The Covariance Structure of Earnings in Ireland:
1994-20011
Aedín Doris*, Donal O’Neill** & Olive Sweetman*
Preliminary - Please do not cite
Abstract
It is well known that Ireland’s earnings inequality is high compared to other countries.
In this paper we use an analysis of the covariance structure of earnings in Ireland to
decompose inequality into a component that is due to individuals’ permanent
characteristics, and one that is due to transitory shocks. Using panel data for the years
1994-2001, we show that permanent inequality in Ireland is also at the upper end of
the range of international estimates. Separate analyses of the public and private
sectors show that, although earnings inequality is significantly lower in the public
sector, the wage structure is much more rigid, with higher persistence in individual
wages over time.
1
We would like to thank Daniele Checchi, Steve Haider, Gary Solon and participants at the NUI
Maynooth seminar series for helpful comments on an earlier draft of this paper. This research benefited
from financial support provided by the Irish Research Council or the Humanities and Social Sciences.
**
Department of Economics, National University of Ireland, Maynooth and IZA, Bonn.
*
Department of Economics, National University of Ireland, Maynooth.
1. Introduction
Much of the existing work on inequality has focussed on aggregate inequality,
examining both comparisons across countries in the levels of inequality and trends
over time within countries. However it is well known that the resulting picture of
aggregate inequality can mask fundamental differences in its underlying sources.
Accordingly, several recent studies (e.g. Haider (2001), Gottschalk and Moffit (1995),
Ramos (2002), Baker and Solon (2003), Gustavsson (2004 and 2007), Daly and
Valetta (2007)) have distinguished between two components of inequality: inequality
that reflects differences across individuals or groups that are due to permanent
characteristics (so called permanent inequality) and inequality arising from temporary
shocks, which cause disadvantage at a point in time but are have limited persistence
over time (transitory inequality). From a policy point of view it is important to
distinguish between these two components as they may require different policy
responses. If, for instance, permanent inequality were high then the government might
want to promote policies that would increase the human capital of the affected groups,
whereas high transitory inequality would indicate that such a policy would be less
effective in reducing inequality.
In this paper, we decompose inequality in Ireland into its permanent and
transitory components, using the European Community Household Panel (ECHP)
data, and examine the results for any evidence of trends in these components over the
period for which the data is available, from 1994 to 2001.
The Irish analysis is potentially of international interest for several reasons.
First, the Irish economy experienced unprecedented changes over this period, with the
unemployment rate falling from over 14 per cent in 1994 to under 4 per cent in 20012
2
Source: Table 2.1 Statistical Yearbook, CSO, 2004.
1
and GNP growth rates ranging from 6% to 11% over 1994-2000. It is known that
aggregate earnings inequality grew somewhat in Ireland over the sample period (see,
for example, Barrett et al., 2002). However, nothing is known about how the
permanent and transitory components evolved. It might be expected that a period of
rapid growth would be a period of rapid labour market change and instability, and that
this should be reflected in an increasing transitory component.
Second, results for levels of Irish aggregate earnings inequality have shown it
to be similar to levels in the US and the UK, rather than continental or northern
Europe. For example, Nolan (2000, Table 6.2) gives the Irish 90/10 ratio 3 in 1994 as
4.06; this compares with 4.35 in the US, 3.31 in the UK, 3.28 in France, 2.32 in
Germany and 2.13 in Sweden; of the sixteen countries listed in the table, only the US
has higher earnings dispersion. It will be interesting to see if this same ranking applies
to permanent and transitory components, as it might be expected that high inequality
countries are also countries of high mobility, with the implication that a lower
proportion of overall inequality is permanent in high inequality countries. The
analysis of Irish data can thus provide another data point in assessing the validity of
this suggestion.
Thirdly, it has been hypothesized that centralized wage bargaining, which has
operated in Ireland over the last twenty years, reduces earnings inequality (Gottschalk
and Joyce, 1998). However, because of the design of the Irish wage agreements, they
can be expected not only to affect the overall level of inequality, but also the relative
importance of individual components; further elaboration of this point requires more
detail on the operation of these agreements, which is provided in Section 2 below.
3
This is the ratio of the earnings at the top decile to earnings at the bottom decile.
2
After describing the wage agreements, the methodology used in the analysis of
the data is discussed in Section 3; the data are discussed in Section 4; results are
presented in Section 5; some preliminary results on the extension of the analysis to the
public and private sector earnings are reported in Section 6; and conclusions are
outlined in Section 7.
