Migration, labour demand, housing markets and the drought in regional Australia
- Executive summary
- Theoretical model of migration
- Background: Population dynamics in rural and regional Australia
- What is a drought?
- Selected data issues
- Labour market, housing market, migration and drought
- Descriptive analysis of relationships between recent drought and migration
- Multivariate analysis of out-migration, in-migration and net migration
- Drought and mobility in the RRFS
- Lists of tables and figures
- Appendix A: Gross migration flows across drought categories (total population)
- Appendix B: Gross migration flows across drought categories (all workers in the agricultural industry)
- Appendix C: Regression results
Multivariate analysis of out-migration, in-migration and net migration
This section uses data from the 2006 Census of Population and Housing to calculate in-migration, out-migration and net migration between 2001 and 2006. These variables are constructed based on the "usual residence 5 years ago" question and are related back to the 2001 geographic boundaries, using area-based concordances provided by the ABS.
In order to analyse the factors associated with migration, we construct a series of summary data for each SLA from the 2001 Census. The relationship between these variables and both the human capital and modified gravity models of migration is covered in the description of the results. Descriptive statistics for data used in the regression analysis are reported in Appendix C, Table C1. The variables used are:
- index of expected employment growth for 2001-06 (%) - based on the expected employment growth based on a weighted average of sectoral employment growth (measured at a national level), conditioned on industrial composition of an SLA in 2001;
- unemployment rate (% of labour force);
- PCA index of housing payments, calculated using the weekly rental data and monthly housing repayment data from the Census;
- percentage of employed residents in public sector;
- percentage of non-school population aged 15 years and over who completed Year 12;
- percentage of population identified as Indigenous (i.e., excluding those who did not respond to that question);
- percentage of families that have children aged 0 to 15 years;
- percentage of population in various age groups (10-year categories and 55 years and over);
- state indicators; and
- the ARIA++ indicator - the ABS Accessibility/Remoteness Index of Australia (ARIA) was further developed for the Western Australian Aboriginal Child Health Survey (WAACHS) into the ARIA++ index (Zubrick, et al., 2004), which provides a finely disaggregated measure of the accessibility of services in an area.
The above analysis provided an important background overview of labour and housing markets in drought-affected areas; however, all multivariate analyses need to also provide detailed information on the variables and sample used, as this assists in interpreting and understanding the reported results. Table C1 reports the descriptive statistics for all 1,320 SLAs used in the regression analysis (all defined on the boundaries used in the 2001 Census).
Regression analysis of migration, 2001-06
The statistical model used to summarise the relationship between drought and migration across SLAs uses ordinary least squares (OLS). Even though out-migration is bound between 0% and 100%, all measured rates are well within these bounds. The mean level of out-migration is 35% of the original resident population, with a standard deviation of around 14%. The mean measures of in-migration are similar to that for out-migration, although the standard deviation is somewhat higher because more than 100% of the original population can move into an area. Technically, net migration is not bounded because it can take on both negative and positive values. The mean net migration for 2001-06 was 1%, with a standard deviation of 24%. Robust standard errors are used for the analysis in this section.
The omitted categories for the regression analysis are non-drought, accessible SLAs and the state of NSW, with the percentage of the local population aged between 35 and 54 also being excluded. Readers should note that we have chosen the non-drought areas as the base category as these areas are included in the RRFS and hence the following conclusions have an analogous interpretation in that context. Consequently, the findings are directly relevant for interpreting that survey.
Appendix C, Table C2 reports the regression results for out-migration (in 2001-06), while Tables C3 and C4 focus on in-migration and net migration respectively (also for the last inter-censal period). The first two columns of those tables report the coefficients for controls; the next two columns report the coefficients for the drought categories only (measured using the 2003-06 data); and the final two columns report full models that include both controls and drought categories for the three years leading up to the last three censuses.
