YouthUnemployment and

Intergenerational Immobility
BRUCE BRADBURY, PAULINE GARDE and JOAN VIPOND*

The personal and family characteristics of those teenage workers and unemployed
teenagers who were living with one or both parents at the time of the 1981 census
are described and analysed. The 1 per cent sample of household records released by
the Australian Bureau of Statistics was the data source. Unemployment rates were
much higher among teenagers of sole-parent families than among other teenagers.
However, in the case of teenagers who lived with their mothers only, this could be
attributed to their low family incomes rather than their family structures. Teenagers
who lived with both parents also had high unemployment rates when family incomes
were low. Among youth who lived with both parents, high unemployment probabilities
were associated with parental disadvantages, even when the personal characteristics
of the youth, such as their education levels, were controlled. It was concluded that
unemployment distribution among 15- to 19-year-olds reflects intergenerational rigidities
with respect to labour market status.

The literature on
intergenerational mobility suggests that Australia is an open
society in which the path to success is education. Education is seen as a route
. from low status families to high status occupations. Yet, relative to other
countries, few Australians stay long in the education system. What happens
to this majority who begin their working life in their teens? They are not an
undifferentiated mass. In the early years of their working lives their successes
and their failures could be approximated by their experiences of unemploy-
ment, for unemployment massively reduces incomes. This paper analyses the
distribution of unemployment among the teenage labour force. The aim is
to explore the associations between personal and family characteristics and
labour market successes or failures of young people. ,

Two studies of family background were undertaken. First, the differences

.

in unemployment rates of youth who lived in three family structures-dual-

parent families, mother-only families and father-only families-were com-
pared. Second, among youth in dual-parent families, the significance of
statistical associations among unemployment rates and family backgrounds
was measured. Before describing these studies, it is necessary to comment
on the data source and techniques of analysis that both used.

The data and methods of analysis

The recent policy of the Australian Bureau of Statistics of releasing unit records
from the census and other surveys provides new data that describe individuals
* Social Welfare Research Centre, University of New South Wates, PO Box 1, Kensington, NSW
2033. The authors wish to thank R. Ross and an anonymous referee of the JIR for comments
on earlier versions of this work.

191

Downloaded from jir.sagepub.com at NORTH CAROLINA STATE UNIV on March 18, 2015
 192

within their households. The data used in this analysis of youth unemployment
were taken from the per cent household sample file of the 1981 census (ABS,
Census of Population and Housing, 30 June 1981, Household Sample File;
Technical Details, Cat. No. 2167.0 provides details of the file). Throughout
this paper, results have been multiplied by 100 to provide values for Australia
as a whole; the original sample and sub-sample sizes can, therefore, be derived
very easily. ’.
Individuals who were young (aged 15 to 19 years), who were in the
workforce, and who lived with one or both parents were selected for analysis.
All those who were still in full-time education or were unpaid helpers were
excluded. The household sample file provides information from the census
questions on the birthplaces, workforce experience, education, housing, ages
and incomes of both young people and their parents, though it does not
describe the location of their. households. Thus, in our sample, we could
identify, first, whether a young person was employed or unemployed and,
second, we could measure a range of family and personal characteristics for
that youth.
Unfortunately, the parental data were not available for youth living away
from their parents on census night, and so our analysis was restricted to those
74 per cent of youth in the workforce and not in full-time education who
were coded as living with their parent(s).&dquo; It was not possible to ascertain
whether the relationships described below applied to the excluded youth,
though we believe that the general thrust of the discussion below could be
expanded to include them also.
Logistic regression analysis was used to test the associations between the
&dquo;
status of being employed or unemployed and the personal and family
characteristics of the young people. The statistical methods used are described
in the appendix.

Family structure andyouth unemployment

Most young people who lived at home were living with both their parents
on census night. Only 15 per cent were in sole-parent families; 12 per cent
lived with their mothers only, and 3 per cent, with their fathers. Unemployment
rates varied markedly among these youths. Among these living with both
parents, unemployment in June 1981 was 13 per cent. Youth living with their
mothers only had unemployment rates of 24 per cent, and those who lived
with their fathers only had an unemployment rate of 28 per cent (Table 1).
We hypothesized that the differences in unemployment rates could be due
to the family situations of those young people and to other factors associated
with their family status. Possibly, youth in single-parent families’are dis-
advantaged in the labour market because of events that led up to their family
status-the trauma of separation or death. The welfare-dependent nature of
their households may also affect their ability to find work. They may act as
carers of their parents. On the other hand, their disadvantages in the labour
market may be due not directly to their family status but indirectly, through
other social and economic variables that are in turn associated with family
status.

