OF UNEMPLOYMEENT RATES
AS AN INDICATOR OF FAIR EMPLOYMENT
Northern Ireland Economic Research Centre
Professor Richard Breen
Centre for Social Research
The Queen's University of Belfast
Rejoinder: Graham Gudgin and Richard Breen
Murphy's comments in this volume are an abbreviated version of a longer paper, funded by the Fair Employment Commission, to be published in the Economics Department Working Paper Series at University College Dublin. Murphy's full paper contains 15 criticisms of the Gudgin/Breen paper, the most important of which are summarised in his contribution to this volume. The remaining points are largely trivial or specious and we are pleased that they are not repeated here. Murphy's comments include no agreement with any aspect of the Gudgin/Breen paper. He makes little effort to achieve a balanced judgement, and writes in an unfortunate tone which includes accusations of 'naivety' and 'sleight of hand'.
Murphy's one-sidedness is clear in his description of the Gudgin/Breen model in his comments in this volume. He describes us as "assuming no labour market discrimination" despite the fact that this assumption appears nowhere in our paper. Later he describes the simulations as "wholly dependent upon the assumptions made" in spite of the fact that the simulations are heavily constrained by the known facts of population employment and unemployment levels in 1971 and 1991, and by the migration flows between these dates. Nor are these "stylised" facts as Murphy claims, they are abstracted from the Census and are as precise as it is possible to make them.
In the first part of the Gudgin/Breen paper, it is shown that the relative stability of the Catholic:Protestant unemployment ratio is difficult to achieve under conditions in which aggregate unemployment fluctuates widely. This is important because it implies that one or more equilibrating mechanisms must be in operation. Moreover, it is difficult to see how discrimination in any form could operate as a mechanism which targets so precisely an unemployment ratio of between 2.0 and 2.5. It is because of this difficulty that we set up a labour market model to investigate alternative balancing mechanisms.
Murphy completely misses the importance of the stability of the
ratio. Indeed, in the UCD working paper series he goes so far
as to deny that the stability exists. In this commentary he claims
to have an alternative equation in which the Catholic:Protestant
unemployment ratio is independent of the aggregate level of unemployment
in Northern Ireland as a whole. The equation itself is not presented
in Murphy's commentary but is in fact the following:
Since the definition of includes aggregate unemployment (ur), Murphy is plainly wrong in asserting that the unemployment ratio is independent of the aggregate level of unemployment. In this formulation the unemployment ratio is made, independent of the aggregate unemployment rate only by making the Catholic unemployment rate a fixed proportion of the aggregate unemployment rate, ie by making the parameter a constant. Why one should wish to do this is unknown. One could, in principle, investigate how the unemployment ratio varies with changes in the proportionate markup of Catholic unemployment over aggregate unemployment, but this is close to a tautology. Murphy's equation also masks the fact that, in any fixed relation between the unemployment ratio and the proportionate mark-up, the Catholic share in jobs would have to adjust to offset any change in the overall unemployment rate.
Murphy's key criticisms in this volume are that two crucial assumptions used in the Gudgin/Breen model are 'arbitrary and implausible'. One of these assumptions concerns the treatment of migration in the model. The second concerns the treatment of Catholic disadvantage in getting jobs. (Murphy is once again wrong in describing this as an assumption but we return to this below).
On migration, Murphy describes our equation as 'implausible and without empirical foundation'. The equation is in fact based on an empirically estimated migration equation for total migration (Catholics and Protestants together) used in the NIERC econometric model of the Northern Ireland economy. Like most other regional migration equations for the UK or Ireland it estimates migration as a function of the difference in unemployment rates, in this case between Northern Ireland and Great Britain, and also a number of other factors mostly unrelated to unemployment (including the impact of the Troubles). The general form of this empirically estimated equation is preserved in the Gudgin/Breen model. However, the model requires two separate migration equations, one for Catholics and one for Protestants. Although Murphy does not make it clear, it is in the specification of the two separate equations that controversy arises.
The problem is that two parameters must be specified for each equation, but we have only one piece of information with which to differentiate between Catholic and Protestant migration. This information is the total number of Catholic migrants and the total number of Protestant migrants over the 1971-91 period. Since the Catholic migration rate was only 50% higher than Protestant migration rate over the period, while Catholic unemployment was twice as high, one clear possibility is that Catholic migration is less responsive to unemployment than is Protestant migration. The other possibility is that Catholic migration is no less responsive than Protestant migration to unemployment, but that the higher level of migration resulting from higher Catholic unemployment is offset by other Catholic migration streams which. are independent of unemployment. For example, fewer Catholics than Protestants may have migrated due to the Troubles or for social reasons such as marriage.
