Race or religion? The impact of religion on the employment and earnings of Britain's ethnic communities.

attempt to measure each separately. Here the religion

categories are

aggregated regardless of ethnicity. For example, the

`Muslim' group contains

Pakistani, Bangladeshi, Indian and other Muslims. Also

included are

`foreign-born' and `British-born' non-white variables.

These variables measure

any ethnic disadvantage and indicate the importance of

race over and above

religion. The insignificance of the British-born

non-white variable might be

thought to indicate that the labour market situation

of second-generation

non-white migrants has fully assimilated to that of

British-born whites, on

average (following Heath and McMahon 1997, and Leslie

et al. 1998). The

insignificance of the foreign-born variable would

indicate no differences

between first-generation non-white migrants and

whites. Both modes are

corrected for employment selectivity (see Heckman

1979). In subsequent

discussion and tables, these are referred to as Model

1 and Model 2.

Descriptive statistics

Table 1 provides the ethnic composition of the various

religions in the

sample. One can see that Muslims mainly consist of two

groups, Pakistanis and

Bangladeshis, with a smaller percentage of Indians and

African Asians. (11) As

a result, Muslims can be disaggregated into Pakistani,

Bangladeshi, Indian and

other Muslims; this last category consists mainly of

African Asians and a very

small percentage of whites. Hindus can also roughly be

divided into two

groups. These are Indians and African Asians, with the

latter group containing

a small number of others. Sikhs are mainly Indians and

are therefore not

disaggregated further. Christians can be roughly split

into two groups: white

and mainly Caribbean. The `other religion' category

consists of Buddhist,

Jain, Rastafarian, Jewish, Parsi/Zorashia and others.

The ethnicity of this

`other' religious group is too diverse to disaggregate

further. Finally those

with `no' religion are about 60 per cent whites, 20

per cent Caribbean and the

rest are Chinese or other. It is noteworthy that more

whites have no religious

views compared with the other ethnic groups.

Table 2 presents some descriptive statistics

concerning the religious

affiliation of the sample by labour market status and

sex. The sample sizes

refer to everyone over the age of 16 and under the

statutory retirement age

and are unweighted. (12) The second column in Table 2

shows the percentage of

each religious group employed. For both men and women,

Muslims have a much

lower percentage employed than the other groups,

although there are

significant differences within the Muslim category.

The percentages of

Pakistani and Bangladeshi Muslims employed are very

low in comparison to all

the other groups. Also Hindu Indians appear to have a

much higher percentage

of employed than Indian Sikhs, Indian Muslims and

African Asian Muslims. This

is evidence that religion rather than race is the

important discriminator. In

fact, Indian Hindu males have a greater percentage of

employed than all the

other non-white religious groups.

The third column shows the percentage of the sample

who are unemployed. As

might be expected, these are much higher for males

than for females. Hindu and

white males show the lowest unemployment rates. The

fourth column refers to

the percentages of the sample who are `out of

employment'. These are

particularly high for Muslims and especially for

women. Interestingly, white

Christians appear to do better on this measure than

those with no religion.

The final column provides average weekly earnings for

those who are employed.

In general, Muslims earn considerably less than the

other groups. Amongst the

Muslims, Bangladeshis demonstrate the lowest average

earnings, whilst

Pakistani males and other Muslim females earn the

most. Again there are

significant differences in the earnings of Indians,

since Hindus and Sikhs do

better than Muslims. For males, those with `other'

religion earn the most, on

average, followed by white Christians, whites with no

religion and Indian

Hindus.

For females, those with some other religion earn the

most, whilst Bangladeshi

women earn the least. Interestingly both Christian and

non-religious non-white

women earn more, on average, than their white

counterparts. The reverse is

true for males. This provides some evidence of a

matriarchal culture, amongst

Caribbeans, where the female adopts a more dominant

economic role; see

Berthoud and Beishon (1997) for further discussion on

this. In addition, this

demonstrates that ethnic differences exist within

religious categories.

The results

[Graphic omitted]Table 3 shows key results from

employment/unemployment

probits for males and females. (13) These include a

standard set of

independent variables measuring human capital and

other socio-economic

characteristics as well as religion. (14) The first

two columns refer to the

first model, whilst the second two columns refer to

the second model. Model 1

includes religion with the ethnically disaggregated

categories, whilst Model 2

includes religion and ethnicity variables. All results

are relative to whites

with no religion.

