OF UNEMPLOYMEENT RATES AS AN INDICATOR OF FAIR EMPLOYMENT
Northern Ireland Economic Research Centre and Professor Richard Breen Centre for Social Research The Queen's University of Belfast
(1996)
COMMENTS
Comments: Anthony Murphy
Catholic men experience a much higher unemployment rate than Protestant
men in Northern Ireland. This unemployment differential has always
been a headline figure and has attracted a lot of attention about
the factors which account for it. In their paper, Gudgin and Breen
address this issue using a stylised model of labour market flows.
They consider what the unemployment differential would be assuming
no labour market discrimination against Catholics. Their results
are controversial since they imply that the chill factor and religious
discrimination, whether direct or indirect, have little or nothing
to do with either the current or past unemployment differential
between Catholics and Protestants. Gudgin and Breen's reasoning is based upon the observation that the ratio of unemployment rates has remained "roughly constant" in the past two decades despite the doubling of overall unemployment rates[1]. Catholics are still approximately 2¼ times more likely to experience unemployment than Protestants. The "constancy" of this ratio contrasts with the rising level of overall unemployment since 1971.
In view of this, Gudgin and Breen argue that the normal or legitimate
operation of the labour market must naturally predispose the Catholic
community to an unemployment rate which is over twice that for
Protestants[2]. If their argument is correct then discrimination,
whether direct or indirect, past or present, has little or no
role to play in explaining the unemployment differential and the
higher levels of deprivation experienced by Catholics. Gudgin and Breen use what is commonly known as a simulation model to isolate key features of the male labour market in Northern Ireland over the past two decades. They focus upon labour market flows or changes in labour market status such as labour turnover, migration and entry and exit to the labour market. The simulation model is used to examine "what if" type scenarios over the period 1971 to 1991 using certain stylised facts and assumptions. Thus the model results are wholly dependant upon the assumptions made.
When correctly formulated simulation models offer a great deal
of insight at relatively little cost. However the value of these
models is crucially dependant upon the validity of the assumptions
made. Inappropriate or invalid assumptions lead to incorrect or
misleading results. Many readers will find it difficult to identify
the key assumptions in the Gudgin-Breen model because of the way
it is presented. In fact there are only two key assumptions in
their model. Upon examination these two assumptions turn out to
be arbitrary and implausible. As a result, the findings in the
Gudgin and Breen paper are not particularly interesting and are
certainly not robust. When plausible assumptions are used, very
different results are obtained. Gudgin and Breen argue that their approach is novel and challenges existing research findings. Of course, the novelty of the Gudgin and Breen methodology does not necessarily mean that they are contributing anything new to the employment equality debate. Existing research "explained" or attributed approximately half of the unemployment differential by factors such as age, education, fertility, location and other so called "structural factors", which apart from religion, differentiate Catholics and Protestants[3]. The other half is "explained" by religion, which includes the effects of the chill factor and discrimination, whether direct or indirect, past or present.
Gudgin and Breen argue that all, or nearly all, of the unemployment
differential is explained by an augmented list of structural factors.
These "new" factors are a higher Catholic rate of labour
turnover, a lower Catholic migration response to unemployment
and a higher Catholic rate of entry into the labour market. According
to Gudgin and Breen, these "new" factors interact with
the traditional structural factors to explain the high unemployment
differential.
The Gudgin and Breen paper is based upon two important assumptions.
Firstly, Catholic migration is less responsive to unemployment
than Protestant migration. In other words Catholics do not respond
sufficiently to their high levels of unemployment.' This observation
is made despite the fact that Catholic migration rates were 50%
higher than Protestant rates over the two decades to 1991.
Secondly, Gudgin and Breen build into their model a Catholic disadvantage
measure which dictates that a Catholic job seeker is only 70%
as likely to be successful as a Protestant job seeker[4].
Their simulation results are crucially dependant upon this assumption.