We have investigated the impact of both possibilities. The results were reported in appendix 5, and are unaccountably not referred to by Murphy. The results show that alternative specifications of the migration equation do not greatly alter the conclusions. In his UCD Working Paper, Murphy himself admits that alternative migration specifications do not greatly alter the results, but this point is not included in his comments in this volume. Murphy's point on the migration equation was first raised by Professor Nickell at the Malone House seminar at which the Gudgin/Breen paper was launched. After seeing the results of alternative specifications given in appendix 5, Professor Nickell has decided not to comment further. Murphy nowhere entertains the possibility that Catholic migration could be less responsive to unemployment than is the case for Protestants. Even without compelling evidence this is surely a possibility to be set alongside discrimination as a potential cause of high Catholic unemployment.
Murphy goes on further to question the view that high Catholic unemployment results from a higher fertility rate combined with an emigration rate which is relatively low. He also questions whether if Catholic migration were higher, Catholic unemployment would be lower. Murphy asserts that these views are not well grounded in research. However, given the extremely high natural increase in the Catholic labour force, unemployment (or alternative forms of inactivity) can only be avoided either by migration, by an equally rapid increase in total jobs, or by a rising Catholic share of jobs (ie a falling Protestant share). High Catholic birth rates, unmatched by similarly high migration, will always cause high Catholic unemployment unless Catholic employment increases to whatever level is demanded by the growth of the Catholic labour force. This might happen if wages were fully flexible and Catholic wages fell to whatever level was consistent with full employment. However, this is hardly of any relevance to the actual wage-setting process in Northern Ireland. Since employment growth has never kept pace with population increase in the Catholic areas of Northern Ireland, some of which have among the best records of job creation in the UK, it is obvious that aggregate unemployment will be high unless offset by out-migration. This in turn implies that Catholic unemployment will be high, unless Catholics are able to displace Protestants from jobs. If Murphy really takes the view espoused in his commentary, it would be interesting to know why he thinks unemployment is so high in the Republic of Ireland despite its good record of economic growth, and why southern unemployment rises when net out-migration falls.
The second 'crucial' assumption claimed by Murphy in the Gudgin/Breen model concerns the 'assumption' that Catholics will have a success rate in applying for jobs which is only 70% as high as that of Protestants. This is completely incorrect. The model contains no assumption whatsoever on Catholic success rates. Indeed, a major purpose of the model is to calculate what the Catholic success rate must be in order to attain the observed Catholic:Protestant unemployment ratio for 1991. With observed natural increase, migration and employment changes, the cakulated answer is that the Catholic success rate is 70% of that for Protestants. This is clearly a calculation and we are at a loss to know why Murphy persists in misleadingly referring to it as an assumption. This calculation of 70% success is identical to Murphy's own calculation using Labour Force Surveys for 1986-91 (Murphy and Armstrong, 1994). One might have thought that this confirmation of Murphy's earlier result, in this case using a simulation model, would have increased his confidence in our methodology, but this appears not to be the case.
Despite finding agreement with his own result, Murphy asserts that our own results include double counting. This, he asserts, is because his own results from cross-sectional logit analyses include the impact of natural increase, job turnover and migration "to some extent". His claim is disingenuous to say the least. Part of the claim is that by including an age variable in a cross-sectional logit analysis one has taken account of the impact on unemployment of natural increase over the preceding two decades. This claim is completely erroneous as can be easily shown. The impact of an age variable in a logit equation explaining unemployment depends completely on the association between age and unemployment (eg do younger people have a higher probability of being unemployed?). This association need have nothing whatever to do with natural increase, and hence it cannot be claimed that the inclusion of an age variable measures the impact of natural increase.
Murphy continues by making a distinction between average and marginal influences on unemployment, a point he illustrates by an analogy with cold water going into a hot bath. Although this distinction is correct it has no relevance to our model which makes almost no use of parameters based on cross-sectional econometric analysis. Once again Murphy misunderstands our approach. As can be seen, we view most of Murphy's points as contentious and unhelpful, but not damaging to our arguments. The key point and, in our view, the only one really worth debating concerns the interpretation of the (agreed) finding that Catholic success in applying for jobs is only 70% of Protestant success.
Our approach has been to isolate this single factor and to run the model with only this factor. The resulting calculation was that a 70% success rate in obtaining jobs was associated with a Catholic unemployment rate 50% above that of Protestants. This 50% figure accords with that found by Murphy and Armstrong (1994), Smith and Chambers (1991), and Gudgin (1994) as a measure of structural factors alone. It is thus reasonable to deduce that the 70% Catholic success rate is due to structural factors. We were careful in the paper to say that this does not mean that no discrimination occurs in Northern Ireland, but merely that alternative explanations are available to explain why Catholic unemployment is persistently above that for Protestants.
