1. INTRODUCTION

Modelling regional economic performance has long been a concern of regional scientists. In recent times, there has been considerable emphasis on focusing on endogenous growth (see, for example, Stimson, Stough and Roberts, 2006; Johansson et al., 2001; Stimson, Stough and Nijkamp, 2011; Stimson and Stough (with Salazar), 2009) providing a framework for measuring and modelling spatial variation in endogenous regional economic performance over time.

The modelling approach requires:

(a) specification of a dependent variable that measures how change in economic performance over time (both growth and decline) might be attributable to factors and processes that are endogenous to the region; and

(b) deciding on a set of independent variables that might provide explanation for the variation across regions in the incidence of that dependent variable, which is achieved using spatial econometric modelling.

That approach proposed by Stimson and Stough [with Salazar] (2009) has been adopted in several studies investigating the endogenous economic performance of regions across Australia over successive inter-census decadal periods (see Stimson, 2012; Stimson, Robson and Shyy, 2009a; 2009b; 2011; Stimson, Mitchell, Rohde and Shyy, 2011), and it is continued in this paper for the decade 2001-2011. It has also been used in the paper by Plummer et al. (2014). Importantly, in the research reported here, functional regions rather than de jure regions are used as the spatial base for the modelling. This has been shown to largely overcome the issue of spatial autocorrelation that is inherent in spatial econometric modelling based on using de jure regions as the spatial base for regional demarcation. The modelling reported in this paper employs a framework in which:

(a) the spatial base is 134 Functional Economic Regions (FERs) across both the capital city metropolitan regions and the non-metropolitan regions of Australia that have been derived by the authors (and reported in Stimson et al., 2016);

(b) the dependent variable, measuring endogenous regional employment performance, is the regional (or differential) component derived from a shift-share analysis of employment change over the decade 2001-2011; and

(c) the independent (explanatory) variables that potentially might explain variation in the dependent variable, are a set of 27 measures derived from census data that relate to factors and processes that regional scientists have been suggesting might influence endogenous regional performance, plus five locational variables.

The paper is structured as follows. The next section briefly reviews past approaches to research investigating regional economic performance in Australia. That is followed by an outline of the data and methodology used in the analysis. Next, the spatial patterns of endogenous regional performance--the dependent variable--over the decade 2001-2011 are mapped and described. The bulk of the paper then presents the results of the spatial econometric modelling performed to identify those factors that might explain the variations in endogenous regional employment performance across Australia's FERs. Finally, there is a brief discussion of the policy implications of the model findings.

2. OVERVIEW OF APPROACHES TO RESEARCH INTO REGIONAL PERFORMANCE IN AUSTRALIA

Since the 1970s Australia has undergone as series of significant structural economic transitions. The impacts of these shifts have not been homogeneous over space and there is considerable variation in the economic performance of regions across Australia, both within the major large metropolitan cities and beyond into regional Australia.

Stimson (2012) has provided a detailed review of research investigating regional economic performance in Australia, most of which has been based on using de jure rather than functional regions--such as Local Government Areas or Statistical Local Areas (SLAs), or aggregations of them--as the spatial unit of analysis.

The nature of those economic and social 'divides', as they were emerging in the decade or so up to the late 1990s, was discussed in a book by O'Connor et al. (2001) on Australia's changing economic geography. Divides have also been identified in other studies (such as Baum et al, 1999; Baum et al, 2006). Spatial mismatches were shown to be evident in regional shares of population and population change and in shares of investment in economic activity.

The O'Connor, et al. (2001) study raised a series of challenging implications for people-based and for place-based policy responses in addressing those spatial disparities. In particular, the infrastructure needs required to enhance the performance of those segments and places in the space economy that are significant contributors to national wealth and competitiveness were a focus.

Stimson (2012: p. 162) pointed out that:

"... Understanding the dynamics underlying the spatial differences that exist in the economic development and performance of Australia's regions is a complex task."

Regional research studies conducted over the last two to three decades have identified a range of factors influencing patterns of regional development and performance for specific periods of time. But:

"... the specific conclusions reached and the relationships identified are not necessarily consistent because of the different focus of the studies and their different methodologies, variations in the spatial units of analysis used, and the different time periods that are analysed" (p. 162).

Stimson pointed out that many studies have investigated:

"... regional differentials in, and inter-relationships between, regional population size and growth, employment changes, structural shifts in industry employment, income levels, resource endowments, and the locational characteristics of regions" (p. 162), along with aspects of human capital.

Among other things, they had shown that a region's industry structure, its occupation mix, and its human capital structure are affected not only by the size of the region's economy and its resource endowments, but also by its level of remoteness in the context of the nation's settlement system. Some examples of such regional research include the following:

* work by the Commonwealth Government's Bureau of Transport and Regional Economics (BTRE, 2004a; 2004b) has modelled relationships between regional shifts in industry structure diversification/specialisation, structural change in employment, unemployment, human capital, and the size of regional economies;

* research by Trendle and Shorney (2003) investigated the relationship between regional industry diversity and unemployment in the State of Queensland;

* studies by Bradley and Gans (1998) and Hogan et al. (1999) focused on analysing increasing regional industrial diversity;

* a study by Garnaut et al. (2001) investigated regional influences on employment and population growth;

* studies by Harrison (1997) and Garnett and Lewis (2000) focused on relationships between regional education participation rates and qualifications, migration, and labour markets;

* a study by the National Centre for Social and Economic Modelling (Lloyd et al., 2000) investigated the 'hollowing out' of income across regions; and

* a study by Plummer, et al. (2014) investigated uneven development and local competitiveness across Western Australia's regional cities.

However, as Stimson (2012) has noted, relatively few studies have:

"... incorporated an explicit focus on the nature of occupational structure, occupational status, and education qualifications and skills, all of which are important components in the consideration of human capital differentials in regional development and performance" (p. 162).

Much of the published research on regional performance has been on the level of income and tended to focus on modelling variation in aggregate employment change or the incidence of unemployment. Some of the research on differentials in regional performance has been restricted to measuring and modelling an aspect of patterns of economic performance within Australia's capital city metropolitan regions or within a specific city, while other research has been focused specifically on the nation's non-metropolitan regions.

A variety of methodologies have been used to investigate aspects of regional economic performance in Australia, but predominately the preferred approach has been to use a multiple regression model, most typically an Ordinary Least Squares (OLS) model, and sometimes a backward step-wise regression model.

Modelling approaches other than straight regression analysis have also been used in research investigating regional performance in Australia. Examples include the following:

* using a binomial logit model in a cross-sectional study investigating the relationship between education, skills and qualifications and the economic performance of the five mainland States of Australia (Lawson and Dwyer, 2002);

* using Principal Components Analysis (PCA) to model regional variations in human capital (Stimson, Baum, Mangan, van Gellecum, Shyy and Yigitcanlar, 2004);

* developing typologies of community opportunity and vulnerability and using Multi-Discriminant Analysis (MDA) to describe the characteristics of the categories in those typologies e.g. the study by Baum et al. (1999) and Baum et al. (2006);

* developing typologies to produce functional classifications of regional cities and towns across Australia and showing how those have evolved over time (Beer, 1999; Beer and Maude, 1995); and

* using Shift-Share Analysis and the national shift component of employment change by industry sector to shed light on the proposition that an explanation of differences in regional employment growth was that some strongly performing regions are more specialized in rapidly growing industry sectors--like mining--across the Australian Bureau of Statistics Labour Market Regions over the period 1996-2001 (BTRE, 2004a; 2004b).