Downloaded from jir.sagepub.com at NORTH CAROLINA STATE UNIV on March 18, 2015
 193

’1’abie 1 Youth aged 15-19 years, not in full-time education and living with at least one parent
on census night: employment status by family type

Note: Youth in families with only one parent present who was now married are excluded from
this table. Youth employed as ’unpaid helpers’ have been included with ’not in workforce’
because their workforce status is ambiguous.
Source: Australian Bureau of Statistics, 1981 Census of Population and Housing, 1 % Household
Sample File. -

The most obvious economic difference among families was in incomes. The
mean family income (income of head and spouse, if present) was approxi-
mately $7,000 per annum for mother-headed families; $11,000 per annum for
father-headed families; and $17,800 per annum for dual-parent families. These
figures reflect both the lower number of income earners in single-parent
families but also, particularly for mother-headed families, the lower income
of the parent.
The lower workforce participation rates of parents with child-care
responsibilities (at the time of the census and before it) and the associated
lower incomes may be of major importance. They could negatively influence
youth’s employment prospects in several ways:
1. a lower level of education attainment for the youth;
2. the family living in a low-rent, low-employment suburb, remote from job
’

opportunities;
3. lack of workforce ’contact’ through workplace experience of parents and
reduced access to informal job-seeking networks; and
4. lack of other resources helpful in finding jobs (good clothes or cars, for ’

example).
Current income differences among parents are a very narrow measure of
their economic situations and their abilities to transfer labour market
advantage to the next generation. For instance, the children of parents with
low current incomes but high educational qualifications may be very successful
themselves in educational attainment because of family encouragement and
example. Their parents may have influential contacts in job-information
networks despite their low incomes. As well as the education levels of parents,

Downloaded from jir.sagepub.com at NORTH CAROLINA STATE UNIV on March 18, 2015
194

their occupations and birthplaces may also influence their ability to help their
children find jobs.
The influence of their occupations could not be tested, since many sole
parents have only marginal or intermittent attachment to the workforce. While
our analysis showed some relationship between family type and the other
parental background variables (single mothers were more likely to be
Australian-born, but mothers of both types had similar education and
qualifications; single fathers were more likely to have been born in other
’English-speaking’ countries, were less likely to have acquired trade
qualifications, and were more likely to have no qualifications), these
relationships did not help explain the different unemployment rates in different
family types. That is, the relationship between family type and youth
unemployment remained essentially the same, whether or not these family
background characteristics were controlled.
The variable that did influence the relationship between family type and
youth unemployment was family income. When logistic regression models
were fitted including family income, it was found that unemployment
probability was strongly associated with family income (t 9.5) with youth =

from low income families much more likely to be unemployed.

Further, when holding family income constant we found that unemployment
among youth of dual-parent families was not significantly different from
unemployment among youth of mother-headed single-parent families. We
found, however, ~hat youth in father-headed, single-parent families had a
higher probability of unemployment than other youth, even when controlling
for income level (t 2.5). =

We have no explanation for the disadvantages found among youth who

lived with their fathers only. With respect to mother-headed single parent
families, however, the pattern of concentration of disadvantage seems clear.
The poverty of many women single parents continues its impact into the next
generation, with higher unemployment levels for their children. It was not
the family situations of these children that mattered. It was their income levels.
Youth in both dual-parent families and who lived with their mothers only
had equal probabilities of unemployment when their family incomes were
the same.

Personal, family and locational characteristics and youth unemployment

The demonstration of a significant association between low parental incomes
and high probabilities of unemployment among teenage workers contrasts
with the results of other studies of intergenerational mobility, which have put
more stress on the openness of Australian society (Blandy & Richardson 1982;
Broom el al. 1980; Richardson 1985). One reason for this difference may be
that our model was too simple: it analysed the family circumstances of young
people but ignored their own personal characteristics, such as their education
and skills. In this section we report on a more complex analysis of youth
unemployment, which aimed to show the association of youth unemployment
probabilities and family background characteristics when variations in
personal characteristics were controlled.
’

Downloaded from jir.sagepub.com at NORTH CAROLINA STATE UNIV on March 18, 2015
 195
’

