Research Report No 2
Northern Ireland Economic Research Centre
This chapter investigates the relationship between religion and unemployment amongst men aged between 20 and 59. It was shown in Chapter 3 that according to the LFS and CHS samples, Catholic men were approximately two and a half times as likely to be unemployed as non-Catholic men. This chapter investigates the extent to which this unemployment differential can be explained in terms of factors such as age structure, geographical location, qualifications, etc. For example, it could be argued that Catholics are more likely to be unemployed than non-Catholics because they are more likely to live in the West of the Province where, for a variety of reasons, unemployment is high. Likewise, it could be argued that Catholics are more likely to be unemployed than non-Catholics because the Catholic population tends to be younger than the non-Catholic population and, for a variety of reasons, younger people are more likely to be unemployed than older people. In this chapter, therefore, we assess how strong the relationship is between religion and unemployment after a range of relevant factors have been taken into account.
Labour Force Survey (LFS) and Continuous Household Survey (CHS)
data are used to model the incidence of unemployment by religion
for men aged 20 to 59 in Northern Ireland. We present a range
of estimates of the effect of religion on the incidence of unemployment.
The estimates imply that religion directly accounts for about
half of the difference in unemployment rates between Catholics
and others. 'Structural factors', ie differences in the personal
and other measured characteristics of the two groups, account
for the rest of the difference in unemployment rates. These findings
are robust and imply that, ceteris paribus, Catholic men are significantly
more likely to be unemployed. We discuss the interpretation of
these findings at length.
In order to measure the contribution of religion and other factors to the incidence of unemployment, a number of quite sophisticated statistical (econometric) models have been estimated. These models estimate the effects of religion and a range of other factors on unemployment simultaneously. The particular type of statistical models which are used are called logit and probit models and they are described in detail in Appendix 4.2.
The models include a range of factors, other than religion, which have been shown in other studies to have an important effect on unemployment. The factors, which are included are as follows:
In terms of the present study it is important to not e that these
factors affect unemployment and many of them are correlated
with religion. For example, it is a standard finding that, all
other things being equal, people with large families are more
likely to be unemployed than people with small families. This
is because the amount of unemployment-related benefit to which
a person is entitled increases with family size. This means that
the income that some people can get when they are unemployed is
similar or even greater than that which they could get if they
were in a job, and this can be attributed, at least in part, to
family size. For example, more than one-fifth of Catholics have
three or more children, compared to slightly more than one-tenth
of non-Catholics. Therefore, it might be the case that part of
the difference in unemployment rates between Catholics and non-Catholics
could be attributed to the fact that Catholic families tend to
be larger and people with large families are more likely to be
unemployed. The correlation between all of these factors and religion
is shown in Appendix 4.5.
Unemployment and Religion in the LFS
Table 4.1 shows the incidence of unemployment by age group in
the LFS sample. The unemployment differential is 2.5 in our sample.
This means that, in aggregate Catholics are two and a half times
more likely to be unemployed than non-Catholics.
UNEMPLOYMENT RATES BY AGE GROUP
In Table 4.2, variables with positive coefficients increase the probability of unemployment and variables with negative coefficients reduce the probability of unemployment. For example, in Table 4.2 the variable called "Married" has a negative coefficient and this means that married men are less likely to be unemployed than non- married men. The coefficients are generally significant (t statistics of 2 or higher) and correctly signed according to our prior expectations. The effects of all the other variables on unemployment are discussed in detail in Appendix 4.5. However, the main point of interest is that these results show that Catholics are significantly more likely to be unemployed even after taking into account the factors listed in Table 4.2.
This, however, has still not answered the question of how much more likely Catholics are to be unemployed compared to non-Catholics after accounting for the other relevant factors. The results of the econometric models, like those presented in Table 4.2, can be used to derive a number of different estimates of this. These estimates are described and discussed in Appendix 4.5. A priori, there is no particular reason for using one measure instead of another. For this reason all three measures have been estimated and, reassuringly, each of them gives a similar result.
