Quantitative Spatial Analysis of Micro and Macro Data: Methodological Challenges and Solutions 3 |
|
| Chair | Professor Henning Best (TU Kaiserslautern ) |
| Coordinator 1 | Professor Corinna Kleinert (Leibniz Institute for Educational Trajectories) |
| Coordinator 2 | Mr Tobias Ruettenauer (TU Kaiserslautern) |
| Coordinator 3 | Dr Michaela Sixt (Leibniz Institute for Educational Trajectories ) |
Modern time series- as well as spatio-temporal-methodologies are compared in their forecasting-performance for employment data at the regional level.
It is investigated which spatial dependencies in Germany’s labour markets exist, and to what extend the knowledge about spillover-effects improves the forecasting of employment numbers at the regional and the aggregated national levels.
Surely, data and forecasts of regional employment are important for regional politics. But also to forecast national data, regional forecasts based on pure time series-methods or on spatio-temporal approaches can be cross-sectionally aggregated. Admittedly, this rather makes sense, if these delicate procedures provide a superior forecasting-performance compared to plain time series-forecasts at the national level. So the forecasting-power of each method is proven with several measures.
Additionally, time series-models which are calibrated for national data are applied to forecast regional employment data - again the forecasting-performance is evaluated.
The procedures are performed in a rolling-window framework to resolve from random effects.
There is growing recognition both that where people live can have a profound effect on their attitudes and behaviour and that survey data provides a valuable tool with which to explore spatial variation at the individual level. It is also important to consider that people’s attitudes and behaviour towards surveys themselves may vary spatially with, for example, people in some areas more likely to respond than others. This in turn can lead the quality of survey data to vary geographically and potentially bias any conclusions drawn from those data.
This paper uses a dataset combining the UK sample for the European Social Survey 2012/13 with contextual data from the 2011 Census, to explore spatial variation in survey nonresponse. It tests the assumption that, as well as survey response rates varying geographically (for example, being lower in densely populated urban areas), the underlying drivers of response propensity - and therefore the nature and extent of any nonresponse bias - may also vary depending on where someone lives. After allowing for potential spatial variation in survey nonresponse through regional dummies and interactions, the paper employs geographically weighted regression to more fully explore the extent of this spatial variation.
Geographically weighted regression (GWR) is a relatively new technique, especially in the field of survey methodology. It provides a way to explore the relationship between a dependent variable and one or more explanatory variables by fitting a separate regression model for each spatially located data point. The value of GWR coefficients can vary continuously over geographical space and are not constrained to follow particular sets of (possibly nested) areal units, as is the case, for example, with standard regression models reliant on including dummy variables for geographic units or multi-level models. Thus, GWR provides a way of exploring variation in regression coefficients in a way that is not limited by a hierarchical or fixed geographical structure.
Results demonstrate that there is a clear spatial dimension to survey response behaviour in the UK and that global models of response propensity are likely to be insufficient. Accommodating geography as a discrete entity defined by regional (NUTS1) boundaries is an improvement but still fails to capture the full picture. Geographically weighted regression show that associations between response propensity and a range of common predictors, including population density, unemployment and crime rates, vary significantly both within and across administrative boundaries.
The novel application of GWR to study nonresponse to the European Social Survey showcases what this technique can offer as a means of analysing spatial variation in survey data for either substantive or methodological purposes. By demonstrating the importance of taking into account spatial variation in response behaviour, this paper also points to a way in which survey methodologists and practitioners could improve their ability to understand and predict patterns of survey nonresponse. This in turn would improve the efficiency and effectiveness of survey data collection and provide those interested in analysing these data for the purposes of substantive research with a more robust source of data.
Regional mobility is seen as a vital mechanism to align spatially unequal distributed supply of and demand for labour. In Germany – as in other European countries – there are considerable regional disparities in regional employment levels and low shares of internal mobility. Previous research suggests that entrapment effects of unfavourable regional contexts could be to blame, because they hinder an individual’s mobility thus perpetuating disparities (e.g., Windzio 2004).
Research on the influence of contextual unemployment on mobility in individual job search behaviour finds positive, negative, or no effects at all. We argue, that these heterogeneous findings are attributable to three deficits: Firstly, studies suffer from selectivity present in the labour market. Only certain workers receive job offers attractive enough to make mobility an attractive option. Inference based on realized mobility only will tell us little about the underlying causal links. Secondly, the regional level that constitutes relevant context is often theoretically undefined, and operationalisations dictated by data availability. This leads often to high levels of aggregation that are vulnerable to confounding influences. Particularly normative effects at the micro- and meso-level could counteract supply and demand factors at the aggregate level. Thirdly, studies use regional units whose borders were chosen arbitrarily and take only direct regional effects into account. These studies suffer from the modifiable area unit problem and fail to incorporate indirect spatial effects.
We analyse the effects of regional context factors on the willingness to relocate for given interregional job offers, by combining an factorial survey module (FSM) with data from the German “Labour market and social security” (PASS) panel study. By employing an experimental design, we observe not only selective realised behaviour but can draw our inference from the whole spectrum of the decision making process. The detailed panel data enable us to consider moderating household level and individual effects on the regional influences. The individual level data is enriched with contextual data at different levels, ranging from house-block to federal state. We combine multilevel modelling with spatial regression to take account of both the hierarchical data structure and context influences that go beyond just the local region. In particular, we consider SLX- models (Elhorst 2014) with different spatial lags of the regional context variable as well as cross level interactions between the contextual, individual, and experimental level.
Our contribution provides an innovative in-depth analysis and helps to answer the following research questions: What role does contextual unemployment at different levels of aggregation play in influencing individuals’ decisions about out-migration? How are these context effects moderated by social environment factors or the household level? Are unemployed individuals, which could profit from job-related mobility in particular, especially prone to regional entrapment effects? Can spatial econometrics help in modelling the contextual structure?
Our research can foster the understanding of the dynamics of contextual effects in decisions about mobility that ultimately can help to improve matching on the labour market thus increasing welfare.
The spatial concentration of socially disadvantaged groups can lead to an additional disadvantage and reinforcement of their socially underprivileged status. We use registry data of unemployment and welfare benefit recipients to analyze these effects and present findings from two studies with different methodological approaches to describe the spatial variations of poverty.
We will present findings on the relation between social segregation and housing prices from the North Rhine-Westfalian poverty and wealth report 2016. Pseudonymized, georeferenced individual data from the registry of the German Federal Employment Agency (Bundesagentur für Arbeit) was aggregated on a small area level (about 15.000 regions) and analyzed together with commercial demographic data from microm and information on housing prices from Immoscout24, an online market place for real estate. We used these data in order to analyze social segregation and housing prices on the macro level in a longitudinal perspective (e.g. with geo-additive regression models).
It is often assumed that the concentration of socially disadvantaged students in a school has an effect on the learning environment and strengthens the direct effect of poverty on individual educational achievements. We use kernel-density-estimates on a 100x100 grid to describe the spatial variations of the share of underage welfare benefit recipients in the neighborhood of schools. It can be shown that the combination of local kernel-densities and share of students with a migration background is a suitable proxy for the social compositions of schools. Further, analyses show that this proxy describes relevant background features for the learning success of students.