Looking at Model 1 in the first two columns, religion

is an important

determinant of employment for both men and women. (15)

The most notable

differences are between the Indian religions, whilst

the most disadvantaged

group appears to be Muslims. Sikhs are the only Indian

males to display a

reduced propensity for employment, relative to those

with no religion. Amongst

the male Muslims, only Pakistani males exhibit a lower

propensity for

employment. This confirms the findings of Brown

(2000). Interestingly males of

some other religion are less likely to be employed and

white Christian males

are more likely to be employed, relative to whites

with no religion. This of

course may in part reflect some unmeasurable

socio-economic factors. A similar

story holds for females, except that Bangladeshi

Muslim females are also less

likely to be employed.

The final two columns refer to Model 2. This contains

the aggregated religious

groups (regardless of ethnicity). Also included are

separate `foreign-born'

and `British-born' non-white variables. The

significance of these variables

would indicate an ethnic penalty over and above

religion, where the reference

category is non-religious whites. It should be

remembered that there are a

large number of whites in the Christian category and

hence there is a

comparison to be made here also.

All Muslims and male Sikhs display significantly lower

propensities for

employment, whereas Christian males demonstrate higher

propensities for

employment, relative to those who are not religious.

This further demonstrates

the importance of religion. For both males and

females, being Muslim implies

the biggest employment penalty. For non-whites, being

British-born non-white

(regardless of religion) also implies a lower

propensity to be employed.

Interestingly, foreign-born non-white males appear to

do better than

non-religious whites, although they do significantly

worse than Christians

(who have a high percentage of whites). (16) For

females, the Muslim religion

is important, although ethnicity is not. There is no

significant employment

penalty to non-fluent non-whites, over and above this

Muslim effect. (17)

In the light of Table 3, one might want to measure the

religious and the

ethnic penalties. Since Table 3 shows the Hindu

variable to be insignificant,

then the religious focus here is on Muslim and Sikh

penalties. To do this, a

counterfactual analysis can be undertaken that

predicts what average

employment rates would be if there were no Muslim

penalty or no Sikh penalty

to non-whites on the basis of the coefficients from

Model 2 in Table 3. These

predicted average probabilities can then be compared

with the actual average

employment rates. This gives a measure of the overall

Muslim and Sikh

penalties to non-whites. In addition to this, average

employment rates can be

predicted by imposing the additional restriction that

there is no ethnic

penalty to employment. This roughly isolates the

Muslim, Sikh and ethnic

penalties, so that the remainder gives the difference

accounted for by other

characteristics. The relative magnitudes of the Muslim

effect, the Sikh

effect, the ethnic effect and the characteristic

effect can then be compared.

Table 4 presents these comparisons. Each cell contains

two figures. The first

refers to the employment/unemployment dichotomy of

Table 3, whilst the second

(in square brackets) refers to an employment/out of

employment dichotomy. The

first row shows the total difference in unemployment

rates, so for example

whites enjoy a 21.5 per cent lower unemployment rate

based on the average of

the predictions from Model 2 in Table 3. For males,

non-white unemployment

falls by 5.1 per cent should all in the sample be

non-Muslim, whilst it falls

a further 1 per cent should all be non-Sikh. The

fourth row shows that the

male non-white employment would fall by 11.0 per cent

if all in the sample

were white. Hence the ethnic penalty (at 50 per cent

of the total

differential) is far greater than the religious

penalties (30 per cent

combined) and indeed that explained by differences in

other characteristics

(20 per cent). Here other characteristics include

English language fluency. So

there exists a relatively large racial discrimination

or non-assimilation

effect over and above controlled characteristics, even

when religious

association is included within the analysis.

Looking at females, there is a noticeably smaller

white/non-white unemployment

differential than that for males. The religious

penalties are quite similar to

those for males (at 34 per cent of the total

differential), although the

ethnic penalty is also noticeably smaller (at 17 per

cent). Most of the female

non-white unemployment penalty can be attributed to

differences in

characteristics (49 per cent).

A further interesting picture is provided by the

female `out of employment'

differential. Unlike those for males these provide

quite different results.

