Case Study 4 and 6 Questions Need to Be Answered
Caiden
assignment_2.pdf
Ann Reg Sci (2010) 45:401–421 DOI 10.1007/s00168-009-0315-6
O R I G I NA L PA P E R
Productivity and employment—Is there a trade-off? Comparing Western European regions and American states 1950–2000
Kerstin Sofia Enflo
Received: 24 October 2007 / Accepted: 24 June 2009 / Published online: 17 July 2009 © Springer-Verlag 2009
Abstract This article report findings from a new dataset that consists of productivity and employment variables from 89 Western European regions and 51 American states and districts from 1950 to 2000. Distribution dynamics is used to investigate conver- gence in labor productivity, Gross Value Added (GVA) per capita, and employment ratios. Main findings are that European labor productivity and GVA per capita have converged more or less continuously since 1950, but that the European employment ratios show divergence after 1970. Compared to US, the European regions have faced an employment-productivity trade-off since the 1970s. This trade-off appears to be related to country-specific factors.
JEL Classification J21 · O47 · R11 1 Introduction
Because the seminal articles written by Barro and Sala-i-Martin (1991, 1992, 1995), many studies have focused on the question of regional convergence. The regional framework has often been proposed to test the neo-classical convergence hypothesis, because institutional factors are more similar across regions within a country than between nations. In Western Europe, the convergence hypothesis has also received much attention after the formation of the European Union, which specifies equaliza- tion of regional income differences as one of its pronounced goals. However, recent studies have cast a doubt on the idea of regional convergence in Western Europe, instead reporting a slow-down in convergence after 1980 (Tondl 1999; Fagerberg and Verspagen 1996)or arguing that regions are converging into different regional clubs
K. S. Enflo (B) Department of Economic History, Box 7083, 220 07 Lund, Sweden e-mail: [email protected]
123
402 K. S. Enflo
(Quah 1996a). For the United States on the other hand, there is rather strong consensus about convergence in Gross Value Added (GVA) per capita (Barro and Sala-i-Martin 1992; Johnson 2000; Rey and Montouri 1999).
Most previous empirical convergence studies focus only on convergence in GVA per capita, despite the fact that the theoretical growth models actually make predic- tions about the distribution of GVA per worker (Solow 1956; Romer 1990). To avoid the assumption of constant unemployment and participation rates, this article focuses on the variables that make up the productivity differentials across the regions; GVA and employment. These variables are studied simultaneously since 1950, which is a longer time period than covered by most previous regional convergence studies. The simultaneous focus on productivity and employment variables also put focus on the relationship between the two in making up convergence in GVA per capita. Recently, some authors have suggested that there might be a trade-off between employment and productivity growth (Gordon 1995; McGuckin and van Ark 2005) and that this may lie behind slow European regional convergence rates in per capita GVA (Meliciani 2006). Meliciani especially argues that productivity increase have taken place in the least productive regions at the expense of employment by stating that “…increased competition through trade liberalization with low labour mobility can have forced convergence in labour productivity by means of a reduction of employment rates in low productivity regions” (2006, p. 88).
This study employs the distribution dynamics approach to convergence suggested by Quah (1993a, 1996b) that consists of non-parametric Kernel density diagrams and transition probability matrices. The distribution dynamics approach has also become a popular tool to investigate whether convergence has taken place between the regions of EU. Pioneering these regional convergence studies, Quah (1996a) found that the GVA per capita distribution of 78 NUTS two regions in Western Europe displayed a similar pattern of increased bipolarization between 1980 and 1989. The absolute majority of these regional distribution dynamics studies focuses on convergence in GVA per capita in various time periods and sub-periods after 1980 and utilizes data on different NUTS-levels (Benito and Ezcurra 2005; Rey and Janikas 2005). The Quah-approach has also been used to study convergence in GVA per capita across the American states 1929 to 1993 (Johnson 2000). Fewer studies have focused on productivity and employment variables, but Overman and Puga (2002) investigate the European employment distribution and report polarization in regional unemployment rates since the mid 1980s. For the labor productivity distribution, Lopez-Bazo et al. (1999) find fast continuous regional convergence since 1980.
The objective of this article is first to analyze the European convergence process in comparison to USA since 1950. Secondly, the article analyzes whether there has been a trade-off between employment and productivity growth, as suggested by Meliciani (2006), and to what extent this potential trade-off is a distinctively European feature. For this purpose, a novel regional dataset that covers 89 Western European regions from 10 countries and 51 American states and districts has been put together. This data- set allows for a convergence analysis that starts with the formation of the EU in 1950, whereas most previous European regional convergence studies have focused on the period after 1980, because that was when Eurostat started to collect regional data more systematically in the Nomenclature of Territorial Units for Statistics (NUTS). The 89
123
Productivity and employment 403
Western European regions are taken from the following 10 countries: France, Western Germany, Italy, Belgium, Netherlands and Luxembourg (the founding countries of EU), Denmark, Ireland, UK (the first expansion), and Spain (the third expansion).
2 Employment and productivity—a trade-off?
Several studies have documented a trade-off between employment and productivity growth at the national level. However, earlier studies have focused on how European increases in the participation rates since the mid 1990s have affected productivity neg- atively. McGuckin and van Ark (2005) have for example argued that increased labor force participation have brought in elements of the workforce with lower marginal productivity that are off-setting productivity gains—at least temporary. Using a panel of OECD countries, they find that especially female labor force participation has neg- ative productivity effects. These negative effects are however specific to certain ages and cohorts and thus likely to diminish over time, according to the authors.
There are some theoretical reasons to expect a negative correlation between employ- ment and productivity growth. First of all, increasing employment could have adverse effects on productivity simply because the labor demand curve has a negative slope. The trade-off also works in the opposite direction: when there are decreasing returns to labor worker, layoffs will increase labor productivity. Increase in employment ratios could also be followed by lower labor productivity growth, because there would be a need to equip workers with scarce capital. As countries become more integrated over time and capital flows increase, there are reasons to expect the employment pro- ductivity trade-off to disappear. Thus, statically one may expect a trade-off between employment and productivity growth, but in a dynamic framework capital accumu- lation will flow to productive regions and offset the diminishing returns to labor. Therefore, one would suspect that increased integration and capital mobility would lead to a moderation of the negative correlation between employment and productivity over time.
