Trade Unionism and Growth: A Panel Data StudyThis is the html
version of the file
http://www.rdg.ac.uk/Econ/Econ/workingpapers/emdp429.pdf.
G o o g l e automatically generates html versions of documents as we
crawl the web.
To link to or bookmark this page, use the following url:
http://www.google.com/search?q=cache:UQoYv9mNTVUC:www.rdg.ac.uk/Econ/Econ/workingpapers/emdp429.pdf+trade+unionism&hl=en&ie=UTF-8
Google is not affiliated with the authors of this page nor
responsible for its content.
These search terms have been highlighted: trade unionism
Page 1
Trade Unionism and Growth: A Panel Data StudyDimitrios Asteriou*Vassilis
MonastiriotisThe University of Reading London School of EconomicsAbstractThis
paper investigates the long-run relationship between trade unionism andeconomic
growth using a panel data set comprising of 18 OECD economies. Much ofthe
existing evidence on the effects of unionism on productivity derives from
micro-economic studies, with little attention to the dynamics of this
relationship and theeconomy-wide effects. Using the recently developed mean
group and pooled meangroup estimation techniques on cross-country panel data,
the paper offers support tothe "enhancing-worker-morale face of unionism"
hypothesis, revealing a positiverelationship between trade union density and
labour productivity.J.E.L. Classification: C23, J51, O4Keywords:Trade unions,
Economic growth, Panel data econometrics.*Corresponding Address: Dimitrios
Asteriou, Department of Economics Faculty of Letters & SocialSciences, PO Box
218, Whiteknights, Reading RG6 6AA, England, Tel: (0118)9875123 (4056);Fax:
(0118) 9750236; E. Mail: D.Asteriou@reading.ac.uk. We are grateful to Dr
EngelbertStockhammer for making available to us the dataset used here. All
responsibility for errors andomissions lies with the authors.
Page 2
1. IntroductionResearch on the productivity effects of unionism over the last
two decades hasbeen lively, offering new insights to the theoretical and
empirical relationshipbetween these two labour market aggregates. Following the
pioneering work ofBrown and Medoff (1978) and inspired by the controversial work
of Freeman andMedoff (1979 and 1984), numerous empirical studies have examined
the extent anddirection of the union productivity effects, mainly for the cases
of the UK and theUSA. There is a rather widespread consensus in the literature
about unionism having anegative impact on productivity and output, although a
number of authors haveestimated positive union productivity differentials (Brown
and Medoff, 1978; Clark,1980; Nickel et al., 1989; Gregg et al., 1993).It is
standard in this literature to investigate productivity differentials
betweenunionised vis-à-vis non-unionised firms using industry or firm level
data.Consequently, there is little attention on the economy-wide and dynamic
effects ofunionism, while other sources of productivity differentials, like
managementstrategies and production efficiency, are difficult to be accounted
for anddistinguished from the direct union effects. There is comparatively
little research onthe issue using aggregate national data. Among the few
studies, the OECD (1997) hasfound evidence of negligible effects of unionism and
the structure of wage bargainingon productivity and productivity growth. Nickel
and Layard (1998) have estimated anegative union effect on growth for a panel of
OECD countries. A negative impact onoutput or productivity has also been found
by earlier economy-wide studies (deFina,1983; Lovell et all., 1988; Koedijk and
Kremers, 1996).In this paper we investigate the economy-wide effects of unionism
onproductivity and productivity growth at an aggregate level, for a large panel
of data. A
Page 3
long time-series (1960-1992) for 18 OECD countries allows us to investigate
theshort- and long-run dynamics of unionism within an economic growth
framework,while controlling for country-specific effects.1The source of the data
is theComparative Welfare States Data Set (Huber et al., 1997), which includes
data fromvarious sources.2In addition to traditional panel data techniques, we
utilise newlydeveloped econometric methods for the estimation of dynamic panel
models. Havinga set of 576 observations and using an auto-regressive distributed
lags (ARDL)specification, we can identify a common-across-countries long-run
coefficient for theunion productivity effects, while allowing different
short-run dynamics for eachcountry. Hence, our estimates are largely unbiased
from any business-cycle andcountry-specific effects. Apart from the relative
novelty of the applied econometricmethodology, our investigation of union
productivity effects based on a large panel ofdata and controlling for short-run
dynamics and country-specific effects is to ourknowledge unique.In the next
section we make some theoretical considerations and derive anestimating model.