To take account of the spatial autocorrelation problem that is inherent in using aggregated spatial data, especially where it is based on de jure regions, it is important for modelling to use a spatial dependence test, along with a multicollinearity test. In addition, a Spatial Error Model (SEM) and a Spatial Autoregressive (SAR) Model should be employed. This was the approach used by Stimson, Mitchell, Rohde and Shyy (2011) in their study of variations in endogenous regional employment performance of FERs over the decade 1996-2006, and it is the approach used in this paper.

There have also been attempts to forecast the growth of regions into the future (Adams, 2002; Beer, 2002) looking at the effects of:

(i) national shifts in employment on regional growth;

(ii) initial industry structure on regional growth;

(iii) industrial diversity; and

(iv) the level of education, skills and qualifications.

It is only during the last decade that regional research in Australia has specifically focused on measuring and modelling endogenous regional performance, initiated by Stimson, Robson and Shyy (2004; 2005; 2006), and used also by Plummer, et al. (2014).

It is worth noting that, while traditionally it has been common for economists to theorize about regional convergence occurring over time in phenomena such as income, it is very clear from the empirical research investigating regional economic performance across Australia that, rather than regional convergence occurring, there appears to be divergence with a considerable degree of unevenness of performance being the rule across regions.

3. DATA AND METHODOLOGY

The analysis of endogenous regional employment performance across Australia over the decade 2001-2011 reported in this paper follows the methodology used by Stimson, Mitchell, Rohde and Shyy (2011) in the previous analysis for the decade 1996-2006.

Spatial Units

The spatial unit of analysis used are Functional Economic Regions (FERs) that have been compiled by the authors, as reported by Stimson et al. (2016), using the Intramax procedure developed by Masser and Brown (1975). The building blocks for the FERs are the Australian Bureau of Statistics' Statistical Areas Level 2 (SA2s). It uses the 2011 census journey-to-work data to analyse commuting patterns with FERs being demarcated to maximise within-region coincidence between where people live and where they work. The advantage of using FERs is that they tend to overcome, or at least minimise, the spatial autocorrelation problem encountered in the use of de jure regions (such as Local Government Areas), as has been demonstrated in the Stimson, Mitchell, Rohde and Shyy (2011) paper.

It is worth noting that the method used to demarcate the FERs was not constrained by restricting them to be fall within State and Territory borders, as is the case with the Australian Bureau of Statistics Labour Market Regions. That means some FERs cross over the state border between New South Wales and Victoria, along the eastern part of the border between New South Wales and Queensland, and along some of the border between Victoria and South Australia.

Within and around the capital city metropolitan city regions there are multiple FERs: 10 across the Sydney-Newcastle-Wollongong conurbation; seven across the Melbourne-Geelong region; six across the Brisbane-South East Queensland region that extends north to the Sunshine Coast and south to the Gold Coast; four across the greater Perth region; but only two across the greater Adelaide region. That reflects the emergence over time of a multi-centre spatial structure for Australia's big cities and the regionalisation of metropolitan labour markets.

Beyond the metropolitan city regions, the FERs tend to become much larger as the degree of remoteness and sparsity of settlement (and thus remoteness) increases. In addition, they are often elongated in shape along the main roads, which is not surprising.

Model Variables

Following the framework proposed by Stimson and Stough [with Salazar] (2009) and Stimson et al. (2009b), the modelling approach here uses, as the dependent variable, a surrogate measure of endogenous regional employment performance over time. This is measured as the differential (or regional) component derived from a Shift-Share Analysis of regional employment change over the decade 2001-2011, standardised by the size of the regional labour force at the beginning of the period.

The set of independent (or explanatory) variables used is the same set of 32 variables used in the previous studies cited above, 27 of which are derived from census data, and five of which are explicit locational variables (see Table 1).

Modelling Approaches

As per the Stimson, Mitchell, Rohde and Shyy (2011) analysis for the decade 1996-2006, a range of models are applied to investigate the potential causes of the spatial variation in endogenous regional employment performance over the decade 2001-2011:

1. First, an OLS full model was run without allowing for spatial effects. Spatial dependence tests were then carried out, including the Lagrange Multiplier (LM) tests, and the Moran 's I test which was run on residuals (see Anselin et al, 1996; Anselin, 1988). A multicollinearity test was also completed.

2. Second, a backward step-wise regression (based on AIC) was employed to derive an OLS specific model. Again, spatial dependence tests and a multicollinearity test were implemented.

3. Third, using the same variables, a Spatial Error Model (SEM) was run, which includes a lagged spatial error term.

4. Finally, for completeness and for comparison, a Spatial Autoregressive (SAR) Model was carried out on the same variables, which includes a lagged dependent variable.

4. SPATIAL PATTERNS OF ENDOGENOUS REGIONAL EMPLOYMENT PERFORMANCE

It is important to understand the economic context of the decade 20012011 in Australia. The decade began just after the 2000 Sydney Olympics. The long-boom years of economic growth that had begun following the recession of the early 1990s continued until later in the first decade of the new millennium.

Not surprisingly then, one would expect there to be marked variations in the direction and the magnitude of endogenous regional employment performance over the decade 2001-2011 across Australia's FERs, and that is the case as clearly shown in Figure 1.

When the world was impacted by the sharp downturn of the Global Financial Crisis (GFC), fortunately it had a relatively low aggregate impact on Australia. But it did have significant variable regional impacts. The decade was also characterised by the resources boom led by high commodity prices and an escalation in mining investments, output and exports, which was to create circumstances for what has been described as a 'two-speed economy'.

Following the elimination of some of the very remote and barely inhabited areas characteristic of a vast continent such as Australia, 134 FERs remained. Over the decade 2001-2011 only 46 of these FERs (or 34%) recorded a positive score on the endogenous regional employment performance measure, with only seven of those FERs having a strong positive performance. Thus, the big majority of the FERs (88 or 66%) recorded negative scores on the endogenous regional employment performance variable, with four of them having a strong negative performance.

Table 2 lists the FERs that were the top 25 positive performers and those that were the bottom negative performing FERs on the endogenous regional employment performance measure for the decade 2001-2011.

Some distinctive characteristics are evident from the pattern, across Australia, of positive and negative performance of FERs on the endogenous regional employment performance variable over the decade 2001-2011, and these are discussed below.

Positive Endogenous Regional Employment Performance

The positive performing FERs are located across much of the capital city metropolitan regions, including Melbourne, Brisbane, Perth, Darwin, Hobart and across the ACT. But this was not the case for Adelaide or Sydney.

Positive performance was also spread across some areas of coastal NSW, Queensland, eastern Victoria, Western Australia and much of the nation's inland wheat-sheep belt. Furthermore, positive performance is found in some of the mining regions and in some of the indigenous settlement regions of Western Australia, Queensland and the Northern Territory. The existence of positive endogenous regional employment performance for some of the FERs that are indigenous settlements is surprising, but probably explained by the concerted public policy efforts of governments to create indigenous employment, which is in fact an exogenous factor but is picked-up by default in the measure of the REG_SHIFT dependent variable.