2. Institutional Background
Centralized wage bargaining in its current form, known as ‘Social Partnership’, was
introduced in 1987 in response to a serious fiscal crisis. Wage restraint was the
primary goal in the early years; in exchange for low bargained wage increases, the
government committed to reducing the income tax burden, so that net pay increases
could be achieved without damaging competitiveness.
Since 1987, agreements have been negotiated every two or three years
between employer organisations, trade unions and the government to award fixed
percentage wage increases to employees at set dates. Higher percentage awards are
typically made to very low-paid workers.
Because the wage agreements are specified in terms of percentage increases
that should be awarded by any given employer, they serve to ‘freeze’ the pre-existing
wage distribution in place. Thus, if there is some external force encouraging an
increase in inequality – such as skill-biased technical change or increasing
international trade – the pay agreement system will indeed reduce inequality below
what it would otherwise be, as suggested by Gottschalk and Joyce (1998), amongst
others. However, it is interesting to note that any external force reducing inequality
would be resisted by the wage bargaining system.
3
Moreover, in the absence of job changes, the wage bargaining system would
freeze not only the overall distribution, but also the position of individuals within that
distribution. Thus, we expect that the system leads to a higher proportion of
permanent inequality than would otherwise prevail.
In practice, however, the agreements may have less ‘bite’ than suggested by
the above description. Firstly, not all employees are covered, as not all employers are
aligned with the employers’ organisations. Secondly, employees can move jobs in
order to escape wage restraint; since negotiated wage increases apply to posts that
already exist within an organisation, it is open to an employee to move between
employers and secure a ‘promotion’ through the distribution.
Thirdly, as the labour market became increasingly tight during the economic
boom, employers began to give ‘top-up’ wage increases to retain staff. Compliance
with early wage agreements was reported to be very high, even in the multinational
sector where employers are typically not affiliated with employers’ organisations, and
unions are weak. However, by 2000, there were widespread reports of top-ups being
awarded. It is important to note, however, that public sector employees would have
been firmly bound by the agreements; as one of the three major players in the
negotiations of the agreements, the government could not be seen to be breaking its
own terms, even in the face of labour shortages in some sectors. If top-ups were
offered across the board, this would lead us to expect that the proportion of inequality
that was permanent would be reducing for private sector workers, but not for public
sector workers, towards the end of the sample period. On the other hand, if top-ups
were correlated with permanent characteristics, no reduction in permanent inequality
would be expected.
4
3. Methodology
To model earnings over the life-cycle we write log-earnings as a function of labourmarket experience, X it and a residual, yit :
Log Yit  g ( X it ,  t )  yit
(1)
In addition we assume that the residual component, yit, can be written as the sum of a
permanent component,  i , due for example to fixed characteristics such as level of
education, and a transitory one, vit , reflecting temporary shocks that hit the individual
or the labour market. That is
yit   i  vit
(2)
where  i and vit are random variables with mean zero variances  2 and  vt2
respectively. Our objective is to identify the separate roles played by the permanent
and transitory shocks in determining inequality. To do this we first estimate yit using
the residuals from an OLS regression of equation (1).4 These residuals are then used
to model the covariance structure described by equation (2). Modelling the dynamics
of earnings through the residual term allows us to abstract from any common growth
trends or life-cycle effects.
We use two approaches to estimate the relative contributions of permanent and
transitory shocks. The ‘non-parametric’ approach uses data from two time periods t
and s, that are sufficiently far apart so that Cov(vit , vis )  0 . In this case the variance of
the permanent component is identified from the covariance of earnings, Cov( yit , yis ) =
 2 and the proportion of total inequality that is permanent is given by
4
In the empirical application g is a simple quadratic in experience.
5
Cov ( yit , yis )
Var ( yit )
(3)
However, this approach relies on some arbitrary assumptions about the
evolution of individual earnings. In particular, it assumes that that the returns to
permanent characteristics accounted for by  i are constant over time and also that the
persistence of transitory shocks is of an order no higher than s  t . To examine the
robustness of our findings to these assumptions, we also consider an alternative
approach that models these features explicitly using a parametric model of earnings
dynamics.
In particular we write yit as:
yit  pti  t vit
(4)
where pt and t are factor loadings that allow the permanent and temporary variances
of earnings respectively to change over time. 5 Identifying the degree of persistence in
the transitory shocks requires a model for vit . We considered a number of alternatives
processes but the preferred specification assume that vit follows an ARMA (1,1)
process. That is
vit   vit 1   it 1   it .
