretical literature. A possible explanation is that reducing labour costs (i.e. competition on costs) is not the main purpose of the adoption and integration of ICT technologies in the production process. Their main purpose could be to 14

ACCEPTED MANUSCRIPT improve the reliability of production process, the quality of the final product or the overall efficiency of production and distribution or adapting to new market needs or regulations (i.e. competition on quality). 6.

Conclusions

This paper examines the labour substitution effects of ICT using firm-level data for manufacturing and services sectors in seven European countries, during the period 20072010. The data come from a unique dataset of harmonised and linked micro data compiled by

CR IP T

National Statistical Offices, containing across-country comparable data based on the production statistics linked with the ICT use survey. This is one of the very few studies that use specific measures of ICT use within the firm, and it is the first that uses a longitudinal dataset containing comprehensive and representative data from several European countries. We estimate the labour “substitution” effect of ICT, i.e. the effect due to ICT substituting for

AN US

labour and to ICT increasing productivity and hence reducing demand for inputs, for constant values of output. We adopt a standard conditional labour demand model and control for unobservable time-invariant firm-specific effects, which may be correlated with both labour demand and ICT use.

M

Our results indicate that the substitution effects of ICT are statistically insignificant. These results are consistently robust across countries, sectors and measures of ICT use. The

ED

insignificant substitution employment effect of ICT intensity may coexist with a positive "compensation" effect, which in our specification, is captured by the coefficient of real output. Overall, our results suggest that, on average, increased intensity of ICT use has at

PT

least a non-negative effect on employment at firm level. These results suggest that the current policy support for the take up and integration of ICT in

CE

enterprise processes should not be affected by fears of the possible negative employment effects of these technologies. Overall, they lend support to the current EU policies that

AC

encourage the take up of various ICT as a means of achieving growth and job creation. The study has several limitations. First, it estimates only the short-term effects of ICT and it covers a short and quite specific time period, characterised by the economic crisis. Estimation of the long-term effects of ICT would provide a useful future extension. Second, the results are very robust across countries, but the country coverage is limited, in particular for Southern and Central Eastern European countries and it is not clear whether the result could be easily generalised to these EU countries. Third, these firm-level results cannot be directly generalised to aggregate level, as they do not account for possible indirect effects of ICT, such as business-stealing or spillover effects. Fourth, the study does not examine the effects 15

ACCEPTED MANUSCRIPT of ICT use on different skill groups, which may be affected differently by the increased use of ICT. Future work should address these limitations.

Acknowledgments: The authors thank Marc Bogdanowicz, Martin Falk and Antonio Vezzani for detailed comments and suggestions. The authors also thank participants and discussants at Italian Association of Labour Economists conference, Esslait Project Workshop 2014, ICT Skills,

CR IP T

Jobs and Growth Workshop, 2014, Jornadas de Economia Laboral 2015, European Regional Science Association conference, 2015, Jornadas de Economia Industrial 2015 for helpful

AC

CE

PT

ED

M

AN US

comments. Any errors and omissions remain the responsibility of the authors.

16

ACCEPTED MANUSCRIPT

References Acemoglu, D. and D. Autor (2011). “Skills, Tasks and Technologies: Implications for Employment and Earnings”, in Handbook of Labour Economics, Volume 4b, Elsevier B.V. Autor, D. H., Frank, L. and J. M. Richard (2003). “The Skill Content of Recent Technological Change: An Empirical Exploration”. Quarterly Journal of Economics,

CR IP T

118, (4), 1279-1333.

Autor, D., Dorn, D. and G. H. Hanson, (2013). “Untangling Trade and Technology: Evidence from Local Labor Markets”, IZA Discussion Papers 7329.

Airaksinen, A., J. Berezowska, N. Djahangiri, E. Edelhofer, M. Redecker, and G. Zupan

AN US

(2013). “Study on Input Data Quality” in Final Report of the ESSnet on Linking of Microdata to Analyse ICT Impact, Eurostat.

Bartelsman, E. J. (2004). “The Analysis of Microdata from an International Perspective”, STD/CSTAT, 12, OECD.

M

Bartelsman, E. J. and Z. Wolf (2014), “Forecasting Productivity Using Information from Firm-Level Data”, The Review of Economics and statistics, 96 (3), 745-755.