Unlike Richardson (1985) we had no information on individuals’ IQ levels

and subjective measures such as personal motivation scores, so we could not
replicate her tests on our larger sample. Nevertheless, the census did provide
many details on the personal and family characteristics of the young people
who lived in 1 per cent all Australian households. In order to make full use
of this information and to take into consideration the occupations of parents
as well as their income levels, we reduced our sample to that of youth who
lived with both parents on census night. As we have noted, many sole parents
have intermittent attachment to the workforce. The omission of their children
reduced our sample size by only 15 per cent.
In our analysis, we identified the characteristics we expected would be
associated with unemployment probabilities. These variables are listed in three
categories in table 2: personal characteristics, family characteristics and
location characteristics. (There were only two of the last, a problem associated
with our data source.) Also in table 2 we show the number of unemployed
young people in Australia with each characteristic, to show when our data
were based on few people, and we list the unemployment rate among the
youth workforce with that characteristic. There was considerable variation
.

among these unemployment rates.

All of the variables listed in table 2 were included in logistic regression tests,
but several were found to be not-significantly
associated with unemployment
probabilities when all other variables were held constant. For variables that
were significantly associated, the logistic regression parameters 0 and their
t values are presented in the appendix.
Modal categories of variables (those categories containing the largest
number of youth) are noted in table 2 by single asterisk. As noted in the
appendix, the results of the logistic regression analysis can be most easily
interpreted through the fitted probabilities of unemployment. These, listed
in column 3 of table 2, show the level of unemployment to be expected if
a person had all characteristics, except the one for which the probability is
being calculated, fixed at their mean and modal values.
Thus, to find the fitted probability of youth unemployment according to
father’s occupation when all other variables were held constant, the following
reference categories were chosen: the family income equalled $17,800; the
mother was not in the workforce; it was an owner/purchaser household; there
were 1.34 rooms per person; the youth had lived in Australia for five or more
years; English language ability was ’well’ or better; the youth was male, and
left school at age 16; the household was in a major urban area and had not
moved in the last five years. In analysing the association between the
probability of unemployment and mother’s occupation, the father’s occur
pation was set to manual employee and all other variables except mother’s
occupation took the values listed above. Similar procedures were followed for
each other variable. The fitted probabilities of unemployment for different
levels of parental income are shown in figure 1.
Our tests suggested that all three categories of variables-personal, family
and location-were significantly and separately associated with labour market
outcomes.

Downloaded from jir.sagepub.com at NORTH CAROLINA STATE UNIV on March 18, 2015
196

Table 2 Youth aged 15-19 living with both parents and not in full-time education: comparative
unemployment rates,

Downloaded from jir.sagepub.com at NORTH CAROLINA STATE UNIV on March 18, 2015
 197

Table 2 Continued

Downloaded from jir.sagepub.com at NORTH CAROLINA STATE UNIV on March 18, 2015
 198

*
Notes: Modal categories, that is, those categories containing the largest number of youth.
**
Unemployment rates significantly different from the rate for the modal category,
all other things held constant.
nss Not statistically significant (given the other variables) at the 5 per cent level.
a. For definition see text. For reasons explained in the text, this variable was not included
in the logistic regression model when determining the other fitted probabilities. The
figures in brackets in column 3 represent the fitted probabilities obtained when this
variable is included in the model.
b-If 1981 or 1980 address not stated, cases are coded as having moved in the last year.
If 1976 address (only) not stated, and haven’t moved in last year, cases are coded
as having moved in the last five years.

Souree: Australian Bureau of Statistics, 1981 Censiis ofPopulation and Housing, 1 % Sample
File.

Personal characteristics and youth unemployment

Our information on personal characteristics was limited to the sex, education
level, age, race, period of residence in Australia and English language ability
of young people. Unemployment rates varied markedly according to these
characteristics.
In 1981, the female youth in our sample had a substantially higher
unemployment rate than males, 16.4 per cent compared with 10.5 per cent.
These higher unemployment rates had existed for the previous two decades
(Bureau of Labour Market Research 1983, 13). As Windschuttle (1979) points
out, the primary cause of this is the segregated labour market, which severely
limits the job choices for young women. Not only is direct discrimination
involved, but also the processes that determine the narrow range of job choices
which are socially appropriate for young women to consider (see Koller et

Downloaded from jir.sagepub.com at NORTH CAROLINA STATE UNIV on March 18, 2015
Downloaded from jir.sagepub.com at NORTH CAROLINA STATE UNIV on March 18, 2015
200

