Composition and context drivers of residential property location value as a socioeconomic status measure

Abstract

Research applying residential property value as a socioeconomic status measure is increasing. The literature includes several measures of residential property value socioeconomic status, all of which highlight location as an important component. This paper examines the drivers of the location component of residential property value that form the basis of its application as a socioeconomic status measure. The metropolitan area of Adelaide, South Australia, is used as a study area to analyse the composition and context embodied in residential property location value. The focus of this paper is to provide an understanding of the drivers of residential property value calculated as the relative location factor, deliberately constructed to reflect the effect on value due to location. The analysis reduced the traditional composition measures of social structure into a smaller number of factors using principal component analysis and regressed these against relative location factor. A spatial lens was applied to the results using Moran’s I to visualise the composition and context influence embodied in relative location factor. The results provided a significantly enhanced understanding of both the composition and context of socioeconomic status wealth that may be a more suitable socioeconomic status measure than the traditional composition measures of income, education and occupation. This paper provides an original interpretation of the contribution and use of residential property location value enabling a broader understanding of socioeconomic status, concluding that relative location factor provided a more informed measure of socioeconomic status, capable of enhancing social science and health research and policy formation.

Keywords

Relative location factor socioeconomic status composition and context spatial analysis principal component analysis

Introduction

Socioeconomic status (SES) representing social standing or the class of an individual or group has long been used in social science research on a wide range of issues. SES is a complex and multidimensional concept (Haghdoost, 2012), though paradoxically its meaning has been taken as accepted without the need for a clear description (Galobardes et al., 2006a, 2006b). The traditional SES measures applied in research are income, education and occupation (Braveman et al., 2005, 2011). While these three measures are used most frequently, other measures are applied and there is a developing discussion around residential property value as a wealth SES measure (Drewnowski et al., 2014, 2015; Leonard et al., 2016; Lockwood et al., 2018; Rehm et al., 2012; Vernez Moudon et al., 2011). The basis for this is the significance of residential property value as a proportion of the asset wealth of an individual or family.

Income, education and occupation are composition measures that reflect attributes of an individual, or when used at an area level, the attributes of the people nearby. Context includes proximal facilities and services such as shops, parks, schools, medical, finance, employment and transport. Residential property value is a complex mix representing the influence of the surrounding environmental attributes (context) and the characteristics of the people living nearby (composition). The central argument of this paper is that the traditional measures of SES only provide composition and the use of residential property value can provide a more nuanced SES measure as it includes composition and context.

Using residential property value to express SES has additional advantages to the broader representation of composition and context. Composition measures are typically collected via a population census which is expensive to collect or via infrequent social surveys which are expensive and have limited sample sizes. Residential property value data are available on a continuous time line, are not aggregated to a priori spatial units and provide a means of overcoming modifiable areal unit problem (MAUP) (Openshaw, 1984b) through the use of interpolated continuous spatial surfaces that represents the composition and context of location at an individual residential property, subject to the caveat of sufficient representation of property sales across the study area.

The original and significant contribution of this paper is to demonstrate that the location component of residential property value includes context (the influence of environment proximity) and composition (characteristics of proximal people) and that residential property value can be used to model spatial socioeconomic status (SSES). It is hypothesised that residential property location value represents geographic wealth via an individual’s ability to purchase at a specific location. This reflects the importance of location expressed as “where” you live rather than “what” you live in as important in understanding SSES.

The literature includes several definitions of residential property value SES measures, but all highlight the location component as representing the context attributes of property wealth. The focus of this paper is to examine composition variables and their association with residential property value calculated using the relative location factor (RLF), expressed as a continuous surface and deliberately constructed to tease out the effect on value due to structural improvements (Lockwood et al., 2018).

The analysis reduced a number of traditional social structure composition measures into a smaller number of factors using principal component analysis (PCA). Then, it applied a hedonic regression model to calculate RLF and interpolated a continuous raster surface to derive property level RLF ratios. RLF was used in a regression as the dependent variable and the PCA factors as the independent composition variables to identify those characteristics of the social structure that were driving spatial variations in the residential property location values. Moran’s I was calculated to model the context influence embodied in the residential property value location component.

Literature review

The impact of SES to social, health and behavioural issues is supported by more than two centuries of research (Julia and Valleron, 2011; Waitzkin, 2006). It is not the focus of this paper to review the literature supporting this association, but to highlight the importance of how SES is measured. Income, education and occupation are the most frequently applied measures of SES (Braveman et al., 2005; Lampert et al., 2013; Tyrrell et al., 2016) either at the individual level or area level via census data using aggregate spatial units. Many measures have been proposed and used to represent SES, but the focus for this research is the use of wealth and property value.

Wealth as an SES measure has grown in popularity over the last two decades (Bond Huie et al., 2003; Duncan et al., 2002). Indeed, wealth and family income have been shown to have stronger associations with mortality rates than education and occupation (Duncan et al., 2002). A study concluded that policies aimed at closing the health disparities gap in the US may be poorly conceived if they ignored the impact of wealth on premature adult mortality (Bond Huie et al., 2003). Pollack et al. (2007) concluded that health studies not including the influence of wealth could underestimate the impact of SES on health. A study on housing wealth decline during the 2008 global financial crisis concluded that psychological distress increased and self-rated health decreased with declining housing wealth (Yilmazer et al., 2015). Wealth measures (including real estate and the family house) account for between 25 and 50% of a family’s net worth (Di et al., 2003). On this basis, it is not surprising that researchers have included residential property value as an SES measure.