Overall, when the different estimates are compared, it is the
case that approximately one-half of the observed unemployment
differential can be explained in terms of religion and the other
half can be explained in terms of other factors in the model such
as geographical location, the age structure of the population
etc. It should be noted that this is not the same as saying that
half of the unemployment differential can be explained in terms
of Northern Ireland employers discriminating against Catholics.
A range of interpretations of the 'religion effect' can be given
and these are discussed in detail in the final section of this
INCIDENCE OF UNEMPLOYMENT
ESTIMATED COEFFICIENTS AND MARGINAL EFFECTS (SLOPES)
LFS SAMPLE (N = 9940)
Log Likelihood LL = -3396.91
Unemployment and Religion in the CHS
All of the analysis reported in the previous section is based
on LFS data. This section presents the results of a similar analysis
which was conducted using CHS data. Table 4.3 shows the unemployment
rates for Catholics and others in different age groups in the
CHS sample. The figures show that the unemployment differential
does not vary greatly by age group.
As in the analysis of the LFS, one of the main purposes of the
present analysis is to estimate how much more likely Catholics
are to be unemployed compared to non- Catholics after a range
of relevant factors have been taken into account. To do this probit
models have been estimated for the incidence of unemployment using
a similar set of explanatory variables to those used to model
the incidence of unemployment in the LFS sample. All of the explanatory
variables, except one, are defined in the same way in the two
INCIDENCE OF UNEMPLOYMENT
PROBIT RESULTS - BASIC MODEL
CHS SAMPLE (N = 5805)
The results of the basic probit model for the incidence of unemployment are given in Table 4.4. These are very similar to the results obtained using our LFS data which is very reassuring. As in the analysis of the LFS, we have used these results to derive an estimate of how much of the unemployment differential can, be explained directly in terms of religion and how much can be explained in terms of the other observed factors. Again, a number of different estimates of the size of the religion effect have been derived. These are summarised in Appendix 4.6, which also contains the results from some more sophisticated econometric models.
When we exclude socioeconomic group (SEG) from our models, our results suggest that between 55% to 60% of the observed unemployment differential is directly accounted for by religion. The remainder is accounted for by observed differences in the characteristics of Catholics and others. These results, therefore, are broadly similar to the results obtained with the LFS sample. In our LFS sample, we cannot readily include SEG in our models because this information is not available for a large proportion of the unemployed. When we include SEG in our CHS models, our results suggest that between 47% to 54% of the observed unemployment differential is accounted for by religion.
In conclusion, therefore, our model results suggest 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 to be robust. Generally speaking, these results
provide support for Smith and Chambers' (1990) findings rather
than Compton's (1990) findings.
In our econometric models of the incidence of unemployment religion accounts for about half of the unemployment differential in our two samples; differences in the observed personal and other characteristics of the two groups account for the rest. However the large and significant Catholic effect on the incidence of unemployment must be interpreted carefully and, for example, cannot simply be equated with discrimination against Catholics. When interpreting the findings a number of factors have to be taken into account.
A large number of explanatory variables are included in our models. As outlined above, our choice of explanatory variables is based on a standard set of variables which have been used to model the incidence of unemployment in the applied econometrics literature. It is possible, however, that some important variables have been omitted from the models.
However, even if some relevant variables have been omitted from our models, the estimated Catholic effect is not necessarily larger than it should be. On the one hand, some of our explanatory variables such as housing tenure or the presence of other unemployed individuals in the household may be endogenous, ie simultaneously determined with unemployment. If this is the case, these variables will tend to reduce the estimated Catholic effect. On the other hand, our choice of explanatory variables is limited to those for which we have data. For example, there is no data on motivation and little or no data on subject mix at school or college in either of our datasets. If motivation or subject mix are correlated with being Catholic and contribute to the incidence to unemployment then their omission tends to increase the estimated religion effect.
The large and significant Catholic effect on the incidence of unemployment may be explained by factors which are not in our models because they are not measured in our data. These factors must be both correlated with religion and have a large effect on the incidence of unemployment. A number of such factors are commonly mentioned. These include differences in labour force growth, subject mix at school or college, motivation as well as direct or indirect discrimination and the 'chill factor.
