Portal de Periódicos da CAPES

A spatial microsimulation approach to economic policy analysis in Scotland

2013; Elsevier BV; Volume: 5; Issue: 3 Linguagem: Inglês

10.1111/rsp3.12009

1757-7802

Malcolm Campbell, Dimitris Ballas,

Spatial and Panel Data Analysis

Regional scientists have increasingly been playing a very important role in the development and application of spatial microsimulation models for policy analysis. It has long been argued that spatial microsimulation modelling has enormous potential for the evaluation of the socio-economic and spatial effects of major developments in the regional or local economy. This paper aims to add to this rapidly expanding work, by presenting a new spatial microsimulation model (SIMALBA) for Scotland (the development of which was co-funded by the Scottish Government) and by demonstrating how it can be used to perform what-if policy analysis in Scotland. The focus of the paper is on economic aspects of social and spatial inequality in the capital of Scotland, Edinburgh. The paper shows how spatial microsimulation modelling can address previously unanswered research questions in Scotland, particularly those relating to fiscal policy. The SIMALBA model has estimated income data for Scotland at output area level geography and this is the focus of the various 'what-if' policy scenarios. Simulated data has been created using a deterministic reweighing algorithm to build a spatial microsimulation model by combining UK Census data for 2001 and Scottish Health Survey (SHS) data for 2003. The analysis demonstrates the importance of geography by examining trends at OA level in Scotland. The paper concludes with a discussion of the simulated data and resulting policy scenarios as well as the impact of this analysis for policy formation in Scotland. Resumen. Los investigadores de ciencias regionales llevan tiempo jugando un papel cada vez más importante en el desarrollo y aplicación de modelos de microsimulación espacial para el análisis de políticas. Durante mucho tiempo se ha argumentado que la modelización de microsimulación espacial tiene un potencial enorme para la evaluación de los efectos socioeconómicos y espaciales de los principales cambios en la economía regional o local. Este artículo tiene por objeto contribuir a esta corriente de estudio en rápida expansión, mediante la presentación de un nuevo modelo de microsimulación espacial (SIMALBA) para Escocia (cuyo desarrollo fue co-financiado por el Gobierno de Escocia) y una demostración de su utilización para realizar un análisis 'what-if' para políticas en Escocia. Este artículo se centra en los aspectos económicos de las desigualdades sociales y espaciales de Edimburgo, la capital de Escocia. El artículo muestra cómo la modelización de microsimulación espacial puede abordar cuestiones de investigación hasta ahora sin responder en Escocia, en particular las relativas a las políticas fiscales. El modelo SIMALBA ha estimado los datos de ingresos de Escocia a nivel geográfico de área de salida (OAC en inglés) y éste es el objetivo principal de los diferentes escenarios de políticas 'what-if'. Los datos de simulación se crearon utilizando un algoritmo de reponderación determinista para construir un modelo de microsimulación espacial mediante la combinación de datos del censo de 2001 del Reino Unido y datos de 2003 de la Encuesta de Salud de Escocia (SHS por sus siglas en inglés). El análisis demuestra la importancia de la geografía mediante el examen de tendencias a nivel de área de salida en Escocia. El artículo concluye con una discusión de los datos de simulación y los escenarios políticos resultantes, así como del impacto de este análisis para la formulación de políticas en Escocia. It has long been argued that there is a need for a geographical approach to social policy and welfare analysis. Although there have been long very important theoretical debates about the relationships between social context, social norms and human needs (e.g., see Runciman 1966; Sen 1987; Townsend 1987; Doyal and Gough 1991; Gordon and Pantazis 1997; Frank 2005; Wilkinson and Pickett 2009; Dorling 2011) there have been relatively limited attempts to add a geographical dimension. Among the first comprehensive attempts to add a geographical dimension to these debates is the seminal work of Smith (1973) exploring the geography of social well-being in the United States and adopting a geographical approach to the analysis of welfare (Smith 1977). In addition, Bennett's (1980) comprehensive work on the geography of public finance strongly argued for a spatial dimension to the traditional public finance analysis by including (with the basic public finance distribution between people question) another set of issues concerned with the geographical dimension of this distribution and putting the question of how burden and public expenditure vary as a function of spatial location. Another very good and more recent example of a comprehensive geographical approach to the analysis of social