Studies by Vernez Moudon et al. (2011) and Drewnowski et al. (2014, 2015) calculated residential property value based upon the property tax assessed value to represent structural as well as the residential property value location component. The mean assessed value per residential unit was calculated as an individual wealth measure, and the focal mean (833m buffer around each respondent’s property) was calculated as a neighbourhood measure. Vernez Moudon et al. (2011) concluded that their property SES measures, if replicated elsewhere, could replace area-based SES measures and simplify the effects of context for health research. Another research isolated a neighbourhood measure, using the residual from a hedonic price regression model blind to neighbourhood attributes, to examine the neighbourhood context effect on various health outcomes (Leonard et al., 2016).

Location and value have been recognised as a continually varying surface across space by real estate researchers and operationalised as location value response surfaces (LVRS) (O’Connor, 1982; O’Connor and Eichenbaum, 1988). LVRS provided a basis for controlling the location effect without the use of a priori spatial units (O’Connor, 1982) and was recognised as a more realistic method to account for change in real estate value due to location (O’Connor and Eichenbaum, 1988; Ward et al., 1999). LVRS methods preserved the concept of location in residential property value, by inclusion as part of the hedonic price modelling process and removed the need for submarket boundaries (Bourassa et al., 2003; Clapp, 2003; Fik et al., 2003; McCluskey et al., 2002; Pryce and Evans, 2007). The common theme in these methods was the recognition that every property had a unique location factor. In extending these principles, Gallimore et al. (1996) argued that the environmental context of residential location value encapsulated the intangible locational influences of real estate value. Wyatt (1996) proposed that the location components of property wealth including “access”, “neighbourhood” and “environment” cannot be changed by an individual property owner in the manner that structural attributes can be thereby emphasising the importance of “where you live” (location) rather than “what you live in” (house and land). Orford (1999) argued that the attributes describing these location components of wealth were almost infinite and impossible to collect and maintain, while Gallimore et al. (1996) claimed they were better observed as a whole in the residual of a hedonic predictive model.

One method to isolate the locational aspect of the spatial variation in SES is the RLF, based on the Gallimore et al. approach (Lockwood et al., 2018). The RLF model uses property attributes but is deliberately “blind” to location attributes. In the RLF model, the error term is assumed to contain the relevance of location to the total price of the property and is expressed as a ratio of actual price to predicted price (Lockwood et al., 2018).

The strong association between SES and social issues and the use of residential property as an SES measure provides a basis for a better understanding of the composition and context drivers of RLF. The literature describes the traditional SES components but does not provide a methodology to quantify the components of real estate value that are being used to describe SES. The objective of this paper is to address this gap by identifying the composition wealth attributes of the RLF construct and analyse their spatial distribution with the context wealth attributes of location to provide a better understanding of property wealth as an SES measure.

Methodology

The study area

The literature highlights the importance of the spatial context of SES and the relationship to the location component of real estate value as demonstrated by RLF. In exploring this complex relationship, this research utilised the urban component of the Adelaide Metropolitan Area as the study area (Figure 1). This area was selected due to Adelaide’s urban character, 1.6 million (median age of 39 years) population in 2016 and 366,090 single detached residential properties on large land parcels within 2689 Statistical Area 1 (SA1) units (the smallest spatial data unit used to disseminate the Australian Population Census results) (Australian Bureau of Statistics, 2018). Adelaide was considered an appropriate geographic scale as it reflected a “contained” housing market characterised by sub-markets (McGreal et al., 2016). The majority of sale transactions were through private treaty with owner occupancy the main tenure, similar to many international jurisdictions. The social construct of the Adelaide Metropolitan Area was comparable to other Australian and international cities reinforcing its validity as a representative study area with potential transferability of results.

Figure 1.

Study area.

The data

The data for this study were drawn from two sources, the 2016 Australian population census and residential sales transactions from the South Australian Lands Title Office. The census data were based on variables used in previous research to characterise social structures (Jackson et al., 2007; Reed, 2013; Rossini and Kupke, 2015), and included, SES, Familism and Ethnicity, and have been validated in several studies of Australian cities as summarised by Burnley (1980) and more recently by Rossini and Kupke (2015). Data representing education, income, occupation, family structure, ethnicity and dwelling tenure were included (Table 1). These data were aggregated (to protect individual confidentiality) to spatial areas by the Australian Bureau of Statistics (SA1 level). For this analysis, these variables were reported as a percentage of the population for comparison purposes at the SA1 level (Table 1).

Table 1.

Socio-demographic variables.