Negative Endogenous Regional Employment Performance

The negative performing FERs are located widely across regional Australia beyond the capital cities and especially across the nation's vast remote areas. That includes farming and grazing regions of western Victoria, South Australia, parts of central and western Queensland, and Western Australia. Those regions largely form Australia's extensive wheat-sheep belt. Some of the negative performing FERs are also found in parts of coastal New South Wales, Queensland, Victoria, South Australia and Western Australia. Most of the FERs in Tasmania also had negative performance.

Within the capital city metropolitan regions, negative performance on the endogenous regional employment performance dependent variable was also present across all of Adelaide, all of Sydney, in eastern Melbourne, and in the north-west of Brisbane.

5. THE MODELLING AND RESULTS

As indicated earlier, several approaches were used to model the role the independent variables might play in explaining the variation in the dependent variable across the FERs. These models and their results are discussed below.

The Ordinary Least Squares Full Model

First the OLS regression full model was run without allowing for spatial effects (see Table 3). The [R.sup.2] is quite high (0.9414). Only nine of the independent variables are significant in explaining the variation in the dependent variable, six having a positive and three having a negative influence on the endogenous regional employment performance of FERs. The positive effects result from variables relating to:

* industry diversification/specialization at the beginning of the decade and the decade 2001-2011;

* the region's structural change index at the start of the decade;

* population change over the decade;

* the level of unemployment at the beginning of the decade;

* change in the incidence of information jobs; and

* the measure of remoteness.

The negative effects are from the variables relating to:

* the initial level of income; and

* change in the incidence both of people with bachelor qualifications and of people with technical qualifications.

It is noteworthy that some variables have a different direction (positive/negative) to model outcomes in the Stimson, et al. (2011) paper analysing the decade 1996-2006.

The Anselin Lagrange Multiplier (LM) tests--both the original and the robust tests--were used to test for spatial dependence, both error and lag (Table 4). In addition, the Moran's I test on residuals was run (Table 5). The full model shows no evidence of spatial dependence (lag or error) according to the LM tests or Moran's I on residuals.

A test for multicollinearity using variance inflation factors (VIF) was run (Table 3). Obviously, some of these are very high.

The Backward Step-Wise Regression Model

A backward step-wise regression (based on AIC) was run to derive an OLS specific model (Table 6). This reduces, to 15, the number of independent variables that are relevant to explaining the dependent variable. Once more, the [R.sup.2] is quite high (0.937).

Twelve variables are significant, 10 having a positive effect and two having a negative effect. Again, some variables have a different direction (positive/negative) to model outcomes in the Stimson, Mitchell, Rohde and Shyy. (2011) paper analysing the decade 1996-2006.

The independent variables having a significant positive effect on FER endogenous regional employment over the decade 2001-2011 relate to:

* industry diversification/specialisation at the beginning of the period and for the change in it over time;

* the region's structural change index at the beginning of the decade;

* population change over time;

* the level of unemployment at the beginning of the decade;

* the change in the degree of concentration of jobs in information and in finance;

* the incidence of volunteering (as a surrogate measure of social capital); and

* the degree of regional remoteness.

Negative effects on endogenous regional employment performance are related to:

* the level of income at the beginning of the decade; and

* the change in the incidence both of people with bachelor qualifications and of people with technical qualifications.

Results of the Anselin's LM and Morans I tests are presented in tables 7 and 8, respectively. There is no spatial dependence according to the Anselin LM tests, but there is significant (at the 0.05 level) error due to spatial dependence using the Moran's I. Multicollinearity test results are reported in Table 6. According to most of the literature, these results are quite acceptable, though some authors advocate for VIFs less than 6.

Spatial Regression Models

Spatial Error Model (SEM)

Given the possibility of spatial error dependence evidenced from the Moran's I test on residuals, the same variables were run in a Spatial Error Model (SEM), which includes a lagged spatial error term (Table 9).

The results show there is little difference between the OLS and the SEM. Those variables that were significant in the OLS are still significant in the SEM, all coefficient directions are the same, and there are only minor variations in magnitudes. Regarding the SEM, the spatial error coefficient, lambda, is significant (p value 0.028) and the AIC is slightly lower, but the Likelihood Ratio test is not significant (p value 0.0733), thus pointing to the OLS as the preferred model.

Spatial Autoregressive Model

For completeness and comparison, the Spatial Autoregressive (SAR) Model was also run, which includes a lagged dependent variable (Table 10).

In the SAR model, LQ_MAN_CH becomes significant, but otherwise the results are similar to those for the OLS model and the SEM. Importantly, the lagged dependent variable coefficient (rho) is not significant (p value 0.548) and the AIC is higher than for the OLS model.

Summary

In summary, the OLS does appear to be the best model, negating the need for spatial models. This is an interesting finding, and it confirms the supposition that the use of a functional rather than a de jure spatial base should help overcome the issue of spatial autocorrelation.

By way of an aside, the Moran's I statistic for the REG_SHIFT variable is significant, meaning the REG_SHIFT itself shows some spatial dependence. However, it is no longer significant in the SAR model (Table 10), which is interesting.

6. POLICY ISSUES

Regional Policy Interventions in Australia

In Australia it has been common to have government involvement in implementing explicit regional development policy, but that has waxed and waned over time. Such involvement has tended to have been focused almost exclusively on non-metropolitan regions and rarely on metropolitan regions. Such policies are what O'Connor, et al. (2001) have referred to as 'place-specific' policies. The interventions have typically been about, inter alia, the following:

* investments in infrastructure (including transportation projects, dams and irrigation systems);

* grants for community facilities;

* providing higher education facilities; and

* industry attraction schemes, which are essentially about 'picking winners' and which have a long history of failure.

Regional economic development policy and programs are almost exclusively the concern of the state/territory governments.

In addition, some government policy and programs that are 'people-specific' can have regional impacts, one of the most notable being the immigration policy of Commonwealth governments, with immigrants overwhelmingly choosing to live in the major cities, especially Sydney and Melbourne and in specific areas within them.

Over the last two to three decades regional development policy has tended to be focused more on developing local capacity and enhancing competitiveness, which is about bolstering-up endogenous processes. But often the implementation of such policy approaches has been characterised by 'picking winners' as illustrated by the Western Australia experience discussed by Plummer, et al. (2014).

It is often the case that attention has been directed towards intervening in poorly performing/lagging regions, rather than making investments to further enhance the performance of successful regions.

Implications of the Modelling for Policy

What might be the implications for policy of the modelling undertaken for this paper? Several are evident if the purpose of regional policy programs is to enhance the endogenous performance of regions.

It is evident that marked differences persist in the pattern of endogenous regional employment performance across Australia, with the large majority of FERs displaying negative performance over the decade 2001-2011. There are marked divides between the regional and some of the metropolitan region FERs, but that is not universal. Across regional and rural Australia there are pockets of positively performing FERs, so it is not all 'gloom-and-doom' across Australia's regional and rural areas. But nor is it all booming across the metropolitan regions.