Labour Force Growth
As we discussed in Chapter 2, differences in labour force growth between Catholics and others have both micro and macro effects. The micro effect has to do with the benefit trap and the age structure of the population. In our models of the incidence of unemployment we include age, the number of children, qualifications and so on. These variables are significant and it is likely that they are capturing the micro labour force growth effects. The macro story appears to have more to do with segregation in employment.
In order to examine the macro labour force growth effect we construct a simple model of the labour market in Appendix 2.1. In order to focus on the labour force growth issue, Catholics and others are assumed to have the same chances of getting and leaving jobs ie equal engagement and separation rates. However, it is assumed that the Catholic labour force grows at the rate of 1% per annum and that the Protestant labour force is constant. The rate of unemployment is assumed constant in the long- run. Using this model we do not find that labour force growth has a large effect on the unemployment differential.
We have also modified the assumptions, and we find that when we do this we can still not generate a large unemployment differential effect. In fact, some combination of low labour turnover, zero or negative employment growth, high Catholic labour force growth and segregation in employment, appears to be required to obtain large unemployment differential effects. Such assumptions appear to us to be implausible.
Differences between Catholics and others in the mix of subjects studied at school or college may be part of the explanation of the unemployment differential. However, there is very little evidence in Northern Ireland or elsewhere that differences in subject mix are a major explanation of the unemployment differential. Murphy and Shuttleworth (1994) do not find a subject mix effect for school leavers. Miller et al (1990) examine the relationship between degree subject and earnings for graduates. However, this evidence relates to a very small group in the labour force and to earnings rather than the incidence of unemployment. Also, Smith and Chambers (1990) note that the difference in unemployment rates between Catholic and Protestant men in their CHS data is highest for those with no qualifications. All of this would tend to suggest that little of the unemployment differential can be attributed to differences in subject mix. In support of this, according to our LFS data, the second largest difference in unemployment rates is for those with no qualifications; the largest difference is for the small category with 'other' qualifications. It is also the case that the difference in unemployment rates is highest for those aged 20 to 24, the age group with the smallest difference in subject mix. However, in these comparisons other factors are not being held constant. On the whole, therefore, there is little evidence that subject mix has a big effect on the incidence of unemployment.
There is little or no evidence of differences in motivation, flexibility or attitudes to work between Catholics and Protestants. The limited evidence in Miller (1978) and McWhirter (1984,1989) does not suggest that Catholics are less work orientated. The evidence from the LFS and the Northern Ireland Social Attitudes Survey suggests that Catholics are as flexible as Protestants in terms of the types of job they are looking for and would accept, retraining, moving and so on. In the absence of good Northern Ireland based studies with motivation data, such as the one analyzed by Gallie and Vogler (1990), we cannot say that there are large differences in motivation between Catholics and Protestants.
Discrimination and the 'Chill Factor'
These factors are hard to measure. However, suppose for the sake of argument we believe that some mix of direct or indirect discrimination or the 'chill factor' is important in explaining our results. With the data available, it is not possible to split this effect into direct discrimination, indirect discrimination or 'chill factor' components. In addition, ft is difficult to apportion these effects into current and past components. The reason is that past unemployment increases the risk of current unemployment. This is likely to be due to the loss of human capital or 'stigma' or 'scarring' effects. For example, a long-term unemployed man is far less likely to leave unemployment for a job than an otherwise similar recently unemployed man. This is the case, either because the long-term unemployed man has actually lost job skills or because employers perceive him as being less employable. Thus, the current incidence of unemployment depends, in part, on the past incidence of unemployment. As a result, it is rather difficult to disentangle the effect of past factors from ongoing factors.