Variable list
1. Persons aged 0–4	21. Families one parent family
2. Persons aged 5–14	22. Dwellings fully owned
3. Persons aged 25–29	23. Dwellings with mortgage
4. Persons aged 30–34	24. Dwellings rented real estate agent
5. Persons aged 45–49	25. Dwellings rented housing authority
6. Persons aged 65–69	26. Dwellings total separate house
7. Persons aged 75–79	27. Dwellings total semi, row, terrace
8. Persons Australian born	28. Dwellings total flat, unit, apartment
9. Persons born elsewhere	29. Dwellings monthly loan payment $3000–3999
10. Persons speak English only	30. Dwellings monthly loan payment ≥$4000
11. Persons speak other language at home	31. Persons same address 1 years ago
12. Persons University/tertiary qualification	32. Persons different address 1 years ago
13. Persons year 8 schooling	33. Persons same address five years ago
14. Persons income $200–299	34. Persons different address 5 years ago
15. Persons income $1500–1999	35. Workers managers
16. Persons income $2000 or more	36. Workers professionals
17. Persons in group household	37. Workers clerical and administrative
18. Persons lone person household	38. Workers labourers
19. Families couple no child	39. Persons did not go to school
20. Families couple plus child under 15	40. Persons left school at year 10

The second data set used 6666 single residential dwelling sale transactions provided by the South Australian Valuer General. These sales represent the 2016 calendar year. As the Adelaide metropolitan market was relatively stable during this period, the full year’s data could be used, enabling a greater number of sale transaction points. The data associated with each sale property provided by the Valuer General are listed in the online supplementary material.

Analysis

PCA was used to reduce the variables to a set of components (combinations of the variables) to explain the maximum variability. These components characterised the more general themes of the social construct of the urban population discussed in the literature review. PCA has been successfully applied in other Australian studies using similar variables to describe the social structure (Lockwood and Coffee, 2006; Reed, 2013; Rossini and Kupke, 2015). Components were summarised using the rotated factors from the analysis and given a label based on the factor scores. The cutoff for the number of factors identified was based on eigenvalues greater than 1 with the variables and their factor loadings (only those greater than or equal to 0.3) shown in the Rotated Component Matrix. For each factor, those variables with factor loadings above 0.5 were used to create a suitable descriptor for each factor.

RLF (Gallimore et al., 1996; Lockwood et al., 2018) was calculated using a global hedonic regression model specified with structural housing attributes (without location) as independent variables (equation 1). The spatially structured component of the error term reflects the locational effect to the total value of the property. In this regard, the RLF is not an absolute measure of location but measures the relationship of the structural component of the property value to the property’s total value including the value of the omitted location attributes. Importantly, this model captures the effect of the aggregation of all omitted variables in the residual alleviating the necessity to individually identify and measure the infinite number of individual locational attribute data (Orford, 1999), which, if available, would be at differing spatial scales. This methodology, by requiring only the combined, and not individual, effect for the RLF construction, is considered as the most effective approach.

\begin{array}{l} Price = β_{0} β_{1}^{LandArea} β_{2}^{BuildArea} β_{3}^{LAsqrd} β_{4}^{BAsqrd} β_{5}^{Framewall} β_{6}^{Stonewall} β_{7}^{Era_Pre19 00} β_{8}^{Era_1900 to192 0} \\ \times β_{9}^{Era_Twenties} β_{10}^{Era_Thirties} β_{11}^{Era_Fourties} β_{12}^{Era_Fifties} β_{13}^{Era_Sixties} β_{14}^{Era_Seventies} β_{15}^{Era_Eighties} \\ \times β_{16}^{Era_Ninties} β_{17}^{Era_Noughties} β_{18}^{GoodQuality} β_{19}^{PoorQuality} β_{20}^{TiledRof} β_{21}^{GovtHousing} β_{22}^{Archit} \\ \times β_{23}^{Contemp} β_{24}^{Colonial} β_{25}^{Reproduction} β_{26}^{Mansion} β_{27}^{Austerity} β_{28}^{Shack} Error \end{array}

(1)

Equation (1) shows the specified hedonic model (calculated in log form).

For a detailed list of variables see the online supplementary material.

The RLF is the sale price/predicted price calculated for each sale property location point. A continuous surface was interpolated using kriging in a geographic information system (GIS) from each sale point ratio across the whole study area (iRLF), from which a point iRLF for each individual property in the Adelaide study area was extracted (piRLF). This was necessary to match with the census data which is aggregated to the SA1 to protect confidentiality. The SA1 includes approximately 200 dwellings and although their geographic area varies, the number of dwellings remains relatively constant. There were 2689 SA1s in the study area; the number of sales transaction points was 6666. To calibrate a reliable model as specified in equation (2), a new variable, the mean piRLF value, for each SA1 was calculated. The census data were based on households and the dwelling extracted piRLF provides a means of creating a spatial unit mean built from the same basis as the census data. The ABS consider a household to equal a private occupied dwelling (Australian Bureau of Statistics, 2016).

Tobler’s (1970) first law of geography posits that near points are likely to be more similar than distant points and this simple premise underpins interpolation techniques. Many interpolation techniques, as discussed by Lockwood et al. (2018), are available to interpolate values from known data points to provide a continuous surface. A variety of techniques are available, all working from Tobler’s (1970) law regarding proximity and near being more likely to be similar than distant points. This study used Empirical Bayesian kriging to interpolate the RLF continuous raster surface from the known sales point ratios (6666). Empirical Bayesian kriging is used as it incorporates local small-scale variation (local semi-variograms) while adhering to the assumption of stationarity for optimal kriging. An overlap factor of 2 is used, as this allows each point to be located in two of the adopted 100 local semi-variograms subsets to provide a smoother surface (Krivoruchko and Krause, 2012).