Of special concern from the experience of the 2000s is the negative performance of FERs in the Sydney metropolitan region, which was a reversal of the situation for prior inter-census decades. Was this a post-Olympics effect? And was it an outcome of planning policy restricting land release and a reaction to the Labor Premier of New South Wales, Bob Carr, declaring that Sydney was closed to expansion? For Australia's global city, this negative endogenous growth employment performance was a disturbing outcome.

It is also disturbing that the Adelaide metropolitan region continues to be a negative performer.

The modelling certainly highlights the difficulty for regional development policy to be formulated in a global sense. This suggests the need for a region-specific policy approach rather than a 'one-size-fits-all' approach.

It is evident that using functional in contrast to de jure regions as the target for regional policy would make more sense than continuing the common practice of directing programs and investment to Local Government Authorities. It is understandable that that there has been a focus on Local Government Authorities as the third-tier of government in Australia, and they are in fact the creatures of State governments. However, these de jure regions are largely historic in origin, although it is not uncommon for State governments to force Local Government amalgamations. At the least, regional development strategies and the investments associated with them should recognise that it is not often the case that a single local government entity equates with a functional economic region (a functional labour market). As a result, there should be an insistence that there be collaboration between the local government entities that might equate with a functional economic unit, and that a regional development strategy be formulated for such a functional entity.

From the modelling undertaken for this paper, it is clear key factors that seem to underpin positive endogenous regional employment performance relate to a region's industry diversity/specialisation, its structural characteristics, and the degree of concentration of employment in information jobs and in finance jobs. In addition, population growth seems to be positively associated with positive endogenous regional employment performance. Not surprisingly, the remoteness of a FER also seems to be a factor that enhanced the performance of some FERs as the decade 2001-2011 coincided with the remarkable resources boom experienced by Australia, with mining activities especially, being highly concentrated in very remote locations and with many highly productive agricultural and pastoral regions that are in relatively remote inland areas. As a hallmark of what was referred to as Australia's 'two-speed' economy, the resources boom sucked jobs out of the non-mining sector adversely impacting non-resource regions.

The reality is there is little that government interventions could do with respect to these factors that related to structural transition in the economy, with some regions being 'winners' while others were 'losers'. However, since the end of the 2000s, the resources boom ended abruptly, so it might be expected that future modelling focusing on endogenous regional employment performance for the current decade will reveal perhaps a different role being played by those regional factors.

The level of unemployment at the beginning of the decade 2001-2011 does appear to assume some significance in a positive way. This is possibly because many of the more remote regions would have experienced jobs growth relating to the resources boom over the decade, with many such regions traditionally having somewhat higher than normal unemployment.

It is interesting that the modelling for the decade 2001-2011 did not reveal population size per se to be a statistically significant factor enhancing endogenous regional employment performance, while population growth over the decade did significantly affect positive performance.

Thus, positive performance was not necessarily the prerogative of large regional labour markets, and nor were small size FERs necessarily poor performers.

Similarly, and perhaps also surprisingly, regional income levels at the start of the decade were shown to, in fact, have a negative effect on endogenous regional employment performance.

But the most surprising result from the modelling was that factors relating to levels of regional human capital were not significant in explaining positive endogenous regional employment performance. Indeed, the modelling showed that change in the incidence of people with bachelor qualifications and change in the incidence of people with technical qualifications, in fact, had a negative influence on regional endogenous performance (at least for the decade 2001-2011). These results are counterintuitive to much of the research on regional development which postulates that improved levels of human capital will improve economic performance. This finding poses questions about outcomes of public policy to encourage engagement in tertiary education, including; the increasing investment that has been occurring in post-school education and training, the massive rise that has occurred in the number of students attending tertiary education institutions and, as a result, the very large increase that has been occurring in the number of tertiary-educated young people of workforce age.

In contrast, the supposed positive effect of increasing social capital--as measured, albeit inadequately, by the incidence of volunteering--does appear to be a factor that is a significant positive factor for enhancing endogenous regional employment performance. Enhancing social capital has been receiving some attention by governments.

Perhaps the most important lesson to take from the modelling for the decade 2001-2011 is that policy interventions to enhance endogenous regional employment performance might be those that relate to the structural characteristics of a region and enhanced diversification of employment, along with enhancing social capital. Interventions to enhance human capital might be worthwhile goals in themselves, but do not appear to be having a positive impact on endogenous regional employment performance.

7. CONCLUSION

This paper continues the research thrust initiated a decade or so ago to investigate endogenous regional employment performance across Australia's regions. That work has operationalised a model framework along the lines proposed by Stimson and his collaborators, and as set out in detail in Stimson, and Stough [with Salazar] (2009) and Stimson et al. (2009b).

The study reported here has focused on:

(a) analysing the patterns of endogenous regional employment performance for the decade 2001-2011; and

(b) modelling the potential determinants of variations in that performance.

A functional as against a de jure spatial base was employed using a new functional geography of FERs (developed by the authors and reported in a published paper by Stimson, et al. (2016)) across both Australia's major metropolitan regions and beyond across the vast expanses of regional Australia. The econometric modelling described in this paper indicates that using FERs appears to overcome the spatial autocorrelation issue inherent in using a de jure regional demarcation, which was also found to be the case in the earlier work by Stimson, Mitchell, Rohde and Shyy (2011) which modelled the endogenous regional employment performance of FERs for the decade 1996-2006. A series of econometric models were run:

(a) first a full OLS model and then a backward step-wise regression OLS spatial specific model (for both models the Anselin Lagrange Multiplier (LM) and Moran's I spatial dependence tests were run, along with a multicollinearity test); and

(b) second spatial regression models were run, both a Spatial Error Model and also a Spatial Autoregressive Model.

The results from these models were discussed.

We judge that an OLS model would be the preferred modelling approach when using a functional spatial base to investigate potential factors explaining the positive or negative performance of FERs in regard to endogenous regional employment over the decade 2001-2011. This finding confirms what Stimson, Mitchell, Rohde and Shyy (2011) also found in their analysis for the decade 1996-2006. It certainly leads us to conclude that a functional spatial base is preferable to a de jure spatial base that has more commonly been used in econometric modelling investigating regional economic performance in Australia. Modelling based on de jure regions has typically been the focus for regional development policy interventions, which is probably not a suitable policy approach.

It is evident from the empirical findings of research investigating regional economic performance in Australia that considerable regional differentiation persists. The gaps are wide. That is particularly evident from the research explicitly focusing on measuring endogenous regional employment performance across the nation's FERs as reported in the paper by Stimson, Mitchell, Rohde and Shyy (2011) for the decade 1996-2006 and in this paper for the decade 2001-2011.

Regarding the findings from the modelling, and depending on which model is used, it appears that a positive influence on regional endogenous employment for FERs over the 2001-2011 period is significantly related to factors to do with:

* regional industry diversification/specialisation at the beginning of the decade;

* the structural change index for the region; population change over time;

* the incidence of employment in information jobs and possibly in finance jobs;

* the initial level of unemployment;

* the level of social capital as measured by the incidence of volunteering; and

* regional remoteness.

A negative influence is significantly related to factors to do with:

* the initial level of regional income; and

* the incidence of people with bachelor and technical qualifications.