Sections 3 and 4 present the empirical results. In section 3 we applytraditional
econometric techniques, while in section 4 we briefly present andconsequently
apply the dynamic panel data methodologies. The final sectionsummarises the
results and concludes.2. A model of changes in unionisation rates and
growthUnion productivity differentials can arise through a variety of
mechanisms. Ata firm-level, unionism can affect the organisation and efficiency
of production, the1The sample countries are Australia, Austria, Belgium, Canada,
Denmark, Finland, France, Germany,Ireland, Italy, Japan, Netherlands, Norway,
New Zealand, Sweden, Switzerland, UK and USA.2The original data sources are: for
employment, OECD Labour Force Statistics (various years); forunion membership
figures, Visser (1996); and for GDP and investment, Penn World Tables (Mark
5.6).
Page 4
pace of technological innovation and capital accumulation, training and
manninglevels and so forth. At a wider level, it can impact upon average wages
and wageinflation, with further effects on inflation, interest rates, investment
and outputgrowth, as well as on the national comparative advantages and
international trade. Atthe level of theory, negative union productivity effects
can be assumed if unionsimpose rigidities in the introduction of new
technologies and working practices, or ifthey reduce profitability and
investment. On the other hand, unions can increaseworkers' participation and
involvement and, hence, production efficiency, while theirincreasing labour
costs function can foster innovation and quality-based competitionfrom the side
of management.Traditionally, however, a more direct effect of unionism is
assumed, as thelatter can impact directly upon the productivity of the
individual workers. In this caseit is the marginal product of labour that
differs between unionised vis-à-vis non-unionised workers. Nevertheless, despite
their theoretical distance, all thesemechanisms exhibit some practical
equivalence, both in technical and empiricalterms.3In building our model we
follow Brown and Medoff (1978) assuming thatboth unionised and non-unionised
labour have the same coefficient in the productionfunction, with unionised
labour being discounted by a factor fwhich reflectsproductivity
differences:baf)*(untLLKAY+=(1)where the standard notation is used with the
exception that Luis unionised labour andLnis non-unionised labour. The
coefficient fcan be greater or smaller than unity,3A technical exposition of
this and relevant empirical evidence can be found in Monastiriotis (2001).
Page 5
implying higher or smaller productivity of unionised labour, respectively. By
addingand subtracting Luin (1) and further manipulating, we
get:baf)]}1(*)/(1[*{--=LLLKAYut(2)or, taking logs and using
log(1+x)ªx,TUDlktalygba+-++=-)1()((3)where lower case letters denote logarithms,
g= b(f-1) andTUD=Lu/L.Thus,productivity growth will be a function of changes in
union density:TUDlkaly+-++=-gba)1()((4)Further, we assume a standard capital
accumulation process (dis depreciation rate)1)1(--+=tttKIKd(5)so that the growth
of capital will bedd-=--+=----)/(/}))1({(1111ttttttttKIdKKKKIdK(6)Using DkªdK,
approximating (It/Kt-1)-dwith It/Ytand finally introducing an error termwe
obtain4tititititieTUDlYIaly,,,,,)()1()/()(++-++=-gba(7)Equation (7) is our main
estimating model, relating productivity growth to the shareof investment to GDP,
employment growth and changes in union density. In thetraditional panel
estimation techniques (section 3) the one- and two-way errorcomponent
(de-meaned) and dummy variables (DVLS) transformations of thisspecification are
used. In section 4 we estimate the above specification using theMean Group (MG)
and Pooled Mean Group (PMG) estimators with a dynamic ECMequation that has (7)
as a long-run solution.4The approximation of capital by GDP is introduced mainly
to avoid data-quality problems related tothe capital series. Data availability
was however an additional problem, since using the capital serieswould further
shorten the time-dimension of our sample and, hence, generate problems in
theapplication of the MG and PMG estimations on the ARLD specification.