Secondly, institutional factors may also affect this trade-off. Gordon (1995, p. 25) analyzes the relationship between labor productivity and employment and argues that any institution, policy, or event that boosts the real wage will move the economy northwest along the labor demand curve, simultaneously raising unemployment and the marginal and average product of labor. This effect would introduce a negative correlation between productivity (which is rising since fewer people are working) and employment. In this case, the marginal returns to capital are decreasing and will be adjusted downward as older capital vintages will not be replaced by new ones and productivity will be decreasing. Until then, however, the negative correlation is likely to persist.
This article hypothesizes that any negative correlation between employment and productivity growth should have been disappearing between 1950 and 2000 due to capital markets’ integration, unless emerging specific institutions sustained the trade-off. Employment protection laws (EPL:s) and union power may be examples of such institutions. These institutions are found in many European countries, which raises the question of whether European countries are more likely to experience a trade-off than the United States.
123
404 K. S. Enflo
0
0.5
1
1.5
2
2.5
3
3.5
4
1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000
Denmark France Germany Italy UK USA
Fig. 1 Index of Strictness of Employment Protections Laws in a sample of European countries and USA, Source: Allard (2003); Allard and Lindert (2006)
Figure 1 displays an index of the strictness of Employment Protection Laws in a selection of European countries and the US for the period 1950—2000 created by Allard (2003). As seen in the picture the USA and the UK display lower worker pro- tection compared to other countries. The figure also shows that employment protection had a tendency to become stricter during the 1970s.
However, contradictory to the expectations of a disappearing trade-off as econo- mies become more integrated, recent findings at the level of industrialized countries suggest that the trade-off increased over the period 1960–1997 (Beaudry and Collard 2002). This evolution has been explained by the role that employment increases had in lowering productivity during the 1990s (McGuckin and van Ark 2005).
This calls for an analysis of the trade-off for earlier historical periods, because employment ratios in many European countries were lower than the American in 1980 and 1990. In 1990, only Denmark and Luxemburg had higher employment to population ratios than the US. Between 1990 and 2000 the Netherlands’ employment ratio also surpassed USA in due to various labor market reforms, such as reductions in social transfers and unemployment benefits as well as a policy of wage moderation.1
With a narrow focus on the negative productivity effects of employment increases, an American trade-off in employment and productivity would be expected between 1980 and 1990. However, if the trade-off was instead caused by an artificial productivity maintenance at the expense of employment, as hypothesized by Meliciani (2006), or by institutional forces specific to Europe, as hypothesized by Gordon (1995), we may expect a European trade-off during this earlier period as well. The employment ratios for 10 European countries and the United States can be found in Table 1.
1 The Dutch experience has been called the Delta-model and has generally been labeled a success although some authors have pointed out that it has come at a productivity cost (van Ark and de Haan 2000, p. 312).
123
Productivity and employment 405
Table 1 Employment to population ratios
1950 1960 1970 1980 1990 2000
Belgium 0.38 0.38 0.37 0.37 0.37 0.38
Denmark 0.46 0.44 0.47 0.48 0.51 0.50
France 0.46 0.40 0.40 0.40 0.39 0.40
West Germany 0.41 0.47 0.44 0.44 0.45
Ireland 0.41 0.37 0.35 0.34 0.33 0.44
Italy 0.36 0.42 0.37 0.38 0.40 0.40
Luxembourg 0.43 0.42 0.41 0.43 0.49 0.60
Netherlands 0.36 0.36 0.38 0.39 0.42 0.50
Spain 0.41 0.38 0.37 0.31 0.33 0.38
U.K. 0.44 0.45 0.44 0.44 0.46 0.46
U.S.A 0.40 0.36 0.38 0.44 0.48 0.49
Source: Groningen Growth and Development Centre and the Conference Board, Total Economy Database, August 2005, http://www.ggdc.net
3 Methodology
3.1 Distribution dynamics
This article will utilizes the distribution dynamics approach formulated by Quah (1993a, 1996a) to assess the strength of regional convergence in the shape of the distributions of GVA per capita (which is also referred to as “regional income”), pro- ductivity and employment.
The vast majority of earlier empirical growth and convergence studies have evolved from standard cross-country regression analyses that take a negative correlation between initial income and average annual growth in subsequent periods as evidence of the β-convergence hypothesis. These regressions often include a set of conditioning parameters to control for differing steady states, so-called conditional convergence. However, the appropriateness of this method has been criticized on several grounds. First, Cheshire and Carbonaro (1995) do not agree with the common practice of includ- ing control variables, because they inevitably imply that the convergence measure will be dependent on the choice of conditioning variables. If these control variables are in fact proxies for forces leading to divergence, the β-coefficient may still turn positive and significant. Cheshire and Magrini (2000) show that the estimates of β-convergence are sensitive to the choice of conditioning variables added to the model. Second, Quah (1993b) has pointed out that, not only does the regression analysis approach suffer from Galton’s fallacy of regression towards the mean, but also inappropriate whenever the underlying process of growth is unstable. Regression-based studies focus on the average economy making the implicit assumption that all observations follow a smooth homogenic transition toward their own steady-state. However, Quah (1996a,b) shows that cross-country income displays strong instability in the underlying growth process. In addition, Canova and Marcet (1995) have pointed out that the 2% convergence rate
123
406 K. S. Enflo
that are commonly found in regression-based studies may well arise from reasons that are independent from the dynamics of economic growth.
To overcome the above-mentioned problems, Quah suggests that the issue of con- vergence should be related to the evolution of the whole cross-sectional income dis- tribution by using so-called distribution dynamics. The methodology has clear merits in the case where the analyzed distribution exhibits more than one peak, a feature that ordinary regression analysis fails to detect, but is also useful for convergence analysis for the reasons outlined above. In the case of the present dataset, there are reasons to expect that dome of the distributions may be multiple-peaked. Earlier research has for example noted a group of high-productivity regions to cluster at the very top-end of the European income distribution, forming a second peak between 1979 and 1990 (Magrini 1999). Multimodality of the income distribution has also been found on a national level. Employing a distribution dynamics approach, Quah describes the evo- lution of the cross country per capita income distribution between 1960 and 1988 as increasingly polarized into “twin peaks” of rich and poor countries. Epstein et al. (2003) extend the approach to cover income distributions from 1870 to 1992, suggest- ing that convergence was a temporary phenomenon mainly found in the period after the World War II, but that it gave way to polarization after 1970.