Gudgin and Breen generate their migration results by using an implausible model of migration which has no empirical foundation. Migration statistics are not collected by religion. Gudgin and Breen get their results by imposing and not estimating a lower Catholic propensity to migrate in response the unemployment rate. This is an arbitrary assumption - other assumptions give very different results. As Professor Nickell of Oxford University pointed out, one cannot determine the slope of a line given one data point. Despite their best efforts, Gudgin and Breen cannot get around this fundamental problem.
The Gudgin and Breen migration model implies that the high Catholic
unemployment rate is the result of their higher fertility rate
combined with their unwillingness to emigrate in sufficiently
large numbers. In other words, given the situation in Northern
Ireland, Catholics should "get on their bikes" more
than they have done. If they did this then, according to Gudgin
and Breen, the unemployment differential would be much lower.
This claim is not well grounded in research since little or no
comprehensive research into the determinants of emigration from
Northern Ireland by religion has been carried out. In addition,
since those in non-manual occupations are more mobile than those
in manual occupations, it is possible that higher Catholic migration
from Northern Ireland could raise the unemployment differential[5].
The second and even more crucial assumption in the Gudgin and Breen model has to do with the relative chances of a Catholic job seeker being successful. There is no denying that, because of structural factors, Catholics are less likely to obtain a job. Catholics tend to be younger, have larger families, have lower qualifications and live in rural areas and these are some of the accepted structural factors which lead to their higher unemployment rate. However, there is no evidence that Catholic job seekers are only 70% as likely to get a job solely as a result of these structural factors. Gudgin and Breen claim that the 70% figure "corresponds closely with existing and widely accepted measures of structural disadvantage". They cite the work of Smith and Chambers (1991) and Murphy and Armstrong (1994) who find that structural factors account for about half of the unemployment differential. Gudgin and Breen assert that the degree of Catholic disadvantage built into their model "accounts for the widely observed structural disadvantage of Catholics" and that "it appears to exclude any substantial level of direct or indirect discrimination or so called 'chill factors'". These claims are incorrect.
Smith and Chambers (1991) and Murphy and Armstrong (1994) obtain
their results by examining the stocks of employed and unemployed.
The estimated religion effects which they find are average effects.
However, Gudgin and Breen are looking at labour market flows so
the religion effects they are interested in are marginal ones.
There is no simple relationship between stock/average and flow/marginal
effects[6]. As a result, one must be very careful when
inferring marginal effects from average effects, especially when
controlling for other factors. This indirect approach is avoided
by looking at the direct evidence on labour market flows.
Gudgin and Breen argue that, solely as a result of structural
factors, Catholic job seekers are only 70% as likely to be successful
in getting a job as Protestant job seekers. The only direct evidence
on this is to be found in Murphy and Armstrong (1994). The 70%
figure used by Gudgin and Breen is close to the actual Catholic:Protestant
ratio for the probability of getting a job obtained in recent
Labour Force Surveys. It is true that, Catholic job seekers are
only about 70% as likely to get a job as Protestant job seekers.
However this raw ratio is made up of two components - one component
due to structural factors and the other due to religion. The econometric
results in Murphy and Armstrong (1994) suggest that the structural
component accounts for only about one third and not the whole
of the 30% gap between Catholics and Protestants. This means that
the correct figure for Catholic disadvantage in getting jobs due
to structural factors is about 90% and not the 70% figure used
by Gudgin and Breen. This conclusion is robust and holds even
when emigration is taken account of. Turning to the indirect evidence, Gudgin and Breen misinterpret the econometric results in Smith and Chambers (1991) and Murphy and Armstrong (1994). The estimated contribution of structural factors to the unemployment differential in these two papers already includes the effects of higher Catholic rates of natural increase, job turnover, and migration to some extent. They are captured, inter alia, by the age variables. High rates of natural increase are reflected in the age structure of the Catholic population and age is the single most important determinant of job turnover and migration. Gudgin and Breen do not take account of this. As a result they double count the effects of higher Catholic rates of natural increase and job turnover. Some other indirect evidence is provided by Murphy and Shuttleworth (1994) who look at the first destinations of recent school leavers in Northern Ireland. Recent school leavers are an interesting group to examine for a number of reasons. School leavers have no previous labour market history and, other things being equal, their experience should only reflect what is currently happening in the labour market. If the arguments of Gudgin and Breen are correct, one should not find any large or significant religion effects when modelling the first destinations of these school leavers.