(5)
where the  it ’s are iid random error terms with mean zero and variance  2 . In
the model given by (4) and (5), and with eight years of data (described below), there
5
Equation (4) can be generalized further to allow for individual heterogeneity in the slope of the
earnings profile; this will be considered in future research.
6
are 19 parameters to estimate (  2 , ρ, 2  8 , p2  p8 ,  v20 ,  2 , θ)6. In this paper we
estimate these parameters using a Generalized Method of Moments (GMM) estimator.
Intuitively, this entails choosing the parameters of the model so that the moments of
the theoretical model outlined in (4) and (5) are matched as closely as possible to their
empirical counterparts.
To do this the residuals, yit , are used to calculate the empirical variancecovariance matrix for the years for which earnings are available, Ĉ . Denote the
population variance-covariance matrix by C. In (4) and (5) above, the variancecovariance matrix C  f ( ) has the typical diagonal element:
t 1
Var ( yt )  pt2 2  t2 (  2 v20  K   2 w )
w0
and typical off-diagonal element,
t 1
Cov( yt , yt  s )  pt pt  s 2  t t  s (  2t 1 vo2   K   2 w   2 ) ,
w 0
where K   2 (1   2  2  ) .
With C  f ( ) specified, the parameter vector  is then chosen to minimize
(Cˆ  f ( ))'W (Cˆ  f ( )) , where W is the optimal weighting matrix (Chamberlain,
1984). Following Altonji and Segal (1996), we set W equal to the identity matrix, I. In
our model there are 19 parameters to estimate and 36 ( t (t  1) 2 ) unique moment
conditions. Because of the nature of the panel data used, the calculation of standard
errors is not straightforward; we followed the procedure outlined in Appendix A of
Haider (2001) and in Haider (2000).7
6
7
For identification, we normalise λ1 and p1 to 1.
We are grateful to Steve Haider for providing a copy of the unpublished 2000 paper.
7
4. Data
The data used in the analysis are the eight waves of Irish data in the European
Community Household Panel (ECHP), which contains data on 14 EU countries.
These are the only panel data with appropriate earnings variables available for
Ireland. The years covered by the survey are 1994-2001.8
In the Irish data, the initial sample size in 1994 was 9,904 individuals;
however, just 4,023 were in the sample by 2001. This reflects substantial sample
attrition in the Irish data; Watson (2003) reports that by 1998, attrition in the Irish data
was the highest of any country, with just 57% of the original sample remaining five
years into the panel. In her analysis of this attrition, Watson concludes that although
there are some correlations between attrition and economic variables, such correlation
explains only a very small part of the attrition.
Following the practice in the previous literature, the sample chosen for the
present study is comprised of men aged 17-65 whose labour market behaviour does
not indicate characteristics likely to be associated with erratic earnings. Thus, anyone
who experiences unemployment or time out of the labour market on ‘home duties’ at
any stage during the sample period is omitted from the sample altogether. Anyone not
reporting earnings in any year for which he was employed is also dropped from the
sample, as is anyone with earnings data missing due to attrition.
The sample is not, however, a fully balanced panel. It is necessary, in order to
identify the effects of age separately from time effects, that the average age of the
sample should not increase one-for-one with time. Thus, individuals are allowed to be
8
Not all of the 14 countries collected the ECHP data in all these year, but Ireland did.
8
in the sample for some years but not for others if their absence is due to either
retirement or being in full-time education. The assumption here is that retirement or
schooling are not indicative of either stability or instability in labour market
attachment. Of course, younger workers are known to have more earnings instability
than older workers, but this appears to be due to time taken to find a good job match
rather than lack of labour market attachment.
Outliers, defined as earnings in the top or bottom 1% of the sample
distribution, were also excluded. Finally, the sample was restricted to those who had
at least two years of observed earnings in the sample, in order to reduce the difference
between the samples used to calculate the variances and those used to calculate the
autocovariances.
Table 1 gives an indication of the degree to which our panel is unbalanced.
The first row gives the number of individuals who satisfy the selection criteria and are
not outliers in each year. This number changes as individuals enter and exit the
required age range. Of these, some will be in education and some will retire, and so
will not report earnings; subtracting the number in these labour market states gives the
‘number of workers’ figure. It is clear from the last two rows of the table that
allowing individuals to enter and exit the sample as they complete education or retire
means that neither mean age nor mean potential experience increase one-for-one with
time.