The main message from Appendix C, Table C2, is that drought variables are associated with lower out-migration - even after other regional factors are accounted for. Employment demand for workers in an SLA is captured by the expected growth of jobs available. In Table C2, increasing labour demand increases out-migration, but the result is only significant for the model that controls for drought during the 2003-06 period. The result can be rationalised because people with jobs will tend to have sufficient resources to move out. Higher local unemployment rates are associated with higher out-migration; however, this out-migration coefficient is only significantly higher when drought is also controlled for (and then only at the 10% level). Public sector employment is also significantly positively associated with out-migration (irrespective of whether the drought categories were included in the regression). Not only do public sector jobs have a wage premium compared to other employment, but large employers such as governments are more likely to have work options for individuals in other parts of the country, thereby directly affecting the capacity for out-migration. Interestingly, education levels (as captured by the proportion with Year 12 education) are significantly associated with higher out-migration rates. However, having more Indigenous people in an SLA is significantly associated with lower out-migration rates. This observation is consistent the fact that Indigenous people tend to be more likely to move within the local area and less likely to move between such areas (Biddle & Hunter, 2006).
As expected, the areas with a higher proportion of families with children aged 0 to 15 years tended to have lower out-migration rates. This is likely to be related to the costs of migration in that it is more difficult and disruptive to organise migration for dependents, such as when making new schooling arrangements. Older age groups (aged 55 and over) were also associated with lower out-migration rates, presumably because of the higher economic, social and psychic costs associated with moving for this group; for example, the opportunity costs of higher wages associated with age and the costs of moving away from social networks established over a longer period.
Another set of explanatory variables is the group of variables that control for the residual geographic factors not elsewhere accounted for. The patterns across state boundaries are consistent with expectations (e.g., Tasmania has low out-migration). Remote and very remote areas seemed to have higher out-migration, probably due to the lack of social connection that many (non-Indigenous) people have with such areas.
A particularly noteworthy finding is that despite the apparently, systematic difference in housing prices across drought areas noted earlier, there is no systematic relationship between the effect of housing payments for any of the regressions in Table C2. If housing prices are playing a role, it may be being picked up by one of the other controls (but not the drought variables, as housing payments are not significant even if such variables are omitted). However, it is also possible that drought causes endogenous falls in housing prices that are only perceptible in later censuses. That is, housing prices may be caused by drought rather than vice versa.
Overall, there appear to be some important interactions between drought and the labour market data - especially labour demand, as this factor only becomes significant when the incidence of drought is controlled for. The converse of this is that the effect of drought is enhanced by controlling for local labour demand. Given that the recent drought is rather pervasive, and therefore has even affected formerly prosperous areas, this observation is understandable.
The analysis of in-migration between 2001 and 2006 is presented in Appendix C, Table C3. Overall, drought regions are associated with lower in-migration rates than non-drought areas, but the effect is only significant at the 5% level after all the regional controls are taken into account.
Having a buoyant local labour market, as captured by the expected job growth in an SLA, is significantly associated with higher rates of in-migration. Obviously, having some available jobs is an attraction for migrants to move into an area. The unemployment rate does not significantly affect in-migration, but completion of Year 12 education was a significant factor in explaining in-migration once local labour market conditions were controlled.
Areas with higher rents and housing payments (as indicated by the PCA index) were also associated with higher in-migration. Given that the main asset of most Australians is their home, it is not surprising that people are attracted to areas with buoyant housing markets where housing prices might be expected to increase in future.
In contrast to the out-migration analysis, having more jobs in the public sector was not a significant factor driving in-migration. While government jobs generally enjoy relatively good pay and conditions, the size of the sector has not grown significantly in recent decades (Commonwealth of Australia, 2007), and in some areas the proportion in such employment may have declined. Hunter (2007a) documented the substantial overall decline in public sector employment since the early 1980s. Consequently, if people wanted to move to find work, they would be more likely to be following employment in the private sector.
The proportion of the SLA population who identify as Indigenous was negatively associated with in-migration in 2001-06. As well as reflecting the generally lower rates of mobility for Indigenous Australians between SLAs, another factor may be that non-Indigenous people may be reluctant to move into areas where there are large numbers of Indigenous Australians. This reluctance may vary from a mild preference for living with people "like" yourself to outright discriminatory attitudes. However, both sets of attitudes can lead to highly segregated outcomes, as was demonstrated in Schelling's (1978) famous model based on moving black and white pieces around a chess board. Note that the desire to live with similar people is embedded in the modified gravity model of migration.