There is a need for further work to be undertaken to enhance our understanding of endogenous regional employment performance across FERs in Australia. For example, it would be worthwhile to explicitly focus the modelling exclusively on FERs beyond the major metropolitan regions. It might also be worthwhile partitioning Australia into groups such as the capital city metropolitan regions and for regional Australia into size category or remoteness category FERs to explicitly analyse endogenous processes in the FERs encompassing the larger and smaller regional cities and towns. Additionally, segmenting the analysis into the two five-year inter-census periods that comprise a decade might be worthwhile.

We also need to be aware that, over time, there will be changes in the boundaries of FERs due to both improvements to the transport infrastructure and changes in the distribution of employment across space. In addition, we need to be cognisant that the macro-economic conditions within which the processes of endogenous regional performance play out do change over time, will be specific to an inter-census period and will have exogenous impacts on regional performance.

ACKNOWLEDGEMENT: The research on which this paper is based is funded by the Australian Research Council (ARC) Discovery Project Grant #DP150103437.

REFERENCES

Adams, P. D. (2002). Prospects for Australia's Regions, Sustaining Regions, 1, pp. 4-16.

Anselin, L. (1988). Spatial Econometrics: Methods and Models. Kluwer Academic Publishers, Dordrecht.

Anselin, L., Bera, A., Florax, R. and Yoon, M. (1996). Simple Diagnostic Tests for Spatial Dependence. Regional Science and Urban Economics, 26, pp. 77-104.

Baum, S., O'Connor, K. and Stimson, R. (2006). Fault Lines Exposed: Advantage and Disadvantage across Australia's Settlement System, Monash University ePress, Melbourne.

Baum, S., Stimson, R., O'Connor, K., Mullins, P. and Davis, R. (1999). Community Opportunity and Vulnerability in Australia's Cities and Towns: Characteristics, Patterns and Implications. University of Queensland Press for the Australian Housing and Urban Research Institute, Brisbane.

Beer, A. (1999). Regional Cities within Australia's Evolving Urban System, 1991-96. Australasian Journal of Regional Studies, 5, pp. 329-348.

Beer, A. (2002). Editorial. Sustaining Regions, 1, p. 3.

Beer, A. and Maude, A. (1995). Regional Cities within the Australian Urban System, 1961 to 1991. Urban Policy and Research, 13, pp. 135-148.

Bradley, R. and Gans, J. (1998). Growth in Australian Cities, The Economic Record, 74(226), pp. 266-78.

Bureau of Transport and Regional Economics (BTRE) (2004a). Focus on Regions No. 1: Industry Structure. Information Paper 49, BTRE, Department of Transport and Regional Services, Canberra.

Bureau of Transport and Regional Economics (BTRE) (2004b). Focus on Regions No. 2: Education, Skills and Qualifications. Information Paper 51, BTRE, Department of Transport and Regional Services, Canberra.

Garnaut, J., Connell, P., Lindsay, R. and Rodriguz, V. (2001). County Australia: Influences on Employment and Population Growth. ABARE Research Report, 2001.1, Canberra.

Garnett, A. M. and Lewis, E. T. (2000). Population and Labour Mobility in Rural Australia. Australasian Journal of Regional Studies, 6, pp. 157-172.

Harrison, H. (1997). Trends in the Delivery of Rural Health, Education and Banking Services, National Focus. National Farmers Federation Research Paper, vol. II, Canberra, February.

Hogan, L., Berry, P. and Thorpe, S. (1999). Regional Australia: Incomes, Industry Location and Infrastructure. Australian Commodities, 6, pp. 674-687.

Johansson, B., Karlsson, C. and Stough, R. R. (2001). Theories and Endogenous Growth: Lessons for Regional Policy. Springer-Verlag, Heidelberg.

Lawson, J. and Dwyer, J. (2002). Labour Market Adjustment in Regional Australia, Reserve Bank of Australia, Sydney.

Lloyd, R., Harding, A. and Hellwig, O. (2000). Regional Divide? A Study of Incomes in Regional Australia. Australasian Journal of Regional Studies, 6, pp. 271-292.

Masser, I. and Brown, P. J. B. (1975). Hierarchical Aggregation Procedures for Interaction Data. Environment and Planning A, 7, pp. 509-523.

O'Connor, K., Stimson, R. and Daly, M. (2001). Australia's Changing Economic Geography: a Society Dividing. Oxford University Press, Melbourne.

Plummer, P., Tonts, M. and Martinus, K. (2014). Endogenous Growth, Local Competitiveness and Regional Development: Western Australia's Regional Cities, 2001-2011. Journal of Economic and Social Policy, 16(1, article 7), pp. 1-29.

Stimson, R. J., Stough, R. R. and Nijkamp, P. (Eds.) (2011). Endogenous Regional Development: Perspectives, Measurement and Empirical Investigation. Edward Elgar, Cheltenham.

Stimson, R. J. (2012), Endogenous Factors in Regional Performance: A Review of Research in Australia. In B. Johansson, C. Karlsson, and R. R. Stough (Eds) Knowledge and Talent in Global Regional Contexts. Edward Elgar, Cheltenham, pp. 159-193.

Stimson, R. J., Mitchell, W., Rohde, D. and Shyy, T.-K. (2011). Using Functional Economic Regions to Model Endogenous Regional Performance in Australia: Implications for Addressing the Spatial Autocorrelation Problem. Regional Science Policy and Practice, 3(3), pp. 131-144.

Stimson, R. J., Robson, A. and Shyy, T-K. (2004). Shift-Share Analysis and Modelling Endogenous Growth Across Queensland's Regions. North American Regional Science Association International, Annual Conference, Seattle, November.

Stimson, R. J., Robson, A. and Shyy, T.-K. (2009a). Measuring Regional Endogenous Growth. In R. Capello and P. J. Nijkamp (Eds) Regional Growth and Development Theories in the XXI Century: Theoretical Achievements and Future Challenges, Edward Elgar, pp. 354-372.

Stimson, R. J., Robson, A. and Shyy, T-K. (2009b). Modelling Regional Endogenous Growth: An Application to the Non-Metropolitan Regions of Australia. Annals of Regional Science, 43, pp. 379-398.

Stimson, R. J., Robson. A. and Shyy, T.-K. (2011). Modelling Endogenous Regional Employment Performance in Non-Metropolitan Australia: What is the Role of Human Capital, Social Capital and Creative Capital? In K. Kourtit, P. Nijkamp and R. R. Stough (Eds) Drivers of Innovation, Entrepreneurship and Regional Dynamics, Springer, Berlin, pp.179-204.

Stimson, R., Baum, S., Mangan, J., Van Gellecum, J. Y., Shyy, T-K. and Yigitcanlar, T. (2004). Analysing Spatial Patterns in the Characteristics of Work, Employment and Skills in Australia's Capital Cities and Across its Regional Cities and Towns: Modelling Community Opportunity and Vulnerability, Main Report. Prepared for the Australian National Training Authority (ANTA), CR-SURF/ANTA 2003 National Project, Centre for Research into Sustainable Urban and Regional Futures, University of Queensland, Brisbane.

Stimson, R., Mitchell, W., Flanagan, M., Baum, S. and Shyy, T.-K. (2016). Demarcating Functional Economic Regions Across Australia Differentiated by Work Participation Categories. Australasian Journal of Regional Studies, 22, pp. 27-57.