However, this is not what Murphy and Shuttleworth find. They find
large and significant religion effects for young Catholic men
who leave school early, even after controlling for a range of
relevant factors including family background, qualifications and
subject choice. Catholic males who leave from the fifth or lower
sixth form are, other things being equal, significantly less likely
to be employed and more likely to be on a youth training scheme.
It is well known that this early disadvantage tends to be compounded
over time. As a result, the same individuals who were on youth
training schemes after leaving school tend to be unemployed a
few years later. Thus the evidence in Murphy and Shuttleworth
is not consistent with the claims of Gudgin and Breen.
Gudgin and Breen argue that they obtain their 70% figure by trial
and error - it is the figure which generates the actual 1991 unemployment
differential in their model. This is how they originally arrived
at the 70% figure but one should not confuse how this figure was
arrived at and how it is subsequently justified. In the absence
of any supporting evidence, the 70% figure is just an arbitrary
figure. However, Gudgin and Breen claim that the 70% "closely
corresponds to the widely accepted measure of Catholic structural
disadvantage". Thus, their original "what if' 70% figure
is transformed by sleight of hand into a "this is" figure.
Put simply, Gudgin and Breen should either use their 70% figure
and exclude the other factors they consider, or use the 90% figure
based on the results in Murphy and Armstrong (1994) and include
the other factors in their model. However, Gudgin and Breen use
the 70% figure and include the other factors in their model and
so Catholics are modelled with a double disadvantage. This is
because the raw Catholic disadvantage in hiring of 70% already
includes the effects of these other factors. The outcome of this
is that Gudgin and Breen double count the structural disadvantage
of Catholics. The key simulation result in the Gudgin-Breen paper is the following one. Gudgin and Breen contend that, using their model and plausible parameters values which only reflect Catholic disadvantage due to structural factors, it is possible to simulate the actual change in the unemployment differential between 1971 and 1991. In other words, a high unemployment differential is a long run, equilibrium or natural feature which has little or nothing to do with discrimination. Obviously if this were true, one would draw some very strong policy implications.
However, the key Gudgin-Breen simulation results are not robust.
They obtain their results by (i) imposing a lower Catholic migration
propensity in response to the unemployment rate and by (ii) setting
the relative chances of a Catholic job seeker getting a job, due
to structural disadvantages, at 70%. When the migration equations
and the relative chances of a Catholic job seeker being successful
are based upon more plausible assumptions, their results breakdown.
Figure 1 illustrates this. The simulated unemployment differential
based on the 70% and 90% figures are labelled GB and MA respectively.
Gudgin and Breen (GB) use the 70% figure. The 90% figure is a
more realistic one based on the research evidence in Murphy and
Armstrong (MA). In the Gudgin-Breen model, when the 70% figure
is replaced by the 90% figure, the simulated unemployment differential
in 1991 is no longer over 2. Instead it is about 1.5[7]
In other words, there is nothing to suggest that the normal working
of the labour market produces a high unemployment differential.
Reasonable assumptions support previous research findings that
the chill factor and discrimination, whether direct or indirect,
have an important role in explaining why Catholics experienced
a considerably higher unemployment rate than Protestants. In conclusion, Gudgin and Breen ask an interesting and important question - why are Catholics more than twice as likely to experience unemployment? They also use an interesting approach to answer this question - a simulation model of labour market flows. However, they misinterpret existing research findings and thereby use implausible assumptions which double count the factors which contribute to Catholic disadvantage. As a result, Gudgin and Breen provide the wrong answer to an important question.