Quah suggested methodology makes use of stochastic kernels to describe the con- vergence process. The stochastic kernels are either discretized into transition probabil- ity matrices or analyzed in the continuous case by looking at kernel density diagrams. The kernel density diagram can be thought of as a smoothed histogram, but it has the advantage over histograms of being continuous and not sensitive to the choice of bin widths. All kernels are based on the Epanechnichov weighting function and their band- width has been chosen in accordance with Silverman’s optimal method (Silverman 1986).
Transition probabilities measure whether regions with above median income or employment figures are more likely to stay above median throughout the time period or whether there is evidence of mobility in the distributions. Five income, productiv- ity, and employment states are defined (much above median, above median, median, below median, and much below median) to count the number of regions transiting from one state to another during a time span of 10 years.2 The transition matrices reported give the average probability of regions moving from one income, employ- ment or productivity state to another and inform us about the mobility and persistency of the distributions.
In addition to computing the average probability of one region moving from one state to another, the transition matrices are also helpful in calculating the long-run equilibrium under the mobility pattern in question. Continuous iteration of any tran- sition probability matrix will yield a steady state distribution where further iteration does not have any effect, called the ergodic distribution of the system. The ergodic distribution tell us something about the inherent dynamics of our estimated system and is useful in quantifying what the regional map of Europe would look like if the probabilities of transition would remain unchanged until steady state was reached. For
2 The time span of transitions is limited by the specific nature of the data set that only allows for 10-year transitions.
123
Productivity and employment 407
technical details about the kernels, transition probabilities and the ergodic distribution of a transition matrix, the reader is referred to Quah (1996a) or to Epstein et al. (2003).
4 Variables and data
The simultaneous study of the three distributions for each decade since 1950 allows for the detection of the sources of growth during different economic regimes. In par- ticular, we can decompose per capita income into the product of labor productivity and employment per capita:
GDP
population = GDP
employee × employee
population
This simple decomposition allows for the assessment of which variable that has played the largest role in regional growth and convergence and whether one variable has improved at the expense of the other. Following a number of other studies (Allard and Lindert 2006; McGuckin and van Ark 2005) the article focuses on employment ratio rather than labor force participation or unemployment. This is due to several reasons apart from data availability at the regional level. First, the decision to enter into the labor force is not completely independent of the unemployment rate. Second, the measure is less subject to measurement differences between countries and regions.
The collection of regional Western European data on employment, GVA and pop- ulation draws upon previous study by Molle et al. (1980) for the regions from UK, France, Netherlands, Belgium, and Western Germany from 1950 to 1970. The Italian data was collected from Molle et al. (1980) as well, but has been quality adjusted in a few cases using data from CRENoS databank.3 Regional data for 17 Spanish regions from 1950 was taken from Alcaide (2003). From 1980 and onwards all regional data was collected from Cambridge Econometrics, a database that builds heavily on Euro- stat’s regional accounts. Because Molle et al. (1980) only report data for the benchmark years 1950, 1960, and 1970; and Alcaide (2003) only reports data every fifth year since 1930, the constructed dataset is not an annual time series, but a series of cross-sectional observations measured in every tenth year since 1950. This feature does not pose any problem for the distribution dynamics analysis at hand.
Given that three datasets have been merged to create the European database, there are potential problems with inconsistency in the definitions of population, employ- ment, and GVA in the underlying sources. To overcome the problem of large national biases due to such inconsistencies, the underlying sources have been used to calculate the regional shares of the national total for each variable. The obtained regional shares have thereafter been multiplied with internationally comparable national data from the Maddison dataset.4 This means that regional GVA is measured in millions of 1990 US dollars (converted at Geary Khamis PPP) and employment and population are
3 University of Cagliari, www.crenos.it 4 Groningen Growth and Development Centre and The Conference Board, Total Economy Database, January 2005, http://www.ggdc.net
123
408 K. S. Enflo
measured in thousands.5 If one is willing to accept that differing national definitions of employment and GVA do not affect the internal regional distribution of the variables within each country and that the definitions are stable over time, this procedure will yield internationally comparable and time-consistent regional estimates.
Molle et al. (1980) regional dataset ends in 1970 whereas Cambridge Econometrics only starts in 1980, so overlapping years from the two main datasets are unfortunately unavailable for the majority of the regions. In the Spanish case Alcaide (2003) provides regional data for the whole period up to 2000 and in the Italian case the CRENoS6 data bank provides regional data from 1950 to 2000. These two overlapping datasets have been used to make sure that there are no too-large fluctuations, for the Spanish and Italian provinces at least, between Molle’s and Cambridge Econometric’s estimations. For most regions, using either Molle/Cambridge or CRENoS does not seem to affect the regional shares of the variables notably. Further information about the construction of data is found in Enflo (2008).
For the American states, data on population comes from US Census Bureau7 and data on employment comes from the Bureau of Labor Statistics (BLS). For Total Gross Domestic Product by State, regional shares in national income were taken from Bureau of Economic Analysis (BEA) from 1950 to 1970, adjusted with 1970 differ- ence between national income and GSP8. For the period 1970 to 1990 data on Total Gross Domestic Product by State come from BEA using the SIC standard, whereas GSP for 2000 has been chained to the SIC standard using 1997 difference between NACE and SIC. Again, regional shares have been multiplied with national figures from Maddison’s dataset.
A problem with the regional dimension is that some boundaries have changed since 1950. For UK and Belgium this means that some regions have been included at higher aggregation level than the conventional NUTS 2-classification and that Denmark and Ireland are measured at the country level. More information about the regional disag- gregation and boundary changes of Britain and Belgium is found in the appendix.