issues is the work of Pacione (1995a, 1995b, 1997). There have also been important attempts to provide evidence-based analysis of social and spatial inequalities (e.g., see Burrows and Rhodes 1998; Madden 1993; Mitchell et al. 2000; Dorling et al. 2007). Overall, it can be argued that regional scientists are particularly well suited to address these issues. There is a long history of modelling work in geography and regional science that focuses on the assessment of the various short-term and long-term effects of major socio-economic regional or local developments. But, there has been relatively limited work on the development and application of regional science models for the estimation of the geographical impacts of national public policy. Nevertheless, over the past thirty years there has been a rapidly increasing number of regional scientists around the world who have been involved in the development and application of spatial microsimulation models (for recent reviews see Ballas and Clarke 2009; Birkin and Clarke 2011). Spatial microsimulation is a methodology that attempts to estimate the demographic and socio-economic characteristics of human behaviour of individual people or households (Clarke 1996; Ballas et al. 2005). Such data, although routinely collected by governments from censuses and population surveys are not typically available to researchers and policy-makers due to privacy and confidentiality concerns. Spatial microsimulation builds on a long successful history of traditional microsimulation models, which have been used widely to analyse re-distributive effects under different policy scenarios. In particular, microsimulation has been broadly developed and used by economists over the last 40 years.1 Microsimulation methods aim to examine changes in the lives of individuals within households and to analyse the impact of government policy changes for each individual and each household (Hancock and Sutherland 1992; Harding 1996; Mitton et al. 2000). In an economic and social science context, microsimulation can be defined as a type of computer program that simulates how a social policy would operate under proposed changes and how particular types of individuals would be affected or react. Recent examples of using microsimulation for policy analysis in Europe include the EUROMOD microsimulation model used for various types of analysis (Lelkes and Sutherland 2009). Static microsimulation involves the analysis of a population microdata set at one point in time for policy analysis. For instance, economists have been involved in the development of static microsimulation models that are capable of analysing the impacts of a particular social policy scheme upon different types of households and individuals as well as the redistributional impacts of the government budget changes. The results of microsimulation models are widely quoted in the media when covering the possible impact of government budget changes upon different types of households (Clarke 1996; Ballas et al. 2005). Spatial microsimulation adds a geographical dimension to traditional economic microsimulation models. In particular, adding spatial detail to traditional microsimulation involves creating a microdata set, as well as using it. Such a microdata set refers to a particular locality, to a geographically well-defined and restricted area such as census output areas (OA). There are very few sources of geographically detailed microdata sets, so there is a need to create these datasets using static spatial microsimulation techniques. The latter involve the merging of census and survey data to simulate a population of individuals within households (for different geographical units), whose characteristics are very close to the real population. They can then be used to answer questions pertaining to the geographical, as well as the socio-economic impacts of urban, regional or national government policies (Clarke 1996; Ballas et al. 2005). In particular, over the past decade there has been considerable progress in the development and application of spatial microsimulation models for national policy analysis by regional scientists in a wide range of fields including social policy (e.g., Ballas and Clarke 2001; Chin et al. 2005; Ballas et al. 2007), poverty small area estimation and analysis (Ballas 2004; Tanton 2011), health (Tomintz et al. 2008; Morrissey et al. 2008; Edwards and Clarke 2009; Smith et al. 2011), agricultural policy (Ballas et al. 2006a; Hynes et al. 2009), international migration (Rephann and Holm 2004), educational policy (Kavroudakis et al. 2012) and crime analysis (Kongmuang et al. 2006). The research presented in this paper aims to build on this policy-relevant work by presenting a spatial microsimulation model for the analysis of national public policies in Scotland. This is, to the best of our knowledge, the first attempt to build a spatial microsimulation model specifically for Scotland. It can be argued that efforts to build models, such as the one presented