At this point, it is worthwhile leaving economic and econometric
issues aside in order to consider the Gudgin-Breen paper at face
value. Many readers of the paper will be tempted to ignore these
issues and draw policy conclusions. For example, some may conclude
that Catholics should migrate at over twice the rate of Protestants
because their unemployment rate is over twice as high, that Catholics
should reduce their fertility or that the long term unemployed
should be placed on job schemes, thereby "solving" the
problem of a high unemployment differential. I am sure that Gudgin
and Breen would not wish the results in their paper to be used
as support for these sorts of policies. However, in the present
climate, some readers may choose to draw these conclusions because
the Gudgin-Breen paper is rather naive from a policy perspective.
BIBLIOGRAPHY Gudgin G and Breen R (1995), 'Evaluation of the Ratio of Unemployment Rates as an Indicator of Fair Employment", revised version of paper originally presented at CCRU Seminar on Community Differentials in Unemployment, September 1994. Hughes G and McCormick B (1994), 'Did Migration in the 1980s Narrow the North-South Divide", Economica, 61, 509-27. Murphy A and Armstrong D (1994), A Picture of the Catholic and Protestant Male Unemployed, Employment Equality Review, Research Report No 2, Central Community Relations Unit, Belfast. Murphy A (1996), 'Explaining High Catholic Unemployment Rates. A Critique of Gudgin and Breen", forthcoming as Centre for Economic Research, Working Paper, UCD. Murphy A and Shuttleworth 1 (1995), 'Education, Religion and the 'First Destinations' of recent School-Leavers in Northern Ireland", Centre for Economic Research, Working Paper No 94/19.
Smith D and Chambers G (1991), Inequality in Northern
Ireland, Clarendon Press, Oxford.
Comments: Robert Rowthorn In the Northern Ireland context, such a finding is politically controversial and, not surprisingly, has been vigorously attacked. The most cogent criticism comes from Mr Anthony Murphy and is contained in abridged form in his comment in the present volume. He makes two central points. The first concerns Gudgin and Breen's migration equation, which he says is mis-specified and has no empirical foundation. However, as he concedes elsewhere nothing of importance really hangs on this issue, and the form of the migration equation in the GB model is not the really crucial element in generating their results (Murphy, 1995, pp33-34). Indeed, this is demonstrated by Gudgin and Breen in their appendix 5, which is a response to critical remarks by Professor Nickell at the conference where their paper was originally presented. The main point of substance in Murphy's attack concerns Gudgin and Breen's claim that their estimate for Catholic disadvantage can be entirely explained by so-called structural factors, without little or no contribution from job discrimination. Gudgin and Breen's justification is as follows. Assuming equal quit rates, population growth and migration propensities in the two communities, they estimate that Catholic disadvantage alone will generate an unemployment ratio of 1.45. Catholic disadvantage plus higher quit rates together generate a ratio of 1.66. Since the ratio when all factors are taken into account is 2.2, these estimates imply that Catholic disadvantage alone accounts for 37.5% of the difference in unemployment rates between Catholics and non-Catholics, and 55% in combination with differential quit rates. Note that the higher quit rate for Catholics arises mainly from the younger age of Catholics and is therefore a "structural factor". Gudgin and Breen point out that most writings on Northern Ireland estimate that structural factors account for around half of the unemployment differential between Catholics and non-Catholics. Since their own estimate for Catholic disadvantage (together with quits) also explains around half of this differential, they conclude that disadvantage must be a reflection of structural factors alone. They admit that the chill factor and discrimination may still exist, but the presumption must be that their effects are not quantitatively significant. It is interesting to note that Murphy and Armstrong (1994) conclude that "about half of the difference in the unemployment rates between Catholic and non-Catholic men is due to differences in the characteristics of the two groups, and the other half is accounted for by religion. This result appears robust" (p25). Thus, on the basis of Murphy and Armstrong's own findings about unemployment differentials, Gudgin and Breen are justified in interpreting their disadvantage parameter as purely structural. However, there is other evidence which contradicts this interpretation. Buried away in appendix 8.2 of Murphy and Armstrong there are some econometric estimates of the factors which influence flows into employment. If these are accurate, they provide a better source of evidence than the studies of unemployment differentials which Gudgin and Breen cite in their support. According to the econometric estimates in appendix 8.2 of Murphy and Armstrong, only one-third of the Catholic disadvantage in hiring is due to structural factors, with the remaining two-thirds being a pure religion effect.
