RDP 2002-04: Labour Market Adjustment in Regional Australia 5. Explaining Regional Migration in Australia

While employment has grown in some regions and contracted in others, the final unemployment outcome was previously shown to be dependent on inter-regional migration. In this section, we discuss some economic characteristics of regions that may influence regional migration. We then attempt to assign a probability to these characteristics being present in Australian regional labour markets.

5.1 Factors Influencing Regional Migration

While many non-economic factors will influence inter-regional migration, the decision to relocate is not taken independently of economic factors. Consequently, in the standard model of regional migration, an individual will choose to relocate if there is ‘net economic advantage’ in doing so.[32] One of the most influential studies is by Harris and Todaro (1970), who emphasise that migration is dependent on relative wages, relative employment prospects, and housing and relocation costs, so that these factors are typically considered in addition to non-economic factors. Based on the small area data available to us, we nominate a set of factors that may be expected to influence regional migration.[33] In fact, many of the factors considered in our explanation of regional employment growth also have bearing on regional migration, primarily through their effect on relative employment prospects.

Unemployment rates are a key indicator of employment prospects, so we might expect out-migration to be more evident from regions with initially high unemployment rates, and in-migration to occur to those regions with initially low unemployment rates (Greenwood 1975; Debelle and Vickery 1998).

Similarly, the extent to which the industry composition of employment in a region is changing may influence employment prospects. Diversity may also have bearing on migration decisions, with diverse regional economies, which are less vulnerable to industry-specific shocks, being more able to attract or retain people than specialised regions.

The age and skill level of a region may influence migration, with some researchers claiming that younger, more educated people will have a greater tendency to migrate when subjected to negative shocks (Greenwood 1975). This is because the gains from relocations are likely to be greater for those with higher levels of skill or a longer expected working life over which gains can be realised.

Finally, regional amenity and access may also affect the incentives to migrate by raising the level of utility derived from a location (Glaeser, Scheinkman and Schleifer 1995). For example, if the probability of finding work in two regions is identical, an individual may prefer to relocate to the region with high amenity or access to markets.[34]

5.2 A Modelling Strategy

Regional labour market outcomes were characterised as falling into four quadrants: two that displayed the expected inverse relationship between employment growth and unemployment, and two that did not (as outlined in Figure 5) due to inter-regional migration. These quadrants are mutually exclusive states, and we are interested in the characteristics that influence the probability of a region being in a given quadrant. A multinomial logit specification defines these probabilities as:

where one of the alternatives, the base category, has Bj = 0 as a normalisation restriction. The estimated equations then provide a set of probabilities for the j choices for a region with characteristics X.

In order to directly compare regions in which the migration response differed, we estimate the multinomial logit model using two separate base categories. In the first case, we choose Quadrant 1 as the base category and ask, ‘which regional economic characteristics influenced the relative strength of out-migration?’[35] From Equation 4 we can derive the following odds ratio for Case 1:

In the second case, we choose Quadrant 3 as the base category and ask, ‘which regional characteristics influenced the relative strength of in-migration?’ The odds ratio for Case 2 is:

5.3 Results

Table 3 presents the results from the estimation of the multinomial logit model for Cases 1 and 2. In Section 3.3, we showed that for most regions, differences in the strength of the migration response to shocks determined the path of unemployment. Consequently, in Case 1, we interpret an odds ratio greater than unity as indicating that a 1 percentage point increase in that variable increased the odds that a region had a larger adjustment through out-migration. In Case 2, we interpret an odds ratio greater than unity as indicating that a 1 percentage point increase in that variable increased the odds that a region had a larger adjustment through in-migration.

Table 3: Results from the Multinomial Logit Models
Variable Case 1: Out-migration
Quadrants 1 and 2
  Case 2: In-migration
Quadrants 3 and 4
  Odds ratio Significance   Odds ratio Significance
Coastal 0.40 **   3.10 ***
Size 0.89   1.04  
Remoteness 1.24 ***   0.80 ***
Unemployment 1.34 ***   0.69 ***
Aged 0.83 ***   1.16 ***
Structural change 0.91 ***   1.04 **
Diversity 1.14   1.06  
Human capital 0.80 ***   0.96  
Number of observations = 637
LR Chi2 (24) = 405.1
Probability that the LR > χ2 = 0.00
Pseudo R2 = 0.24
Note: ***, ** and * represent significance at 1, 5 and 10 per cent levels.

The results for Case 1, which seeks to explain influences on out-migration, illustrate some strong implications of the model. They suggest that high initial unemployment rates are an important factor increasing the probability that a region experienced out-migration. This is an important result. It indicates that after we control for the dominant effect of the direction of employment growth, inter-regional migration does play a role in narrowing unemployment differentials.

Low access to markets and low regional amenity also emerged as significant factors increasing the likelihood of leaving a region. And regions with younger populations were also more likely to adjust through out-migration than those with older populations. Each of these results accords with the reasons for inter-regional migration advanced in the literature.

However, regions with fewer skilled workers, which we interpret as relatively low human capital, had stronger rates of out-migration than others. At first glance, this appears inconsistent with the expectation that educated people are the most mobile. One interpretation follows Glaeser and Shapiro (2001), who argue that skilled workers are more likely to leave regions with low levels of human capital than other regions, because of diminished expectations of future growth and employment opportunities.

Regions with lower rates of structural change also had stronger rates of out-migration. This result is consistent with our earlier findings that regions with low rates of structural change also tended to experience contractions in employment because new growth industries did not emerge, and suggests that people are more likely to leave regions where this type of structural change has not occurred.

The results for Case 2, which seeks to explain influences on in-migration, are, broadly speaking, the flip side of those for Case 1. Just as people were likely to leave regions with high initial unemployment, they are likely to move to regions of low unemployment. Similarly, they are more likely to move to accessible, high amenity regions. Regions with an older population in 1986 subsequently had higher rates of in-migration, possibly reflecting a role for retirement related migration.

Again, in the obverse of Case 1, higher rates of structural change were associated with stronger in-migration, perhaps suggesting that rapid structural change was associated with more employment opportunities in the future. However, unlike Case 1, we could not find a statistically significant role for regional human capital influencing relative rates of in-migration.

Finally, it is worth remembering that our focus in this section has been on assigning probabilities to the influence of initial characteristics on migration patterns between 1986 and 1996, rather than on events that occurred during this period. Both of course will have had a bearing on migration patterns. For example, we know that employment declines will have been driven by long-run structural factors in some regions (such as rising labour productivity in broad-acre farming), and by cyclical factors (such as the early 1990s recession) in others. However, such differences in the reason for and timing of employment declines may have different implications for the strength of the migration response. This should be borne in mind when interpreting the results.


As argued by Hicks (1932) in his assessment of migration patterns. [32]

We do not discuss a role for the relative cost of housing or relative wages, both of which may influence regional location decisions, because the appropriate data are unavailable for SLAs. [33]

In the regional science literature, differences in regional amenity drive a wedge between regional unemployment rates. For example, in regions with high unemployment, attractive physical and cultural characteristics may compensate for reduced employment opportunities, reducing the incentives for out-migration. [34]

This inference follows naturally from our analysis in Section 3.3, where we demonstrated that it was differences in the strength of the migration response that most often determined the path of unemployment following a shock. However, it should be remembered that the statistical question that our model is addressing is, ‘which regional characteristics influenced whether a region's unemployment rate increased or fell when the level of unemployment fell?’ [35]