RDP 2002-07: An Exploration of Marginal Attachment to the Australian Labour Market 5. Factors Affecting the Transitions of the Unemployed and Marginally Attached
November 2002
- Download the Paper 647KB
5.1 Method and Specification
Without a more sophisticated statistical approach it is not possible to determine the relative importance of factors which may be associated with the labour market transitions of the unemployed and marginally attached. For both these groups, there are four possible labour force states they could be observed in after 12 months. Therefore, the appropriate framework is the multinomial logit model, which allows the dependent variable to take one of four mutually exclusive and exhaustive values j = 1, 2, 3, and 4:
Yit = 1 if person i is employed at time t;
Yit = 2 if person i is unemployed at time t;
Yit = 3 if person i is marginally attached at time t; and
Yit = 4 if person i is outside the broadly-defined labour force (other NILF) at time t.
The multinomial model for respondents who were unemployed at the initial point in time is given by:
For those who were marginally attached at the initial point in time, the multinomial model is given by:
The sample includes respondents who were unemployed or marginally attached at the time of recruitment into the survey, and includes individuals from both the JS and PRG samples. As discussed earlier, the data used in the regression analysis are not weighted because the objective is to understand the relative importance of different factors underlying labour force transitions rather than provide population estimates. The estimates of the transitions from marginal attachment exclude transitions to other NILF since the numbers making this transition are so small that it is impossible to obtain reliable estimates of the determinants of this transition.[9]
These models are reduced form and the estimated effects should not be interpreted as estimates of a structural labour supply model. The specification of the multinomial logit model includes a number of variables which economic theory suggests will be related to labour force status or which previous empirical studies have shown to be important determinants. Care has been taken to exclude potentially endogenous variables, in particular, variables that are likely to change value over time in response to changes in labour force status. All explanatory variables are measured at the wave 1 interview, which is approximately mid-way between the two transition dates. While the details of the construction of the variables can be found in Appendix C, the remainder of this section provides a rationale for the empirical specification used. The omitted category of each set of dummy variables is also listed in Appendix C and summary statistics are provided in Appendix D.
As a starting point, the variable choice is based on the specification used in standard employment equations and labour supply studies. Almost all analysis of employment and labour market prospects control for age, sex, education, geographic factors and family circumstances (including migrant status) as a matter of course. SEUP studies of labour force status also tend to include a control for the effect of individual disability because the data set includes information on this potential impediment to employment. Le and Miller (2000) provide a detailed background to the pertinent literature.
Age is included to capture lifecycle effects and an age squared term is included to allow for a potentially non-linear relationship. The highest level of educational attainment is also included to capture differences in the human capital, which will affect both the chances of finding employment and the probability of participating in the conventionally-defined labour market. The highest level of educational attainment is specified as a set of dummy variables indicating degree or diploma level qualification, vocational qualification, and not having left school prior to completing secondary schooling.[10]
Relationship status, which is included to capture family structure, is measured by whether a person is in a couple relationship, rather than being single, and whether the respondent has dependent children. Differences in the traditional gender roles regarding work and family responsibilities and the implications this has for the value placed on time outside of the work force, means that the effects of relationship status and the presence of dependent children are likely to differ by gender. Consequently we interact family structure and the presence of dependent children with gender.
The discussion of the theoretical literature suggests that local labour market conditions could be an important factor in determining the labour market behaviour of marginally attached workers, particularly those who are discouraged workers. While this literature points to the importance of both the level of and change in the unemployment rate, there appears to be very little change in the local unemployment rates over the period examined.[11] We therefore only use estimates of the level of the local unemployment rate to capture regional differences in labour demand. The 1996 census data are used because they provide the most reliable estimates of small area unemployment rates at a point in time. To allow for labour demand conditions to affect males and females separately, the local unemployment rate variable is also interacted with gender.
Having a disability can severely limit a person's chances of finding employment. We therefore include a variable that measures whether the respondent has a disability. Because the nature of a disability is likely to differ between younger and older people, the disability variable is also interacted with a dummy variable for being aged 45 years or older. Other variables control for whether a migrant comes from either an English speaking background (ESB) or an non-English speaking background (NESB) and the number of years since arrival in Australia – all of which have been found to be important determinants of labour market outcomes (Le and Miller 2000).