However, We may still obtain some indication of whether these
factors operated only in the past or continue to operate more
recently. We obtain this evidence by examining the incidence of
unemployment for young males, say those aged 20 to 24, and by
examining flows to and from employment and unemployment. In both
cases, as we show in later chapters, we find significant negative
Catholic effects. This indicates that some current disadvantage
is present for Catholics, in addition to the 'structural' disadvantages
We construct a series of econometric models of the incidence of
unemployment by religion and carry out a range of statistical
tests to check the robustness of our results. Our models control
for a large number of relevant factors including age, number of
children, housing tenure, educational and other qualifications
and area of residence. We find that religion accounts for about
half of the unemployment differential in our two samples. Differences
in the personal and other characteristics of the Catholic and
non-Catholic populations account for the rest of the unemployment
differential. Ceteris paribus, Catholic men are significantly
more likely to be unemployed. These findings are robust and are
consistent with the results of Smith and Chambers (1991).
 One might argue that some of the explanatory variables may be endogenous, eg housing tenure and others employed/unemployed in the household. Individuals with a high probability of unemployment are more likely to be NIHE tenants rather than owner occupiers and to be in households with other unemployed members. These variables are likely to pick up part of any religion effect. We do not find that housing tenure is endogenous when we control for other relevant variables. In general, there is not too much that we can do about the endogeneity of the explanatory variables. In any case, they are standard explanatory variables which are widely used in the literature.
 The one exception is the variable which relates to health problems. In the CHS the health problem question refers to a long standing illness which lirni% activity. In the LFS the health problem question refers to a long standing illness which limits economic activity.
 We do not have to consider security-related or black economy jobs. The reason is that those employed in security-related jobs or engaged in the black economy are very unlikely to respond to household surveys such as the LFS or CHS.
 As Teague (1993) notes in relation to the link between qualifications and earnings, any observed relationship between subject mix and the incidence of unemployment may just be a screening or statistical discrimination' effect.
In the previous chapter we show that, ceteris paribus, Catholic
men are more likely to be unemployed than non-Catholic men. Are
they also more likely to be outside the labour force? In this
chapter we examine differences in economic inactivity or non-participation
between Catholic and non-Catholic men. Since we are dealing with
prime aged men, ie those aged 20 to 59, the vast majority of men
in our samples are in the labour force. The main reasons for economic
inactivity are long term or permanent illness or disability, full-time
study and discouragement. Discouragement, ie the belief that no,
presumably suitable, jobs are available is an important reason
for not actively searching for a job. We examine the incidence
of discouragement since the discouraged are often considered to
be disguised unemployed. We also consider the claimant status
of the economically inactive.
Differences Between LFS and CHS Economic Inactivity
In the LFS economic activity is more precisely defined than in
the CHS. For example, the unemployed must not only be looking
for work but (i) have looked for work in the past four weeks or
(ii) be waiting to start a job already obtained and (iii) be ready
to work in the next two weeks. In our OHS sample, the definition
of unemployment is broader. As a result we expect that some individuals
classified as inactive in the LFS would be classified as unemployed
in the CHS. We find higher economic inactivity rates and discouragement
rates in the LFS than in the CHS which supports our conjecture.
However, apart from these two differences, the results obtained
using the two sources are fairly similar.
Economic Inactivity In the IFS
Table 5.1 presents inactivity or non-participation rates by age group. The figures show that Catholic men are more likely to be economically inactive than non-Catholic men. In particular, Catholic men are on average, nearly twice as likely to be inactive than non-Catholic men.
The reasons for economic inactivity are shown in Table 5.2. The three main reasons are long term/permanent sickness and disability, discouragement and studying. The most striking difference between Catholic and other men is that Catholic men are significantly more likely to be discouraged than other men - the rates are 21.8% and 13.1% respectively. However, the significantly higher rate of Catholic discouragement only accounts for about one third of the overall difference in non-participation rates between Catholic and other men.
NON PARTICIPATION BY AGE GROUPS
REASONS FOR ECONOMIC INACTIVITY
WHETHER ECONOMICALLY INACTIVE WOULD LIKE A JOB
CLAIMANT STATUS OF ECONOMICALLY INACTIVE
Table 5.3 shows that, despite the higher level of discouragement,
significantly more inactive Catholic men would like a job. In
Table 5.4 significantly more inactive Catholic men claim unemployment-related
benefits. However, this is almost wholly explained by the higher
level of Catholic discouragement.