Murphy finds this deduction unacceptable for two reasons. Firstly, his own econometric work with Armstrong for 1986-91 indicated that only one-third of the lower Catholic success rate in getting jobs could be accounted for by structural factors. The remainder could not be explained in this way and remains unexplained. Murphy argues that only the structural (explained) part of Catholic disadvantage in hiring should be used in any simulation model. He argues that a 90% Catholic success rate in applying for jobs should be used instead of the 70% observed rate. This is once again technically incorrect. Since we make no assumption about Catholic disadvantage, it is not possible to any particular rate. To repeat an earlier point, the disadvantage rate is calculated and its value is determined by the data. Murphy's point on Catholic disadvantage affects the interpretation of the model, not the model itself.
For the sake of argument we can however follow Murphy's logic. Murphy inserts a 90% success rate into a model and unsurprisingly obtains a low unemployment ratio. His result cannot however be reproduced with our model, and has presumably been produced by his own different, but unspecified, version of our model. Until we know what his model is we are unable to comment further. What can easily be done using our own model is to use a 90% Catholic disadvantage factor alone, ie independently of any other influence. The resulting Catholic:Protestant unemployment ratio is 1.10. This is much lower than the ratio of 1.5 obtained with a Catholic disadvantage of 70%. The following question, not posed by Murphy, then needs to be answered. Why does a Catholic success-rate of 90% relative to Protestants, caused by structural factors alone, not produce an unemployment ratio of 1.5 as Murphy, Smith and Chambers and others assert it should? If Murphy has insufficient confidence in our model, he should develop his own to overcome this obvious inconsistency in his logic. His attempt to claim that only his own work is correct is something which requires more than mere assertion.
We intend to pursue further research ourselves to attempt to resolve this issue. One approach is to simulate annual changes over the 1971-91 period, instead of using the period averages as in this paper. Preliminary work indicates that with annual data, Catholic success rate in job applications above 70% is consistent with the observed unemployment ratios. This would close some of the gap with Murphy's economic results on Catholic success rates.
Another approach is to think more carefully about the usefulness of the kind of cross-sectional analyses carried out by Murphy, and Smith and Chambers. Unemployment rates are determined by the magnitude of various flows of people in and out of the labour market and between jobs and unemployment. Discussions of Fair Employment should, therefore, be concerned with differences between the two communities in these flows and the reasons for them. This is the approach we have taken in developing our model. Cross-sectional analyses, by contrast, cannot shed any light on these underlying flows. This is because, as Murphy himself points out, there is no necessary relationship between the marginal or dynamic parameters and the average, or cross-sectional, parameters. In practice this implies that the impact of a particular factor on the chance of being unemployed at a given point in time (as measured in a cross-sectional analysis) may be quite different from its effect on the chances of becoming unemployed. Furthermore, precisely because such analyses are confined to a single point in time they cannot take into account dynamic factors such as natural increase and migration. One consequence of this is that in cross-sectional analyses the measurement of the effects of structural factors are likely to be conflated with the influence of these (necessarily) omitted dynamic factors. In addition, the direction of causality is unclear in cross-sectional analyses. It is unclear for instance whether location is exogenous and can thus be considered to be a structural influence.
Murphy concludes by suggesting that 'many readers of the paper will be tempted to draw policy conclusions'. We ourselves hope that this will be so - though the main policy conclusion we would wish to be drawn is somewhat different from the rather offensive (particularly to the Catholic population) ones to which Murphy points. It is simply that, to refer to the title of our paper, the ratio of unemployment rates is indeed not a useful indicator of Fair Employment in Northern Ireland. Murphy attempts to widen the issues by attributing to us policy recommendations including a reduction of the Catholic birth-rate. We nowhere attempt to do this, but since Murphy raises this issue he may wish to reflect on the fact that the birth-rate in the Republic of Ireland has halved since 1971 and is now close to the European average. As this feeds through to the labour market unemployment in the Republic is widely expected to converge to the European average unemployment rate in the early years of the next decade.
In contrast to Murphy's comments, we find ourselves in agreement
with almost all of Rowthorn's points. We note in particular that,
like ourselves, Rowthorn views it reasonable to interpret our
findings on Catholic disadvantage as likely to indicate structural
influences alone. The only comment which may be worth making is
on Rowthorn's final implication that discrimination and chill
factors in engineering and shipbuilding industries would have
affected the unemployment ratio in the early 1970s. This may be
less obvious than it seems. The well known engineering firms which
employed few Catholics were in sharp decline during the 1970s,
and it is not obvious that any change in recruitment practice
would have had much impact on Catholic employment or unemployment.
At Shorts, where strenuous efforts have been made to recruit Catholics,
including setting up a new factory in West Belfast and relocating
their Training Centre, the Catholic workforce remains small and
will have achieved only a small impact on the balance of unemployment.