Unemployment rates fell as the age of youth increased, yet age’ was not
significantly associated with unemployment when holding the other variables
constant. This was due to the.strong association between age and education
level; by definition, 15-year-old workers, employed ’Or unemployed, must have
a low education level. Thus knowing the youth’s age contributes little extra
information if we already know their education level.
The data suggest that, with their 21 per centunemployment rate, youth
of Aboriginal or Torres Strait Islander descent were much more likely to be
unemployed than other young people. When other variables were held
constant, however, race was not significantly associated with unemployment.
It is difficult to interpret this result as it was based on a small sample.
Recent migrants to Australia can generally be expected to suffer from
numerous disadvantages in the labour market, such as unfamiliarity, language
problems, unrecognized qualifications and direct discrimination. Among the
few youth who had arrived in the last year, four out of the five in the sample
were unemployed. Miller (1981), using the 1976 census, and Brooks and Volker
(1983), using the household sample file, found similar tendencies. Even after
standardizing for the other variables, period of residence in Australia con-
tinued to be a significant predictor of unemployment rates: those who had
arrived in the previous year had an unemployment probability of 66 per cent.
The 4 per cent unemployment probability for the one to four year residents
was not, however, significantly different from the rate for the long-term
residents.
English language ability was associated with unemployment rates in the
expected manner, with those speaking English ’not well’ or ’not at all’ having
7 out of 10 unemployed. English language ability remained a significant
predictor of unemployment probability when all other variables were held
constant. Those speaking ’not well’ or not at all had an unemployment
probability of 43 per cent. This reduction from the 70 per cent in the raw
data was probably due to the association of English language ability with
period of residence in Australia, income levels of newly arrived migrants and
other factors.

Family characteristics and youth unemployment .

Family characteristics were measured by income and eight other variables,

five of which were statistically associated with unemployment probabilities.
Both father’s and mother’s incomes were higher among employed than
unemployed youth. On average, fathers of employed youth earned $13,500
per annum and mothers’ earnings were $4,700. Unemployed youth had fathers
who earned, on average, $10,700 and mothers who earned $3,600. The effects
of father’s income and mother’s income on the probability of unemployment
were found to be very similar, so in the models described here they were
summed together as ’family income’. In theory family income should include
any board or rent paid to parents by youth and may thus simply reflect the
labour force status of the youth. However, given the self-reported nature of
the census data, it is not clear whether people would have included such
incomes with their total income.

Downloaded from jir.sagepub.com at NORTH CAROLINA STATE UNIV on March 18, 2015
 201

When controlling for the other variables, the

probability of unemployment
was found to drop, at a decreasing rate, as the level of family income rose
up to $38,000 per annum. The relationship is illustrated in figure 1, which
shows the fitted probability of unemployment ranging from 23 per cent at
$5,000 per annum to 9 per cent at $30,000 a year. This curve illustrates the
smoothed relationship derived from the logistic regression model described
in the appendix.

Figure 1 Fitted youth unemployment rates with and without other variables

Notes. The ’family income model’ curve models youth unemployment as a function of family
income (and family income squared) only.
The’general model’ curve models youth unemployment as a function of family income
together with the other variables listed in appendix table 1.
Source: The ’general model’ curve is derived from the parameters in appendix table 1. The ’family
income model’ curve is derived from a similar model including only the family income
variables.

Figure 1 also contains the curve derived from fitting unemployment

probability as a function of family income (and family income squared) only.
As one would expect, this curve shows an even stronger relationship between
the two variables. The difference between the two curves, however, is not all
that great, suggesting that there was a fair degree of independence between
the effect of income and parental occupation on youth unemployment. Of
the aspects of parental occupation other than income that affected youth’s

Downloaded from jir.sagepub.com at NORTH CAROLINA STATE UNIV on March 18, 2015
202