Page 6
3. Empirical resultsThe panel nature of the sample offers a wealth of
information that can beinvestigated. Before applying a number of pooled
regression methodologies, it isinteresting, therefore, to obtain first some
information from the cross-sectional andtime-series dimensions.(i)
Cross-sectional and time-series analysesA first impression about the relation
between productivity growth and changesin unionisation rates can be obtained by
looking at scatter plots. Hence, we plottedthese two aggregates against each
other once for each of our sample years and oncefor each of our sample
countries. The general pattern was surprisingly one of apositive relationship,
although the graphs were often very sensitive to the inclusion ofsome
observations. Moreover, the scatter-plots revealed some cyclicality,
withpositive and negative slopes alternating every around five years. These two
pieces ofinformation seem to suggest that the relationship under investigation
has not beenstable over the years and across countries in the period of our
focus. This furthersuggests that country and time specific effects must be
significant.To get some more formal indication for this, we also run
cross-country andtime-series regressions based on the model derived earlier. The
coefficients obtainedfor the unionisation variable were quite stable in the
cross-country regressions (overthe years), but less so in the time-series
regressions (among countries). Moreover, thecoefficients from the cross-country
regressions also seemed to follow a time trend.Figure 1 plots these estimates
over time.
Page 7
As it can be seen, the estimated union effects have been rather volatile with
amean value close to zero, especially during the 1970s. The logarithmic trend
fitted,suggests an original negative union effect which in the 1980s turns
positive. Overall,however, the results obtained are not statistically
significant, as Table 1 reveals.As the overall performance of the estimated
regressions is not satisfactory, thisis as much of inferences as we can make
from the uni-dimensional regressions. Thenext step is to look at the panel
dimension, by pooling the data together.(ii) Traditional panel data analysisIn
order to get as much information as possible from the panel, we ran allpossible
regression specifications, both for productivity growth and for output
growth.The results are presented in Table 2. The union coefficient is
significant in all casesfor both the output and productivity growth regressions,
with the exception of the timerandom effects model (second last column), which
also performs very poorly, with theimplication that the time effects are fixed
and not random. However, all specificationtests show both time and country
effects to be significant, suggesting that the correctspecification is a two-way
error component model (last column).This last model suggests that the net effect
of unionism on productivity,controlling for time and country specific effects,
has been positive in the threedecades and for the 18 countries of our sample.
The interpretation of the theoreticalmodel in (7) suggests that unionised labour
has -other things equal- been by around19% (=0.214/1.115) more productive than
non-unionised labour. The coefficients onthe capital and labour variables have
the expected signs and are significant. Moreover,the results are very stable
across the different specifications. The results from the
Page 8
traditional analysis on the panel seem to suggest that unionism enhances
theproductivity of labour.On the other hand, the country specific effects are
always highly significant. Itis interesting to investigate the determinants of
these effects. One possible explanationis that the productivity effects of
unionism might differ among countries depending onthe strategies employed by the
national trade unions and the structure of wagebargaining (centralisation and
co-ordination) in each country. To test for that, weexamined the relationship
between the estimated country-fixed effects and someindicators of union
co-ordination and centralisation of wage bargaining, produced bythe OECD (1997).
Correlation analysis between these variables returned a statisticallysignificant
correlation coefficient of 0.34, suggesting that the estimated positive
unionproductivity effect is weaker in countries with more rigid wage bargaining
structures.4. Robustness of the empirical findingsThe unconventional result
obtained above, of a positive relationship betweenunionism and productivity
requires further investigation, since the data used here havecomplex dynamics
and are characterised by strong trends and non-stationarity. Theidentified time
and country effects could be possibly capturing country specific long-or
short-run dynamics which the traditional pooled estimators, such as the fixed
andrandom effects, cannot estimate. Therefore, these methods might not be
appropriate inour case. New estimation techniques are now available in the
literature that allow sucheffects to be controlled for and measured.(i) The MG
and PMG estimation methodology
Page 9
It has become conventional to view long-run parameters as
reflectingcointegrating relationships among a set of I(1) variables. The
standard methodology insuch cases first establishes the order of integration of
the variables in question, andthen - having established that the variables are
of the same order of integration - testswhether there is at least one linear
relationship among these variables.Our analysis follows a different approach.