Regression analysis was used to determine which PCA factors significantly contributed to the RLF measure. A linear regression model with the mean piRLF for each SA1 as the dependent variable and the PCA factor scores as independent variables was used. The model used the general form

Y = β_{0} + β_{1} X_{1} + β_{2} X_{2} + \dots + β_{n} X_{n} + ε

(2)

where Y = mean piRLF for each SA1 and β₀ to β_n are the regression coefficients associated with each of the factor scores. X₁ to X_n are the PCA factor scores for each SA1 and ε is the random error term.

From the regression model, the composition drivers were identified as those that were statistically significant and made the most important contribution in explaining the variation in the dependent variable (piRLF).

The PCA factors were analysed for significant spatial clustering using Moran’s I at the global level to establish if the factors were spatially clustered and to test for significant local clustering of high surrounded by high factor scores (HH) and low surrounded by low factor scores (LL) (Anselin, 1998). For each factor, the HH and LL were plotted enabling visual comparison of the spatial distribution with the RLF. Both tests used ESRI GIS ArcMap V10.4.

Results

Forty census attributes were deemed suitable for the factor analysis, as the Kaiser–Meyer–Olkin measure of sampling adequacy was greater than 0.6 and the Bartlett test of sphericity was significant. Seven factors were identified explaining 71% of the total variance in the data (Table 2).

Table 2.

PCA extraction resulting in seven factors.

	Total variance explained
Factor	Rotation sums of squared loadings
Factor	Total	% of variance	Cumulative (%)
1	7.035	17.588	17.588
2	5.375	13.437	31.025
3	5.275	13.187	44.213
4	3.451	8.628	52.841
5	2.993	7.483	60.323
6	2.224	5.561	65.884
7	2.084	5.209	71.093

This exceeded the 60% threshold considered adequate for this type of analysis (Hair et al., 2009). The cutoff in the number of factors identified was based on eigenvalues greater than 1 and interpretation of the Scree Plot showing seven or eight to be the optimum. After Varimax Rotation, variables with factor loadings above 0.5 were used to designate and describe each factor (Table 3).

Table 3.

Factor analysis output showing seven factors.

Variable number and description	Component
Variable number and description	1	2	3	4	5	6	7
31. People same address 1 year ago	.838
26. Dwellings total separate house	.831
23. Dwellings with mortgage	.823			–.305
2. People aged 5–14	.785
8. People Australian born	.737				–.460
33. People same address 5 years ago	.728	–.412
20. Families couple plus child under 15	.696		.307				–.362
10. People speak English only	.638				–.601		.356
15. People income $1500–1999	.618
37. Workers clerical and administrative	.573					.344
1. People aged 0–4	.569						–.442
5. People aged 45–49	.552
34. People different address 5 years ago		.857
32. People different address 1 year ago		.828
24. Dwellings rented real estate agent		.815
3. People aged 25–29		.795
4. People aged 30–34		.684
17. People in group household		.626
28. Dwellings total flat, unit, apartment	–.310	.560
12. People University/tertiary qualification			.890
36. Workers professionals			.871
16. People income $2000 or more	.438		.738
29. Dwellings monthly loan payment $3000–3999			.721
30. Dwellings monthly loan payment ≥$4000			.694
35. Workers managers			.647
40. People left school at year 10	.401		–.595
38. Workers labourers			–.592	.408
22. Dwellings fully owned	.382		.426			.404	.374
25. Dwellings rented housing authority				.820
14. People income $200–299				.779
21. Families one-parent family			–.415	.637
27. Dwellings total semi, row, terrace house				.629
18. People lone person household		.371		.575		.454
11. People speak other language at home		.360			.847
9. People born elsewhere		.451			.760
39. People did not go to school					.692
7. People aged 75–79						.816
13. People year 8 schooling				.404	.443	.476
6. People aged 65–69							.740
19. Families couple no child		.377				.529	.534

Extraction method: principal component analysis. Rotation method: Varimax with Kaiser normalisation.

Factor 1 (Familism) had high positive loadings on census variables describing Australian born couples with young children living in a mortgaged separate dwelling for at least five years. This indicated a stable family structure explaining 17.6% of the variation in the original data set. The variables associated with Factor 2 (Mobility) characterised residential mobility, younger working age populations, renting, without children (potentially) accounting for 13.4% of the variation. Factor 3 (Traditional SES) represented personal wealth with the top income bracket, professional occupations, tertiary education, higher mortgages and a negative association with non-skilled occupations. This factor most closely aligned with traditional SES (income, education and occupation) and accounted for 13.2% of the variation. Factor 4 (Marginalised Housing Choice) was associated with low-cost rentals, one-parent families describing a marginalised group of households with a low financial capability and limited housing choice/social rental. Factor 5 (Ethnicity), non-English speaking and minimal schooling. Factors 6 and 7 appear to represent two retiree groups distinguished by age. Factor 6 (Older Retirees) was characterised by a higher positive loading on the 75–79 age cohort, no children, lone household and fully owned house. Factor 7 (Younger Retirees) had the higher positive loading on the 65–69 age cohort, no children and may represent the group referred to as “grey nomads”.

These factors characterise general constructs of socioeconomic, family and ethnic status and indicate that the structure of the study area was consistent with other national and international metropolitan urban areas (Burnley, 1980) and for the study area (Lockwood and Coffee, 2006; Stimson and Cleland, 1975).

The results of the calibration of equation (1) gave an adjusted R² value of 0.59 and an F value of 345.9. The resulting RLF calculation was numerically and spatially distributed as shown in the online supplementary material.