ED

Bartelsman, E. J. and M. Barnes (2001). “Comparative Analysis of Firm-level Data: a Low Marginal Cost Approach”, Research Report for the OECD.

PT

Bartelsman, E. J., E. Hagsten and M. J. Polder (2013). Cross-country Analysis of ICT Impact using Firm-level Data: Micro Moments Database and Research Infrastructure, Eurostat.

CE

Beaudry, P., D. A. Green, and B. M. Sand. (2014). “The Declining Fortunes of the Young since 2000.” American Economic Review 104 (5), 381–386.

AC

Berman, E., J. Bound and Z. Griliches (1994). “Changes in the Demand for Skilled Labor within US Manufacturing: Evidence from the Annual Survey of Manufacturers”, Quarterly Journal of Economics, 109, 367-397.

Biagi, F. (2013). “ICT and Productivity: a Literature Review”, Digital Economy Working Paper 2013/07, JRC-IPTS. Bloom, N., M. Draca and J. Van Reenen (2011). “Trade-induced Technical Change? The Impact of Chinese Imports on Innovation, IT and Productivity,” NBER Working Papers 16717. 17

ACCEPTED MANUSCRIPT Brynjolfsson, E. and A. McAfee (2011). Race against the Machine: How the Digital Revolution is Accelerating Innovation, Driving Productivity, and Irreversibly Transforming Employment and the Economy, Digital Frontier Press. Brynjolfsson, E. and A. McAfee (2014). The Second Machine Age: Work, Progress, and Prosperity in a Time of Brilliant Technologies, W. W. Norton and Company. Brynjolfsson E., L. M. Hitt and S.Yang (2002). “Intangible Assets: Computers and

CR IP T

Organizational Capital”, Brookings Papers on Economic Activity, 1, 137-198. Brynjolfsson E. and L. M. Hitt, (2003). “Computing productivity: Firm-Level Evidence”, The Review of Economics and Statistics, 85, (4), 793-808.

Chennells, L. and J. Van Reenen (2002). “Technical change and the structure of employment

AN US

and wages: A survey on the micro-econometric evidence”, in Greenan, N., L’Horty, Y. and Mairesse J. (eds.), Productivity, inequality and the digital economy, MIT Press. Dachs, B. and B. Peters (2014). “Innovation, employment growth, and foreign ownership of firms a European perspective”, Research Policy, 43, 214– 232.

Data from Different Sources.

M

Eurostat (2008). Final Report, Information Society: ICT Impacts Assessment by Linking

ED

Eurostat (2012). Final Report, ESSNet on Linking of Microdata on ICT Usage. Eurostat (2013) Final Report, ESSnet on Linking of Microdata to Analyse ICT Impact.

PT

Evangelista, R. and A. Vezzani (2011). “The impact of technological and organizational innovations on employment in European firms”, Industrial and Corporate Change,

CE

21(4), 871–899.

European Commission (2014) A Digital Single Market Strategy for Europe - Analysis and

AC

Evidence Accompanying the document Communication from the Commission, European Commission Staff Working Documents.

Fazio, G., K. H. Lam and F. Ritchie (2006). “Sample Bias in Microeconometric Analyses of Official Microdata”, Report to the Department of Trade and Industry URN 06/737.

Falk, M. (2001) “Organizational change, new information and communication technologies and the demand for labor in services” ZEW Discussion Papers 01-25, ZEW.

18

ACCEPTED MANUSCRIPT Falk, M. (2014). “Employment Effects of Technological and Organizational Innovations: Evidence

based

on

Linked

Firm-level

data

for

Austria”,

Jahrbücher

f.

Nationalökonomie und Statistik, 23. Falk M. and Biagi F. (2016). “Relative Demand for Highly Skilled Workers and Use of Different ICT Technologies”, Applied Economics, forthcoming. Giuliodori, D. and R. Stucchi (2012). “Innovation and job creation in a dual labor market:

CR IP T

evidence from Spain”, Economics of Innovation and New Technology, 21 (8), 801-813. Hagsten, E. (2015). “Broadband connected employees and labour productivity: a comparative analysis of 14 European countries based on distributed Microdata access”, Economics of Innovation and New Technology, 1-17.