Suppose we accept this finding. What implications does it have
for the Gudgin and Breen simulations? Assume that one-third of
the Catholic disadvantage is explained by structural factors and
the rest is a pure religion effect. This implies that in the absence
of a pure religion effect, the parameter 6 would be 0.93 instead
of 0.80. At my request, Dr Gudgin has kindly undertaken a new
simulation using a value of 0.93 for , whilst maintaining unchanged
the original differentials in population growth, migration propensities
and quit rates. The result indicates how unemployment behaves
in the absence of a pure religion effect, but with all other intercommunal
differences unchanged.
It is clear that with =0.93 a substantial part of the unemployment ratio remains unexplained. Suppose that the difference between =0.80 and = 0.93 is a pure religion effect which captures the impact of discrimination and the so-called chill factor. Then, in the absence of discrimination and the chill factor, the ratio of Catholic to non-Catholic unemployment would be in the region of 1.68. This is very high, but it is well below the original figure of 2.2.
The above discussion suggests that there is a great deal of uncertainty
about the influence of discrimination and the chill factor on
unemployment in Northern Ireland. I find it difficult to believe
that Gudgin and Breen are right in implying that these factors
have had virtually no effect on the evolution of the Catholic
to non-Catholic unemployment ratio since 1971. Discrimination
and even the chill factor may be less important than they used
to be, but the estimates of Gudgin and Breen refer to the entire
period 1971-91 and should therefore capture the effects of these
factors in the earlier years when they are clearly significant
in sectors of the economy such as engineering or shipbuilding.
On the other hand, I am not convinced that there is still a large
Catholic disadvantage in hiring which is not explained by structural
factors such as location, house ownership, age, health and education.
There is room for more empirical investigation of this issue.
BIBLIOGRAPHY Murphy A and Armstrong D (1994), A Picture of the Catholic and Protestant Male Unemployed, Employment Equality Review, Research Report No 2, Central Community Relations Unit, Belfast.
Murphy A (1995), Changing Structures Within the Labour Force
and Their Effect on the Unemployment Differential, Belfast:
Report to the Fair Employment Commission, May.
Notes: [2] According to their model, the unemployment differential should fall as the aggregate unemployment rate rises, other things being equal. However, Murphy (1995) presents a simple and plausible alternative model in which the unemployment differential is independent of the overall unemployment rate. [3] These structural factors include age, marital status, number of children, housing tenure, the economic activity of others in the household, educational and vocational qualifications and location. In other studies these factors are important determinants of the incidence of unemployment. For example, married middle-aged men with no children, who own or are buying their own house, whose wives work, who have a degree are far less likely to be unemployed than other men. Of course all of these factors are not necessarily exogenous. [4] Gudgin-Breen measure Catholic disadvantage in hiring () relative to the average job seeker. Their preferred figure for is 0.8 or 80%. When Catholic disadvantage in hiring is measured relative to the average Protestant job seeker, the figure is 0.7 or 70%. [5] For example, Hughes and McCormick (1994) examine inter-regional migration in Britain in the 1980s using Labour Force Survey data and econometric models. Other things being equal, they find that migration helped to reduce regional unemployment differentials for non-manual workers but not for manual workers. Non-manual workers were much more mobile than manual workers. [6] A simple analogy helps to illustrate this. Consider a bath being filled with water. The statement that the water in the bath (the stock) is cold (average effect) does not imply that the water coming from the tap (the flow) is cold (marginal effect).
[7] The simulated unemployment differential is 2.2 in the key
Gudgin-Breen simulation and 2.5 in Murphy (1995) when a 70% figure
is used. The slightly different results are generated by the different
migration equations used.
| ||||||||||||||||||