Table 4 shows the results from the regression analysis of piRLF with the calibrated standardised regression coefficients from the specification in equation (2). The adjusted R² value indicates that the PCA factors explained 62% of the total variance in the mean piRLF. All factors significantly contributed to the variation in mean piRLF with four factors contributing more with higher standardised beta coefficients as shown in Table 4.

Table 4.

Results of regression analysis and Global Moran’s I.

Standardised coefficients for average piRLF as dependent variable
Factor	Beta	Moran’s I
(Constant)	***
Factor 3 – Traditional SES	0.722***	0.86^a
Factor 6 – Retirees (older)	0.207***	0.27^a
Factor 7 – Retirees (younger)	0.124***	0.28^a
Factor 1 – Familism	–0.100***	0.48^a
Factor 2 – Mobility	0.074***	0.52^a
Factor 4 – Marginalised Housing Choice	–0.036***	0.43^a
Factor 5 – Ethnicity	0.028**	0.58^a

Note: the model explained 62% of the variation (R² = 0.62).

Less than 1% likelihood that the clustered patterns could be the result of chance alone.

Statistically significant at the 5% level; *** Statistically significant at the 1% level.

Spatial autocorrelation (global and local) of the seven factors was calculated to test for significant spatial clustering. The results of the global spatial autocorrelation analysis showed that each of the factors returned a global Moran’s I that indicated a less than 1% likelihood that the observed clustered patterns could be the result of chance alone (Table 4).

To demonstrate the spatial relationship between clustering of the significant factors and piRLF, the top three (bounded green) and the bottom three (bounded black) piRLF deciles are overlain on the local Moran’s I for each of the four factors that significantly contribute to RLF and explained most of the variation through their higher standardised beta coefficients (Figure 2).

Figure 2.

Spatial distribution of significant factor clusters (1, 3, 6 and 7) and top and bottom RLF deciles.

Discussion

The analysis revealed important features in the relationship between piRLF (context) and the association with piRLF of the PCA factors summarising composition social structure. All seven factors significantly contributed to the variation in the piRLF at the 99% level except Factor 5 (95% level) (Table 4). Of the seven factors four were more important in their prediction of the piRLF (1, 3, 6 and 7) as indicated via their standardised beta coefficients (Table 4). Factor 1 (Familism) was negative with a standardised beta coefficient of –0.10, while Factors 3, 6 and 7 were positive with standardised beta coefficients of 0.72, 0.21 and 0.12. The negative direction for Factor 1 indicated that a high value was associated with low piRLF. Factor 1 captured younger families who were not necessarily wealthy and may not be able to purchase a property in better locations.

Of the four factors, the most influential was Factor 3 (Traditional SES) with a standardised beta coefficient of 0.72 and statistical significance at the 99% level. This was greater than the other three factors (1, 6 and 7) combined. Factor 3 characterised traditional SES (income, education and occupation) and was the most significant in explaining the variation in piRLF. Importantly, piRLF included an additional three significant factors that made contributions to explaining the variance. Factors 6 and 7 described a form of wealth not measured by traditional SES, retiree’s wealth. This may be associated with the equity contained in the family home (asset rich). The other factors did not contribute much to the variation in piRLF, but did explain some of the piRLF variation and represented wealth not included with the traditional income, education and occupation measures.

Traditional composition SES (Factor 3) and the other wealth dimensions (Factors 1, 6 and 7), explained 62% of the variation in piRLF leaving 38% as the unexplained component, inferring omitted variable bias in the specification shown in equation (1). Omitted variables are generally considered to encapsulate the intangible location influences of real estate value such as culture, accessibility, amenity, social status, employment, traffic, leisure facilities and other neighbourhood amenities (Gallimore et al., 1996) and the “location” component of property wealth described by Wyatt (1996) including “access”, “neighbourhood” and “environment”. It would be impossible to identify and measure all these attributes individually (Orford, 1999), but their overall combined effect can be spatially captured and visualised.

The piRLF top and bottom three deciles (Figure 2) provide the geographic overlap of context (location wealth SES) with the composition social construct dimensions providing a spatial understanding of both. Locations of high and low piRLF deciles that do not contain significantly high or low factor clusters (Figure 2) may represent a stronger context influence than composition. The analysis of the four most important factors (1, 3, 6 and 7) with piRLF (Figure 2) indicated a distinct spatial pattern contributing to an understanding of the socio-spatial construct as represented through the principal components. Factor 1 (Familism) made a significant contribution to piRLF describing a negative relationship between the desirability of place (“where you live”) and the predominance of families with young children and a mortgage. This may reflect financial capability when choosing a location to live during the early stage of the family life-cycle. The spatial distribution of Familism is significant and showed a spatial association between the clustering of the higher (HH) Factor 1 scores and the bottom three piRLF deciles in the northern and southern suburbs, coincident with less wealthy locations. The more desirable locations of the eastern and coastal suburbs (top three piRLF deciles) with the lower clustering (LL) of Factor 1 scores is consistent with the financial capacity to locate in more desirable locations.

Factor 3 (Traditional SES) is the most significant contribution to the variation in RLF through both its statistical significance (99%) and the highest standardised beta coefficient. It reflects the traditional SES attributes of income, education and occupation. Clustering of high (HH) scores aligns well with the eastern and coastal suburbs of the Adelaide Metropolitan Area and the top three piRLF deciles. Clustering of low (LL) occurred in the less desirable northern and southern suburbs. This also highlighted the negative association with Factor 1 to live in more desirable locations.