The `out of employment' differential is much larger

for females than is the

unemployment differential. Second there is no ethnic

penalty over and above

religious effects. In fact there is a penalty to white

women of 21 per cent of

the total white/non-white differential. After

differences in other

characteristics (which account for 69 per cent), the

only other penalties are

explained by differences in religion. This is

especially high for Muslim women

(at 43 per cent). This provides evidence to support

increased female

non-participation as a direct result of Islamic faith,

relative to the other

religious groups.

Table 5 shows the key results for employment

selectivity adjusted earnings,

(18) Again all results are relative to whites with no

religion and the first

two columns refer to Model 1, whilst the second two

columns refer to Model 2.

Looking at the first model, one can see that there are

significant differences

between Indian males, with only Sikhs demonstrating a

significant earnings

penalty. (19) All Muslims display an earnings penalty,

with Bangladeshis

experiencing the highest, followed by other Muslims

(mainly Indian and

African) and Pakistani Muslims. The second column

refers to female earnings.

The only females who experience an earnings penalty

are those who are not

fluent in English.

The final two columns refer to Model 2. Only Sikh and

Muslim males and white

Christian women exhibit significantly lower earnings,

whilst males with

another religion exhibit higher earnings, relative to

those with no religion.

For males, there is a significant earnings penalty to

immigrant non-whites

regardless of their religion. This provides evidence

of a foreign-born ethnic

disadvantage (but not a British-born one) which is

over and above religion and

English language fluency. (20) Of course this could

suggest that British-born

non-white earnings are assimilating to those for

whites. For females there is

an earnings premium to British-born non-white females,

relative to white

non-religious females. This further supports the

assimilation of earnings.

Only white Christian females and those non-fluent in

English experience

significantly lower earnings, relative to those with

no religion.

The final row in Table 5 shows the correlation

coefficients between the error

terms of the employment selection equation and those

of the earnings equation.

These show that the correlation coefficients are

insignificant for males and

significantly negative for females. For males this

suggests that correcting

for selectivity has only a marginal effect on the

estimated parameters in

Table 5. For females, earnings would be higher for

those who are unemployed,

should they gain employment, relative to those who are

already in jobs.

Blackaby et al. (1999) suggest that this occurs since

the unemployed have

higher reservation wages than the employed and in turn

would require a greater

reward to enter into employment.

Islamic disadvantage

Clearly Tables 3 and 4 provide little evidence of

ethnic assimilation in

unemployment rates. Furthermore, Muslims are the most

disadvantaged of all the

groups. It would be interesting to know how much of

this disadvantage is a

direct result of differences in characteristics of the

group and how much is

the result of other factors such as anti-Islamic

discrimination and

differences in the attitudes and aspirations of

Muslims, relative to other

ethnic religions.

By estimating a separate employment probit equation

for Muslim and non-Muslim

non-whites, the Gomulka and Stern (1990) method can be

used to decompose

unemployment differences into a characteristics

component and a coefficients

component. The analysis leads to two alternative

decompositions, which are

(1) [I.sup.NM] - [I.sup.M] =

[bar]P([[alpha].sup.NM][X.sup.M]) -

[bar]P([[alpha].sup.M][X.sup.M])] -

[[bar]P([[alpha].sup.NM][X.sup.NM] -

[bar]P([[alpha].sup.NM][X.sup.M])]

(2) [I.sup.NM] = [I.sup.M] =

[[bar]P([[alpha].sup.NM][X.sup.NM]) -

[bar]P([[alpha].sup.M][X.sup.NM])] -

[[bar]P([[alpha].sup.M][X.sup.NM]) -

[bar]P([[alpha].sup.M][X.sup.M])]

where NM refers to non-Muslim and M refers to Muslim,

with [[alpha].sup.M] and

[[alpha].sup.NM] the vectors of estimated coefficients

from the probit

equations, [I.sup.NM] and [I.sup.M] are the respective

predicted average of

the employment probabilities of group NM and group M.

(21)

[bar]P([[alpha].sup.M][X.sup.M]) is the average across

the sample of the

predicted probabilities using group M coefficients and

group M characteristics

and similarly for the other terms. The first term in

square brackets measures

the difference in means due to differences in

coefficients and the second term

the differences arising from differences in the

individual characteristics of

group M and NM. Equation (1) decomposes around average

group M characteristics

and equation (2) decomposes around average group NM

characteristics. Because

non-whites are split according to whether they are

Muslim or not and ethnic

controls are included as regressors, coefficient

differences can be identified

as the pure `Islam' effect. In this way it can be

investigated whether Muslims

are additionally penalised in having poorer average

characteristics, which

contribute to a lack of employability.