in this paper, would be particularly useful and important in the changing public policy context since devolution and especially given the current debate about a referendum on Scottish independence (for more information on the changing policy context see Keating 2009, 2010; Danson et al. 2012; Danson and Lloyd 2012; Law and Mooney 2012). This paper presents SIMALBA,2 a spatial microsimulation model of the Scottish population and it shows how it can be used to model the geographical impacts of national fiscal policies, focusing on the capital of Scotland, Edinburgh. The paper proceeds as follows: Section 2 briefly discusses the UK and Scottish policy context and outlines examples of policies that could be modelled using a spatial microsimulation model. Section 3 presents the data and methods that were used to develop the SIMALBA model. Section 4 shows how SIMALBA was used to estimate the spatial impacts of the policies discussed in Section 2. Section 5 offers some concluding comments. Before discussing our spatial microsimulation model and how it was used for policy analysis it is useful to provide a brief discussion of the current situation of government in the UK and Scotland in particular. It should be noted that there is a multi-level governance system at work, with regional (Welsh, Scottish and Northern Irish) parliaments responsible for devolved matters. The devolved administrations do not have any control over defence or international relations but over specific devolved areas such as health and education for example (Cairney 2006). In addition, there is an unusual situation at present at the national (UK) level, with a coalition government in place with a set of proposals for welfare reforms which have been described as "perhaps the most radical reshaping of the British welfare system since its introduction post-1945" (Hamnett 2011, p.147). Fiscal policy is still driven primarily by the UK government, although the Scottish government does have some tax varying powers. The tax varying powers extend to adjusting the basic rate of income tax by three pence in the pound if desired (Mair and McAteer 1997), the so called 'tartan tax'. The economic policy which a government pursues can have real and lasting consequences in people's lives, as it affects the levels of employment, unemployment benefits, the income distribution and education and training. These consequences also have important geographical dimensions (Ballas and Clarke 2001). A fiscal policy which stimulates demand in areas most adversely affected is likely to improve the level of employment and income in these areas. It is interesting to provide a brief overview of the policies that have been debated in the last UK parliamentary election drawing on the main three political party manifestos. Prior to the 2010 British general election political parties published a series of manifestos filled with a set of promises in the event that the party concerned are elected to serve in government. It is these manifestos from the Labour Party, the Liberal Democrats, the Conservative party and for Scotland, the Scottish National Party (see The Labour Party 2010; Liberal Democrats 2010; The Conservative Party 2010; The Scottish National Party 2010 respectively) that contain the likely alterations to existing policy or the new policies which could be introduced during a period of government by the respective political party. One of the key proposed policies was to increase the threshold of income tax to £10,000, meaning that the first £10,000 of earned income would not be taxed. In other words the tax rate on earnings up to £10,000 would be 0 per cent. However, there will still be deductions, for example national insurance and so forth. The idea behind this policy proposal was that it would provide an incentive to 'make work pay' over and above the level of state benefits an individual or family would be receiving. This would thereby remove incentives encouraging individuals to remain on benefits, as their absolute income was potentially lower or equal when in employment. Additionally, the effective marginal rates of tax for some individuals moving from benefits to employment are high. Governments in the United States, Canada, UK and New Zealand have resorted to measures such as in work benefits, for example tax credits to try to alleviate poverty without creating adverse incentives for participation (Brewer et al. 2009). Another report notes the problem of working a minimum wage job that, "compared to an income at 40 hours of work, a couple on Jobseeker's Allowance will only be £29.06 better off" (Kay 2010, p.7). This kind of scenario is known as the 'poverty trap' (Kay 2010) in the UK. At the other end of the income scale the planned rise in the tax rate on those earning over £150,000 will help to address income inequality to some extent, particularly if the goal is to redistribute this wealth towards the 'poorest' in Scotland (or the UK). Other policies, mainly from the Conservative party