RDP 1999-07: Job-Search Methods, Neighbourhood Effects and the Youth Labour Market 4. Estimation Framework and Results

The dependent variable for the analysis below indicates which of the three alternative job-search methods has been chosen by the unemployed respondent. Because the choices of main job-search method are mutually exclusive and exhaustive, it is necessary to use an estimation technique which imposes this restriction. Labelling the choices available to an individual from 1 to M, the multinomial logit specification defines the probability of choosing option m as:

There will be M such probabilities, which will sum to unity, as required. The coefficients will not be uniquely identified, however, without a further restriction. The standard restriction for multinomial logit models is to assume that the vector of coefficients for one alternative, known as the base category, is normalised to zero. The normalised coefficients are then interpreted as the effect of a given characteristic on the probability of choosing a given job-search method, relative to its effect on the probability of choosing the base category. For example, if the base category is choosing to search through newspapers, a positive coefficient on age for the CES category indicates that older people are more likely to choose to search using the CES than to look in newspapers.

Because multinomial choice models are non-linear functions of their coefficients, the estimated coefficients provide information about the direction of the effect, but not about its size. The size will depend on the values taken by all the independent variables. Another way of presenting the results so that they are more meaningful is to express them in terms of the marginal effect a given variable has on the probability of choosing an outcome, given that all other variables are set to their mean value. In terms of Equation 1, this can be written as:

The coefficients on dummy variables can be interpreted as the change in the probability of observing a given outcome if the value of the dummy variable is changed from zero to one. For example, in Table 4, the variable male takes the value one if the individual is male and zero if female. Therefore, being male increases the probability that the individual will use the CES as their main job-search method by 6.8 percentage points, all other variables held at the average values. Dummy variables are indicated by an asterisk. For continuous variables such as age, the coefficient is interpreted as the increase in the probability of the outcome if the explanatory variable increases by one unit. Therefore, if age increases by one year, the probability of using the CES as the main job-search method increases by 5.6 percentage points.

Table 4 presents the results of estimating the multinomial logit model for the choice of main job-search method. It should be noted that section of state, state and year indicators were included in the estimation, but have not been reported to minimise the volume of results. The sample includes all respondents who were unemployed at the time of the interview and provided information about the main job-search method being used. For each choice, the first column presents the marginal effects calculated at the sample averages, and the second presents the t-statistics.

Gender and age are both important for explaining the observed job-search behaviour of unemployed teenagers. Males are 4.9 per cent more likely to choose direct methods, and are 6.8 per cent more likely to choose the CES than females. Because the predicted probabilities must sum to unity, this implies that females are 11.6 per cent more likely than males to use newspapers as their main job-search choice. Older teenagers are more likely to use direct methods and are less likely to use the CES.

Parents' characteristics have some effect on observed job-search behaviour. Unemployed teenagers are more likely to use direct methods and are less likely to use the CES as the occupational status of their parents, especially their father's, job increases. If either parent was not employed when the respondent was 14 years old, the respondent is more likely to use direct search methods and less likely to use newspapers. The effects of parents' education are difficult to interpret. More educated mothers, however, generally have children who are more likely to use direct search methods, and are less likely to use the CES.

Education and employment histories are also important determinants of the main job-search method. Respondents who attended government schools and/or left school in Year 10 or earlier are significantly more likely to use the CES as their main job-search method. The combined effect of having both these characteristics is to increase the probability of choosing the CES by 10.6 percentage points.

Teenagers who have been in the labour force for longer are more likely to be searching through the CES and are less likely to be using newspapers. Having controlled for the years since leaving school it is interesting that current unemployment duration significantly increases the probability that newspapers are the main job-search method and decreases the probability of using direct search methods. This provides some limited support for the hypothesis that the unemployed are less likely to be observed using direct methods because the expected benefits of these methods diminish over the duration of unemployment.

Perhaps the most important single variable for explaining the choice of main job-search method is the indicator for unemployment benefit receipt. Individuals receiving unemployment benefits are almost 20 percentage points more likely to use the CES as their main method of job search. One explanation for this is that individuals receiving unemployment benefits are required to demonstrate that they are looking for work and registering with the CES offers an easy way of doing this. However, even if this were the case, the CES would not necessarily be reported as the main method of search.

Although receiving unemployment benefits would be expected to increase the flow of benefits to being unemployed all else being equal, unemployment benefits in Australia are subject to a means test. Therefore, individuals who receive benefits are likely to come from more financially constrained backgrounds and the net effect of these two financial considerations could easily be that benefit recipients have lower flows of income while unemployed on average. In light of this, another explanation for the significant effect of benefit receipt on the probability of using the CES is that eligible individuals have a lower flow of benefits to being unemployed on average. Following the model outlined in Appendix A, this would lead eligible individuals to search harder, increasing the probability that they are observed using indirect methods. However, this explanation also implies that the probability of observing that newspapers are the main method of job-search should be higher for eligible individuals. In fact teenagers receiving unemployment benefits are less likely to be observed using newspapers by around 8.6 percentage points.

The two explanations for the significance of the effect of benefit receipt on job-search behaviour are not inconsistent, and there is likely to be some truth in each. However, the argument for the work-test explanation is perhaps the most consistent with the effects of eligibility on the probability of choosing other job-search methods.

Two neighbourhood characteristics appear to be important. As hypothesised, a higher local unemployment rate decreases the probability that an unemployed teenager will choose direct search methods. An increase in the local unemployment rate of one percentage point will decrease the probability of using direct search methods by 1.1 percentage points. The marginal effect of a one percentage point increase in the local unemployment rate on the probability of using either the CES or newspapers is around 0.5 of a percentage point. Consequently, the degree of competition for jobs at a local level and the lack of access to a local job-information network, as proxied by the local unemployment rate, can help explain why unemployed teenagers are less likely to be observed using direct search methods although they have proved to be the most successful methods of finding work for employed teenagers.

The other significant neighbourhood characteristic is the average level of personal income. Given that we have controlled for an extensive array of background characteristics as well as the proportion of the neighbourhood with either academic or vocational post-school training, it is puzzling that coming from a neighbourhood with higher average personal income seems to reduce the use of direct methods. Since high income is likely to be correlated with unobserved characteristics measuring success, the average level of personal income in the neighbourhood might have been expected to have the opposite effect.