Stimson, R., Robson, A. and Shyy, T.-K. (2005), Modelling Determinants of Spatial Variations in Regional Endogenous Growth: Non-Metropolitan Regions in the Mainland States of Australia. 45th Congress of the European Regional Science Association, Amsterdam, August.

Stimson, R., Robson, A. and Shyy, T.-K. (2006). Modeling Regional Endogenous Growth: an Application to the Non-Metropolitan Regions of Australia. International Workshop on Creativity and Smart Policy as Signposts for Innovative Development, Tinbergen Institute, Amsterdam, May.

Stimson, R. J. Stough, R. R. and Roberts, B. H. (2006). Regional Economic Development: Analysis and Planning Strategy. Springer, Berlin (2nd edition).

Stimson. R. J. and Stough, R. R. (with Salazar, M.) (2009). Leadership and, Institutions in Regional Endogenous Development. Edward Elgar, Cheltenham, pp. 151.

Trendle, B. and Shorney, G. (2003). The Effect of Industrial Diversification on Regional Economic. Proceedings of the Australian and New Zealand Regional Science Association, Annual Conference, 2003, Perth.

Robert J. Stimson

Emeritus Professor, School of Geography, University of Melbourne, Vic, 3052 and School of Earth and Environmental

Sciences, University of Queensland, Qld, 4072, Australia.

Email: [email protected].

Michael Flanagan

Senior Research Assistant, Centre of Full Employment and Equity, University of Newcastle, Callaghan, NSW, 2308,

Australia.

Email: [email protected].

William Mitchell

Emeritus Professor, Centre of Full Employment and Equity, University of Newcastle, Callaghan, NSW, 2308, Australia.

Email: bill. [email protected]. au.

Tung-Kai Shyy

Research Fellow, School of Geography, Planning and Environmental Management, University of Queensland, Qld, 4072,

Australia.

Email: t. shyy@uq. edu. au.

Scott Baum

Professor of Urban and Regional Analysis, Griffith University, Brisbane, Qld, 4111, Australia. Email:

[email protected].

Caption: Figure 1: Map of Positive and Negative Scores on the Endogenous Regional Employment Performance (REG_SHIFT) Dependent Variable Measure, 2011-2011. Source the Authors.

Table 1. The Variables Used to Model Change in Endogenous Regional
Employment Change over the Decade 2001-2011 Across Australia's
Functional Economic Regions.

                     DEPENDENT VARIABLE

REG_SHIFT      Regional Shift component of a Shift-Share
               Analysis of Employment change (2001 to 2011)
               /Labour Force (2001)/1000

                    INDEPENDENT VARIABLES

Derived from Census Data

SPEC_01        Specialisation Index for 2001 (Herfindahl-
               Hirschman Index)

SPEC_CH        Change in Specialisation Index from 2001 to
               2011 (Herfindahl-Hirschman Index)

SCI            Structural Change Index (2001 to 2011)

SCI_CH         Change in the Structural Change Index (from
               2001-2006 to 2006-2011)

L_INC_01       Median Individual Income-2001 Annual (Log)
               (real)

L_INC_CH       Change in Median Individual Income-2001 to
               2011 Annual (Log) (real)

UNEMP_01       Unemployment rate in 2001

UNEMP_CH       Change in Unemployment rate from 2001 to 2011

L_POP_01       Log of population (2001)

L_POP_CH       Change in Log of population (2001 to 2011)

LQ_MAN_01      Location Quotient for the Manufacturing
               Industry in 2001

LQ_INF_01      Location Quotient for the Information media
               & telecommunications Industry in 2001

LQ_FIN_01      Location Quotient for the Financial &
               insurance services Industry in 2001

LQ_PRO_01      Location Quotient for the Professional,
               scientific & technical services Industry in
               2001

LQ_MAN_CH      Change in the Location Quotient for the
               Manufacturing Industry, 2001 to 2011

LQ_INF_CH      Change in the Location Quotient for the
               Information media & telecommunications
               Industry, 2001 to 2011

LQ_FIN_CH      Change in the Location Quotient for the
               Financial & insurance services Industry,
               2001 to 2011

LQ_PRO_CH      Change in the Location Quotient for the
               Professional, scientific & technical
               services Industry, 2001 to 2011

POSTGRAD_01    Proportion of labour force with a
               Postgraduate Degree or higher in 2001

BACHELOR_01    Proportion of labour force with a Bachelor
               Degree or higher in 2001

TECHQUALS_01   Proportion of labour force with technical
               qualifications in 2001

POSTGRAD_CH    Change in the Proportion of labour force
               with a postgraduate degree or higher, from
               2001 to 2011

BACHELOR_CH    Change in the Proportion of labour force
               with a bachelor degree or higher, from
               2001 to 2011

TECHQUALS_CH   Change in the Proportion of labour force
               with technical qualifications, from 2001 to
               2011

SYMBA_01       Proportion of Symbolic Analysts (Managers +
               Professionals) in Employment in 2001

SYMBA_CH       Change in the proportion of Symbolic
               Analysts (Managers + Professionals) in
               Employment from 2001 to 2011

VOLUNTEER_11   Proportion of Volunteers in Working Age
               Population (15-64) in 2011

Location Variables

A_COAST        Border is adjacent to coastline
               (No = 0; Yes = 1)

P_METRO        Border is adjacent to metropolitan
               statistical division (No = 0; Yes = 1)

D_URBAN        Classified as Urban under Australian
               Classification of Local Governments system
               (1 = Yes, 0 = No)

D_REMOTE       Classified as Remote under Australian
               Classification of Local Governments system
               (1 = Yes, 0 = No)

W_METRO        Border is within metropolitan statistical
               division (No = 0; Yes = 1)

Source: the Authors.

Table 2. The Bottom 25 Negative Performing FERs on the Endogenous
Regional Employment Performance Variable for the Decade 2001-2011.

Bottom 25 negative
  performing FERs             State/Territory,             REG_SHIFT
    (name given)               type of region                score

1. Petermann--       NT: Remote                          -437.2739166
Simpson

2. APY Lands         SA: Remote indigenous lands,        -367.6052307
                     north west

3. Bourke--          NSW: Inland far west, remote        -359.6419387
Walgett

4. Coober Pedy       SA: Inland, remote                  -320.0340477

5. Renmark--         SA: Inland, irrigation area         -290.1754182
Loxton

6. Victoria          NT: Remote, indigenous              -287.7707821
River

7. Dorset            TAS: Coastal, rural, north          -281.1617911
                     east

8. King Island       TAS: Island, rural remote           -265.9318968

9. Swan Hill--       NSW-VIC: Inland irrigation          -263.4774996
Deniliquin--         area
Wentworth

10. Snowy            NSW: Inland south, rural            -248.6266220
Mountains

11. Griffith--       NSW: Inland, irrigation area        -244.0103606
Narrandera

12. Manjimup--       WA: Coastal, rural, south west      -243.3089350
Bridgetown

13. Longreach and    QLD: Inland far west, pastoral      -217.2412607
surrounds

14. Carnarvon--      WA: NW remote mining                -214.1387132
Exmouth

15. Moree--          NSW- QLD: Inland north, rural       -213.3415212
Inverell
--Goondiwindi

16. Charters         QLD: Coastal and inland north       -204.8155484
Towers--Ayr

17. Brookton--       WA: Inland, rural                   -202.4857934
Narrogin
--Katanning

18. Parkes--         NSW: Inland western plains,         -196.4080787
Cobar                rural

19. Cape York        QLD: Far North Cape, remote         -195.2428712
Peninsula

20. Ingham--         QLD: Coastal north, rural           -194.9280890
Innisfail

21. York--           WA: Inland, rural                   -193.7412466
Dalwallinu
--Merredin

22. Carpentaria      QLD: Coastal, far north,            -192.9291385
                     remote

23. Tennant          NT: Inland remote                   -189.3799572
Creek--Barkly

24. Northern         NT: Metropolitan suburban           -187.5386101
Darwin suburbs

25. Mildura and      VIC-NSW: Inland, irrigation         -185.2242600
surrounds            area

  Top 25 positive
  performing FERs             State/Territory,             REG_SHIFT
    (name given)               type of region                score