5. Results
For each year of data, current monthly log gross earnings are regressed on potential
experience and its square, as described above, and the residuals saved. The variance-
9
covariance matrix of these residuals is then calculated. This matrix is reported in
Table 2.
Looking along the diagonal, we observe that the variance is relatively stable
over the period, ranging from a high of 0.239 in 1994 to a low of 0.191 in 1996.
Looking down each column, the autocovariances decline over time, with the most
significant decline at the first order; after the first period, there is a smooth, but very
gradual decline. This has also been found by many other authors, including, for
example, Daly and Valetta (2007), for the US, Germany and the UK; Haider (2001)
for the US; Baker and Solon (2003) for Canada; and Gustavsson (2004) for Sweden.
As a preliminary estimate of the proportion of inequality that is due to
permanent factors, we use the non-parametric estimate given by (2) above. Following
Moffitt and Gottschalk (2002), we choose covariances that are five years apart to
estimate Var (i ) . Of course, given the short panel available, this means that only for
three pairs of years (1994/1999, 1995/2000 and 1996/2001) can the required
covariances be calculated. These indicate that in 1994, the permanent component of
inequality was 66% of the total; in 1995, the figure was 67% and in 1996, permanent
inequality was 69% of the total. These figures are suggestive of a slight rise in
permanent inequality, but with estimates for only three years available, no
significance can be read into the trend.
The estimated parameters of our full model are given in Table 3. The results
suggest some increase in the t , the factor loadings on the transitory shocks, towards
the end of the sample period, and a fall in the pt , the factor loadings on the permanent
characteristics. However, none of these is individually significantly different from one
10
at the 5% confidence level, and Wald tests also fail to reject the hypotheses that the
pt are jointly equal to one, or that the t are jointly equal to one, so no conclusions of
any trend can be drawn. ρ, the parameter indicating the degree of persistence of
transitory shocks, is estimated to be 0.65, and θ is estimated to be –0.2, which reduces
the magnitude of the first order correlations.
These estimates can be used to calculate permanent inequality, transitory
inequality and predicted total inequality. The decomposition results are presented in
Table 4 and Figure 1. Looking at the first two columns of Table 4, we see that the
model does well in predicting the actual total variance. Of more interest is the relative
importance of the transitory and permanent components in total inequality, which is
given in the last column of the table. At the beginning of the period, permanent
inequality accounted for approximately 67% of total earnings inequality. In
subsequent years, it ranged from 57% (2000) to 71% (1998), but generally stayed
around the average level of 65%. It is striking how similar this estimate is to the nonparametric one reported above.
Given the absence of significant trends in the above analysis we also report the
results for a more parsimonious specification of the model in which all the factor
loadings are set equal to one. The results are given in Table 5 and, as expected, are
similar to the more general model. The average proportion of inequality due to the
permanent component is 67%, as compared to the 65% reported earlier.
Although it can be difficult to compare studies across different countries
because of differences in time period, sample construction, and measures of earnings,
as well as in the details of the models of earnings dynamics used, it is interesting
nevertheless to try and compare our results to previous studies. In doing so we restrict
11
our comparisons where possible to studies that use a similar methodology to the one
we adopt. Haider (2001) finds that permanent inequality accounts for on average,
about two-thirds of total variance in the US between 1968-1992. Baker and Solon
(2003) conduct a similar analysis for Canada from 1976-1992, and find that the
permanent component fell from about 70% to 64% over that period. Gustavsson’s
(2007) results indicate that permanent inequality varied between approximately 6370% for males aged 40 in Sweden during the 1990s. Cervini Plá and Ramos (2006)
find, for Spain, that the permanent component varies substantially by age cohort, but
for men born from 1954-1963, it rose from about 60% to about 75% between 1993
and 2001. Moffitt and Gottschalk (2002) find the permanent component varying from
about 37% to about 63% over the 1969-1996 period in the US. And Ramos (2003)
reports a permanent component that averages about 40% during the 1990s in Britain.
The study that is the most similar to ours is, however, Daly and Valetta
(2007). They report a permanent contribution of 55% for the US, 58% for Germany
and 53% for Great Britain over the 1990’s. Thus our estimate of 65% is at the upper
end of the range of previous estimates, and is particularly high in the most relevant
comparison.9
6. Extension to Public and Private Sector Earnings
In this section, we report preliminary results of the decomposition of public and
private sector earnings inequality into permanent and transitory components. For now,
the analysis is restricted to the non-parametric estimate of the permanent component
given by equation (2).