5 Results
5.1 Kernels—Europe
In Figs. 2, 3, and 4 the kernel diagrams of the distribution of the three economic vari- ables for Western Europe during each decades from 1950 and onwards are displayed. To adjust for the trend in the data, all regional values have been normalized on the sample average, represented by a 1 on the horizontal axis in the graphs. The kernel
5 When collecting internationally comparable national figures, some problems were encountered with former West Germany, because Maddison’s West German estimates only cover the period up to 1997. However, the Statistiches Landesamt Baden-Wurtemberg has estimated the proportion of West and East Germany’s GDP, employment and population (including Berlin) for 1997 to 2000. These proportions have been used to calculate a national figure for 2000 in this thesis. 6 University of Cagliari, www.crenos.it 7 Table No. HS-4. Resident population by state: 1900–2002. 8 No. HS-35. Personal income and personal income per capita by state: 1929–2001.
123
Productivity and employment 409
0
0,5
1
1,5
2
2,5
0 0,5 1 1,5 2 2,5 3
income50
income60
income70
income80
income90
income00
Fig. 2 Relative GVA per capita in the 89 regions 1950–2000
0
0,5
1
1,5
2
2,5
3
3,5
4
0 0,5 1 1,5 2 2,5
prod50
prod60
prod70
prod80
prod90
prod00
Fig. 3 Relative GVA per worker (labor productivity) in the 89 regions 1950–2000
density diagrams thus describe the relative distribution of regions around the sample average during the six analyzed decades.
In Fig. 2 we see how convergence in GVA per capita has been a continuous process in Western Europe since the 1950s. The largest converging force took place between 1960 and 1970 although the 1980s also seem to have brought about a relatively large increase in the density mass of regions around the sample average.
Interestingly, enough income convergence measured as the height of the mode of the distribution came to a complete standstill between 1990 and 2000 and we also see a tendency for a pronounced group of regions to cluster around income levels of
123
410 K. S. Enflo
0
0,5
1
1,5
2
2,5
3
3,5
4
4,5
0 0,5 1 1,5 2
emp50
emp60
emp70
emp80
emp90
emp00
Fig. 4 Relative employment per capita (employment structure) in the 89 regions 1950–2000
1.5–2 times the European average. This result is similar to Magrini (1999) finding of a tendency for bipolarization of the income distribution after 1979. The cluster- ing top-income regions found in the present dataset are Luxemburg, Hamburg, Paris, and Brussels. All these regions are dominated by a big European metropolis city, and this emerging pattern could confirm the NEG prediction9 of a rising core-periphery relation among the Western European regions.
Although, visual inspection of the kernel density diagrams gives us an indication that convergence in regional income has been a continuous process, at least between 1950 and 1990, we have yet to test whether the convergence process is actually statis- tically significant. This is done by using a non-parametric test for the comparison of two unknown distributions proposed by Li (1996) and Fan and Ullah (1999). The test is based on the kernel-method and to estimate the confidence intervals and the standard errors of the test statistic, a bootstrap approximation is employed, see Li and Racine (2007). The number of bootstrap replications is set to 1000 and the critical values for the test statistic are calculated at the 5% significance level. The null hypothesis of the test is H0: f (x ) = g(x ) for all x against the alternative, H1: f (x ) �= (x ) for some x . This means that the income distribution of 1950 is first tested against the distribu- tion from 1960 to see whether these distributions are statistically different from each other. Thereafter, the income distribution of 1950 is tested against the distribution of 1970 and so on. To save space, the subsequent distributions are only tested against the
9 Theoretical literature under the heading New Economic Geography (NEG) has focused on the relationship between economic integration and the determinants of the location of production (for example Krugman 1991). In Krugman (1991), economic integration leads to catastrophic agglomeration in one single region, but more complex recent models tend to emphasize that this relationship is rather bell shaped. The bell- shaped relationship appears since geographic concentration of economic activity arises in a first stage of the integration process when transport costs are reduced, but diminishes in a second stage as transport costs are further reduced.
123
Productivity and employment 411
Table 2 Li-test of similarity between two distributions
Null hypothetis T value Probability
GVA per capita
f (1950) = g(1960) 0.08 0.37 f (1950) = g(1970) 1.94 0.03 f (1950) = g(1980) 3.36 0.01 f (1950) = g(1990) 4.65 0.00 f (1950) = g(2000) 5.58 0.00 Labor productivity
f (1950) = g(1960) 1.24 0.07 f (1950) = g(1970) 5.17 0.00 f (1950) = g(1980) 11.22 0.00 f (1950) = g(1990) 12.31 0.00 f (1950) = g(2000) 15.8 0.00 Employment ratio
f (1950) = g(1960) 0.26 0.26 f (1950) = g(1970) 2.08 0.03 f (1950) = g(1980) 0.18 0.32 f (1950) = g(1990) 1.24 0.07 f (1950) = g(2000) 0.69 0.16
“initial” year 1950. This is because we mainly want to examine whether there has been significant convergence between 1950 and 2000, since this longer sample period has not been investigated using distribution dynamics before. However, the test can natu- rally be used to trace the convergence process between any two decades in our sample. The test results for the income distributions are given in the first section of Table 2. As seen from the table, we are not able to reject the null hypothesis that the income distribution of 1950 and 1960 are statistically different (p value is 0.37). Thereafter, all tests are however rejected at the 5% level, thus ensuring that the convergence process of regional income has been statistically significant from its initial year.
Figure 3 displays the kernel distribution of labor productivity since 1950 and looks relatively similar to the picture of GVA per capita. However, the convergence in labor productivity took place at a rather fast pace between 1960 and 1980, but was followed by a pronounced slowed-down in the 1980s. Thereafter it took off again, which is seen in the increased density of regions clustering around the sample average between 1990 and 2000. The 2000 distribution also shows some tendency for a group of regions to distance themselves from the distribution of the rest at labor productivity levels about 1.4 times the European average. Again the above-mentioned metropolis regions con- stitute the high-end peak, but this high-end peak is positioned relatively closer to the sample average (only 1.4 times the average) compared to the peak in to the income distribution (1.5–2 times). The fact that differences in income per capita are larger between the metropolis regions and the rest of the sample compared to income per worker, is most probably due to the amount of commuting that takes place from
123
412 K. S. Enflo
neighboring regions into the metropolis, inflating the income per capita numbers for certain city regions. In the mid-section of Table 2 the non-parametric tests of the labor productivity distribution are found. Again, we are not able to detect any significant convergence between 1950 and 1960. After 1970 however, we find that all distributions are statistically different from the distribution of 1950.