[Graphic omitted]Table 6 presents a summary of the

Gomulka-Stern

decompositions. The first set of figures in the

columns show the decomposition

for the employment/ unemployment dichotomy, whilst

those in square brackets

show the larger sample of those in employment or `out

of employment'.

The first row of Table 6 confirms that Muslim

non-whites are disadvantaged

relative to non-Muslim non-whites in terms of

unemployment rates. The female

Muslim unemployment penalty is everywhere greater than

that for males. One can

see that Muslim males display a greater unemployment

penalty, whereas Muslim

females demonstrate a greater out-of-employment

penalty. For both men and

women, there is a substantial component of the total

non-Muslim/ Muslim

differential that can be attributed to differences in

coefficients. However,

the coefficient (unexplainable) component is greater

for male unemployment and

female out-of-employment. This substantial

`unexplainable' component can be

attributed to being Muslim, although one should take

care in labelling this

entirely as `Muslim discrimination'. This component

may contain other cultural

and attitudinal differences, for example the Muslim

tradition whereby there is

no expectation that women should engage in market

economic activity. Moreover,

there is a strong characteristic effect and this is

especially large for

Muslim women. This includes such things as poor

English language fluency and

undervalued overseas qualifications. Such

significantly lower

employment-enhancing characteristics suggest labour

market assimilation

problems amongst Britain's Muslims.

Concluding comments

This study builds on Brown (2000) by including whites

in the sample for

comparative purposes. The results confirm those found

in Brown (2000) and have

shown significant differences within ethnic groups,

depending on religious

association. The most notable of these differences

applies to Britain's Indian

population. Hindu Indians appear to fare better in the

labour market than do

Sikhs. As a result, local changes in the labour market

will have different

implications for Indian communities, depending on

whether they are Sikh, Hindu

or Muslim. Second, this study has shown that Muslims

and Sikhs generally

experience greater labour market disadvantage

regardless of ethnicity. So it

would seem that religion is an important issue, as

well as ethnicity.

Unlike Brown (2000), this study controls for religion

and ethnicity

separately. The results demonstrate that religion

alone cannot fully explain

the difference between white and non-white

unemployment. There is evidence of

unemployment disadvantages to British-born non-whites

over and above

differences in religious affiliation. Moreover, the

results suggest that both

British- and foreign-born non-whites do worse than

most whites, in terms of

employment prospects. (22) For earnings, lack of

fluency and religion appear

to be the dominant factors (rather than ethnicity) in

terms of determining

average weekly earnings. Only foreign born non-white

males experience a

significant earnings penalty over and above religion.

Since there is no

earnings penalty to British born non-whites once

religion has been controlled

for, this provides some evidence of non-white earnings

assimilation towards

those of whites.

[Graphic omitted]Gender differences are also apparent.

Unemployment

disadvantage is more important for ethnic minority

men, whereas non-white

women exhibit a greater propensity to be `out of

employment'. Given that

ethnic and religious differences exist in the

attitudes towards female labour

market participation, one cannot directly ascribe this

observation to

discrimination. Interestingly though, nonfluent

females tend to be penalised

in terms of lower earnings, rather than through

employment prospects. There is

no unemployment disadvantage to non-fluent females,

although there is a

significant earnings penalty. In short, the results

for females further

support the assimilation of ethnic earnings, whilst

highlighting non-fluency

in English as an important determinant of observed

ethnic labour market

disadvantage.

Finally, this study shows that Muslims do experience

some unexplainable

employment penalty, relative to other non-white

religions, over and above all

other characteristics (including ethnic differences

and language fluency).

This supports the existence of religious

discrimination towards Muslims,

although such unexplained differences may well contain

unmeasurable components

such as motivation and attitudes towards employment.

Again this is especially

true for Muslim females who are `out of employment'.

To ensure equal

employment opportunities to all ethnic groups,

employers should be prohibited

from indirect racial discrimination. Religious

discrimination can occur

through the failure to allow for individual religious

requirements such as

prayer facilities, dress requirements, and the

requirements of religious

festivals such as fasting etc. Approximately half of

the Muslim/non-Muslim

unemployment differential can be explained by

differences in characteristics.

Hence Muslims appear to be less assimilated and/or

have less transferable

human capital than other non-whites.