The following estimates are based on the transitions over a 12-month period. The determinants of the transitions over 3 months and 24 months reveal similar patterns and can be obtained from the authors on request.
5.2 Multinomial Logit Results
This section presents the results of the estimates of the determinants of labour force status for those who were marginally attached and those who were unemployed at the point of selection into the survey. The validity of the estimated multinomial logit model depends partly on whether the assumption of Independence of Irrelevant Alternatives (IIA) is acceptable. This can be tested using a Hausman test, which suggests that the following models are well specified, at least in terms ofHA (Greene 2000).
As the multinomial logit model results themselves are not straightforward to interpret, the estimated marginal effects are presented for the unemployed sample in Table 7, and the marginally attached sample in Table 8.[12] The marginal effect is usually calculated as the effect of a one unit change in an explanatory variable from its sample average on the probability of being in each of the labour force states after 12 months, holding all other variables at their average value. In the case of binary variables, the marginal effect is the effect of having the characteristic, given that all other variables are at their average value. The marginal effects for each variable sum to zero across the labour market states since each respondent must be in one, and only one, labour force state.
| Employed | Unemployed | Marginally attached | Other NILF | |
|---|---|---|---|---|
| Age | 1.6* | −1.0 | −0.2 | −0.4 |
| Degree or diploma qualification | 7.6* | −9.7* | −0.2 | 2.3 |
| Vocational qualification | 4.1 | −6.1* | −1.8 | 3.9 |
| Incomplete secondary education | −8.1* | 4.8 | 1.2 | 2.1 |
| ESB migrant | 3.8 | −5.8 | 3.5 | −1.5 |
| NESB migrant | −8.6* | 7.0* | 1.5 | 0.1 |
| Year of arrival in Australia | −11.7* | 8.5* | 0.2 | 3.1 |
| Male(a) | −3.5* | 5.0 | 0.2 | −1.7 |
| Male × couple family | 2.8 | −2.6 | 0.5 | −0.8 |
| Male × dependent children | 1.9 | 1.4 | −1.6 | −1.6 |
| Male × local unemployment rate | −0.6 | 0.7* | 0.0 | 0.0 |
| Female(b) | 3.1 | −4.9 | −0.3 | 2.2 |
| Female × couple family | 1.0 | −7.8 | 4.1* | 2.7 |
| Female × dependent children | −7.9* | −5.8* | 2.6 | −0.6 |
| Female × local unemployment rate | −1.0* | 0.7 | 0.4 | −0.1 |
| Younger × has a disability(c) | −12.1* | 8.6* | 1.9 | 4.5* |
| Older × has a disability | −14.0* | 6.2 | 1.7 | 6.1* |
| Probability | 43.6 | 42.7 | 5.7 | 7.9 |
|
Notes: Marginal effects are derived from the estimates of the determinants of
labour force status and are calculated using numerical methods (see Stata 2001a,
pp 333–334). * indicates that the marginal effect is statistically
significant at the 5 per cent confidence level. |
||||
For the results reported in Tables 7 and 8, the marginal effects for the variables interacted with gender are calculated as the effect of changing the characteristic given that all other variables are set to the average values for the male or female sample as appropriate. The marginal effects of the interaction terms for disability are calculated from the average of the younger and older samples respectively.