Thus, convergence in labor productivity appears to have been stronger than in per capita incomes since 1950, and interestingly enough labor productivity convergence has continued after 1990 although income convergence halted. To understand these differences, we need to turn to the employment distribution.
From Fig. 4, we get a rather different picture than from the proceeding two graphs. It is clear that the distribution of employment structures was at its peak around the sample average in 1970. Thus, employment structures converged at a fast rate in between 1960 and 1970. However, in 1980 this peak fell down to levels similar to 1950, indicating strong divergence during the 1970s. Between 1980 and 1990 the distribution fell to its lowest point ever since 1950, whereas it regained converging speed in the 1990s. The non-parametric tests in the last section of Table 2 corrobo- rate these findings. Although, we cannot find statistically significant convergence in employment ratios between 1950 and 1960, the null hypothesis for the distribution of 1970 is rejected with a p value of 0.03. This finding underlines that the employment distributions of 1950 and 1970 are clearly statistically different from each other, as the 1970 distribution displays a higher mode and smaller variance. However, because the employment distribution drops dramatically up to 1980, the tests are no longer rejected (p value of 0.32). In subsequent decades the p values remain above the critical 5% level, suggesting that the employment distribution was not statistically different from the one 1950. This result emphasizes that there was no convergence in employment after 1970. Instead, the distribution diverged back to levels that were not even statistically significant from 1950.
The finding of convergence in labor productivity combined with post-1970 diver- gence in employment ratios tends to suggest that these two variables have increasingly been opposing each other since the 1970s. These findings are surprising, because economic integration has been a continuous process since 1950 and, given constant institutional arrangements, we would expect the trade-off mainly to occur in periods when integration is low and capital mobility is slow.
The transition probability matrices in Table 3 tell a similar story about the trade-off in the long run. As mentioned in Sect. 3.1, five classes are chosen for the discretization of the income and employment that describe a region’s position from lowest to highest in the sample. The initial division of regions into classes has been chosen so that there are similar numbers of regions in the initial classes, as advised in Quah (1993a).10
Table 3 displays the transition probability matrices for the productivity and employ- ment distributions and are calculated every tenth year for the whole time period so
10 The income states are derived on purely empirical grounds. The observations are ranked from highest to lowest and split into five equally large income states, containing the same number of regions. This gives different values for partitioning for each state.
123
Productivity and employment 413
Table 3 Transition probability matrices, 10-year transitions during 1950–2000
Labor productivity Employment ratio
1 2 3 4 5 1 2 3 4 5
1 0.64 0.28 0.07 0.01 0.00 1 0.84 0.14 0.01 0.01 0.00
2 0.10 0.68 0.12 0.09 0.01 2 0.16 0.55 0.19 0.10 0.00
3 0.06 0.12 0.68 0.14 0.00 3 0.02 0.24 0.49 0.25 0.00
4 0.01 0.04 0.25 0.61 0.09 4 0.00 0.06 0.23 0.52 0.19
5 0.00 0.03 0.09 0.28 0.61 5 0.01 0.02 0.02 0.15 0.79
Ergodic distribution
0.135 0.268 0.316 0.220 0.061 0.239 0.199 0.179 0.199 0.184
Note: Initial income on vertical axis, probability of being in class at t + 10 on horizontal axis
that the total number of transitions are 87 × 5 = 435.11 State 1 represents the lowest category (much below median) and state 5 gives the highest. The diagonal elements represent the probability of a region remaining in its initial class, whereas the upper triangular elements measure upward mobility (regions transiting to higher classes) and the lower triangular elements represent downward mobility.
Several comments can be made about Table 3. First of all, we see that the transition probabilities on the diagonal are more evenly distributed for the productivity variable (between 61 and 68% probability of remaining in every state) than the employment variable which shows larger persistence at the top and bottom end of the distribution (84% probability of remaining in employment category 1 and 79% of remaining in category 5).12 The ergodic distribution shows a pattern consistent with what we have learned from the Kernel diagrams: if the system was left to iterate towards steady state, around a third of the sample would be located in the middle income category whereas only 6 and 13% would end up in the highest respectively lowest categories.
The mobility in the middle categories of the employment variable is relatively high, only around half of the regions in employment categories 2, 3, and 4 remain in their position over a 10-year period. The persistence at the ends of the distribution in combi- nation with mobility in the middle is consistent with the long-run behavior of employ- ment polarization that is observed in the ergodic distribution. If the system is iterated toward steady state, 24% of all the regions would be found in the lowest employment category, whereas only 18% of all the regions would have a median employment struc- ture. This picture confirms earlier evidence of diverging employment structures since 1970 and shows that the system is bound for polarization in the long run. It thus appears as if the regions with low employment have difficulties moving up the employment
11 Groningen and Ceuta y Meilla have been excluded from the analysis, the former because of fluctuations in the price of North See Oil which affects the Groningen’s Value Added dramatically, the latter due to the region’s specific nature as an administrative unit in North Africa. For the ergodic distribution to converge properly approximately 300 observations is needed. It would of course be nice to divide the sample before and after 1970, but the number of observations would unfortunately be too small for that. 12 However, it shall be noted that the employment distribution has a smaller variance relative to the average than the income distribution, which means that the employment classes are more narrowly defined.
123
414 K. S. Enflo
0
0.5
1
1.5
2
2.5
3
3.5
0 0.5 1 1.5 2 2.5 3 3.5
income1950
income1960
income1970
income1980
income1990
income2000
0
0.5
1
1.5
2
2.5
3
3.5
0 0.5 1 1.5 2 2.5 3
prod1970
prod1980
prod1990
prod2000
0
1
2
3
4
5
6
7
8
0 0.5 1 1.5 2 2.5
1970
1980
1990
2000
Fig. 5 Income, productivity, and employment kernels for the 51 American states
ladder, while low productivity regions still manage to converge in productivity: the upward mobility from employment category 1–2 is 14%, whereas the probability of going from productivity 1–2 is 28%. Parallel, regions with high employment tend to stay in the highest positions, but productive regions tend to converge.