employment prospects, the most notable seems to be the increased ability of

self-employed or employer parents~to help their offspring find work. When
all other variables were held constant, the lowest unemployment rates were
found among the children of independent. tradesmen.
Generally the pattern of unemployment rates showed that parents in
occupations that enabled them to directly place their children in work or
allowed them to transfer social resources, such as education and self esteem,
to their children were least likely to have unemployed children. Thus fathers
in professional, managerial or clerical occupations had children with
significantly lower unemployment than manual workers’ children. Fathers and
mothers who were employers or self-employed also had children with lower
unemployment rates.
A notable result was that almost half of the youth with a mother or father
unemployed were themselves unemployed. In establishing the magnitude of
this effect, however, it should be borne in mind that relatively few of the fathers
and mothers of youth in this age group were unemployed, with unemployment
rates of 2.4 per cent and 1.3 per cent respectively. When all other variables
were held constant, the unemployment rates of children with unemployed
parents were not significantly different from those of the children of manual
or trades workers. No doubt this was due to the association between parents’
unemployment and low family income.
The level of parental education was also associated with youth
unemployment rates, youth with more highly educated parents having a lower
probability of unemployment. In addition, parental education was associated
with occupational status of parents and their income levels. As a result it did
not appear as a statistically significant variable in the logistic regression. In
other words the effect of parental education on youth employment was
mediated by the other variables in the model, most notably parental
occupation and income, and youth’s education level.
Although the children of migrants had higher unemployment rates than
those of the Australian born, the birthplace of the father was not significant
when all other variables were held constant. This result is consistent with that
of Inglis and Stromback (1984), who argued that migrants are less likely to
be unemployed the longer they have been in Australia. The important
controlling variable was length of residence in Australia. There was, of course,
a positive correlation between being born outside Australia and being a
resident here for less than a year. For the small number of youth born outside
Australia, it was found that their unemployment was not significantly different
to Australian youth both when controlling and not controlling for other
variables.
The class position of youth, as determined by their access to social resources
of various kinds, may be indicated by family wealth as well as by parental
occupations and education. Broom and his co-authors (1980), for instance,
found their indicators of (parental) family wealth to be strong predictors of
later occupational status and incomes-better indicators than any of the other
family background variables.
Housing situations have been used here as an indicator of family wealth
because no direct information was available and because the assets of most

Downloaded from jir.sagepub.com at NORTH CAROLINA STATE UNIV on March 18, 2015
 203

families are concentrated in housing. The ’nature of occupancy’ variable

distinguishes home owners and purchasers, private renters and public renters
as major tenancy groups. Household density, defined as rooms per person,
was also included in our data. Both these variables indicated associations with
youth unemployment. Youth in renting households, particularly public renters,
had over twice the unemployment rate of youth in owner-purchaser
households and employed youth lived in less cramped accommodation than
unemployed.
When all other variables were held constant, these indicators of wealth were
significantly associated with unemployment. Youth in private rental
accommodation had similar unemployment rates to those in owner-occupied
dwellings, but youth who were housing-authority tenants had significantly
higher unemployment (17 per cent compared with 11 per cent among families
that were owners and purchasers). Teenage unemployment among all families
increased as rooms per person in the dwelling decreased.
The final variable measuring family background referred to the
unemployment experience of siblings. Although this was a statistically
significant variable it was not easy to interpret because it can have two
meanings. It may imply that the employment status of one youth directly
affects that of another. For example an elder, employed sibling may be able
to help a younger sibling find a job in ways that he or she could not do if
they were unemployed. If this interpretation is correct, the variable ’sibling
unemployment’ should be included in the equations if we wish to measure
the independent effect of the other variables.
On the other hand, it may measure the impact of variables that are missing
from our equations, or that are poorly measured in our data, and that affect
all the family. Locational variables are particularly important, since all family
members are part of the same local labour market. Unemployment is higher
in rural areas than in the big cities (see below). Moreover, intra-urban
unemployment probabilities are not equal. We know from other studies that
living in the outer suburbs of large cities may lead to higher unemployment
than being in the middle or inner areas (Vipond 1985).
In the absence of precise measures of locations, the impact of place could
appear as an impact of sibling unemployment. In addition, our incomplete
measures of the social class characteristics of the family could also show up
as a similar communality between the youth. In this case, including the variable
’sibling unemployment’ would produce confusing results, as we would be
looking at the effect of variables while controlling for other indicators of those
effects.
Although we have provided parameter values in appendix table 1 for
equations with and without the variable describing sibling unemployment,
we tended to favour the argument that it mainly measured family components
in the variation in youth unemployment that our model had failed to measure
precisely. Consequently in presenting probabilities of unemployment in table
2 we used the equation that excluded the sibling unemployment variable. Some
fitted probabilities for sibling unemployment itself were included (in brackets)
in order to describe the relationship of this variable to youth unemployment
when controlling for other variables.