AN US

Hagsten, E. and Sabadash, A. (2016). “A Neglected Input to Production: The Role of ICTSchooled Employees in Firm Performance," International Journal of Manpower, forthcoming in 2016 Hamermesh, D. (1993). Labor Demand, Princeton University Press, Princeton, New Jersey. Hall, B. H., F. Lotti and J. Mairesse (2008). “Employment, innovation, and productivity:

M

evidence from Italian microdata”, Industrial and Corporate Change, 17, 813–839. Harrison, R., Jaumandreu, J., Mairesse, J., Peters, B., (2014). “Does Innovation Stimulate

ED

Employment? A Firm-Level Analysis Using Comparable Micro-Data from Four European Countries”. International Journal of Industrial Organization 35, 29-43.

PT

Iancu, D., E. Hagsten and P. Kotnik (2013). “Quality of Linked Firm-level and MicroAggregated Datasets: The Example of the ESSLait Micro Moments Database”, in Final

CE

Report of the ESSnet on Linking of Microdata to Analyse ICT Impact, Eurostat Lachenmaier, S., and H. Rottmann (2007). “Employment Effects of Innovation at the Firm

AC

Level”, Jahrbücher für Nationalökonomie und Statistik, 227(3), 254–272.

Lachenmaier, S., and H. Rottmann (2011). “Effects of Innovation on Employment: A Dynamic Panel Analysis”, International Journal of Industrial Organization, 29, 210– 220. Lichtenberg, F. (1993). “The Output Contributions of Computer Equipment and Personnel: A Firm-Level Analysis”, NBER Working Paper Series, WP No. 4540, November.

19

ACCEPTED MANUSCRIPT Liu, T.-K. & J-R. Chen, C. C. J Huang and C-H. Yang (2013). "E-commerce, R&D, and Productivity: Firm-level Evidence from Taiwan," Information Economics and Policy 25(4), 272-283. Lorenzani, D. and Varga, J., (2014)'The Economic Impact of Digital Structural Reforms', European Commission Economic Papers No 529, 2014. Meriküll, J., (2010). "The impact of innovation on employment: firm- and industry-level

CR IP T

evidence from a catching-up economy," Eastern European Economics, 48(2), 25–38. Pianta, M., (2005) "Innovation and employment." in The Oxford Handbook of Innovation (2005), 568-598.

Polder, M., Leeuwen, G. V., Mohnen, P., & Raymond, W. (2010). Product, process and

Scientific Publications 2010s-28.

AN US

organizational innovation: drivers, complementarity and productivity effects. CIRANO-

Peters, B., Dachs, B., Duenser, M., Hud, M., Koehler, C., and C. Rammer (2014), Firm Growth, Innovation and the Business Cycle. Background study for the European Competitiveness Report 2014.

M

Sabadash, A. (2013). “ICT-induced Technological Progress and Employment: A Literature

ED

Review” Digital Economy Working Paper 2013/07, JRC-IPTS. Severgnini, B. (2009). Growth Accounting, ICT and Labour Demand, in Burda M. (ed). The Impact of ICT on Employment, Humbold-Universitat zu Berlin.

PT

Van Reenen, J. (1997). “Employment and Technological Innovation: Evidence from U.K.

CE

Manufacturing Firms,” Journal of Labor Economics, 15(2), 255-284. Van Reenen J., N. Bloom N, M. Draca, T. Kretschmer, R. Sadun, H. Overman and M. Schankerman (2010). “The Economic Impact of ICT”, Final Report for the European

AC

Commission.

Vivarelli, M. (2007) “Innovation and Employment: a Survey”, IZA Discussion paper, 2621. Vivarelli, M. (2014). “Innovation, Employment and Skills in Advanced and Developing Countries: A Survey of Economic Literature”, Journal of Economic Issues, 48(1), 123154.

20

ACCEPTED MANUSCRIPT Zimmermann, V. (2009). “The impact of innovation on employment in small and medium enterprises with different growth rates” Jahrbücher für Nationalökonomie und

AC

CE

PT

ED

M

AN US

CR IP T

Statistik, 229, 313-326.