Factors 6 and 7 (Retirees) made a significant (99% level) contribution and higher standardised beta coefficients to the piRLF than Factors 2, 4 and 5. In this case, the spatial distribution of retirees is more visually aligned with high clusters (HH), predominately in the older eastern and coastal suburbs (top three piRLF deciles) for the older retirees and low clustering (LL) in the newer northern locations (bottom three piRLF deciles). Retirees were not confined to one location but had a relative wealth advantage in the area in which they lived and accumulated wealth through capital gain and were mortgage free. This is not measurable using income, as retirees, although asset rich, often have a relatively low income.

It is interesting to note that Factor 2 (Mobility), which was significant in explaining RLF at the 99% level, was not defined by traditional SES measures. Although a relatively low standardised beta coefficient, this factor explained that 13.4% of variation in the original compositional census data may be attributable to the high spatial clustering associated with both the upper three (inner northern and eastern suburbs) and lower three piRLF deciles (northern suburbs), suggesting a dichotomy in the wealth of this group (Figure 3). This can only be appreciated by observing the location of the low clustering in the northern and southern suburbs (lower three piRLF deciles) that may describe a young population unable to purchase a home and living in rental accommodation in lower RLF locations. The location of the high clustering in and around the central business district (CBD) as well as the inner eastern suburbs (higher three piRLF deciles) may indicate a wealthier and more mobile younger lifestyle group. This dichotomy highlights the clustering in some coastal and inner suburbs (high piRLF) and in the northern and southern suburbs (low piRLF). Factors 2, 4 and 5 show the influence of location, reflecting the factors as a mix of higher and lower wealth represented in high and low piRLF deciles (Figure 3).

Figure 3.

Spatial distribution of significant factor clusters (2, 4 and 5) and top and bottom RLF deciles.

The piRLF top and bottom three decile locations (Figure 2) provided the geographic overlap of the location wealth or context with the composition social construct dimensions and provided greater spatial understanding. Locations of high and low piRLF deciles that do not contain significantly high or low factor clusters (Figure 2) are indicative of a stronger context influence of wealth than composition factors.

The literature identified fields of social science research that are important to government policy, all requiring a range of specialised services. The contribution of piRLF to policy is through an informed identification of locations associated with societal issue hotspots. The top and bottom three piRLF deciles overlap with the spatial (HH) clustering of Factor 3 illustrating that Mobility may be spatially segregated into two wealth components that may require different location specific policy solutions. The older suburbs with spatial (HH) clusters of Factors 6 and 7 (Retirees) identified locations of wealth that were not apparent using census-based income, education and occupation. This could assist policy planning for aged care services through identifying clusters of Retirees.

The strength of piRLF is a methodology to help overcome the issues associated with traditional SES measures, such as reporting bias, collection and reporting issues or the use of administrative spatial units that introduce the MAUP (Openshaw, 1984a). The derivation of piRLF from individual georeferenced real estate sales transaction data analysed as representing the market value by property professionals avoids the potential bias of survey data. Interpolated surfaces using piRLF help overcome MAUP which, as highlighted in the literature, is associated with a priori aggregated spatial units. In this analysis, piRLF was aggregated to the SA1, but it can be used at the individual property or calculated for any spatial units providing an adaptable measure. As piRLF was calculated using residential property sale transactions, it is not restricted to census cycles, does not require additional data capture and can be derived at any point in time. Potential limitations of piRLF are the availability of suitable spatially enabled land transaction data and property expertise to specify a global hedonic model that represents the market conditions of the study area.

Conclusion

The objective of this paper was to analyse the compositional drivers of RLF to contribute to a better understanding of the wealth measures inherent in residential property location value as a contribution to providing an enhanced SES indicator. This paper found that the composition drivers of RLF have a fundamental relationship with context as expressed through the single residential property market structure. The originality of this study is the increased understanding of the composition drivers of SES as expressed through the residential property location value. This paper demonstrated that a broader range of composition drivers (beyond the traditional measures of income, education and occupation) and context drivers (aspects of access and proximity) are embedded in the value of residential property via location.

The composition drivers of piRLF included income, education and occupation as well as measures of family structures ranging from lower income stable families through to asset-rich, income-poor retirees. A key outcome of this research was that the census composition variables explained 62% of the piRLF measure, leaving 38% representing context via omitted variable bias. These include broader context wealth parameters associated with location including environment, access and neighbourhood attributes which are not captured in traditional composition SES. This establishes the basis for piRLF to be applied in social science research as a more complete SSES measure representing composition and context at a spatial scale not available from census measures.

The composition and context attributes of residential property location value have a critical role in explaining a residential property value wealth-based SES as input to housing policy that can be spatially targeted based on differences in spatial social structure and amenities. For example, housing affordability and aged care are complex issues, and using piRLF as an SES measure may provide additional dimensions of understanding, both in composition and context terms, through introducing location as an important consideration when providing specific social services.

In the literature concerning composition and context, this paper provides an innovative and original interpretation of the contribution of the use of residential property to a broader understanding of SES and the potential to contribute to all tiers of government policy across a broad range of social issues. This paper contends that piRLF provides a more informed and broader measure of SSES, capable of enhancing social science and health research, understanding and policy formation.