1. Ashburton         WA: North west, remote mining       1475.2003860

2. Thamarrurr        NT: Indigenous area, remote         991.9831003

3. Port Hedland--    WA: North west, remote mining       751.6731816
Newman--East
Pilbara

4. Karratha--        WA: North west, remote mining       708.2806127
Roebourne

5. Rockingham--      WA: Metropolitan outer              443.3334812
Mandurah             suburban, south

6. Anindilyakwa      NT: Indigenous land council         439.2604002
                     area, remote

7. West Arnhem       NT: Remote, indigenous              309.9875564

8. Darwin City--     NT: Metropolitan inner              271.3055835
Inner suburbs

9. Western           SA: Inland, remote                  253.3676169

10. Weipa            QLD: Coastal, far north, remote     243.8628282

11. Mackay--         QLD: Coastal, regional city         232.8954608
Whitsunday           and rural

12. Gladstone        QLD: Coastal, north                 218.6443992
and surrounds

13. Sunshine         QLD: Metropolitan outer             197.4943849
Coast                suburban northern

14. Palmerston--     NT: Inland, remote                  183.0851687
Litchfield

15. Melbourne West   VIC: Metropolitan outer             162.6978148
--North West--       suburban, north and west
Bacchus Marsh

16. Gold Coast       QLD-NSW: Metropolitan outer         158.5154890
--Tweed              suburban, south

17. Bunbury--        WA: South west coastal,             142.9204609
Margaret River       regional city and wine area

18. Ipswich--        QLD: Metropolitan outer             138.8049684
Springfield          suburban, west

19. Hervey Bay--     QLD: Coastal, regional city         136.4600045
Maryborough          and rural

20. Midland--        WA: Inland rural                    134.8232042
Mundaring--
Gingin

21. Brisbane North   QLD: Metropolitan suburban,         128.5096330
--Moreton Bay        north
Region

22. Mornington       VIC: Metropolitan outer             98.1366192
Peninsula--          suburban, southeast
Dandenong--
Pakenham

23. North Perth--    WA: Metropolitan outer              98.0774790
Joondalup            suburban, north

24. Greater          QLD: Coastal regional city          91.2981568
Townsville

25. Fremantle--      WA: Metropolitan suburban,          90.9848153
South Eastern        south
Perth

Source: the Authors.

Table 3: Full OLS Model Results and Multicollinearity Test Results.

Coefficient     Estimate      Std.       t       p value
                             Error     value

(Intercept)      510.084    412.418    1.237    0.219025
SPEC_01          542.682    249.199    2.178    0.031756 *
SPEC_CH          725.606    372.764    1.947    0.054366
SCI              799.754    192.563    4.153    6.87e-05 ***
SCI_CH           67.315     212.352    0.317    0.751898
L_INC_01        -369.200    131.331    -2.811   0.005929 **
L_INC_CH         114.551    182.396    0.628    0.531398
UNEMP_01         13.761      5.238     2.627    0.009956 **
UNEMP_CH         -6.573      5.943     -1.106   0.271349
L_POP_01         24.596      23.474    -1.048   0.297240
L_POP_CH        3441.275    153.338    22.442   < 2.0e-16 ***
LQ_MAN_01        19.628      25.393    0.773    0.441342
LQ_INF_01        160.601     40.129    0.669    0.505141
LQ_FIN_01        402.623    203.362    1.980    0.050444
LQ_PRO_01       -133.266    124.184    -1.073   0.285772
LQ_MAN_CH        -67.364     39.632    -1.700   0.092259
LQ_INF_CH        73.053     280.635    2.042    0.043759 *
LQ_FIN_CH       -170.392    218.090    0.781    0.436459
LQ_PRO_CH        96.262      98.370    0.979    0.330131
POSTGRAD_01      42.977     1296.986   0.033    0.973631
POSTGRAD_CH     -1179.092   1739.547   -0.678   0.499439
BACHELOR_01      90.571     499.123    0.181    0.856370
BACHELOR_CH     -2557.620   730.100    -3.503   0.000687 ***
TECHQUALS_01     238.953    162.494    1.471    0.144524
TECHQUALS_CH    -1484.946   294.507    -5.042   2.03e-06 ***
SYMBA_01         259.703    315.795    0.822    0.412798
SYMBA_CH         183.322    632.459    0.290    0.772521
VOLUNTEER_11      2.412      2.367     1.019    0.310660
A_COAST          -20.734     15.912    -1.303   0.195511
P_METRO           4.405      21.777    0.202    0.840116
D_URBAN          -4.909      21.746    -0.226   0.821847
D_REMOTE         84.127      42.356    1.986    0.049723 *
W_METRO          -14.090     28.492    -0.495   0.621997

Coefficient     Multicollinearity
                    test (VIF)

(Intercept)             --
SPEC_01              8.907465
SPEC_CH              8.835060
SCI                  8.300719
SCI_CH               2.712929
L_INC_01             7.720642
L_INC_CH             6.698806
UNEMP_01             6.533965
UNEMP_CH            10.313893
L_POP_01             8.789758
L_POP_CH             2.708974
LQ_MAN_01            3.479791
LQ_INF_01           13.368863
LQ_FIN_01           20.924515
LQ_PRO_01           26.257744
LQ_MAN_CH            1.823345
LQ_INF_CH            1.653523
LQ_FIN_CH            5.857960
LQ_PRO_CH            4.123870
POSTGRAD_01         13.713052
POSTGRAD_CH         17.156961
BACHELOR_01         28.413478
BACHELOR_CH          6.886694
TECHQUALS_01         4.083310
TECHQUALS_CH         6.622777
SYMBA_01            14.158755
SYMBA_CH            16.353417
VOLUNTEER_11         5.758360
A_COAST              1.778726
P_METRO              1.281645
D_URBAN              3.626887
D_REMOTE             3.107093
W_METRO              2.633234

Notes: Residual standard error: 65.91 on 101 degrees of freedom;
Multiple [R.sup.2] = 0.9414; Adjusted [R.sup.2] = 0.9229;
F Statistic = 50.73(32, 101); p value < 2.2 [e.sup.-16].
* = significant at 0.05 level; ** = significant at 0.01 level;
*** = significant at 0.00 level. Source: the Authors.