One possible explanation for the pattern of long-run income convergence and employment divergence is that the restructuring of the European economies in com- bination with increased economic integration in the post-1970s increasingly forced regions to face a trade-off between employment and productivity and that income convergence was achieved from the sources of a converging labor productivity distri- bution, while the employment ratio distribution was influencing the income distribu- tion in opposite ways. Thus, diverging employment ratios may explain why income convergence halted during the 1990s. To investigate to what degree such a trade-off between productivity and employment convergence is a typically European feature, we will compare the European Kernel distributions to similar ones drawn for the US during the same period.
5.2 Kernels—comparison with US
In Fig. 5 we compare the European kernels to similar ones obtained for the United States.13 In the top left panel, we see that the process of GVA per capita convergence has not at all been continuous for the States. Rather, the mode of the GVA distribution was at its highest in 1970 and dropped thereafter. However, the European distribution
13 Lack of data on agricultural employment at the state level restricts the sample to 1970–2000 for the employment and productivity variables.
123
Productivity and employment 415
Table 4 Li-test of similarity between two distributions
Null hypothesis T value Probability
GVA per capita
f (EU1950) = g(US1950) 4.38 0.04 f (EU1960) = g(US1960) 8.99 0.02 f (EU1970) = g(US1970) 6.22 0.04 f (EU1980) = g(US1980) 5.59 0.01 f (EU1900) = g(US1990) 2.60 0.05 f (EU2000) = g(US2000) −0.85 0.56 Labor productivity
f (EU1970) = g(US1970) 3.72 0.11 f (EU1980) = g(US1980) 0.75 0.14 f (EU1990) = g(US1990) 11.22 0.89 f (EU2000) = g(US2000) 1.38 0.21 All values below 0.05 are significant
in 2000 exhibits a similar shape and mode as the American distribution in 2000. The American productivity distribution in the top right panel peaked in 1980 and 1990, but fell to its lowest level in 2000.14
Comparing the distributions from Europe and USA show that the European integra- tion with the formation of the EU has indeed fuelled a process of regional convergence in incomes and productivity that does not have a counterpart in the USA. It rather appears that the income and productivity distributions have fluctuated around levels of the mode that are rather similar to the European mode in 2000, suggesting that the 89 European regions of today have reached a distribution of incomes and productivity that is similar to the distribution of the already integrated states of the USA. Testing formally for the difference in the American and European income distributions corrob- orates this finding. Between 1950 and 1990 the test of similarity between distributions is rejected at the 5% level, but in year 2000 the probability of the null hypothesis is 0.56, indicating that the two income distributions are no longer significantly different. The test results are found in Table 4. This suggests that market integration in Europe has spurred a convergence process that reached the shaped of the already integrated American states by the end of the period. The question is whether the European con- vergence process will levels off at this stage, or whether the European regions will obtain a significantly more equal income distribution than the American states in the future.
The results of the Li-test of the productivity distributions in the lower part of Table 4 reveal that similarity of the distributions cannot be rejected at the 5% level for any year. It thus appears that the American productivity distribution is fluctuating at levels that are not completely different from the European ones. The result that the European
14 Unsurprisingly, given the evidence from the visual inspection of the Kernels, we are also unable to detect any significant and continuous convergence in United States for any of the three variables using the non-parametric test.
123
416 K. S. Enflo
productivity distribution was significantly less equal in Europe than in USA for most of the period, whereas no significant differences could be traced for the productivity distribution suggest that the productivity evolution may not have been connected with income growth in many regions.
The American employment distribution (bottom panel of Fig. 5) was at its highest peak in 1970 but fell thereafter, just like the European employment distributions. The American employment distribution is significantly more equal than the European for every year however. To save space the non-parametrical test of equality between dis- tributions test was not presented for this distribution (similarity was always rejected). For some years the mode of the American distribution was actually almost twice as high as the European one.15
Because, there was no clear American convergence in labor productivity, there are no evidence that there might have been an American trade-off between employ- ment and productivity.16 Rather, the trade-off between employment and productivity appears to be a limited European post-1970 phenomenon.
6 Did productivity growth translate into income growth?
The analysis of the productivity, employment, and GVA distribution suggests that there has been a European trade-off between employment and productivity since the 1970s that has not been present in the United States. However, it still needs to be established that the very same regions that were increasing their employment also suffered in pro- ductivity changes and vice versa. One way of analyzing this trade-off is to ask to what extent labor productivity growth really translated into growth in regional incomes. If there is a sharp trade-off between productivity and employment, for example in the respect that productivity is kept up by keeping fewer workers in employment, the productivity gains will not translate to a higher income per capita.
Figure 6 compares the European and American pair-wise correlations between growth in productivity and growth in GVA per capita for every decade.17 For the American states, the correlations are close to one throughout the investigated period indicating that the states with the highest productivity growth also experienced high growth in incomes and vice versa. In Europe, the correlation was rather weak during the 1950s, but increased during the 1960s which was a decade of fast general growth, convergence, and catching-up to the US. However, this correlation drops in the 1970s and is down to 0.5 in the 1980s, exactly at the timing when the employment distribution started to diverge.
Figure 7 presents a picture of the actual employment-productivity trade-off during the 1970–2000. In this pattern, we may also visibly distinguish national borders. Many Spanish regions (denoted with regional codes starting with ES) had positive produc-
15 To save space we do not report the test results for this distribution, although they are available upon request. 16 The outlier in income, productivity, and employment that is visible as a bump at the far right of all distributions is Washington, DC. 17 The American data is limited to the period 1970–2000 due to data limitations on the number of agricul- tural workers before 1970.