This can be justified by two facts.First, there are only a few (and even fewer
statistically satisfactory) tests ofcointegration in a panel data context, while
it is also well known that tests of order ofintegration in panel data do not
reliably distinguish between series that contain a unitroot and those that are
stationary with a "near-unit root". Second, long-runparameters may be
consistently estimated using the traditional autoregressive-distributed lag
(ARDL) approach (Pesaran and Shin, 1998). Moreover, as Pesaran,Shin and Smith
(1999) have shown, this approach yields consistent andasymptotically normal
estimates of the long-run coefficients irrespective of whetherthe underlying
regressors are I(1) or I(0).5Further, it compares favourably in MonteCarlo
experiments with conventional methods of cointegration analysis.Therefore, our
estimates were obtained using two recently developed methodsfor the statistical
analysis of dynamic panel data: the Mean Group (MG) and thePooled Mean Group
(PMG) estimation. These methods are particularly suited to theanalysis of panels
with large time and cross-section dimensions.6MG estimationderives the long-run
parameters for the panel from an average of the long-runparameters from ARDL
models for individual countries (see Pesaran and Smith,1995). For example, if
the ARDL is the following5In our analysis the GDP, Investment, Employment and
Union Density variables are clearly trendedfor all countries and can be assumed
to be I(1), hence become stationary after first differencing(productivity and
employment growth, changes in union density) or taking their ratio
(investmentshare).
Page 10
ititiitiitiezdxLbyLa++=)()((8)for countrly i,where i=1,....,N,then the long-run
parameter for country iis)1()1(iiidb=q(9)and the MG estimator for the whole
panel will be given by==NiiN11qq(10)It can be shown that MG estimation with
sufficiently high lag orders yieldssuper-consistent estimators of the long-run
parameters even when the regressors areI(1) (see Pesaran, Shin and Smith,
1999).The PMG method of estimation, introduced by Pesaran, Shin and Smith(1999)
occupies an intermediate position between the MG method, in which both theslopes
and the intercepts are allowed to differ across country, and the standard
fixedeffects method, in which the slopes are fixed and the intercepts are
allowed to vary. InPMG estimation, only the long-run coefficients are
constrained to be the same acrosscountries, while the short-run coefficients are
allowed to vary.Setting this out more precisely, the unrestricted specification
for the ARDLsystem of equations for t=1,2,...Ttime periods and
i=1,2,...Ncountries for thedependent variable
yisitmjinjjtiijjtiijitxyyemdl+++===--10,,(11)where xijis the (k¥1)vector of
explanatory variables for group iand mirepresents thefixed effects. In principle
the panel can be unbalanced and mand nmay vary acrosscountries. This model can
be re-parameterised as a VECM system6Quah (1993) has referred to such data sets
as "data fields".
Page 11
itmjinjjtiijjtiijtiitiiitxyxyyemggbq++++-=-=-=----1110,,1,1,)((12)where bis are
the long-run parameters and qis are the error correction parameters.The pooled
group restriction is that the elements of bare common across countries,
sothatitmjinjjtiijjtiijtitiiitxyxyyemggbq++++-=-=-=----1110,,1,1,)((13)All the
dynamics and the ECM terms are free to vary. Estimation of this model is
bymaximum likelihood. Again it is proved that under some regularity assumptions,
theparameter estimates of this model are consistent and asymptotically normal
for bothstationary and non-stationary I(1) regressors. Both MG and PMG
estimations requireselecting the appropriate lag length for the individual
country equations. Thisselection was made using the Schwarz Bayesian
Criterion.(ii) The MG and PMG estimation resultsInitially we estimated the model
given in (7) assuming that all of the long runcoefficients are the same across
countries. The estimation results from the MG andPMG methods are presented in
Table 3. The PMG estimates provide further evidenceto our previous finding of a
strong positive relationship between changes inunionisation and productivity
growth, while the MG results are in the same line butless strongly so. The
capital growth variable (investment share) has the expected sign,which for the
PMG model is highly significant. The growth of employment has againa positive
estimated effect, but this is insignificant in the PMG estimation, with
theimplication that labour productivity is constant across different employment
levels.Although the Hausman test for the poolability of this coefficient is
rejected, for boththe unionism variable and investment as a share of GDP, the
pooling restrictions
Page 12
cannot be rejected (p-values 0.31 and 0.34 respectively). Moreover, the
jointHasuman test suggests that the PMG results are more appropriate than the MG
ones.Overall, the results obtained from the ARDL specifications are highly
consistent tothe ones derived from the more traditional methods. The estimated