Table 1. Religion and ethnicity of the FNSEM sample

 

                              Ethnic groups

 

Religious group        Males           Females

 

Muslim            51% Pakistani     53% Pakistani

                  34% Bangladeshi   32% Bangladeshi

                   7% Indian         7% Indian

                   7% African        7% African

                   1% Other          1% Other

 

Hindu             56% African       52% Indian

                  41% Indian        46% African

                   3% Other          2% Other

 

Sikh              80% Indian        84% Indian

                  20% African       16% African

 

Christian         61% White         60% White

                  31% Caribbean     35% Caribbean

                   7% Other          5% Other

 

Other religions   31% Chinese       38% Chinese

                  23% Caribbean     23% White

                  23% African       10% Caribbean

                  23% Other         10% African

                                    19% Other

 

No religion       59% White         62% White

                  27% Caribbean     21% Caribbean

                   8% Chinese        9% Chinese

                   6% Other          8% Other

 

Note: `Other religion' consists of 39% Buddhist, 17%

Jain, 12%

Rastafarian, 6% Jewish, 2% Parsi/Zorashia and 24%

other

religion.

Table 2. Descriptive statistics on religion, FNSEM

sample

 

                                                             

Average

                                                             

weekly

                                       Un-        Out

of     earnings

Religious        Sample   Employed   employed  

employment    [pounds

group             size       %          %           %        

sterling]

 

                                       Males

 

Muslim

  Pakistani        376       32         41          68        

188.67

  Bangladeshi      251       33         40          67        

122.73

  Indian            49       51         29          49        

179.47

  Other             55       62         18          38        

175.91

 

Hindu

  Indian           104       62         14          38        

270.96

  African          148       53          9          47        

256.46

  and Other

 

Sikh

  Indian           161       47         26          53        

185.01

  African           40       45         28          55        

257.69

 

Christian

  White            428       74         11          26        

304.59

  Caribbean        273       58         21          42        

260.86

  and Other

 

No religion

  White            304       66         19          34        

287.20

  Caribbean        215       50         32          50        

250.88

  and Other

 

Other religion      26       50         35          50        

357.81

 

Total            2,430       54         25          46        

244.69

 

                                      Females

 

Muslim

  Pakistani        434        8         12          92        

122.73

  Bangladeshi      256        4          7          96        

100.48

  Indian            56       15          7          86        

113.29

  Other             66       29         14          71        

177.68

 

Hindu

  Indian           149       36          8          64        

132.23

  African          137       49          8          51        

170.58

  and Other

 

Sikh

  Indian           201       30         10          70        

133.97

  African           37       35         11          65        

134.65

 

Christian

  White            675       57          5          43        

140.51

  Caribbean        453       54         18          46        

182.12

  and Other

 

No religion

  White            325       55          8          45        

164.35

  Caribbean        201       49         14          51        

192.91

  and Other

 

Other religion      39       31         21          69        

196.37

 

Total            3,029       39         10          61        

156.02

Table 3. Key results of employment probits for whites

and non-whites,

FNSEM sample (dependent variable = 1 if employed, 0 if

unemployed)

 

                               Model 1

 

                       Males            Females

 

Hindu                   --                --

  Indian           0.036 (0.17)     -0.152 (0.63)

  African Asian    0.225 (1.11)      0.055 (0.25)

 

Sikh              -0.569 (3.49) *   -0.222 (1.03)

 

Muslim                  --                --

  Pakistani       -0.886 (6.16) *   -0.997 (4.84) *

  Bangladeshi     -0.184 (1.09)     -0.890 (1.98) *

  Other            0.073 (0.38)     -0.518 (1.95)

 

Christian               --                --

  White            0.344 (2.55) *    0.274 (1.78)

  Caribbean and    0.137 (1.65)      0.008 (1.65)

  Other

 

Other religion    -0.561 (1.98) *   -0.616 (1.79) *

 

No religion             --                --

  Caribbean and   -0.164 (1.12)      0.176 (0.96)

  Other

 

Foreign-born            --                --

non-white

 

British-born            --                --

non-white

 

Non-fluency       -0.224 (2.18) *    0.087 (0.57)

 

Sample size            1,910             1,495

 

                               Model 2

 

                       Males            Females

 

Hindu              0.293 (1.85)      0.064 (0.37)

  Indian                --                --

  African Asian         --                --

 