123
Productivity and employment 417
0.4
0.6
0.8
1
1950-60 1960-70 1970-80 1980-90 1990-00
Western Europe
USA
Fig. 6 Correlation (y-axis) between labor productivity growth and income growth
BE1_31
BE21_25
BE32_35
DE1 DE2
DE3 DE5
DE6DE7 DE9
DEA DEB DEC
DEF
DK
ES11
ES12ES13
ES21
ES22
ES23
ES24
ES3
ES41
ES42
ES43
ES51
ES52
ES53
ES61
ES62 ES63ES7
FR1
FR21FR22
FR23
FR24
FR25
FR26
FR3FR41
FR42 FR43
FR51
FR52
FR53 FR61
FR62 FR63
FR71
FR72
FR81
FR82
FR83
IE
IT11
IT12IT13
IT2
IT31
IT32
IT33
IT4 IT51
IT52
IT53
IT6
IT71
IT72
IT8
IT91
IT92
IT93
ITA
ITB
LU
NL1
NL2
NL3 NL4
UKCDEUKF123UKG123
UKHIJ
UKK1234 UKL
UKM UKN
0 .5
1 1
.5 2
P ro
d u
ct iv
ity G
ro w
th
-.2 0 .2 .4 .6
Employment Growth
Fig. 7 Employment growth (x-axis) versus productivity growth (y-axis) in the European regions 1970–2000
tivity growth rates, but negative employment growth rates. The French (denoted by FR) and the Dutch (denoted by NL) regions seem to be located at the other extreme of the scale: with increases in employment ratios but modest productivity increases. The national clusters may suggest that country-specific policies and institution could matter in explaining the trade-off. The increase in the trade-off also coincides with a time in Europe when labor market laws were growing more protective and unemploy- ment benefits rose sharply (Allard and Lindert 2006). In addition, if the correlation is solely due to low integration and capital mobility, it is hard to explain why it occurred in the 1970s.
7 Conclusions
In Western Europe, the period between 1950 and 1970 exhibits a continuous and unambiguous convergence in all the three studied variables: incomes, productivity, and employment. This convergence process could be driven by market integration,
123
418 K. S. Enflo
trade, and technological catch-up. The 1960s stand out in its extraordinary virtuous growth pattern in which the regions that went through structural change increased labor productivity without sacrificing employment. In 1970, the mode in the employ- ment structure distribution peaked. The Post-1970s was in contrast characterized by a process of far-reaching changes across Western Europe, including both economic crisis and further economic and political integration. The continuing convergence in regional income seems however to come to a halt in 1990; although, the labor produc- tivity distribution continued to converge until the end of the sample period. In the last decades, these two distributions show a tendency for the emergence of a second peak consisting of four of regions at very high income and productivity levels. These regions are located in different countries but share the trait of being capital and metropolis regions.
United States on the other hand start off at a much higher convergence level already in 1950 and GVA per capita converges continuously until 1970. In 2000, there are no significant difference in the income distribution between Europe and the USA. After 1970, American productivity and GVA per capita show tendencies for divergence. Productivity sharply diverges in the 1980s and the employment ratio’s convergence drop in the 1970s.
Interestingly, although the European development since the 1970s has lead to a trade-off between employment growth and productivity growth at the regional level, we find no similar effects in the USA. In Europe, it appears that the regions with the fastest productivity growth achieved it at the expense of employment growth and vice versa. This practice can be seen in the distribution of employment structures that peaked around the average in 1970 and started to diverge thereafter. The data shows some country effects, the regions of Spain tend to have had high productivity growth at the expense of employment growth whereas the Dutch regions show the opposite pattern.
This finding shall be seen as an empirical result. Dynamic economic theory would suggest that any trade-off between productivity and employment growth to be offset in the presence of capital and labor mobility. In addition, some studies show that pro- ductivity and employment are both positively correlated with economic density and human capital (Overman and Puga 2002; Ciccone 2002). Productivity and employ- ment growth have indeed been positively correlated in Europe before 1970 and in the USA throughout the period. The negative trade-off that has occurred in the Western European regions after 1970 therefore represents something of a limited phenomenon and a paradox. As high-lighted in Gordon (1995), the trade-off may have something to do with national institutional factors that boosts real wage and moves the economy north–west along the labor demand curve, for example protective employment poli- cies. Future research should investigate whether this negative correlation may have a link with European institutional factors that have arisen since the 1970s.
The article highlights the need to understand the mechanism behind employment and productivity in achieving income convergence and calls for policies that do not cause a trade-off between these two variables. The European Union needs to improve labor market mobility in those regions where increased international competition has forced labor productivity convergence at the expense of employment growth. Lastly, from a policy perspective there may be reasons to be alarmed by the finding that the
123
Productivity and employment 419
continuous European convergence process in regional incomes have slowed down after 1990; although, the European collaboration deepened during this period with for example the creation of the Euro.
Acknowledgments I am thankful for comments from Philip Epstein, Peter Howlett, Max-Stephan Schulze, Lennart Schön, Erik Wengström, seminar participants at Oxford Graduate Economic History Workshop and the APHES conference. Financial support from the European Commission’s Sixth Frame- work Programme under the Marie Curie Actions, Contract no. MRTN-CT-2004-512439 is gratefully acknowledged.