9
See also Gangl (2005) who uses a different methodology but also concludes that permanent inequality
is high in Ireland
12
Tables 5 and 6 give the covariance matrices for public and private sector
earnings respectively. In constructing these matrices, the sample was initially
constructed exactly as before – individuals who were missing earnings data for any
year that they were not either in education or in retirement were excluded, as were
those outside the 17-65 age range. Earnings in the top and bottom 1% of the full
sample’s distribution (i.e. public and private sector earnings together) were also
dropped. It was at this stage that the sample was separated into observations on public
sector and private sector earnings. At that point, any individual who did not have at
least two years of earnings in the relevant sector was dropped. It is important to note
that individuals who moved from one sector to the other and had at least two years of
recorded earnings in each sector will be present in both samples. Thus, this is an
analysis of public and private sector earnings, rather than public and private sector
workers.
Looking at the variance-covariance matrix for public sector earnings in Table
6, its most striking characteristic is how much lower the variances (along the
diagonal) are than in Table 2, the equivalent table for all workers. Variances range
from 0.124 in 1998 to 0.153 in 1994, with a typical value of about 0.14; this compares
with a typical value of about 0.20 for all workers. In general, however, the same
pattern of autocovariances holds as before – the biggest drop is in the one-period
covariances, with smaller drops – and indeed some small rises – thereafter. There are
a couple of exceptions, however: in 1996, and particularly in 1998, the one-period
autocovariances are very close to the variances, indicating very high correlations
between earnings in those years and in the following year.
13
For the private sector variance-covariance matrix given in Table 7, the
variances are very similar to those in Table 2, as is the pattern of covariances.
However, the sharp contrast with Table 6, for public sector earnings, is clear. In 1994,
the variance of private sector earnings was 0.24, whereas that for public sector
earnings was 0.15; in 2001, the variances were 0.22 and 0.14 respectively. Thus, wage
dispersion is, unsurprisingly, significantly lower in the public sector than in the
private sector.
The non-parametric estimates of the proportion of total inequality that is
permanent in the two sectors also reveal sharp differences. In the public sector, 76%
of inequality is estimated to be permanent in 1994, 77% in 1995 and 78% in 1996. In
contrast, the permanent proportion in the private sector is estimated to be 60% in
1994, 60% again in 1995 and 63% in 1996. Thus, public sector earnings are less
dispersed, but more persistent than private sector earnings; the distribution is less
unequal, but once assigned a place in the distribution, it is more difficult to change
rank within the distribution.
We have also estimated the parametric model for both these sectors. When
estimating the fully specified model with factor loadings for the private sector the
model had difficulty distinguishing between two competing structures; the first had a
high permanent variance and a relatively low autoregressive parameter, while the
second had a lower permanent variance and a much higher autoregressive parameter
(close to one). The implications of both structures for earnings dynamics are very
similar, highly persistent wages across time, however the interpretation of this
persistence differs. Given this difficulty in the full-model and the fact that the factor
loadings show no significant trend over time we report the results for the restricted
14
model with no factor loadings. These are given in Table 8 for both the private and the
public sector. The autocorrelation function for the transitory component (using 1994
as a base) for both sectors is given in Figures 2 and 3. Both functions show little
correlation in transitory shocks beyond a lag length of 5.
The decomposition of inequality into its permanent and transitory component
is consistent with the non-parametric results above. For the private sector the
permanent effect accounts for 60% of the total variance over the sample, while the
corresponding figure for the public sector is 76%, confirming the view that while
inequality is lower in the public sector the wage distributions are much more rigid
with little mobility within the distribution. This would accord with expectations, given
that pay progression occurs almost exclusively along pre-determined ‘pay scales’ in
the public sector.10 It also concurs with similar conclusions drawn by Postel-Vinay
and Turon (2005), about the public sector earnings in Britain.
7. Conclusions
It is widely known from previous studies that Ireland has a high degree of earnings
inequality, of an order similar to that in North America and the UK. Using the ECHP
to analyse the covariance structure of earnings in Ireland from 1994-2001, we
decompose aggregate inequality into its transitory and permanent components. Our
results show that the degree of persistence of Irish inequality is also at the upper end
of the range of international estimates, with a permanent component of about 65%.