21

ACCEPTED MANUSCRIPT Annexes Tables Table 1 Variable Definition Variable ICT measures Broadband intensity (Broadpct) Mobile internet intensity (Mobpct) E-commerce activities (e-sales) Firm characteristics Employment (Lijt)

Definition Share of broadband internet-enabled workers Share of mobile internet-enabled workers Share of sales through the firm's web-site and through Electronic Data Interchange (EDI) in total sales Full-time equivalents (FTE) in Finland, Poland and Sweden and head-counts in France, Netherlands, Norway and the United Kingdom. Real total wage bill per unit of employment in the firm. Nominal wage bill is deflated using output price index. Real capital stocks were estimated by the ESSlait project using book value, replacement value of assets or depreciation in monetary terms (Airaksinen et al., 2013). Firms’ sales from all products, goods, materials and services, including price subsidies, consumption tax and excise duties (excluding value added taxes), purchase value of goods resold, as well as indirect services. Nominal output is deflated using industry price indexes. Dummy variable indicating whether the firm's age is: less than 3 years old, between 3 and 6 years old, between 6 and 9 years old, between 9 and 12 years old, between 12 and 15 years old or older than 15 years Dummy variable indicating whether the firm has: less than 10 employees, between 10 and 19 employees, between 20 and 49 employees, between 50 and 199 employees, between 100 and 249 employees, between 250 and 499 employees, or above 500 employees Dummy variable indicating whether the firm has foreign subsidiaries and/or it is part of group with headquarters abroad. Dummy variable indicating whether the firm exports any goods and/or services.

CR IP T

Real average wage Real capital stock (Kijt) Real output (Yijt) Age

AN US

Size

Multinational status Exporter status

ED

M

Table 2 Coverage through the merging procedure: share of employment captured in the linked dataset, 2010 Country BR (no. of employees) BR-PS (% of BR) BR-PS-EC (% of BR) Finland 1 113 088 95.6 43.3 France 4 373 323 100.0 78.1 The Netherlands 4 548 025 50.9 19.5 Norway 1 218 653 95.4 41.8 Poland 4 108 381 100.0 55.8 Sweden 1 886 255 100.0 34.1 The United Kingdom 15 422 078 52.0 27.8 Source: Adapted from Iancu et al. (2013), p.8.

AC

CE

PT

Table 3 Coverage through the merging procedure: share of firms captured in the linked dataset, 2010 Country BR (no. of firms) BR-PS (% of BR) BR-PS-EC (% of BR) Finland 182 009 54.1 1.4 France 23 253 0.8 0.3 The Netherlands 937 362 5.0 0.5 Norway 398 577 44.5 0.8 Poland 56 958 100.0 18.0 Sweden 569 478 100.0 0.5 The United Kingdom 1 366 044 2.6 0.2 Source: Adapted from Iancu et al. (2013), p.7.

22

Table 4 Descriptive Statistics: means Country Year

FI

FR Δ(%)

2007

NL

NO Δ(%)

2007

2010

45.4

4.3

47.2

29.8

13.4

22.6

5.8

12.2

6.4

-7.0

109.0

99.3

48.1

-9.2

65.3

66.0

42.4

15.1

32.3

45.7

2010

Δ(%)

2007

2010

2007

Broadpct (%)

46.7

52.5

5.9

44.7

38.8

-5.9

41.1

Mobpct (%)

34.6

48.0

13.4

24.3

37.3

13.0

16.4

E-sales (%)

9.2

10.8

1.7

15.4

11.5

-3.8

Employment

64.7

62.2

-3.8

160.5

149.2

Broadpct (%)

70.0

76.4

6.4

57.3

Mobpct (%)

48.5

69.3

20.9

27.3

2010

PL

Δ(%)

2007

53.7

6.4

44.5

21.9

31.7

34.4

-8.8

50.8

47.8

0.6

63.8

13.4

29.9

6.1

7.7

1.6

7.9

6.0

-1.9

6.7

14.0

Employment

43.9

44.1

0.4

189.1

240.7

27.3

77.4

77.7

7.3 0.4

2007

2010

Δ(%)