Economic Inactivity in the CHS Sample
Table 5.5 shows the rates of economic inactivity or non-participation
in our CHS sample. As in the LFS, the Catholic rate of non-participation
is significantly higher overall and by age group. The rates of
non-participation for both Catholics and others are lower in the
CHS data than in the LFS data. As indicated above, this is because
a stricter definition of unemployment is used in the LFS.
The composition of the economically inactive is shown in Table
5.6. Those who are permanently unable to work make up over 60%
of the total. Fewer Catholic men are studying or have retired
early whilst more Catholic men are in the residual 'other' category.
COMPOSITION OF ECONOMICALLY INACTIVE
For those interviewed face to face, we know the reasons they are
not looking for a job; these are set out in Table 5.7. The most
striking difference between Catholic and other men is, again,
in the rate of discouragement. The Catholic rate of discouragement
is 12.5% as opposed to 5.6% for other men. Table 5.2, based on
the LFS, and Table 5.7, based on the OHS, are not directly comparable
since different definitions of unemployment and inactivity are
used in the two surveys. In addition, the possible responses in
the two surveys to the question of why someone is not looking
for a job are different. As a result the large differences in
discouragement rates between the LFS and OHS, eg 21.8% versus
12.5% for Catholic men, are not too surprising.
For a subsample of the inactive, we know whether they would like
a job or not. As Table 5.8 shows, more inactive Catholic men would
like a job. The sample numbers are small but we obtain a similar
result in our LFS sample.
Table 5.9 shows the claimant status of the unemployed and inactive.
Significantly more inactive Catholic men claim unemployment related
benefits. However, this higher claimant rate is largely accounted
for by the higher rate of discouragement amongst Catholics. This
is also found in our LFS sample.
REASONS ECONOMICALLY INACTIVE NOT LOOKING FOR A JOB
CHS SAMPLE (excluding proxies)
ECONOMICALLY INACTIVE WHO WOULD LIKE A JOB
CHS SAMPLE (excluding proxies)
PERCENTAGE CLAIMING UNEMPLOYMENT RELATED BENEFITS
CHS SAMPLE (excluding proxies)
Econometric Analysis of Inactivity and Discouragement
In our LFS and OHS samples there are important differences in
the rates of economic inactivity and discouragement between Catholic
and other men. In particular, Catholic men are significantly more
likely than others to be both economically inactive and discouraged.
We have estimated a range of econometric models of the incidence
of inactivity and discouragement. Probit model results based on
the LFS and OHS are reported in Appendices 5.1 and 5.2 respectively.
The probit results show that there are still important differences
in economic inactivity rates between Catholics and others, even
after we control for a range of factors such as geographical location,
health status and qualifications. For example, the models which
were estimated using LFS data suggest that between 40% and 60%
of the diffference in economic inactivity rates is accounted for
by religion. The remainder is accounted for by the other factors
included in the models.
In the econometric models, the effect of religion on the incidence
of discouragement is smaller. For example, the models based on
LFS data suggest that only between one-quarter and one-third of
the religious difference in discouragement rates is accounted
for by religion; the rest is accounted for by differences in the
other observed characteristics of Catholic and other men. When
the incidence of discouragement is modelled using OHS data, we
obtain a statistically insignificant religion effect.
Catholic men are significantly more likely to be economically inactive than non-Catholic men. In our econometric models half of the difference in inactivity rates between Catholics and others is explained by religion; the remainder is explained by differences in other characteristics.
In the raw data significantly more inactive Catholics are discouraged
ie they are not actively looking for work because they believe
that there are no, presumably suitable, jobs available. However,
when we control for a range of relevant factors, we do not find
a large religion effect on the incidence of discouragement. When
we model the incidence of discouragement a significant, albeit
small, religion effect is obtained with the LFS data but not with
the OHS data.
More inactive Catholics claim benefits. However, the higher Catholic rate of claiming is largely accounted for by their higher rate of discouragement.
This chapter focuses on differences in the duration of unemployment
between Catholic and other men. It is important to do this because
whether or not someone is unemployed depends on two factors, namely
the probability of entering unemployment and how long the person
remains unemployed (ie unemployment duration). We look at the
raw data and estimate a range of econometric models in order to
disentangle the effects of religion and the other explanatory
variables which are all correlated with, each other.