| Employed | Unemployed | Marginally attached | |
|---|---|---|---|
| Age | 1.7* | −1.4 | −0.3 |
| Degree or diploma qualification | 4.7 | −5.8 | 1.1 |
| Vocational qualification | −10.0 | 7.2 | 2.7 |
| Incomplete secondary education | −12.6* | −2.2 | 14.8* |
| ESB migrant | −7.8 | −0.9 | 8.7 |
| NESB migrant | −6.8 | −3.2 | 10.0 |
| Year of arrival in Australia | −2.1 | 16.2 | −14.1 |
| Male(a) | −41.1* | 6.9 | 34.3* |
| Male × couple family | 12.0 | −1.7 | −10.2 |
| Male × dependent children | −12.1 | −7.3 | 19.4 |
| Male × local unemployment rate | 0.7 | 1.8 | −2.5 |
| Female(b) | 15.5* | 5.1 | −20.6* |
| Female × couple family | −6.0 | −1.6 | 7.6 |
| Female × dependent children | −14.2* | 2.1 | 12.1 |
| Female × local unemployment rate | −1.3 | −0.0 | 1.4 |
| Younger × has a disability(c) | −4.7 | −5.2 | 10.0 |
| Older × has a disability | 9.6 | −12.4* | 2.8 |
| Probability | 21.2 | 15.4 | 63.4 |
| Notes: Marginal effects are derived from the estimates of the
determinants of labour force status and are calculated using numerical methods (see
Stata 2001a, pp 333–334). * indicates that the marginal effect is
statistically significant at the 5 per cent confidence level. (a) Marginal effects calculated using the averages of the male sample. (b) Marginal effects calculated using the averages of the female sample. (c) Marginal effects calculated using the averages of the younger and older samples respectively. |
|||
As an example of the interpretation of the marginal effects, consider the effects of being one year older than the average person in the sample. This raises the probability of moving from unemployment to employment by 1.6 percentage points. Having a degree or diploma qualification decreases the probability of still being unemployed after 12 months by 9.7 percentage points. Being male increases the probability of still being unemployed by 5 percentage points, assuming that all other characteristics are at the sample average for the male sub-sample. The marginal effect of being female would be equal and oppositely signed to the marginal effect of being male if they were evaluated at the same sample averages. However, if this marginal effect is evaluated at the average characteristics of the female sub-group, it is slightly different at −4.9 percentage points.
While educational attainment is a major determinant of transitions of both the unemployed and marginally attached, the pattern differs. For the unemployed, the overall pattern is that an increase in educational attainment significantly increases the chances of moving into employment and significantly decreases the probability of remaining unemployed. For the marginally attached, it is only the lack of a complete secondary schooling that has a statistically significant negative effect on becoming employed, although point estimates suggest that there is a positive relationship between education and employment prospects.
For the unemployed, having a degree or diploma level qualification decreases the chances of remaining unemployed after 12 months by 9.7 percentage points and significantly increases the chances of becoming employed. Having a degree or diploma level qualification is found to have no statistically significant effect upon the labour force transitions of the marginally attached, although the point estimates suggest that there are economically significant effects that are roughly half the size of those for the unemployed sample.
For the unemployed, having a vocational qualification is estimated to decrease the chances of remaining unemployed by 6.1 percentage points, increase the chance of moving to employment by 4.1 percentage points, and increase the chances of becoming other NILF by 3.9 percentage points. As with higher-level qualifications, having a vocational qualification has no statistically significant impact on the labour force transitions of the marginally attached, although the estimates suggest that vocational qualifications have an economically important positive effect on the probability of becoming unemployed and an economically significant negative effect on the probability of becoming employed.
Having an incomplete secondary education (and no post-secondary educational qualifications) is estimated to decrease the chances of moving from unemployment to employment and to increase the chances of remaining unemployed, although these effects are not significant. For the marginally attached it is also estimated to decrease the chances of moving to employment by 12.6 percentage points and to increase the chance of remaining marginally attached by 14.8 percentage points. In contrast, the level of education has no effect on the transition from unemployment to marginal attachment.
For the unemployed, being a migrant from an ESB country is found to have no statistically significant effect, but being a migrant from a NESB country is found to decrease the chances of moving to employment by 8.6 percentage points and to increase the chances of remaining unemployed by 7 percentage points. Year of arrival in Australia is found to have quite a strong effect on the labour force transitions of the unemployed, with more recent arrivals being more likely to remain unemployed and less likely to move to employment. For the marginally attached, being a migrant is found to be unrelated to labour force transitions, with the controls for being a migrant from an ESB or a NESB country being statistically insignificant. Year of arrival in Australia is also found to have no effect.
Gender does not appear to have a statistically significant direct effect on the probability of moving from unemployment to other labour force states. However, marginally attached males are 34.3 percentage points more likely to remain marginally attached and are 41.1 percentage points less likely to be employed than if they were female. On the other hand, females are 15.5 percentage points more likely to gain employment and 20.6 percentage points less likely to remain marginally attached.