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Neil T Coffee is an Associate Professor of Health Geography in the Health Research Institute, University of Canberra, in the Spatial Epidemiology Group and Adjunct Fellow in the School of Health Science, University of South Australia, and School of Architecture and Built Environment, Healthy Cities Research Group, The University of Adelaide. Research interests include in the application of GIS to social science and public health, the impact of the built environment on physical activity and obesity, developing socio-economic GIS databases and the use and adaptation of government collected and maintained administrative data for social and health research, and the integration of GIS into government systems.

Tony Lockwood has worked in the property field over the last 30 years both in South Australia as a valuer and in land data management and internationally on World Bank funded Land Titling Projects in Thailand and Laos. More recently, completed studies at The University of Adelaide in the Masters of Spatial Information Science (course work) and a PhD. Tony is currently an Adjunct at both the University of South Australia and The University of Adelaide. His main areas of interest are Mass Appraisal; Spatial Analysis; Housing Market Structure and Land Administration Systems.

Peter Rossini is a Senior Lecturer in the School of Commerce at the University of South Australia Business School. His research focuses on residential property indexing and market performance, automated valuation methods, mass appraisal, urban renewal evaluation, housing market and property implication, and water markets. Rossini holds a Bachelor of Applied Science from the South Australian Institute of Technology and a Graduate Diploma in Economic Statistics from the University of New England. Peter is a member of the Pacific Rim Real Estate Society (PRRES) regularly presenting at their conferences and is widely published in peer reviewed journals.

Theo Niyonsenga is an Associate Professor of Biostatistics, Faculty of Health and Health Research Institute, University of Canberra. He holds a PhD in Mathematical Statistics with 25 years of experience in statistical data analysis and consulting. His interests are in developing and applying statistical methodology to collaborative health research focusing on various areas of Biostatistics in general and particularly on spatial statistics with applications to spatial epidemiology as well as multivariate data analysis methods and growth curve hierarchical data modelling. Theo has acquired extensive knowledge of modern statistical computing software packages, including SAS and MPLUS, STATA and Geographically Weighted Regression as well as WinBugs and R tools for Bayesian inferential analysis.

Stanley McGreal is a Professor of Property Research in the Belfast School of Architecture and the Built Environment at the University of Ulster and was formerly the Director of the Built Environment Research Institute at the University for over 10 years. He holds an Adjunct Professorship in the School of Commerce, the University of South Australia. Stanley has published widely in the real estate, housing and urban regeneration areas and has over 300 published works, he has received research funding from prestige sources including UK research councils (ESRC, EPSRC), the European Union through Framework 7 and Horizon 2020, various charities such as the Joseph Rowntree Foundation, the RICS Education Trust, the Investment Property Forum, government departments and agencies in the UK and from industry. Professor McGreal was editor of the Journal of European Real Estate Research from 2008 to 2017, he is a past President of the International Real Estate Society and is a Fellow of the Royal Institution of Chartered Surveyors.

References

Anselin

(1998) GIS research infrastructure for spatial analysis of real estate markets. Journal of Housing Research 9: 113–133.

Australian Bureau of Statistics (2016) 2016 Census Dictionary. Canberra: Australian Bureau of Statistics.

Australian Bureau of Statistics (2018) 2016 Population Census QuickStats. Available at: www.censusdata.abs.gov.au/census_services/getproduct/census/2016/quickstat/UCL401001?opendocument (accessed 24 September 2018).

Bond Huie

Krueger

Rogers

, et al. (2003) Wealth, race, and mortality*. Social Science Quarterly 84: 667–684.

Bourassa

Hoesli

Peng

(2003) Do housing submarkets really matter? Journal of Housing Economics 12(1): 12–28.

Braveman

Cubbin

Egerter

, et al. (2005) Socioeconomic status in health research: One size does not fit all. JAMA: The Journal of the American Medical Association 294: 2879–2888.

Braveman

Egerter

Williams

(2011) The social determinants of health: Coming of age. Annual Review of Public Health 32: 381–398.

Burnley

(1980) The Australian Urban System. Growth, Change and Differentiation. Melbourne, Australia: Longman Cheshire.

Clapp

(2003) A semiparametric method for valuing residential locations: Application to automated valuation. Journal of Real Estate Finance and Economics 27: 303–320.

10.

Drewnowski

Aggarwal

Rehm

, et al. (2014) Environments perceived as obesogenic have lower residential property values. American Journal of Preventive Medicine 47: 260–274.

11.

Drewnowski

Aggarwal

Tang

, et al. (2015) Residential property values predict prevalent obesity but do not predict 1-year weight change. Obesity 23: 671–676.

12.

Duncan

Daly

McDonough

, et al. (2002) Optimal indicators of socioeconomic status for health research. American Journal of Public Health 92: 1151–1157.

13.

Fik

Ling

Mulligan

(2003) Modelling spatial variation in housing prices: A variable interaction approach. Real Estate Economics 31: 623–646.

14.

Gallimore

Fletcher

Carter

(1996) Modelling the influence of location on value. Journal of Property Valuation and Investment 14: 6–19.

15.

Galobardes

Shaw

Lawlor

, et al. (2006a) Indicators of socioeconomic position (part 1). Journal of Epidemiology and Community Health 60: 7–12.

16.