Table 4. Anselin Lagrange Multiplier test: Spatial Error and
Spatial Lag Results.

est               [chi square]   df   p value

LM error             1.0049      1    0.3161
LM lag               1.3566      1    0.2441
Robust LM error      2.6684      1    0.1024
Robust LM lag         3.02       1    0.08224

Source: the Authors.

Table 5. Moran's I Test Results

Moran's I   z value   p value

0.062399     1.814    0.06967

Source: the Authors.

Table 6. Backward Step-Wise Regression OLS Specific Model
and Multicollinearity Test Results.

Coefficient     Estimate     Std.       t       p value
                             Error    value

(Intercept)      486.664    257.791   1.888    0.061506
SPEC_01          534.036    206.619   2.585    0.010966 *
SPEC_CH          895.148    288.790   3.100    0.002423 **
SCI             1006.150    128.078   7.856    2.05e-12 ***
L_INC_01        -372.183    101.600   -3.663   0.000374 ***
UNEMP_01         14.410      3.648    3.951    0.000133 ***
UNEMP_CH         -5.928      3.941    -1.504   0.135236
L_POP_CH        3340.952    111.646   29.924   2.00e-16 ***
LQ_FIN_01        195.943    87.959    2.228    0.027800 *
LQ_MAN_CH        -66.334    33.987    -1.952   0.053339
LQ_INF_CH        511.158    232.272   2.201    0.029705 *
BACHELOR_CH     -2704.582   389.691   -6.940   2.26e-10 ***
TECHQUALS_01     201.311    122.432   1.644    0.102783
TECHQUALS_CH    -1531.763   218.177   -7.021   1.51e-10 ***
VOLUNTEER_11      4.172      1.310    3.184    0.001858 **
D_REMOTE         96.245     32.977    2.919    0.004212 **

Coefficient     Multicollinearity
                    test (VIF)

(Intercept)             --
SPEC_01              6.646055
SPEC_CH              5.755288
SCI                  3.985473
L_INC_01             5.015021
UNEMP_01             3.438619
UNEMP_CH             4.922926
L_POP_CH             1.558675
LQ_FIN_01            4.248533
LQ_MAN_CH            1.455343
LQ_INF_CH            1.229364
BACHELOR_CH          2.129349
TECHQUALS_01         2.515898
TECHQUALS_CH         3.944820
VOLUNTEER_11         1.915148
D_REMOTE             2.044094

Notes: Residual standard error: 63.26 on 118 degrees of freedom;
Multiple [R.sup.2] = 0.937; Adjusted [R.sup.2] = 0.9289;
F Statistic = 116.9(15, 118); p value < 2.2 [e.sup.-16].
* = significant at 0.05 level; ** = significant at 0.01 level;
*** = significant at 0.00 level. Source: the Authors.

Table 7: Anselin Lagrange Multiplier Test Results: Backward
Step-Wise Regression OLS Specific Model.

                   [chi
Test              square]   df   p value

LM error          2.3964    1    0.1216
LM lag            0.4289    1    0.5125
Robust LM error   3.7228    1    0.0537
Robust LM lag     1.7553    1    0.1852

Source: the Authors

Table 8. Moran's I Test on Residuals Results.

Moran's I   z value   p value

0.096359    2.0674    0.0387 *

Source: the Authors.

Table 9. Spatial Error Model Results.

Coefficient      Estimate    Std. Error   z value      p value

(Intercept)      505.5062     241.1581    2.0962    0.0360679 *
SPEC_01          586.9817     194.5074    3.0178    0.0025463 **
SPEC_CH         1054.3037     274.7729    3.8370    0.0001245 ***
SCI             1010.0631     115.2328    8.7654    <2.2e-16 ***
L_INC_01         395.5645     94.4138     -4.1897   2.793e-05 ***
UNEMP_01         13.8385       3.4481     4.0134    5.986e-05 ***
UNEMP_CH         -5.0196       3.6979     -1.3574   0.1746421
L_POP_CH        3349.6546     109.5279    30.5827   <2.2e-16 ***
LQ_FIN_01        208.8935     88.2814     2.3662    0.0179705 *
LQ_MAN_CH        -60.5119     31.2101     -1.9389   0.0525192
LQ_INF_CH        582.1127     214.9754    2.7078    0.0067729 **
BACHELOR_CH     -2569.5230    370.7719    -6.9302   4.202e-12 ***
TECHQUALS_01     218.6299     120.1337    1.8199    0.0687761
TECHQUALS_CH    -1442.5091    201.6122    -7.1549   8.376e-13 ***
VOLUNTEER_11      4.4870       1.2194     3.6796    0.0002336 ***
D_REMOTE         83.4578      30.5998     2.7274    0.0063836 **

Notes: Lambda: 0.23906, LR test value: 3.2081, p-value: 0.073276;
Asymptotic standard error: 0.10888; z-value: 2.1956, p-value:
0.028124 *; Wald statistic: 4.8205, p-value: 0.028124 *;
Log likelihood: -735.7494 for error model; ML residual variance
(sigma squared): 3393.7, (sigma: 58.256); Number of observations:
134; Number of parameters estimated: 18; AIC: 1507.5, (AIC for
lm: 1508.7). * = significant at 0.05 level; ** = significant at
0.01 level; *** = significant at 0.00 level. Source: the Authors.

Table 10. Spatial Autoregressive (SAR) Model Results.

Coefficient      Estimate   Std. Error   z value      p value

(Intercept)      463.0629     243.9175    1.8984    0.0576380
SPEC_01          541.2858     194.2695    2.7863    0.0053320 **
SPEC_CH          898.9836     271.0777    3.3163    0.0009121 ***
SCI             1015.7014     121.0489    8.3908    <2.2e-16 ***
L_INC_01        -366.9723     95.3511     -3.8486   0.0001188 ***
UNEMP_01         14.5123       3.4177     4.2463    2.174e-05 ***
UNEMP_CH         -5.9810       3.6942     -1.6190   0.1054405
L_POP_CH         369.3973     114.2799    29.4837   <2.2e-16 ***
LQ_FIN_01        192.5118     82.5667     2.3316    0.0197221 *
LQ_MAN_CH        67.3119      31.8483     -2.1135   0.0345563 *
LQ_INF_CH        469.4857     227.3773    2.0648    0.0389431 *
BACHELOR_CH     -2674.0478    371.5417    -7.1972   6.148e-13 ***
TECHQUALS_01     207.3834     115.3710    1.7975    0.0722507
TECHQUALS_CH    -1533.2985    204.4728    -7.4988   6.439e-14 ***
VOLUNTEER_11      4.2163       1.2306     3.4262    0.0006120 ***
D_REMOTE         96.8746      30.9031     3.1348    0.0017198 **

Rho: -0.026617, LR test value: 0.39186, p value: 0.53132; Asymptotic
standard error: 0.044357; z value: -0.60006, p value: 0.54846; Wald
statistic: 0.36008, p value: 0.54846. Log likelihood: -737.1575 for
lag model; ML residual variance (sigma squared): 3513.3,
(sigma: 59.273); Number of observations: 134; Number of parameters
estimated: 18; AIC: 1510.3, (AIC for lm: 1508.7); LM test for
residual autocorrelat * = significant at 0.05 level; ** =
significant at 0.01
level; *** = significant at 0.00 level. Source: the Authors.