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
Labour market, housing market, migration and drought
There are theoretical reasons for thinking that the labour and housing markets have direct impacts on migration and hence it would be useful to understand how such factors can be measured when attempting to analyse the relationship between drought and migration. Ultimately, the objective of this report is to document the differences and changes in such factors in areas associated with drought vis-à-vis other areas. Since we have some measures of drought back to the early 1990s it may also be possible to examine whether the labour and housing markets became depressed after the experience of drought, or indeed in the lead up to the more recent experiences of drought. Just as we must rely on Census data for migration, we are confined to using Census data to measure the local labour and housing markets as other sources of such data are unreliable. The SLA level is again used as the geographic analysis as this provides reliable data. This decision has the added advantage that it potentially simplifies the interpretation of results. That is, there should be no issue with the fallacy of composition (sometimes called the "ecological fallacy")4 as all geographic data is measured at the SLA level.
Measuring local labour market changes
What is the best way to measure labour market conditions in the context of drought? Census data on unemployment rates provide a direct measure of excess supply of labour in the local labour market. Labour demand can also be measured using changes in regional employment by industry sector, also using Census data. Shift-share analyses are commonly used by geographers to decompose regional employment growth into several components: aggregate growth, growth associated with the industry mix of an area and a residual growth component (see Karmel, McHugh, & Aungles, 1993); the industry mix component of such analysis is closely related to many popular indices of labour demand and hence provides a useful precedent for this report (see Katz & Murphy, 1992). Rather than provide the unnecessary complication of conducting a full shift-share, we simply estimate the expected employment growth for an area given the initial industry structure (at the beginning of the period of interest) and overall growth in industry employment observed at a national level.
The estimates of local employment demand used in this report rely heavily on the assumption that the industrial composition of a group is not changing much over time. If we relax this assumption, then one can estimate a direct measure of sectoral change, the so-called turbulence index that was mentioned above (in the "What is a drought?" discussion). The turbulence index is defined as half the sum of the absolute changes in the share of employment in particular industries (i.e., 1/2 * Σ/Δ[Ei/E]|, where Ei is the employment levels in a particular area, E refers to the aggregate employment levels and Δ Σ is the change in employment levels between two periods). It is a direct measure of sectoral change and an indirect measure of spatial mismatch that enables us to understand how structural change varies across socio-economic status regions (Driver & Dunne, 1992).5
The overall turbulence index can be interpreted as indicating what fraction of all jobs in the region has changed industry. The differences in the magnitude of these indices indicate the proportion of the employed who would have to move across sectors in the later period to restore the region to the industry structure evident in the earlier period. Since this index is additive, the component attributable to the agricultural sector represents the absolute change in agricultural participation as a proportion of all employment change.
Measuring the local housing market
Affordability of local housing is driven by house prices in the local area and the prevailing interest rate charged by financial institutions. Interest rates vary over time with macroeconomic policy, basically affecting all housing markets in a similar fashion, and hence can be largely ignored when focusing on regional variation in housing markets.6
In the absence of consistent local data on real estate sales, housing markets must be measured using a proxy derived from the Census. Censuses do not include direct information on housing prices, but it does include information on weekly rent and monthly housing loan repayments. Both of these are related more or less to local housing prices, but both are imperfect proxies.
When using average weekly rent as a proxy, it needs to be recognised that public housing and community housing in certain areas may involve subsidised rent. One preliminary attempt was made by the authors to focus solely on the private rental market. However, this resulted in a relatively large number of missing observations for average rent and hence a decision was made to use average overall rent for respective SLAs in the analysis.
Average housing loan repayments are also an imperfect proxy for local housing prices because they depend on a number of factors that are difficult, if not impossible, to control for in Census data. The relationship between housing prices and loan amounts depends on the initial equity position of the owners, when the house was purchased and the nature of the loan (e.g., fixed term versus flexible market interest rates). Also, housing prices may have changed since a home was purchased. The upshot is that average housing loan repayments may provide some long-run indication of relative housing prices across SLAs, but it is likely to be problematic for measuring changes in housing prices over time. Given that a similar criticism could be made of the average rent payments (i.e., the relocation of public housing could distort relative rents for areas), these proxies were only measured at a particular point in time - the 2001 Census.7
As average rental payments and monthly housing loan repayments are obviously both less than perfect proxies for housing prices, it seems to make sense to use a combination of the two rather than relying only on one variable in the formal analysis. We chose to use a principle component analysis (PCA) that exploits this correlation to summarise the information contained in the housing variables It transforms a set of correlated data (in this case, the set of housing variables) into a smaller set of uncorrelated components that capture most of the variation of the original set of variables (Dunteman, 1989).
The aim of the PCA method is to maximise the correlation between the components and the original variables. While the procedure produces several new principal component index variables, the first principal component explains the largest amount of variation in the original variables by definition, with the second component explaining less variation or information (and so on). The viability of components can be assessed using the Kaiser criterion (which requires a viable component to have an eigenvalue greater than 1). In this case, given that we are just summarising two variables, one would be surprised if more than one component was important. Indeed, this is the case, with the eigenvalue for the first component being 1.75. Overall, this component explains 88.0% of the variance of average rent and average housing repayments and gives an equal loading of 0.7071 on each variable. While the index generated using the PCA technique is theoretically more robust and more likely to be associated with the underlying concept (i.e., long-run value of housing stock or, more simply, housing prices), the following discussion also reports both rents and monthly housing payments because they are easier to interpret intuitively. Neither variable contradicts the multivariate analysis obtained using the PCA index.
4 The ecological fallacy refers to the incorrect inference that can occur when inappropriately comparing several levels of analysis (e.g., different geographic levels of aggregation).
5 Driver and Dunne (1992) claimed that while the turbulence index is the "obvious way to measure structural change" (p. 123), it has several drawbacks, including: the possible loss of information in representing a vector of changes by a single summary statistic, the difficulties of defining the sectors over which mismatch occurs, and the problem of confusing noise with structural movements when examining changes over short periods. Notwithstanding these problems, they argue that it is the best measure of structural change available for regionally based data.
6 It is possible that the cost of providing capital in some areas (e.g., remote areas) may be more expensive and hence financial institutions might charge different rates for different areas. However, to our knowledge, we are not aware that this scenario is a substantial factor in Australia for private finance. Notwithstanding, there are some interest rate subsidies available for certain categories of loans in drought-affected areas that may be reflected in the value of housing equity. It is not possible to take this into account in a systematic (and relatively straightforward) way and hence this is left for future research.
7 In aggregating housing payments to regional averages, the difference in timing of housing purchases (and associated loans) of various residents may add an extra complexity, especially if the average local housing prices are not highly correlated over time. That is, if the local prices are particularly dynamic or unpredictable and local houses are sold at different rates in different areas (i.e., variable rates of churn in the local housing market over time), then the relationship between average housing payments and local housing prices is further weakened.