Downloaded from jir.sagepub.com at NORTH CAROLINA STATE UNIV on March 18, 2015
 204

Locational characteristics and youth unemployment

The limited information available on the geographic location of the youth
pointed to the relative disadvantage of those youth not living in the major
urban centres. When controlling for the other variables in the model, rural
youth had significantly higher unemployment rates than urban youth.
The effect of geographic mobility on unemployment is at first glance
perhaps more surprising than that of location, as it illustrates a very strong
association between residential mobility and youth unemployment. It is
important to note, however, that we have no information of the type of
mobility from the census. That is, residential movement could mean an
interstate, inter-city, inter-suburb or simply ’up the street’ movement.
The fact that unemployment rates are higher among movers suggests that:
1. people (and families) that are more prone to unemployment are also more .

likely to move (e.g. renting families), or,

2. moving has a causal influence upon the likelihood of a person being
unemployed (e.g. through loss of contacts, unfamiliarity with the local
labour market), or,
3. unemployment causes movement in search of work. This, however, is not
a likely relationship for our sample of youth as they would have to get
their parents to move with them as they looked for work (otherwise they
would be excluded from the sample).
Although the issue is obviously complicated, we suggest that, as the
relationship between the unemployment rates of these youth and their
geographical mobility was very strong, even when controlling for the other
variables, this weakens the argument that it is people of a ’similar type’ who
both move and become unemployed, and correspondingly strengthens the
alternative-that there is a causal link between moving and being unemployed.
As the unemployment rate of recent movers was 24 per cent when all other
variables were controlled, the figures seem to imply that family relocation
was a major disruption to youth’s employment search. To the extent that
relocation was due to parental (most likely father’s) labour mobility, it allows
us to suggest the mixed blessing that such labour mobility holds for the moving
families. Although the father may move to find or improve his job, the
disruption to the youth’s life-particularly to the informal contacts so
necessary to finding work-may be severe. This effect continued beyond the
youths who had moved in the last 12 months to also include those who had
moved in the last 5 years. The fitted probability of unemployment dropped
from 24 per cent for the first group to 15 per cent and then to 11 per cent
for those who had not moved.
’

Conclusions
The main result in this analysis is that family background and location as
well as personal characteristics appear to be associated with the unemployment
levels of 15- to 19-year-old workers. The results may be summarized as follows.
Youth in single-parent families had almost twice the unemployment rate
of youth living with both parents. Youth with the same level of family income,
however, had similar unemployment rates, whether they lived with single

Downloaded from jir.sagepub.com at NORTH CAROLINA STATE UNIV on March 18, 2015
 205

mothers or both parents. This implies that it is the associated economic

disadvantages of single parenthood that disadvantaged the youth in the labour
market rather than the qualities of single parenthood as such.
This explanation, however, does not hold for the youth of single-fathered
families, who had both a higher unemployment rate and higher family income
than those of single-mothered families. ~’Vhy this is so remains something of
a mystery, though the small size of the sample should be noted.
For youth with both parents in the household, a much more detailed analysis
of the effect of family and household characteristics was possible. The results
indicated that the personal characteristics of teenagers were associated with
unemployment risks in predictable ways. Young women had higher
unemployment rates than young men; more education was associated with
less unemployment, as was the ability to speak English well; and
unemployment was high among recent arrivals to Australia. Yet, even when
these characteristics were held constant, there remained a significant
association between family characteristics and unemployment rates.
Social class was found to be strongly associated with unemployment
probabilities, with those youth coming from disadvantaged families (low family
incomes, parents in working class occupations or unemployed, parents with
lower education levels, and families living in poorer quality or rental
accommodation) were more likely to be unemployed. The effect of parental
education, however, was largely an indirect one, acting through its effect on
the other variables mentioned rather than directly. Similarly, youth living in
private rental accommodation did not have higher probabilities of being
unemployed when the other class variables were held constant, though those
living in housing-authority accommodation did. This is probably because the
housing-authority tenure better reflected low family wealth than did tenure
of private rental accommodation.
Rural youth had substantially higher unemployment rates, irrespective of
whether controlled for the other variables in the model. The process of family
movement seemed significantly disadvantageous for the younger members
of the family. As other studies have shown, informal job-seeking networks
are very important in finding work, particularly for youth. These are likely
to become substantially disrupted when families move. A case could thus be
made to include those mobile youth with other disadvantaged groups when
allocating places in youth training and temporary employment programmes,
such as the Community Employment Program.
All these relationships, however, were not sufficient to describe entirely why
youth unemployment was concentrated in particular households, for even
when taking all the above relationships into account, youth were still more
likely to be unemployed if they had siblings unemployed. We believe that this
residual association was most likely due to regional variations in
unemployment rates, which we have been unable to model from the household
sample file, together with other family factors which we have been unable
to measure, or able to measure only imprecisely.
The significant association we have found between high unemployment rates
and disadvantages in family background is in contrast with studies of the
occupational status of employed parents and their employed children, which