Appendix 89 European regions
Belgium Nordrhein-Westfalen Oost
Vlanderen Hessen West
Wallonie Rheinland-Pfalz Zui
Brabant Saarland Spain
Denmark Baden-Wurttemberg Andalucia
France Bayern Aragon
Pegion Parisienne Berlin Asturias
Champagne-Ardenne Ireland Baleares
Picarde Italy Canarias
Haute Normandie Piemonte Cantabria
Centre Valle d’aosta Castilla-la Mancha
Basse Normandie Liguria Castilla-Leon
Bourgogne Lombardia Cataluna
Nord-Pas-de-Calais Trento-Alte Adige Com. Valenciana
Lorraine Veneto Extremadura
Alsace Friuli-Venezia Guilia Galicia
Franche-Comte Emilia Romagna Madrid
Pays de la Lorraine Marche Murcia
Bretagne Toscana Navarra
Poitou-Charentes Umbria Pais Vasco
Aquitaine Lazio Rioja
Midi-Pyrenees Campania Ceuta y Melilla
Limousin Abruzzi UK
Rhone-Alpes Molise North
Auvergne Puglia East Midlands
Langue-doc-Roussilion Basilicata South East
Provence, cote dázur Calabria South West
Germany Sicilia West Midlands
Schleswig-Holstein Sardegna Wales
Hamburg Luxemburg Scotland
Niedersachsen Netherlands Northern Ireland
Bremen Noord
123
420 K. S. Enflo
Appendix continued
Enflo dataset Molle et al. (1980) Cambridge econometrics
Regional boundaries change in the UK
North North + North West + Yorkshire Humber- side
UKC+ UKD + UKE (North East, North West, Yorkshire and the Humber)
East Midlands East Midlands UKF1+2+3 (Derbyshire, Leicestershire, Lincolnshire)
West Midlands West Midlands UKG 1+2+3 (Herefordshire et al, Shrop- shire, West Midlands)
South East East Anglia + South East
UKH + UKI + UKJ (Eastern, London, South East)
South West South West UKK1+2+3+4 (Gloucester et al, Dorset, Cornwall, Devon)
Wales Wales UKL (Wales)
Scotland Scotland UKM (Scotland)
Northern Ireland Northern Ireland UKN (Northern Ireland)
Regional boundaries change in Belgium
Vlanderen Vlanderen BE21+BE22+BE23+BE25
(Antwerpen, Limburg, Oost-Vlaander- en, West-Vlaanderen)
Vallonie Vallonie BE32+BE33+BE34+BE35
(Hainaut, Liege, Luxembourg, Namur)
Brabant Brabant BE1+BE24+BE31
(Brussels, Vlaams Brabant, Brabant Val- lon)
References
Alcaide IJ (2003) Evolución económica de las regiones y provincias españolas en el siglo XX Fundación BBVA
Allard GJ (2003) Jobs and labour market institutions in the OECD, PhD Dissertation, University of Cali- fornia, Davis
Allard GJ, Lindert PH (2006) Euro-productivity and Euro-jobs since the 1960s: which institutions really mattered? NBER working paper 12460
Barro R, Sala-i-Martin X (1991) Convergence across states and regions. In: Brookings papers of economic activity, 1, April 1991. Reprinted in Cukierman A, Hercowitz Z, Leiderman L (eds) Political economy, growth, and business cycles. M.I.T. Press, 1992
Barro R, Sala-i-Martin X (1992) Convergence. J Polit Econ 100:223–251 Barro R, Sala-i-Martin X (1995) Economic growth. McGraw Hill, New York Benito JM, Ezcurra R (2005) Spatial disparities in productivity and industry mix: The case of the European
regions. Eur Reg Urb Stud 12:177–194 Beaudry P, Collard F (2002) Why has the employment-productivity trade-off among industrialized countries
been so strong. NBER working paper 8754 Canova F, Marcet A (1995) The poor stay poor: Non-convergence across countries and regions. CEPR
discussion paper series 1265 Ciccone A (2002) Agglomeration effects in Europe. Eur Econ Rev 46:213–227
123
Productivity and employment 421
Cheshire P, Carbonaro G (1995) Convergence-divergence in regional growth rates—an empty black box. In: Armstrong HW, Vickerman RW (eds) Convergence and divergence among European regions. European research in regional science, vol. 5. Pion Ltd, London
Cheshire P, Magrini S (2000) Endogenous processes in European regional growth: convergence and policy. Growth Chang 31:455–479
Enflo K (2008) Regional and industrial growth patterns in 20th century Western Europe, PhD dissertation. Lund studies in economic history, vol 46. Holmbergs, Malmö
Epstein P, Howlett P, Schulze M-S (2003) Distribution dynamics: stratification, polarization and conver- gence among OECD economies, 1870–1992. Explor Econ Hist 40:78–97
Fagerberg J, Verspagen B (1996) Heading for divergence? Regional growth in Europe reconsidered. J Com- mon Market Stud 34:431–448
Fan Y, Ullah A (1999) On goodness-of-fit tests for weakly dependent processes using kernel method. J Nonparametr Stat 11:337–360
Gordon R (1995) Is there a trade-off between unemployment and productivity growth? NBER Working Paper 5081
Johnson PA (2000) A nonparametric analysis of income convergence across the US states. Econ Lett 69:219–223
Krugman P (1991) Geography and trade. Leuven University Press, Leuven Li Q (1996) Nonparametric testing of closeness between two unknown distribution functions. Econom Rev
15:261–274 Li Q, Racine J (2007) Nonparametric econometrics: theory and practice. Princeton University Press,
Princeton López-Bazo E, Vayá E, Mora AJ, Surinach J (1999) Regional economic dynamics and convergence in the
European Union. Ann Reg Sci 33:343–370 Magrini S (1999) The evolution of income disparities among the regions of the European Union. Reg Sci
Urb Econ 29:257–281 McGuckin R, van Ark B (2005) Productivity and participation: an international comparison. Research
Memorandum GD-78, Groningen Growth and Development Centre Meliciani V (2006) Income and employment disparities across European Regions: the role of national and
spatial factors. Reg Stud 40:75–91 Molle W, van Holst B, Smit H (1980) Regional disparity and Economic development in the European
community. University Press Cambridge, Cambridge Overman HG, Puga D (2002) Unemployment clusters across Europe’s regions and countries. Econ Policy
17:115–147 Quah D (1993a) Empirical cross-section dynamics in economic growth. Eur Econ Rev 37:426–434 Quah D (1993b) Galton’s fallacy and tests of the convergence hypothesis. Scand J Econ 95:427–443 Quah D (1996a) Regional convergence clusters across Europe. Eur Econ Rev 40:951–958 Quah D (1996b) Twin peaks: growth and convergence in models of distribution dynamics. Econ J 106:
1045–1055 Quah D (1997) Empirics for growth and distribution: stratification, polarization, and convergence clubs.
J Econ Growth 2:27–59 Rey S, Janikas M (2005) Regional convergence, inequality and space. J Econ Geogr 5:155–176 Rey SJ, Montouri BD (1999) US regional income convergence: a spatial econometric perspective. Reg Stud
33:143–156 Romer P (1990) Endogenous technological change. J Polit Econ 98:71–102 Silverman BW (1986) Density estimation for statistics and data analysis. London, Chapman and Hall Solow RM (1956) A contribution to the theory of economic growth. Q J Econ 70:65–94 Tondl G (1999) The changing pattern of regional convergence in Europe. Jahrbuch für Regionalwissens-
chaft 19:1–33 van Ark B, Haan J (2000) The delta-model revisited—recent trends in the structural performance of the
Dutch economy. Int Rev Appl Econ 14:307–321
123
Copyright of Annals of Regional Science is the property of Springer Science & Business Media B.V. and its
content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's
express written permission. However, users may print, download, or email articles for individual use.