16.4

25.6

9.1

47.0

66.4

19.4

39.9

50.1

10.2

9.6

16.6

7.0

31.1

60.4

29.2

30.1

45.4

15.3

2.7

5.2

7.4

2.1

34.9

35.1

0.3

9.3

23.6

14.3

-5.9

93.5

86.8

-7.1

67.2

60.7

-9.8

192.4

259.9

35.1

70.2

6.4

38.1

47.3

9.2

65.5

65.3

-0.3

50.5

60.3

9.8

56.2

26.3

22.9

30.9

7.9

44.8

56.5

11.7

42.3

56.4

14.1

28.0

29.0

1.0

3.4

4.8

1.4

25.5

27.6

2.1

7.0

11.4

4.4

31.6

34.3

8.7

70.6

67.1

-4.8

36.9

36.9

-0.1

404.1

545.7

35.0

AC

CE

PT

ED

M

Notes: Own calculations based on Esslait. For Sweden, the change in E-sales is calculated for 2007-2009. The statistics are based on the joint Production and E-commerce Surveys sample. Table 5 Descriptive Statistics: Pearson correlation coefficient Manufacturing Broadpct (%) -0.1188 Mobpct (%) 0.0176 E-sales (%) -0.0755 Employment 1.0000 Services Broadpct (%) -0.3166 Mobpct (%) 0.0683 E-sales (%) -0.2685 Employment 1.0000 Notes: Correlations are calculated for the country- and year-specific means (data constructions do not allow access to the firm-level analysis for descriptive statistics).

UK

Δ(%)

AN US

E-sales (%)

SE

2010

Manufacturing

Services

CR IP T

ACCEPTED MANUSCRIPT

2007

2010

Δ(%)

AC

CE

PT

ED

M

AN US

Table 6 Effect of changes in ICT use (measured as share of employees with mobile internet access) on changes in employment Country FI FR NO SE UK Manufacturing ΔlnWage -0.87 *** -0.77 *** -0.44 *** -0.69 *** -0.41 *** [0.05] [0.01] [0.03] [0.02] [0.02] ΔlnCapital -0.03 * 0.17 *** 0.07 *** 0.11 *** 0.23 *** [0.02] [0.01] [0.01] [0.01] [0.05] ΔlnOutput 0.83 *** 0.32 *** 0.35 *** 0.43 *** 0.37 *** [0.01] [0.01] [0.01] [0.01] [0.02] ΔBroadpct 0.01 0.00 0.00 0.01 0.04 [0.03] [0.01] [0.02] [0.01] [0.04] Obs. 1647 3654 1304 1367 837 R-squared 0.86 0.61 0.52 0.63 0.44 Services ΔlnWage -0.84 *** -0.96 *** -0.29 *** -0.76 *** -0.62 *** [0.04] [0.01] [0.01] [0.02] [0.01] ΔlnCapital 0.01 0.04 *** 0.04 *** 0.05 *** 0.09 *** [0.01] [0.01] [0.00] [0.01] [0.02] ΔlnOutput 0.90 *** 0.59 *** 0.43 *** 0.54 *** 0.39 *** [0.01] [0.01] [0.01] [0.01] [0.01] ΔBroadpct -0.04 0.00 0.01 0.01 -0.01 [0.03] [0.01] [0.01] [0.02] [0.02] Obs. 1786 3900 2809 1773 1810 R-squared 0.81 0.84 0.43 0.62 0.57 Notes: OLS in first differences estimates. Dependent variable is Δln(Employment). All equations include controls for export and multinational status, age and size categories, sector and year fixed effects. Standard errors are reported in the brackets. *, ** and *** indicate significance at 10%, 5% and 1%, respectively.