Moreover, preliminary results show that this figure masks a significant
difference in the importance of the permanent component between the public and
While we are confident that the public sector exhibits significantly more ‘persistence’ than the
private sector the fragility of some of the parametric models when estimated for the public sector
suggests we should exercise caution when labelling this persistence as ‘permanent’ or ‘transitory’.
10
15
private sectors in Ireland. Although earnings inequality is significantly lower in the
public than in the private sector, the permanent component is much higher.
However, we can discern no strong time trend in the proportion of inequality
that is permanent, either in the analysis for all workers, or separately for the public
and private sectors. We argued that there are two forces likely to affect the size of the
permanent proportion of inequality: centralized wage bargaining, tending to increase
permanence, particularly in the public sector; and greater dynamism caused by the
prolonged economic boom of the 1990s, tending to decrease permanence. Perhaps it is
the case that these two forces, in pulling in opposite directions, are cancelling each
other out.
We hope in future research to carry out the decomposition of earnings
inequality into its permanent and transitory components for other countries using a
common methodology and comparable data, which would allow consistent
comparisons between countries.
16
References
Baker, Michael (1997). ‘Growth Rate Heterogeneity and Covariance Structure of
Life-Cycle Earnings’, Journal of Labor Economics, Vol 15, No. 2, pp. 338-375.
Baker, Michael and Gary Solon (2003). ‘Earnings Dynamics and Inequality among
Canadian Men, 1976-1992: Evidence from Longitudinal Income Tax Records’,
Journal of Labor Economics, Vol. 21, No. 2, pp. 289-321.
Barrett, A., T. Callan and B. Nolan (1999).’Rising Wage Inequality, Returns to
Education and Labour Market Institutions: Evidence from Ireland’, British Journal of
Industrial Relations, Vol. 37, No. 1, pp. 77-100.
Barrett, A. J. FitzGerald and B. Nolan (2002). ‘Earnings Inequality, Returns to
Education and Immigration in Ireland’, Labour Economics, Vol. 9, No. 5.
Cervini Plá, Maria and Xavier Ramos Morilla (2006). ‘Permanent and Transitory
Earnings Inequality in Spain, 1993-2000’, Working Paper, Universitat Autonoma de
Barcelona.
Chamberlain, Gary, (1984). ‘Panel Data’ in Z. Griliches and M.D. Intriligator (eds.),
Handbook of Econometrics, Amsterdam: Elsevier.
Daly, Mary and Robert Valletta (2007). ‘Cross-National Trends in Earnings
Instability and Earnings Inequality’, Economic Letters, forthcoming.
Gangl, M. (2005). ‘Income Inequality, Permanent Incomes and Income Dynamics’,
Income Dynamics, Vol. 32, No.2, pp. 140-162.
Gottschalk, Peter and Mary Joyce (1998). ‘Cross-National Differences in the Rise in
Earnings Inequality: Market and Institutional Factors’, Review of Economics and
Statistics, Vol. 80, No. 4, pp. 489-502.
Gottschalk, Peter and Robert Moffitt (1994) “The Growth of Earnings Instability in
the US Labor Market.’ Brookings Papers on Economic Activity , No. 2, pp. 217-254.
Gottschalk, Peter and Robert Moffitt (1995). ‘Trends in the Covariance Structure of
Earnings in the US: 1969-1987’. Working Paper, John Hopkins University.
Gustavsson, Magnus (2004). ‘Trends in the Transitory Variance of Earnings:
Evidence from Sweden 1960-1990 and a Comparison with the United States’,
Working Paper 2004:11, Uppsala University.
Gustavsson, Magnus (2007). ‘The 1990s Rise in Swedish Earnings Inequality –
Persistent or Transitory?’, Applied Economics, Vol. 39, pp. 25-30.
Haider, Steven J. (2000). ‘Generalized Method of Moments with Incomplete Data:
Estimating Earnings Dynamics Models’, Unpublished Manuscript, RAND.
17
Haider, S J. (2001). ‘Earnings Instability and Earnings Inequality of Males in the
United States: 1967-1991’, Journal of Labor Economics, Vol. 19, No 4, pp. 799-836.
MaCurdy, Thomas (2007) ‘A Practioner’s Approach to Estimating Intertemporal
Relationships Using longitudinal Data: Lessons from Applications in Wage
Dynamics,” in Handbook of Econometrics, Vol. 6, Part 1, pp. 4057-4167.