The Duration of Unemployment in the LFS Sample
Table 6.1 shows the duration of unemployment in our LFS sample.
Catholics are significantly less likely to be unemployed for less
than a year and significantly more likely to be unemployed for
four years or more. These figures suggest, therefore, that Catholics
are more likely to be long-term unemployed than non-Catholics
and less likely to be short-term unemployed.
DURATION OF UNEMPLOYMENT
DURATION OF UNEMPLOYMENT GROUPED BY YEAR
The duration of unemployment by age group is shown in Table 6.3.
The numbers in some of the cells are small. However the difference
in the duration of unemployment between Catholics and others aged
20 to 24 is striking.
Table 6.4 shows that the activity of Catholics and non-Catholics
before becoming unemployed is very similar. In the LFS we only
know the reasons the unemployed left their last job if they left
in the past three years. For these individuals Table 6.5 shows
that Catholics are significantly more likely to have their last
job because a temporary job ended.
DURATION OF UNEMPLOYMENT BY AGE GROUP
ACTIVITY BEFORE BECOMING UNEMPLOYED
REASONS FOR LEAVING LAST JOB
IFS Sample of Those Who Had a Job Which They Left in the Past 3 Years
The duration of unemployment in the OHS sample is shown in Table
6.6. The figures show that non-Catholics are significantly
more likely to be in 0-6 months duration band. The figures also
show that there is quite a large difference in the proportions
of Catholics and others in the five years plus duration band.
The differences in the duration of unemployment between Catholics
and others are much smaller in our OHS sample than in our larger
LFS sample. This is because a stricter definition of unemployment
is used in the LFS and so some of the CHS unemployed would be
classified as being economically inactive in the LFS.
DURATION OF UNEMPLOYMENT
The duration of unemployment by age group is shown in Table 6.7.
The numbers in some of the cells are very small. However, generally
speaking the CHS figures, like the LFS figures show that Catholic
men are more likely to be long-term unemployed than non-Catholic
men and less likely to be short-term unemployed. It should be
noted, however, that in our econometric analysis of the CHS data
the effect of religion on unemployment duration is found to be
statistically insignificant. This is discussed below. Finally,
it should be noted that a couple of unusual results are revealed
in Table 6.7, eg the relative proportions of Catholics and others
aged 45 to 54 in two to five years duration band. These results
are probably the result of random sampling variability.
DURATION OF UNEMPLOYMENT BY AGE GROUP
We have estimated a number of econometric models of unemployment
duration; these models are discussed in detail in Appendices 6.1-6.4.
Some of these models are relatively simple, eg probit models for
particular durations, and some of them are quite complex, eg hazard
rate models. Using the LFS the model results generally show that
even when a range of relevant factors are controlled for, Catholics
are still more likely to be long-term unemployed than non-Catholics
and less likely to be short-term unemployed. It is also the case
that high local unemployment rates, living in Belfast DC area,
a large number of children, a health problem, claiming benefit,
others unemployed in the household all increase the probability
of longer durations.
When the CHS is used to model unemployment duration we do not
find any significant religion effects. This is highly surprising
and implausible. If true it implies that, ceteris paribus, differences
in unemployment rates between Catholic and other men are solely
due to differences in entry rates into unemployment. There is
no incidence for this in the larger LFS sample or from other data
sources such as the Census and the Social Attitudes Surveys. Examination
of the data suggests that sampling variability is the reason for
these strange results. Certainly there appear to be significant
religion effects in the OHS duration data for the years 1983 to
Using LES duration data we find significant religion effects.
Catholic men are significantly more likely to be long term unemployed
than other men, ceteris paribus. Also, their exit rate from unemployment
is significantly lower. These effects are found both in the raw
data and in the models which control for a range of factors. With
CHS duration data no significant religion effects are found either
in the raw data or in our models. This is highly surprising and
implausible and we suspect that it can be explained in terms of
 When the CHS data are disaggregated by year, the 1988 duration data appear to be rather different from the data for the other three years in our sample.