manifesto, suggest removing the child benefit payments for those earning over a certain limit as well as placing a cap on the level of Housing Benefit by number of bedrooms. As noted above there is an important geographical dimension to public finance and fiscal policies and there is a need to address the geographical as well as the socio-economic impacts of social policy change. It would be reasonable to expect that geographers would and should be at the forefront of addressing these issues and there has long been strong criticism of the lack of 'policy relevance' of geography as a discipline (Peck 1999; Martin 2001; Dorling and Shaw 2002). Nevertheless and despite such criticisms there have been very few examples of geographers engaging with such issues and it has recently been argued that "not only has the geography of social security and welfare benefits been a much-neglected issue within the subject … but the cultural turn within human geography has arguably redirected interest away from the important material basis of economic and social life, towards issues of representation and identity which are arguably far removed from the everyday experiences and problems of most people's lives" (Hamnett 2011, p.147). As noted in the introduction, there have been considerable efforts by a small but growing number of quantitative geographers and regional scientists to address the geographical implications of public policies at different spatial scales. The work reported in this paper aims to contribute further to these efforts with the development and use of spatial microsimulation techniques to model some of the policy scenarios outlined above. In particular, the focus of the analysis is on two proposed taxation policy changes as well as proposed amendments to child benefit and housing benefits. In particular, one of the policies that we explore is the proposal to increase the threshold of income tax to £10,000. The result is that the first £10,000 of earned income is not taxed (in other words the tax rate on earnings up to £10,000 would be 0%). The second policy under investigation is at the other end of the income scale, the highest earners. The rise in the tax rate on those earning over £150,000 to 50 per cent (from 40%) could arguably help to address income inequality to some extent. There is some debate over the efficacy of such a tax, with a UK Treasury report estimating the amount (HM Treasury 2009) that could potentially be collected. This could be the case if the goal is to redistribute this increased tax towards the 'poorest' in Scotland (or the UK). It is also worth noting that this tax is currently under review and it is very likely that it will be reduced to 45 per cent in 2012 (BBC News 2012). We also explore the geographical impacts of proposed welfare policy changes pertaining to housing and child benefits. In particular, we model and map the geographical impact of the current coalition government's proposals for a maximum level of housing benefit of differing levels depending on the size of the property. We also model the proposed changes to child benefit entitlement for households with earnings over a given income threshold. A simple two constraint example of microsimulation is outlined in Tables 1-4, showing how the process works. What has been done is that the original survey weights seen in Table 1 have been adjusted to new (re)weighted survey weights that now form the microsimulated microdata set (see Table 4). Using the formula the first new weight is calculated by multiplying the weight (Wi), which is 1, by the old owners (CENij) of whom there are 3 (i.e., Table 2 row 2, column 1), divided by the corresponding number (SHSij) which is 2 (i.e., Table 3 row 2, column 1). So in summary that is 1 × 3/2 = 1.5, which is the new weight shown in Table 4. This process is completed iteratively until a suitable level of convergence is reached. The above process was followed to adjust the weights of all records in the SHS to match a selection of small area descriptions (on the so called census Output Areas(OAs), which, on average, have a population of around 125 households), which is shown in Table 5.3 It should be noted that the development of the SIMALBA model was part-funded by the Scottish Government and one of the key aims of the model was to explore the geographical and socio-economic impact of public policies in Scotland and to also investigate linkages between socio-economic issues and health inequalities. This influenced the decisions on the choice of datasets that formed the basis for the model and hence the SHS was selected. Another key reason for selecting the SHS is that it has a relatively large sample size for Scotland compared to other surveys typically used in a spatial microsimulation models in Britain (e.g., see Ballas et al. 2007) such as the British Household Panel Survey (BHPS). As Boyle et al (2009, p. 386) point out: "studies that encompass the whole of Britain rarely have sample sizes that allow for Scotland-specific research". An