effect ofunionisation changes on productivity growth suggests that discount
factor forunionised labour, f, is equal to 1.19, or that unionised labour is 19%
more productivethan non-unionised labour. This result is identical to the one
obtained from the two-way error component model of Table 2. The estimated
returns to capital are also verysimilar to the ones obtained earlier.7Further,
restricting the coefficient of unionism tobe the same in all countries but
removing this restriction for the other two coefficientsdoes not affect
significantly our main conclusions (Table 4). Again the unionismcoefficient is
positive and significant for the PMG estimates (which are againapproved by the
Hausman test for poolability), while now both coefficients forinvestment share
and employment growth become insignificant. Alternative estimates not reported
here for economy of space restricting subsets of the long-runcoefficients gave
similar results with those initially obtained.8The employmentgrowth coefficient
verifies the constant returns to scale long run effect of employmenton labour
productivity, which was not captured in the traditional estimation methodsthat
didn't control for the short-run dynamics.9For capital growth, the
insignificantresult strengthens our earlier conclusion about the poolability of
this coefficient. Thistime however, the union productivity effect increases
further, to around 23%.7The estimated coefficient in all specifications is lower
than theory would suggests, but this is largelydue to a scaling effect caused by
our approximation of capital growth with the investment share.8Tables and
results are available from authors upon request.9The short-run dynamics are very
different from the fixed country-specific effects estimated before. Acorrelation
analysis of the impact of the wage bargaining structure on these dynamics
returned asignificant correlation coefficient of -0.45, which implied that in
countries with more rigid wagebargaining structures unionism had also a positive
short-run effect.
Page 13
This figure is at the margin of plausibility if one assumes that the
unionproductivity effects are solely activated through workers' performance
(Brown andMedoff, 1978). Hence, an explanation suggesting that union
productivity effects areactivated through a number of plausible mediating
factors (production efficiency,capacity utilisation and the extent of labour
hoarding being but a few) cannot be ruledout. In any case, the overall
productivity effect of unionism is robustly found to havebeen positive in the
three decades and for the 18 countries of our sample.5. ConclusionsNumerous
studies at the firm and industry levels have provided evidence of anegative
productivity effect of unionism, although there are cases where a positiveunion
productivity effect has been estimated. Most of the empirical literature
usesAnglo-Saxon data and there are few cross-country studies. Further,
time-seriesanalysis on the issue is rather scarce, with the implication that the
dynamics of therelationship at question have been relatively
overlooked.Attempting to partially fill this gap, in this paper we examined the
long- andshort-run relationship between unionism and productivity using a panel
of 18 OECDcountries over a 32-year period. The MG and PMG estimation techniques
that weused together with more traditional methods are at the forefront of panel
dataeconometrics. Our time-series and cross-country analyses revealed that
thisrelationship has been different among countries and over time. Controlling
forpossible time and country-specific effects, the panel data analyses allowed
theestimation of a common across countries long-run coefficient. The good
performanceof our regressions and the stability of our results, we interpret as
evidence in supportof the appropriateness of the econometric method we employed.
Page 14
Our basic results provide robust evidence of a positive impact of unionism
onproductivity. Both the long- and short-run effects are positive and
statisticallysignificant, although we also offer some evidence suggesting that
country-specificfactors, like the strategies employed by the national trade
unions and the degree of co-ordination among them and between them and the
employers, might play an importantrole at the short-run.
Page 15
References1. Brown C. and J. Medoff (1978), "Trade unions in the production
process",Journal of Political Economy, Vol.86, pp.355-378.2. Clark K. (1980),
"The impact of unionisation on productivity: a case study",Industrial and Labour
Relations Review, Vol.33, pp.451-469.3. DeFina R. (1983), "Unions, relative
wages and economic efficiency", Journal ofLabor Economics, Vol.1, No4,
pp.408-429.4. Freeman R. and J. Medoff (1979), "The two faces of unionism?",
Public Interest,No57, pp.69-93.5. Freeman R. and J. Medoff (1984), What do
Unions do?, Basic Books, NY.6. Gregg P., S. Machin and D. Metcalf (1993),
"Signals and cycles: productivitygrowth and changes in union status in British
companies, 1984-1989", EconomicJournal, Vol.103, pp.894-907.7. Huber E., C.