Moffitt, Robert and Peter Gottschalk (2002). ‘Trends in the Transitory Variance of
Earnings in the United States’, Economic Journal, Vol.112, pp. 68-73.
Nolan, B. (2000). ‘The Distribution of Earnings’, Chapter 6 in B. Nolan, B. Maitre, D.
O’Neill and O. Sweetman, The Distribution of Income in Ireland, Dublin: Combat
Poverty Agency.
Nolan, B. (2003). ‘Income Inequality during Ireland’s Boom’, Studies: An Irish
Quarterly Review, Vol. 92, No. 366.
Nolan, Brian (2004). ‘Low Income Dynamics’, EU Research on Social Sciences and
Humanities - Economic Change, Unequal Life Chances and Quality of Life
(CHANGEQUAL), Project HPSE-CT2002-50028.
Nolan, B., B. Maitre, D. O’Neill and O. Sweetman (2000). The Distribution of Income
in Ireland, Dublin: Combat Poverty Agency.
Nolan, B. and B. Maitre (2000). ‘A Comparative Perspective on Trends in Income
Inequality in Ireland’, Economic and Social Review, Vol. 31, No. 4.
O'Neill, Donal and Olive Sweetman (2001). ‘Inequality in Ireland 1987-1994: A
Comparison using Measures of Income and Consumption’, Journal of Income
Distribution, Vol. 10, No. 3-4, pp. 23-39.
Postel-Vinay, F and H. Turon (2005). ‘The Public Pay Gap in Britain: Small
Differences that (Don’t) Matter’, CMPO Working paper, no. 05/121.
Ramos, Xavier (2002). ‘The Covariance Structure of Earnings in Great Britain, 19911999’, Economica, Vol. 70, pp. 353-374.
Watson, Dorothy (2003). ‘Sample Attrition Between Waves 1 and 5 in the European
Community Household Panel’, European Sociological Review, Vol. 19, No. 4, pp.
361-378.
Whelan, C.T., R. Layte, and B. Maître (2003). ‘Persistent Income Poverty and
Deprivation in the European Union: An Analysis of the First Three Waves of the
European Community Household Panel’, Journal of Social Policy, Vol. 32, No. 1,
pp.1-18.
Whelan, C.T., R. Layte and B. Maître (2004). ‘Understanding the Mismatch between
Income Poverty and Deprivation: A Dynamic Comparative Analysis’, European
Sociological Review, Vol. 20, No. 4, pp. 287-301.
18
Table 1: Details of Sample Size, showing Mean Age and Experience for
Revolving Balanced Panel
1994 1995 1996 1997 1998 1999 2000 2001
Initial Balanced Sample N
490
505
521
540
556
568
574
570
% in Education
7.1
8.1
8.4
7.4
7.6
4.2
1.9
0.7
% Retired
0
0
0.8
1.9
2.9
3.2
3.8
4.0
Balanced 455
464
473
490
498
526
541
543
38.9
39.7
39.8
39.4
39.3
38.7
38.5
39.1
18.9
19.5
19.7
19.3
19.2
18.9
18.9
19.4
Revolving
Sample N
Mean Age
Mean Potential Experience
Table 2: Variance-Covariance Matrix of Residuals from Regressions of Monthly
Gross Earnings on Experience, Male Employees. Cell sizes in italics
1994
1994
1995
1996
1997
1998
1999
2000
2001
.2394
455
1995
1996
1997
1998
1999
2000
2001
.1868
.2030
448
464
.1693
.1681
.1914
444
452
473
.1644
.1603
.1628
.1989
435
446
462
490
.1564
.1563
.1546
.1658
.1977
427
435
452
475
498
.1571
.1448
.1413
.1530
.1728
.2024
423
429
447
470
487
526
.1453
.1360
.1347
.1457
.1551
.1597
.2142
415
422
440
462
481
509
541
.1476
.1325
.1321
.1395
.1504
.1533
.1720
.2066
412
419
433
458
475
507
531
543
19
Table 3: Parameter Estimates: Monthly Gross Earnings, All Male Employees
Parameters
Estimate
Standard Error
 2
0.1594
0.02296

0.6503
0.10638
Factor Loadings – Transitory Shock
( t )
1994
1
1995
0.9908
0.12507