Relationship status appears to be more important for unemployed females and the marginally attached of both genders than it is for unemployed males. Unemployed females who are in a couple are significantly more likely to become marginally attached and are less likely to remain unemployed than females who are not in a couple, although the latter effect is not statistically significant. Marginally attached males in a couple are more likely to become employed and are less likely to remain marginally attached than their single counterparts, and this effect is reversed for marginally attached females in a couple. Again, while these effects are economically significant, they are not statistically significant. Overall, this suggests that females in a couple have a stronger tendency to be marginally attached, which is consistent with the idea that females in a couple are often the second earner.
Unsurprisingly, the effect of having children for females works in the same direction. Females with dependents are more likely to remain marginally attached, or become marginally attached if they are unemployed, at the expense of becoming employed, than females with no dependents. The presence of dependents has a minimal effect on the labour market transitions of unemployed males, but marginally attached males with dependents are significantly more likely to remain marginally attached, and are less likely to become employed, than males without dependents.
Local labour market conditions have a significant but apparently limited impact on the labour force transitions of the unemployed. For both males and females, an increase in the local unemployment rate of 1 percentage point from the average slightly increases the probability of remaining unemployed by 0.7 percentage points and reduces the probability of moving from unemployment to employment. The only statistically significant effect of local labour market conditions on the transitions of the marginally attached is to reduce the probability of males remaining marginally attached by 2.5 percentage points. There is also a small increase in the probability of moving from marginal attachment to employment, which may reflect that fact that marginal attachment is a broader concept and that these individuals are more affected by changes in personal circumstances and incentive structures that are not as closely tied to labour demand conditions.
While the effects of local labour market conditions are found to be quite small, this may also be a product of the relatively indirect measure of labour market conditions used. Analysis using more detailed geographic information may lead to finding an impact from local labour market conditions, but this is not possible using the public release SEUP data set. As mentioned above, the SEUP data were collected over a period of fluctuating employment growth following a recession. Both the labour market dynamics and the determinants of the transitions may differ if the labour force transitions were considered at a different point of the macroeconomic cycle.
As expected, having a disability significantly reduces the prospects of the unemployed moving into employment. Younger disabled people are 12.1 percentage points less likely to become employed, while older disabled people are 14 percentage points less likely. Offsetting this, both groups are more likely to remain unemployed and to leave the labour force entirely (NILF), and the magnitudes of these effects are similar across the two groups.
For the marginally attached, the only statistically significant transition is that older disabled people are 12.4 percentage points less likely to enter unemployment. This is offset by an economically large marginal effect of 9.6 percentage points of entering employment. This suggests that older disabled respondents may have left the work force due to an injury or illness, which was preventing them from working temporarily. For example, being on workers' compensation or leave without pay while recovering from a disability means that they will not be looking for work, and hence are excluded from the conventionally-defined labour force, but are likely to return to their old job. In contrast, younger disabled people who are marginally attached are 10 percentage points more likely to remain marginally attached and have a roughly equal decreased probability of moving into employment or unemployment.
Footnotes
Of the respondents who were marginally attached on 1 June 1995 only 15 were other NILF on 1 June 1996. [9]
There are a small number of respondents who are participating full time in education. We have no other information on the educational attainment of these individuals. They are coded as having incomplete secondary education. While this may introduce some error into the estimates, the small numbers of such respondents will mean that any biases are small. [10]
Estimates of changes in the local unemployment rate are derived using the ABS Labour Force Survey and the former Department of Social Security data. See Appendix C for further details on the construction of this variable. The seeming contradiction of lack of temporal variation in local unemployment rates and changes at the macro level may reflect the experimental nature of our estimates of the former. While the methodology is analogous to that used by the Department of Employment and Workplace Relations (DEWR) in estimating the labour market conditions in statistical local areas, the process of averaging out unemployment rates within the respective deciles of socioeconomic status may highlight the unreliability of the derived estimates. [11]
The parameter estimates are presented in Appendix E. [12]