Ragin and J.D. Stephens (1997), "Comparative Welfare States DataSet",
Northwestern University and University of North Carolina.8. Koedijk K. and J.
Kremers (1996), "Deregulation: a political economy analysis",Economic Policy,
Vol.26, pp.443-467.9. Lovell C., R. Sickles and R. Warren Jr (1988), "The
effects of unionisation on labourproductivity: some additional evidence",
Journal of Labor Research, vol.9, pp.55-63.10. Monastiriotis V. (2000) "Trade
unions and productivity: model specification andempirical evidence", unpublished
manuscript, Department of Geography andEnvironment, LSE, UK.11. Nickel S. and R.
Layard (1998), "Labour market institutions and economicperformance", CEP
Discussion Paper No. 407, LSE.
Page 16
12. Nickell S., S. Wadhwani and M. Wall (1989), "Unions and productivity growth
inBritain, 1974-86: evidence from Company Accounts data", CLE Working Paper
No.1149, LSE.13. OECD (1997), Economic performance and the structure of
collective bargaining,ch.3 in Employment Outlook, OECD, Paris.14. Pesaran M.H.
and Y. Shin (1998), "An Autoregressive Distributed Lag ModellingApproach to
Cointegration Analysis", in S. Strom, A. Holly and P. Diamond(eds.), Centennial
Volume of Ragnar Frisch, Cambridge University Press.15. Pesaran M.H., Y. Shin
and R.P. Smith (1999) "Pooled Mean Group Estimation ofDynamic Heterogeneous
Panels" Journal of American Statistical Association,Vol. 94, pp. 621-634.16.
Pesaran M.H. and R.P. Smith (1995) "Estimation of long-run Relationships
fromDynamic Heterogeneous Panels" Journal of Econometrics, 68, pp. 79-113.17.
Quah D. (1993) "Empirical Cross-Section Dynamics in Economic Growth"European
Economic Review, 37, pp. 326-434.18. Visser J. (1996), "Unionisation Trends: The
OECD Countries Union MembershipFile", University of Amsterdam, Centre for
Research of European Societies andLabour Relations CESAR.
Page 17
Table 1: Estimated union effect on growth (cross-country and time-series
regressions)Year Impact ofTUD Year Impact ofTUD Year Impact ofTUDYear/Country
Impact ofTUDCountryImpact
ofTUD19612.873(2.09)1971-0.347(-1.10)19810.731(0.95)1991-0.009(-0.01)IRE1.848(2.25)1962-1.180(-2.21)19720.101(0.33)1982-0.254(-0.29)19920.648(2.70)ITA-0.343(-0.86)1963-0.331(-0.274)1973-0.085(-0.15)19830.753(1.31)AUL-1.690(-2.50)JPN4.024(2.11)19640.446(0.36)1974-0.018(-0.09)19840.373(1.41)AUS-0.547(-0.38)NET-0.580(-1.24)19650.388(0.55)19750.280(1.89)19850.240(0.43)BEL-0.370(-0.71)NOR-0.429(-0.99)1966-0.945(-1.15)19760.481(0.69)1986-0.247(-0.72)CAN-0.475(-0.77)NZL0.143(0.87)1967-0.547(-0.58)1977-0.693(-0.78)1987-0.466(-1.30.)DEN0.237(0.51)SWE0.095(0.24)1968-0.058(-0.05)19780.017(0.03)1988-0.671(-1.69)FIN0.777(1.63)SWZ-0.820(-0.51)19690.141(0.32)19790.79(0.20)1989-0.344(-0.54)FRA0.970(0.92)UKM-0.475(-1.10)19700.589(1.37)19800.103(0.16)1990-0.397(-0.59)FRG0.312(0.30)USA0.006(0.01)Notes:
All regressions have been estimated by OLS. t-statistics are in
parentheses.Figure 1: The effect of changes in unionism on productivity growth
(cross-country regressions)-1.5-1-0.500.511962 1964 1966 1968 1970 1972 1974
1976 1978 1980 1982 1984 1986 1988 1990 1992yeareffect
Page 18
Table 2: Pooled regressions on productivity and output growthModelNECFE (1) CFE
(2)TFECRETRE(C/T)FEProductivity growthInvestment
share0.145(5.71)0.171(4.22)0.171(4.22)0.127(5.49)0.157(4.78)0.135(2.31)0.129(3.11)Employment