Downloaded from jir.sagepub.com at NORTH CAROLINA STATE UNIV on March 18, 2015
 206

have tended to stress the openness of Australian society. We have shown that
both parental occupations and parental incomes were strongly associated with
the unemployment rates of young workers: that is, that there was consistency
in the labour market disadvantages of parents and the higher probabilities
of unemployment in their children.
As we have noted, our analysis was hampered by lack of data on subjective
measures, such as attitudes to work and success, and on personal qualities,
such as ‘inteIligence’. Nevertheless our statistics were not without important
details on the individual characteristics of employed and unemployed young
people. Moreover, they were drawn from a very large, random sample of the
entire Australian population. The results of this section therefore present a
particularly disquieting picture of inequality in Australian society.
Appendix: Statistical methods
This appendix provides further details of the logistic regression models
described in the text. Logistic regression, or more precisely, binomial regression
with logit link, was used because the dependent variable for the analyses was
binary (youth were defined as being either employed or unemployed).
Formally, the procedure analysed the fitted probability of unemployment, pi,
as equal to

where X;~3 represents a linear function of variables X and parameters ~3 as

in ordinary regression. This linear predictor X~(3 was expressed in the same
way as in conventional regression models with the vector of parameters, 0,
being estimated by maximum likelihood methods. Asymptotic t statistics were
used to determine whether the parameter values obtained were likely to be
due to chance.
Parsimonious models of the effects of personal and family characteristics
(Xi) on unemployment probabilities (pi) were arrived at by excluding variables
that did not improve the fit of the model significantly given the presence of
the other variables. The test used was the likelihood ratio or deviance difference
test (Baker & Nelder 1978).
As with most statistical tests the significance levels obtained assume a
random sample of independent observations. For the models used here,
however, the observations were not independent. The household sample file
contains an independent random sample of households, but we were analysing
a sample of youtlz. In some cases there was more than one youth from a
particular household and they had identical values for their household level
variables and, more than likely, identical values for their family level variables.
As a result, the test statistics showed an inflated level of descriptive signifi-
cance. For example, a variable that contributed significantly to the fit of the
model at the 5 per cent level might really have been significant at only the
10 per cent level, though in practice the effect was not so great.
Direct modelling of this sample structure was too difficult, but there were
ways of getting approximate estimates. One way would have been to select
only one person from each household and build the models on this sub-sample.

Downloaded from jir.sagepub.com at NORTH CAROLINA STATE UNIV on March 18, 2015
 207

This, however, would have had the effect of under-representing the conditions
of youth in large households. The compromise approach used was to explore
the fit of the model (checking significance of terms and looking for inter-
actions) with a sub-sample consisting of only one youth from each household.
For the model for youth with both parents present, this meant dropping 15
per cent of the 3950 cases. The t statistics presented with the results of the
model are from the sub-sample (see appendix table 1). Parameter estimates
for the model, however, were derived from the’model fitted to the full sample.
Also presented with the results are two measures of goodness of fit: the Cragg-
Baxter statistic (Brooks & Volker 1983, 31) and McFadden’s R2 (Amemiya
1981. 1505). The Cragg-Baxter statistic is defined as:

McFadden’s R2 is defined as:

with n number of observations, Lc = the maximum of the log likelihood

=

when only a constant is used and L, = the maximum when all coefficients
are included in the model. These attempt to provide some analogy to the R2
statistic of standard regression. All these statistics were quite tow, which is
not uncommon in models based upon observations of individual persons.
Nevertheless, although our models described only a small proportion of the
pattern of youth unemployment, we were able to identify a range of significant
associations between youth unemployment and personal and family
characteristics.
The appendix table 1 contains the estimated parameters for two models,
which differ by the inclusion or exclusion of the variable sibling unemployment
from the model. This was done because the interpretation of this variable
was ambiguous. This is discussed further in the text.
For continuous variables, the parameters ~3 measure the increase in log odds
of unemployment for an increase of one unit of the independent variable
(given the other terms in the model), whilst for categorical variables they
measure the increase in the log odds of unemployment relative to the reference
category (for which (3 is defined to be zero) of the variable. In the tables here
we have always defined the reference category to be the largest or modal
category for that variable.
Although odds ratios (obtained by taking the exponent of the parameters
given) may be the ’natural’ measure to use in these models, they are not as
easily interpreted as the more familiar probability differences or ratios. For
this reason fitted probabilities at the modal categories and means of
continuous variables are presented in table 2 above. They were derived from
the following formula:

Downloaded from jir.sagepub.com at NORTH CAROLINA STATE UNIV on March 18, 2015
 208

where C = constant term + effect of incomes and rooms per person at the
mean and 13¡ =
parameter of interest.
For the model in this table,
mean family income =
$17,800 a year
mean rooms per person = 1.34
and so C = -2.081
The non-linear nature of the logistic regression model means that the
differences or ratios of these fitted probabilities will vary, depending on which
categories are chosen as reference categories, though the ordering of categories
will remain the same.
Appendix table 1 Logistic regression models of youth unemployment and family background

Downloaded from jir.sagepub.com at NORTH CAROLINA STATE UNIV on March 18, 2015
 209

Appendix table 1 Continued

Note: aFor youth aged 15-I9, not in full-time education, in the workforce and living with
both parents on census night 1981. Dependent variable is youth unemployment status-
equal to 1 if unemployed, 0 if employed. The parameter estimates are obtained from
the full sample. Zero parameter values represent reference categories of variables. These
are also the modal categories.
b Parameter estimate divided
by its standard error. Absolute t values in excess of 1.96
indicate parameters are significantly different from zero at the approximate 5 per cent
level. Note these t values are obtained from the model fitted to the sub-sample of
one youth per household.
Tuning point of quadratic relationship occurs at $39,300 per annum and $38,000
per annum for models with sibling terms included and excluded respectively.
Source: Australian Bureau of Statistics, 1981 Census of Population and Ifolising, 10;0 House-
hold Sample File.

Downloaded from jir.sagepub.com at NORTH CAROLINA STATE UNIV on March 18, 2015
210

References
Amemiya, T. (1981), ’Qualitative Response Models: A Survey’, Journal of Economic Literature,
XIX, December, 1483-536.
Australian Bureau of Statistics, Census of Population and Housing, 30 June 1981, Household
Sample File; Technical Details, Cat. No. 2167.0.
Baker, R.J. and Nelder, J.A. (1978), The GLIM System Manual, Release 3, Numerical
Algorithms Group, Oxford.
Blandy, R. and Richardson, S. (1982), ’The Fate of the Class of ’71’, Australian Bulletin
of Labour, September, Supplement No. 4.
Brooks, C. and Volker, P. (1983), Labour Market Success and Failure: An Analysis of the
Factors Leading to the Workforce Destinations of the Australian Population. Bureau of
Labour Market Research, Conference Paper No. 38, Canberra.
Broom, L., Lancaster Jones, F., McDonnell, P. and Williams, T. (1980), The Inheritance of
Inequality, Routledge & Kegan Paul, London.
Bureau of Labour Market Research (1983), Youth Wages, Employment and the Labour
Force. Research Report No. 3, AGPS, Canberra.
Inglis, P. and Stromback, T. (1984), ’Migrants’ Unemployment: The Determinants of Em-
ployment Success’, Bureau of Labour Afarket Research, Working Paper, No. 49, November,
Canberra.
Koller, K., Wade, R. and Gosden, S. (1980), ’Youth unemployment, the special case of young
women’, Australian Journal of Social Issues, 15, 1, February, pp. 43-8.
Miller, P.W. (1981), ’How Unemployment Rates Differ: The Influence of Education and
Migration’, Centre for Economic Policy Research, Discussion Paper No. 19, Australian
National University, Canberra.
Richardson, S. (1985), ’Intergenerational Mobility in Labour Force Attainment’, Paper Pre-
sented to the 14th Conference of Economists, May, University of New South Wales,
Kensington.
Vipond, J. (1985), ’Unemployment—A Current Issue in Intra-Urban Inequalities’ in
I. Burnley and J. Forrest (eds), Living in Cities, Allen & Unwin, Sydney.
Windschuttle, K. (1979), Unemployment, Penguin, Ringwood.

Downloaded from jir.sagepub.com at NORTH CAROLINA STATE UNIV on March 18, 2015