Sikh              -0.362 (2.34) *   -0.254 (1.26)

 

Muslim            -0.374 (3.07) *   -0.801 (4.76) *

  Pakistani             --                --

  Bangladeshi           --                --

  Other                 --                --

 

Christian          0.318 (3.17) *    0.051 (0.46)

  White                 --                --

  Caribbean and         --                --

  Other

 

Other religion    -0.458 (1.65)     -0.684 (1.99) *

 

No religion             --                --

  Caribbean and         --                --

  Other

 

Foreign-born       0.278 (2.36) *   -0.003 (0.03)

non-white

 

British-born      -0.379 (3.25) *   -0.087 (0.69)

non-white

 

Non-fluency       -0.261 (2.55) *   -0.073 (0.47)

 

Sample size            1,910             1,495

 

Notes: t statistics are in parentheses, where t stat =

[alpha]/

SE([alpha]) and SE([alpha]) is the standard error of

[alpha].

 

* denotes statistically significant at the 5% level.

 

`Other Muslims' consists of 2% white, 4% Caribbean,

44% Indian

and 49% African for males. There are 2% white, 3%

Caribbean, 46%

Indian and 49% African females.

 

In Model 1 the default category is married,

non-religious whites,

who have no qualifications, are in good health, with

no children,

who are owner occupiers, who live in the South of

England

(excluding London) and have access to a car.

 

In Model 2 the default category is married whites,

with no religion,

born in the UK, no qualifications, in good health,

with no children,

who are owner occupiers, who live in the South of

England (excluding

London) and have access to a car.

Table 4. Decomposing differences between white and

non-white employment

probabilities into the ethnic penalty and that caused

by other

characteristics

 

                               (Model 2)

 

                                           Males          

Females

 

Total differences in predicted means   0.215 [0.237]  

0.170 [0.254]

Muslim penalty                         0.051 [0.055]  

0.048 [0.109]

Sikh penalty                           0.012 [0.015]  

0.010 [0.022]

Ethnic penalty                         0.110 [0.098]  

0.028 [-0.051]

Differences due to other               0.042 [0.069]  

0.084 [0.174]

  characteristics

 

Note: Figures in square brackets refer to the

employed/out of

employment dichotomy.

Table 5. Key results of earnings functions for whites

and non-whites,

FNSEM sample (dependent variable log of average

earnings)

 

                                  Model 1

 

                          Males            Females

 

Hindu                      --                --

  Indian              0.378 (0.54)      0.083 (0.61)

  African Asian      -0.159 (2.31) *    0.118 (1.05)

 

Sikh                 -0.216 (3.27) *    0.092 (0.83)

 

Muslim                     --                --

  Pakistani          -0.155 (2.32) *    0.166 (1.01)

  Bangladeshi        -0.289 (3.71) *    0.175 (0.71)

  Other              -0.177 (2.32) *    0.283 (1.49)

 

Christian                  --                --

  White               0.035 (0.89)     -0.082 (1.22)

  Caribbean          -0.063 (1.33)      0.094 (1.22)

  and Other

 

Other religion        0.112 (1.15)      0.271 (1.18)

 

No religion                --                --

  Caribbean          -0.014 (0.26)      0.124 (1.35)

  and Other

 

Foreign-born               --                --

non-white

 

British-born               --                --

non-white

 

Non-fluency          -0.209 (4.51) *   -0.314 (3.15) *

 

Correlation           0.005 (0.05)     -0.767 (12.21)

*

coefficient, [rho]

 

Sample size               1,309             1,186

 

                                 Model 2

 

                          Males            Females

 

Hindu                 0.006 (0.11)      0.037 (0.41)

  Indian                   --                --

  African Asian            --                --

 

Sikh                 -0.156 (2.41) *   -0.053 (0.49)

 

Muslim               -0.122 (2.23) *    0.078 (0.66)

  Pakistani                --                --

  Bangladeshi              --                --

  Other                    --                --

 

Christian            -0.001 (0.06)     -0.078 (0.66) *

  White                    --                --

  Caribbean                --                --

  and Other

 

Other religion        0.204 (2.09) *    0.172 (0.76)

 

No religion                --                --

  Caribbean                --                --

  and Other

 

Foreign-born         -0.128 (2.70) *   -0.031 (0.44)

non-white

 

British-born          0.006 (0.12)      0.170 (2.32) *

non-white

 

Non-fluency          -0.189 (4.04) *   -0.306 (3.08) *

 

Correlation          -0.026 (0.22)     -0.754 (10.90)

*

coefficient, [rho]

 

Sample size               1,309             1,186

 

Notes: t statistics are in parentheses, where t stat =

[beta]/

SE([beta]) and SE([beta]) is the standard error of

[beta].