growth0.149(2.43)0.222(3.57)0.222(3.57)0.034(0.62)0.201(3.30)-0.774(-2.10)0.115(2.08)Change
in union
density0.166(1.86)0.191(2.13)0.191(2.13)0.168(2.05)0.186(2.10)-0.428(-0.74)0.214(2.67)Constant-0.015(-2.28)--0.023(-2.15)--0.019(-2.15)-0.003(-0.22)-Brausch-Pagan
test--4.19#-59.17--Hausman test----11.07--F-test (year
dummies)---7.14--7.60F-test (country dummies)-4.28----5.29F-test ( y+c
dummies)------7.13R-squared0.080.500.080.600.080.000.66Output growthInvestment
share0.152(7.38)0.220(6.79)0.220(6.79)0.116(6.26)0.185(6.92)0.118(2.42)0.124(3.75)Employment
growth0.377(7.65)0.415(8.33)0.415(8.33)0.277(6.28)0.410(8.36)-0.341(-1.11)0.337(7.69)Change
in union
density0.226(3.13)0.229(3.18)0.229(3.18)0.146(2.42)0.235(3.30)-0.280(-0.58)0.179(2.80)Constant-0.018(-3.34)--0.037(-4.32)--0.027(-3.80)-0.002(-0.16)-Brausch-Pagan
test--4.75#-69.90--Hausman test----17.75--F-test (year
dummies)---8.00--8.31F-test (country dummies)-5.17----5.58F-test ( y+c
dummies)------7.91R-squared0.200.630.190.710.190.000.75Notes:t-statistics are in
parentheses. The Brausch-Pagan and Hausman tests are 2tests for the significance
ofrandom effects (against no effects and fixed effects, respectively). The
various F-tests refer to thesignificance of the corresponding dummies. The
abbreviations in the head of the Table are as follows(estimation method in
parenthesis): NE, pooled regressions with no controls for any effects (OLS);CFE
(1), country fixed effects (DVLS); CFE (2), country fixed effects (GLS); TFE,
time fixed effects(DVLS); CRE, country random effects (GLS); TRE, time random
effects (GLS); and (C/T)FE, two-way error component model with both time and
country fixed effects (DVLS).#: Instead of the B-P test, an F-test for zero
variance of the random effects is used in the CFE (2) model.
Page 19
Table 3: Pooled Mean Group and Mean Group Estimates(Dependent Variable:
DDyit)PMG EstimatesMG EstimatesHausman
TestCoef.s.e.t-ratioCoef.s.e.t-ratioHp-val(I/Y)it0.1240.0216.0060.670.0631.0610.910.34DDlit0.0360.0430.8540.2470.1052.3614.860.03DDTUDit0.1900.0672.8460.4350.2521.7261.010.31Joint
Hausman test:5.420.14Error Correction Coefficientsff-0.9790.021-46.791
-0.9880.012 -82.313Short-Run Coefficients not reported for economy of
spaceNotes:The maximum number of time periods and groups are: 32 18SBC (Schwarz)
has been used to select the lag orders for each group.All the long-run
parameters have been restricted to be the same across groups.The mean group
estimates have been used as initial estimate(s) of the long-run parameter(s) for
thepooled maximum likelihood estimation.Table 4: Pooled Mean Group and Mean
Group Estimates(Dependent Variable: DDyit)PMG EstimatesMG EstimatesHausman
TestCoef.s.e.t-ratioCoef.s.e.t-ratiohp-valLong-Run Coefficients Restricted to be
the Same Across all
GroupsDDTUDit0.2310.0643.5820.4530.2521.7260.700.40Unrestricted Long-Run
CoefficientsDDlit0.0540.0660.8180.0670.0631.061(I/Y)it0.1210.01771.0350.2470.1052.361Error
Correction Coefficientsff-0.9880.012-80.809 -0.9880.012 -82.313Short-Run
Coefficients not reported for economy of spaceNotes:The maximum number of time
periods and groups are: 32 18SBC (Schwarz) has been used to select the lag
orders for each group.All the long-run parameters have been restricted to be the
same across groups.The mean group estimates have been used as initial
estimate(s) of the long-run parameter(s) for thepooled maximum likelihood
estimation.