alternative that was considered was the UK Household Sample of Anonymised Records (also used in spatial microsimulation applications in the UK, such as MoSeS; see Wu et al. 2008) which has a very large sample size for Scotland and overall is a very high quality data set. But this did not include information on income, which is a key policy relevant variable (and this was included in the SHS). It should be noted that in the UK the census does not provide any information on variables such as household income, wealth and taxation. With regards to the choice of small area constraint variables, we used a combination of demographic and socio-economic variables that are thought to be correlated with income (a key variable in our model) and overall quality of life, as it is also the case with other spatial microsimulation studies of this kind (e.g., see Ballas et al. 2007; Hynes et al. 2009; Kavroudakis et al. 2012). This process resulted in the creation of a small area microdata set that can be used for what-if policy analysis. This dataset has also been proved to be a reasonably robust estimate of the data (see Figure 1) to within 5 per cent to 10 per cent of the actual census data. Following the conventional approach to validating the outputs of spatial microsimulation models, the SIMALBA outputs were extensively compared against known census totals for both variables that were used as constraints (and which matched extremely well) as well as variables which were not used as small area constraints, using scatter plots such as that shown in Figure 1. In particular, what can be seen from Figure 1 is that an unconstrained variable, ethnicity, has been simulated using the SIMALBA algorithm and then compared with the 6,637 census output areas in Lothian HB as a direct comparison. The figure shows the variability of the estimates, plus or minus 5 per cent and 10 per cent lines. The fit of the data is reasonably robust given the model was not designed to predict ethnicity, most estimates within 10 per cent with a few output areas over predicting non-white ethnicity and therefore under predicting white ethnicity in these same areas, similar to models elsewhere (Smith et al. 2011). Constraint validation for Lothian Health Board In this section we show how the SIMALBA model was used to simulate the geographical impacts of some of the key policy changes discussed in Section 2. In particular, we simulate the impacts of a raise in the bottom income tax threshold, the 50 pence tax rate for every pound earned over £150,000 (which was an increase on the tax rate of 40 per cent and was introduced by a previous Labour government), as well as proposed changes in the housing and child benefit. We also created maps showing the geographical distribution of these impacts at the small area (census output area) level for the greater Edinburgh city region (see Figure 3 and 4), with a reference map of some key areas within Edinburgh city (see Figure 2). Reference map Percentage earning up to £10,400 Lothian region: quintiles Percentage earning up to £10,400 Edinburgh: quintiles As noted in Section 2, there has been some discussion about the advantages and disadvantages of raising the income tax threshold for individuals in the UK as well as more general taxation issues which have been outlined by several different bodies with varying viewpoints. In general, the consensus is that raising the tax threshold will provide an incentive whereby employment will be preferable to living on benefits as the first £10,000 of income earned will be received by the individual directly. It could be argued that this addresses situations whereby an individual is in a more financially secure position in choosing not to be in employment and continue to support him or herself using the welfare entitlements provided to them by the state. The current rules stipulate that the personal allowance is £6,475 for 2009–2011 and will rise to £7,475 by 2011-12 (see HMRC4 ). As discussed previously there is a compromise in the microsimulated dataset in that there is only categorical data for microsimulated income. Therefore, the policy analysis presents a case for assessing people currently in the income categories up to £7,800 compared with those who are in the income categories up to £10,400 to show the effect of the policy if the income tax threshold was raised to the £10,400 level from £7,800. It therefore must be noted that it is a compromise that the cut-off point has been chosen as £10,400 due to the categorical nature of the estimated income distribution from SIMBALBA. The remainder of this section explores the spatial distribution of the individuals live who will 'gain' most from the increase in the income tax threshold to a hypothetical position of £10,400. Figures 4 and 5 show the situation in the Lothian region and in Edinburgh City OAs within the Lothian region and those who will benefit most from the changes. The darkest colour, represents the highest proportions (the areas with proportions in