 

* denotes statistically significant at the 5% level.

 

`Other Muslims' consists of 2% white, 4% Caribbean,

45% Indian

and 49% African males. There are 2% white, 3%

Caribbean, 46%

Indian and 49% African females.

 

In Model 1 the default category is married,

non-religious whites,

who have no qualifications, work for a firm with more

than 500

employees, in the construction industry, are owner

occupiers and

live in a low unemployment area in the South of

England.

 

In Model 2 the default category is married whites,

with no religion,

born in the UK, no qualifications, who work for a firm

with more

than 500 employees, in the construction industry, are

owner occupiers

and live in a low unemployment area in the South of

England.

Table 6. Employment decompositions between Muslim and

non-Muslim

non-whites, FNSEM sample

 

                                             Males          

Females

 

Differences in means                     0.230 [0.198]    

0.295 [0.361]

[I.sup.NM] - [I.sup.M]

 

Differences in coefficients

[[bar]P([[alpha].sup.NM][X.sup.M]) -     0.141 [0.0921    

0.165 [0.1461

  [bar]P([[alpha].sup.M][X.sup.M])]

[[bar]P([[alpha].sup.NM][X.sup.NM]) -   -0.035

[-0.037]   0.186 [0.176]

  [bar]P([[alpha].sup.M][X.sup.NM])]

 

Differences in characteristics

[[bar]P([[alpha].sup.NM][X.sup.NM]) -    0.089 [0.106]    

0.129 [0.215]

  [bar]P([[alpha].sup.NM][X.sup.M])]

[[bar]P([[alpha].sup.M][X.sup.NM]) -     0.265 [0.2351    

0.108 [0.185]

  [bar]P([[alpha].sup.M][X.sup.M])]

 

[chi square] (18 d.o.f. critical

  value)

 

Sample size                               1286 [1695]    

[859] 2018

 

Note: Figures in square brackets refer to the

employed/out of

employment dichotomy.

[Graphic omitted][Graphic omitted]Acknowledgements

The author is grateful for comments from two anonymous

referees and advice

from Derek Leslie, Roger Ballard and Ken Clark. The

Fourth National Survey of

Ethnic Minorities was made available through the ESRC

Data Archive.

Notes

(1) Black Africans were not included in the survey.

(2) Respondents in the FNSEM were asked two questions

concerning ethnicity.

First they were asked which ethnic group they thought

they belonged to,

followed by their family origin. This is similar to

that of other major

surveys such as the Population Census of 1991 and the

Labour Force Survey.

(3) The FNSEM also asks a question on the importance

of religion in the

respondent's life. However, this question was only

asked to 46 per cent of

non-whites.

(4) The Third National Survey of Ethnic Minorities was

conducted in 1982.

Brown (1984) provides a description.

(5) This study also excludes the self-employed. Clark

and Drinkwater (1998,

1999, 2000) explore non-white self-employment.

(6) Since the categories for the economic activity

variables provided by the

FNSEM were not mutually exclusive, the employed and

out-of-employment

variables redefined a small number of those who

answered more than once by

careful interpretation of each duplicated response.

Results presented do not

significantly differ from those derived when

duplicated answers were omitted.

(7) In these probit models, employment status is

governed by a latent index

variable measuring `employability', [E.sub.i.sup.*],

where [E.sub.i.sup.*] =

[alpha][X.sub.i] + [[epsilon].sub.i] is the latent

employment model.

[E.sub.i.sup.*] is explained by a characteristic

vector, [X.sub.i] and an

unexplained residual component, [[epsilon].sub.i]. In

practice one cannot

observe employability. One can only observe whether

the individual is employed

or not. This can be captured by a binary variable,

[E.sub.i], which takes the

value unity if the individual is employed and zero if

they are unemployed. The

observed binary outcome is related to [E.sub.i.sup.*]

in the following way.