CR IP T

ACCEPTED MANUSCRIPT

CR IP T

ACCEPTED MANUSCRIPT

AC

CE

PT

ED

M

AN US

Table 7 Effect of changes in ICT use (measured as share of employees with mobile internet access) on changes in employment Country FI FR NL NO PL SE UK Manufacturing ΔlnWage -0.87 *** -0.77 *** -0.67 *** -0.44 *** -0.34 *** -0.69 *** -0.41 [0.05] [0.01] [0.01] [0.03] [0.04] [0.02] [0.02] ΔlnCapital -0.03 * 0.17 *** 0.05 *** 0.07 *** 0.01 *** 0.11 *** 0.23 [0.02] [0.01] [0.01] [0.01] [0.00] [0.01] [0.05] ΔlnOutput 0.83 *** 0.32 *** 0.27 *** 0.35 *** 0.24 *** 0.43 *** 0.37 [0.01] [0.01] [0.01] [0.01] [0.01] [0.01] [0.02] ΔMobpct 0.00 0.01 0.01 0.01 -0.07 *** 0.02 * -0.01 [0.02] [0.01] [0.01] [0.01] [0.03] [0.01] [0.03] Obs. 1647 3654 1349 1304 1184 1367 837 R-squared 0.86 0.61 0.70 0.52 0.34 0.63 0.44 Services ΔlnWage -0.85 *** -0.96 *** -0.84 *** -0.29 *** -0.26 *** -0.76 *** -0.62 [0.04] [0.01] [0.01] [0.01] [0.04] [0.02] [0.01] ΔlnCapital 0.01 0.04 *** 0.06 *** 0.04 *** 0.01 * 0.05 *** 0.09 [0.01] [0.01] [0.01] [0.00] [0.00] [0.01] [0.02] ΔlnOutput 0.90 *** 0.59 *** 0.33 *** 0.43 *** 0.10 *** 0.54 *** 0.39 [0.01] [0.01] [0.02] [0.01] [0.01] [0.01] [0.01] ΔMobpct 0.01 0.00 0.01 -0.01 0.01 0.01 -0.04 [0.02] [0.01] [0.01] [0.01] [0.02] [0.01] [0.02] Obs. 1786 3900 1503 2809 1073 1773 1810 R-squared 0.81 0.84 0.77 0.43 0.20 0.62 0.57 Notes: OLS in first differences estimates. Dependent variable is Δln(Employment). All equations contain controls for export and multinational status, age and size categories, sector and year fixed effects. Standard errors are reported in the brackets. *, ** and *** indicate significance at 10%, 5% and 1%, respectively.

*** *** ***

*** *** *** **

CR IP T

ACCEPTED MANUSCRIPT

AC

CE

PT

ED

M

AN US

Table 8 Effect of changes in ICT use (measured as share of e-sales in total sales) on changes in employment Country FI FR NL NO PL SE UK Manufacturing ΔlnWage -0.87 *** -0.77 *** -0.68 *** -0.49 *** -0.34 *** -0.83 *** -0.41 *** [0.05] [0.01] [0.01] [0.06] [0.05] [0.03] [0.02] ΔlnCapital -0.03 * 0.17 *** 0.05 *** 0.06 *** 0.01 *** 0.05 ** 0.23 *** [0.02] [0.01] [0.01] [0.01] [0.00] [0.02] [0.05] ΔlnOutput 0.84 *** 0.32 *** 0.28 *** 0.60 *** 0.24 *** 0.44 *** 0.37 *** [0.01] [0.01] [0.01] [0.02] [0.01] [0.03] [0.02] ΔE-sales 0.04 0.01 -0.01 0.00 0.00 0.02 -0.02 [0.04] [0.01] [0.01] [0.02] [0.02] [0.04] [0.03] Obs. 1570 3655 1724 418 1184 352 837 R-squared 0.86 0.61 0.69 0.76 0.34 0.81 0.44 Services ΔlnWage -0.81 *** -0.96 *** -0.83 *** -0.39 *** -0.26 *** -0.54 *** -0.62 *** [0.04] [0.01] [0.01] [0.03] [0.04] [0.04] [0.01] ΔlnCapital 0.01 0.04 *** 0.05 *** 0.03 *** 0.01 * 0.02 0.09 *** [0.01] [0.01] [0.01] [0.01] [0.00] [0.02] [0.02] ΔlnOutput 0.90 *** 0.59 *** 0.37 *** 0.40 *** 0.10 *** 0.36 *** 0.39 *** [0.01] [0.01] [0.02] [0.02] [0.01] [0.03] [0.01] ΔE-sales 0.09 * 0.00 0.01 0.00 0.02 0.04 -0.04 [0.04] [0.01] [0.02] [0.01] [0.04] [0.04] [0.03] Obs. 1696 3900 1961 1000 1073 421 1810 R-squared 0.81 0.84 0.74 0.42 0.20 0.49 0.57 Notes: OLS in first differences estimates. Dependent variable is Δln(Employment). All equations contain controls for export and multinational status, age and size categories, sector and year fixed effects. Standard errors are reported in the brackets. *, ** and *** indicate significance at 10%, 5% and 1%, respectively.