the highest fifth of the distribution) of low income earners, that is up to £10,400 in each OA and the lightest colour represents the least low income earners as a proportion of total population in that OA. The distribution of areas has been split into quintiles, so Q5 represents the areas with the highest proportions of low earners, Q1 the areas with the lowest concentrations. The wider Lothian region exhibits a general trend of those areas with the highest proportion of low income earners being located in OAs around Edinburgh (in the middle of the region) or around urban areas. At the other end of the distribution, the areas with the lowest proportion of low earners are generally speaking, located in the more rural areas of the Lothian region, for example in the east, near North Berwick and to the south west of Lothian region also. It is also interesting to explore the distribution at smaller area level within Edinburgh City. It can be argued that within Edinburgh there is a slight clustering effect, particularly around Muirhouse and Leith (to the north) where there is a concentration of low income persons. There is also some clustering of (Q5) areas to the western edges of the figure. The power of a simple map is that we can see where it is likely that individuals who will benefit most will reside and target particular policies or resources to those areas if desired. The introduction of a tax rate of 50 per cent on personal income in the UK is discussed in detail. SIMALBA allows the spatial distribution of people who will be affected by this change to be mapped at OA geography (or any other spatial scale) for the Lothian region. First, the percentage of people who earn over £150,000 must be calculated in each OA. This can then be mapped as shown in Figures 5 and 6 respectively. What each map represents is the percentage of people in each OA that would be liable for a 50 per cent rate of income tax if it were to be introduced in Scotland. Figure 5 shows the Lothian OA model outputs and Figure 6 shows a version of the same model, but focused in on Edinburgh City. What can be deduced from Figure 5, is that the pattern would appear to be fairly random, such that those parts of Lothian where the microsimulated data has estimated the highest concentrations of those earning £150,000 and therefore will be liable for a new 50 pence tax rate on those earnings is not immediately obvious other than to highlight it is mainly the polar opposite of the figure showing those earning up to £10,400. There are smaller clusters to the south of Edinburgh and the eastern side of Lothian region also has some concentrations, but overall a clear pattern is difficult to deduce. This may be due to the small proportions in each area. Percentage earning over £150,000, Lothian region: quintiles Percentage earning over £150,000, Edinburgh: quintiles Looking within Edinburgh the main concentration of these individuals is around the southern edges of the city. There is also a clustering effect of the highest proportions of areas with those earning £150,000 or more around the financial and central western districts of Edinburgh and Holyrood Park. There are also notable gaps, with areas in the lowest quintile mainly to the north of Edinburgh. The spatial pattern then is not particularly easy to categorize for this particular microsimulated data. It is the areas in the darkest shade of the colour that will be affected by an increase in the tax rate on individuals earning over £150,000 most severely. We now move to another major application area for the SIMALBA model, that of welfare policy change. The focus of this subsection is on the potential changes to housing benefit. The current situation is also modelled where possible to show the 'before' and 'after' results. Both will be explored in more detail in the subsections that follow. The previous analysis on income tax is used as a template for the analysis of policy on welfare, with analyses involving welfare changes described here. The results are of particular relevance for policy debates surrounding issues of who will bear the burden of the cuts most acutely including which areas (and which people) the cuts will effect most directly. The aim of housing benefit is to provide those on low incomes or in receipt of welfare with a way in which to pay their full housing costs, for example rent is the most obvious cost. It is therefore a means tested benefit. In practice, this can be beneficial to both the tenant receiving this benefit, but also to those private landlords who will potentially gain financially from providing accommodation. The rules surrounding the eligibility for housing benefit are complex as when eligibility is established there are varying rates that can be paid as well as for different reasons (Child Poverty Action Group 2003). Briefly, there are several key criteria. Income must be low (which is defined as below a certain threshold) and savings and capital must also be below a certain limit (usually under £16,000)