Demographic factors did throw up some potentially unexpected results for in-migration. While younger age groups (under 25 years old) are associated with higher in-migration rates, areas with more families with children are associated with significantly lower in-migration. One interpretation of this is that once the differences in resources are taken into account (e.g., the buoyancy of the local labour market), the location of educational institutions (which are presumably located where the children live), does not attract additional in-migrants on average. The dominant factor is that, for obvious reasons, it is more costly to move families with dependents.
The accessibility index has a negative association with in-migration in the more remote areas. This is easily explained by the fact that people have fewer reasons to move to an area that is further away from important infrastructure and services.
Table C4 reports the net effect of in-migration and out-migration taken together (for 2001-06). Again we focus on the full model with controls first before providing a detailed scrutiny of the apparent effects of living in drought effects areas, vis-à-vis other SLAs. Overall, net migration follows fairly similar patterns to those identified in the results for in-migration; however, minor differences do arise when factors associated with out-migration are at odds with the in-migration results. A buoyant local labour market is associated with higher net migration, and once this is taken into account, neither the unemployment rate or education factors are significant at the 10% level. SLAs with more Indigenous residents are associated with lower net migration, while younger age groups (under 25 years old) are associated with significantly higher net migration rates, and areas with more families with children are associated with significantly lower net migration, presumably for similar reasons to that identified in the discussion of Table C3. Less accessible areas tend to have significantly lower net migration rates.
Overall, Appendix C illustrates that the controls for observable regional factors overwhelmingly have the effect predicted by the human capital theories of migration referred to above and hence it is crucial that our assessment of the association between drought and migration takes this into account. Table 6 summarises the apparent effect of drought on out-migration, in-migration and net migration for all the drought periods examined in this report: 1993-96, 1998-2001 and 2003-06 (but in reverse temporal order). Given that it is important to control for observable regional characteristics in order to get some sense of the true association between drought and migration patterns, the ceteris paribus assumption is used; that is, other regional factors are held constant. Drought areas are compared with non-drought SLAs defined for the three drought periods examined in this report. The 2003-06 drought areas had lower out-migration and in-migration rates than non-drought areas, but the net effect on migration was not significant for either drought or below average rainfall areas. In the short run, there was no effect of drought on net migration. While the overall level of population was not affected much by the incidence of recent drought, the composition of the resident population may have changed as the types of people moving out and moving in are likely to have beeen systematically different.
|Drought periods 2003–06|
|Drought periods 1998–2001|
|Drought periods 1993–96|
Notes: Standard errors are reported in parentheses. Full regression results are reported in Appendix C for the 2003-06 drought categories; however, it should be noted that the control variables for the other drought categories for 1993-96 and 1998-2001 have a similar relationship with the various aspects of migration reported in this table.
Drought measured using 1998-2001 data was not significantly related to net migration between 2001 and 2006; however, drought and below average rainfall areas again had significantly lower out-migration and in-migration rates.
In the longer run, drought may have a greater effect on the resident population. Table 6 shows that if one focuses on the drought categories for the 1993-96 period, the negative effect on in-migration is relatively strong, but the effect on out-migration is not significant. Living in one of the 22 SLAs that experienced a drought between 1993 and 1996 was associated with lower net migration. In contrast, there was no significant effect of living in below average rainfall areas on any of the three forms of migration.
The significant difference between the effect of living in drought versus below average rainfall areas can be seen in several entries in Table 6. This finding provides an ex poste rationalisation for the decision to distinguish between drought and below average rainfall areas in the sample design of the RRFS (Hunter, 2007b).
Given that the changes in in-migration and out-migration seem to balance out for more recent droughts, it is plausible that the analysis of RRFS data can ignore migration issues, especially if the analysts control for other relevant observable regional characteristics. However, in the long run, drought is more likely to feed back on the regional economy and hence the regional population. While it is arguable that, on average, the feedback between drought and population can be ignored in the short run, it is fairly clear that there were substantial changes in the composition of the local population as a result of drought. Some of the preliminary analyses from Edwards, Gray, Hunter, and De Vaus (2008) are presented in the next section to illustrate some of the relevant issues.