Galobardes

Shaw

Lawlor

, et al. (2006b) Indicators of socioeconomic position (part 2). Journal of Epidemiology and Community Health 60: 95–101.

17.

Haghdoost

(2012) Complexity of the socioeconomic status and its disparity as a determinant of health. International Journal of Preventative Medicine 3(2): 75–76.

18.

Hair

Black

Babin

, et al. (2009) Multivariate Data Analysis. 7th ed. New Jersey: Prentice Hall.

19.

Jackson

Kupke

Rossini

(2007) The relationship between socio-economic indicators and residential property values in Darwin. In: Thirteenth annual pacific-rim real estate society conference, Fremantle, Western Australia, 21-24 January.

20.

Julia

Valleron

A-J

(2011) Louis-René Villermé (1782–1863), a pioneer in social epidemiology: Re-analysis of his data on comparative mortality in Paris in the early 19th century. Journal of Epidemiology and Community Health 65: 666–670.

21.

Krivoruchko

Krause

(2012) Concepts and application of kriging. In: ESRI international user conference, San Diago, California, 23-27 July.

22.

Lampert

Kroll

Müters

, et al. (2013) Measurement of socioeconomic status in the German Health Interview and Examination Survey for Adults (DEGS1) (accessed published on edoc: 2013-05-28T10:50:00Z access: 2018-01-08T03:58:54Z). Available at: http://www.gbe-bund.de/pdf/DEGS1_Soziooekonomischen_Status_E.pdf

23.

Leonard

Powell-Wiley

Ayers

, et al. (2016) Property values as a measure of neighborhoods: An application of hedonic price theory. Epidemiology 27(4): 518–524.

24.

Lockwood

Coffee

(2006) Residential living structure as a basis for the spatial delineation of residential submarkets. Pacific Rim Property Research Journal 12: 350–368.

25.

Lockwood

Coffee

Rossini

, et al. (2018) Does where you live influence your socio-economic status? Land Use Policy 72: 152–160.

26.

McCluskey

Deddis

Lamont

(2002) The application of spatially derived location factors within a GIS environment. In: PRRES Eighth Annual Conference, Christchurch, New Zealand, 21-23 January.

27.

McGreal

Taltavull de La Paz

Kupke

, et al. (2016) Measuring the influence of space and time effects on time on the market. Urban Studies 53: 2867–2884.

28.

O’Connor

(1982) Locational Valuation Derived Directly from the Real Estate Market with the Assistance of Response Service Techniques. Lincoln Institute of Land Policy.

29.

O’Connor

Eichenbaum

(1988) Location value response surfaces: The geometry of advanced mass appraisal. Property Tax Journal 7: 277–296.

30.

Openshaw

(1984a) Ecological fallacies and the analysis of areal census data. Environment and Planning A 16: 17–31.

31.

Openshaw S( 1984b) Concepts and Techniques in Modern Geography Number 38: The Modifiable Areal Unit Problem, Norwick: Geo Books.

32.

Orford

(1999) Valuing the Built Environment GIS and House Price Analysis. Aldershot: Ashgate.

33.

Pollack

Chideya

Cubbin

, et al. (2007) Should health studies measure wealth? A systematic review. American Journal of Preventive Medicine 33: 250–264.

34.

Pryce

Evans

(2007) Identifying Submarkets at the Sub-Regional Level in England. London: Department for Communities and Local Government.

35.

Reed

(2013) The contribution of social area analysis: Modelling house price variations at the neighbourhood level in Australia. International Journal of Housing Markets and Analysis 6: 455–472.

36.

Rehm

Moudon

Hurvitz

, et al. (2012) Residential property values are associated with obesity among women in King County, WA, USA. Social Science & Medicine 75: 491–495.

37.

Rossini

Kupke

(2015) The socio spatial substructure of the Australian city: How well does it explain house price? International Journal of Housing Markets and Analysis 53: 2867–2884.

38.

Stimson

Cleland

(1975) A Socio-Economic Atlas of Adelaide. An Analysis of the 1971 Census. South Australia: The Flinders University of South Australia and Department for Community Welfare.

39.

Tobler

(1970) A computer movie simulating urban growth in the Detroit region. Economic Geography 46: 234–240.

40.

Tyrrell

Jones

Beaumont

, et al. (2016) Height, body mass index, and socioeconomic status: Mendelian randomisation study in UK Biobank. BMJ 352: i582.

41.

Vernez Moudon

Cook

Ulmer

, et al. (2011) A neighborhood wealth metric for use in health studies. American Journal of Preventive Medicine 41: 88–97.

42.

Waitzkin

(2006) One and a half centuries of forgetting and rediscovering: Virchow’s lasting contributions to social medicine. Social Medicine 1: 6.

43.

Ward

Weaver

German

(1999) Improving CAMA models using geographic information systems/response surface analysis location factors. Assessment Journal 6: 30–39.

44.

Wyatt

(1996) The development of a property information system for valuation using a geographical information system (GIS). Journal of Property Research 13: 317–336.

45.

Yilmazer

Babiarz

Liu

(2015) The impact of diminished housing wealth on health in the United States: Evidence from the great recession. Social Science & Medicine 130: 234–241.

46.

Yang

Liu

(2003) The importance of housing to the accumulation of household net wealth. Working Paper W01-6, Joint Center for Housing Studies of Harvard University.