[E.sub.i] = 1 if [E.sub.i.sup.*] > 0 and the

individual is employed or

[E.sub.i] = 0 if [E.sub.i.sup.*] [less than or equal]

0 and the individual is

unemployed. The statistical model that underlies this

is probabilistic. If the

residual term is assumed to be logistically

distributed then the probability

of the ith individual being employed is given by

P([E.sub.i]) = 1/1 + exp - ([alpha] [X.sub.i]) '

If the residual term is normally distributed then a

probit model is used.

Greene (1997) provides an excellent discussion.

(8) The FNSEM asks the interviewer to code the

respondent according to their

English language ability. The categories of these

abilities were fluent, fair,

poor or none. Here non-fluents consist of those with

fair, poor or no

language-speaking ability. See Dustmann and Fabbri

(2000), Leslie and Lindley

(2001), as well as Lindley (2001) for a discussion on

the impact of language

fluency on the employment and earnings of Britain's

non-whites.

(9) There are only 25 (8) employed Indian Muslim men

(women) with available

earnings information.

(10) There are only 18 (13) employed African Sikh men

(women) with available

earnings information.

(11) The FNSEM excludes, among others, those of a

Black African and Middle

Eastern origin. These groups form a growing part of

the Muslim population in

Britain.

(12) At the time of the survey the statutory

retirement age was 65 years for

men and 60 years for women.

(13) Full regression results are available from the

author on request.

(14) Where [E.sup.*] = X'[alpha] + [epsilon] is the

latent employment model,

the coefficients in the probit model can be thought of

as the effect of a

regressor on the latent variable [E.sup.*]. An

algebraic formulation of the

coefficients would just be [alpha] = d[E.sup.*]/dX,

whilst the marginal

effects (which are the responses of the probabilities

p(E = 1) to the

regressors) are given by dp(E = 1)/dX =

[alpha].[PI](X'[alpha]). Note that

[PI] is the standard normal density function and

employability, [E.sup.*] is

in the first expression but E (the binary outcome

variable) is in the second

expression.

(15) The coefficients in Table 3 can be interpreted as

follows. The first term

in the first column for Indian Hindu males is 0.036

(0.17). The coefficient is

0.036 and this is the effect of being Indian Hindu on

the employability of

males, relative to the default of non-religious

whites. The term in

parentheses (0.17) is the t statistic. This is a test

for the significance of

being Indian Hindu on the employability of males,

relative to non-religious

whites. Since 0.17 is less than an approximate t value

of 1.96, we can deduce

that the effect of being Indian Hindu on the

employability of males is

statistically insignificant at the 5% level.

(16) Comparing the coefficient on foreign-born

non-white of 0.278 with that on

Christian of 0.318.

(17) There are endogeneity issues involved when

including language fluency as

a determinant of earnings or employment probabilities.

Lindley (2001)

demonstrates that using single equation techniques

implies that the language

penalty may be underestimated.

(18) Here the semi-log specification of the earnings

function is used. In

practice, for low values (of around 0.10) log points

method and the level

method give the same result. The semi-log

specification is given by lnY =

[alpha] + [beta]X, where lnY is the log of earnings

and X is an ethnic binary

variable. Since X is binary, the effect of X on Y is

[Y.sub.|X = 1] - [Y.sub.|X = 0]/[Y.sub.[X = 0] =

[e.sup.[beta]] - 1,

whereas the log point difference is simply [beta].

(19) The coefficients in Table 5 are the log points

and can be interpreted as

follows. The first term in the first column for Indian

Hindu males is 0.378

(0.54). The first term indicates that Indian Hindu

males would earn on average

37.8 percentage log points more than the default

category of non-religious

whites. The second term in parentheses (0.54) is the t

statistic. This is a

test for the significance of being Indian Hindu on the

earnings of males,

relative to non-religious whites. Since 0.54 is less

than an approximate t

value of 1.96, we can deduce that there are no

significant differences between

the average earnings of Indian Hindu and non-religious

white men, at the 5%

significance level.

(20) There are endogeneity issues involved when

including language fluency as

a determinant of earnings. As a result using single

equation techniques

implies that the language penalty may well be

underestimated for males and

overestimated for females.

(21) In a single equation probit the average

predictions are virtually the

same as the sample averages.

(22) Table 3 shows an employment penalty to both

British- and foreign-born

whites, when these are compared to Christians (who are

mainly white). However,

only British-born non-whites exhibit significantly

lower earnings than

non-religious whites.