1996
1.0241
0.15715
1997
1.0277
0.14647
1998
0.9542
0.13343
1999
1.0422
0.14641
2000
1.2265
0.17577
2001
1.1457
0.16598
Factor Loadings – Permanent Shock
( pt )
1994
1
1995
0.9236
0.05542
1996
0.8831
0.07091
1997
0.9118
0.07695
1998
0.9427
0.08269
1999
0.9140
0.08463
2000
0.8723
0.08670
2001
0.8923
0.08148
 v2
0.1029
0.07142
 2
0.0456
0.01447

-0.1995
0.07895
20
Table 4: Trends in Permanent and Transitory Inequality in Ireland 1994-2001
Year
Actual
Predicted
Transitory
Persistent
Proportion of
Variance
Variance
Inequality
Inequality
inequality due to
Permanent
component
1994
.2394
.2385
.0791
.1594
.6684
1995
.20300
.2037
.0678
.1360
.6674
1996
.1914
.1922
.0679
.1243
.6466
1997
.1988
.1990
.0665
.1325
.6658
1998
.1977
.1983
.0566
.1417
.7144
1999
.2024
.2004
.0672
.1332
.6645
2000
.2141
.2142
.0929
.1213
.5662
2001
.2066
.2079
.0810
.1269
.6104
Table 5: Parameter Estimates: Monthly Gross Earnings, All Male Employees:
No Factor Loadings
Parameters
Estimate
 2
0.1387

0.5812
 v2
0.1705
 2
0.0490

-0.181
Note: Standard errors yet to be estimated.
21
Standard Error
Table 6: Variance-Covariance Matrix of Residuals from Regressions of
Monthly Gross Public Sector Earnings on Experience, Males. Cell sizes in italics
1994
1994
1995
1996
1997
1998
1999
2000
2001
.1525
209
1995
1996
1997
1998
1999
2000
2001
.1331
.1508
206
210
.1158
.1228
.1272
202
205
209
.1167
.1200
.1137
.1455
184
186
190
196
.1101
.1140
.1071
.1139
.1240
177
178
180
179
189
.1164
.1166
.1070
.1148
.1219
.1413
167
169
171
170
174
185
.1129
.1164
.1095
.1210
.1123
.1143
.1404
158
158
160
159
164
166
175
.1086
.1035
.0994
.1104
.1031
.1071
.1161
.1445
162
161
164
166
167
170
168
182
22
Table 7: Variance-Covariance Matrix of Residuals from Regression of
Monthly Gross Private Sector Earnings on Experience, Male Employees. Cell
sizes in italics
1994
1994
1995
1996
1997
1998
1999
2000
2001
.2536
243
1995
1996
1997
1998
1999
2000
2001
.1888
.2003
235
253
.1695
.1615
.1966
232
242
263
.1542
.1484
.1568
.1997
226
236
253
288
.1528
.1502
.1540
.1650
.2033
224
234
248
276
305
.1517
.1348
.1367
.1475
.1693
.2100
221
228
243
269
289
332
.1302
.1208
.1246
.1345
.1562
.1663
.2301
219
228
241
268
288
317
357
.1390
.1157
.1240
.1277
.1511
.1609
.1817
.2173
217
225
236
263
280
308
334
347
23
Table 8: Parameter Estimates: Monthly Gross Earnings, Private and Public
Sector Employees: No Factor Loadings
Parameters
Private Sector
Estimate
Standard
Error
Public Sector
Estimate
 2
.1298
.1068

.5542
.6516
 v2
.2089
.0733
 2
.0595
.0265

-.1055
-.3934
Note: Standard errors yet to be estimated.
24
Standard
Error
Figure 1: Trends in Permanent and Transitory Components of Earnings
Inequality in Ireland, 1994-2001, All Male Employees
Earnings Inequality: Outliers omitted
0.3
Inequality
0.25
0.2
Predicted
0.15
Transitory
0.1
Permanent
0.05
0
1992
1994
1996
1998
Year
25
2000
2002
Figure 2: Autocorrelation Function for Transitory Shocks (Private Sector)
Correlation
ACF Transitory component
Private Sector (Base 1994)
1.2
1
0.8
0.6
0.4
0.2
0
0
1
2
3
4
5
6
7
Lag Length
Figure 3: Autocorrelation Function for Transitory Shocks (Public Sector)
Correlation
ACF Transitory component Public Sector (Base Year
1994)
1.2
1
0.8
0.6
0.4
0.2
0
0
1
2
3
4
Lag Length
26
5
6
7