Rumslig autokorrelation i kommunal arbetslöshet
(2014) STAK01 20102Department of Statistics
- Abstract (Swedish)
- Studien ger en analys av kommunal arbetslöshet i Sverige. Syftet är att undersöka om arbetslöshet kan modelleras med en linjär regressionsmodell, eller om den kräver en utvidgad modell med en rumslig komponent. I detta syfte skattas en Durbinmodell – en modell som utvecklar den multipla linjära regressionen med rumsligt vägda variabler.
Landet delas i fem regioner (Götaland, Norrland, Svealand, Stockholm och Malmö-Göteborg) och för varje region skattas en multipel linjär regressionsmodell och två varianter av Durbinmodellen. Durbinmodellerna är skilda i valet av vikt-matris, det vill säga definitionen av de rumsliga vikterna. Varje modelltyp skattas med två olika variabeluppsättningar: 1) en fullständig uppsättning med tolv förklarande... (More) - Studien ger en analys av kommunal arbetslöshet i Sverige. Syftet är att undersöka om arbetslöshet kan modelleras med en linjär regressionsmodell, eller om den kräver en utvidgad modell med en rumslig komponent. I detta syfte skattas en Durbinmodell – en modell som utvecklar den multipla linjära regressionen med rumsligt vägda variabler.
Landet delas i fem regioner (Götaland, Norrland, Svealand, Stockholm och Malmö-Göteborg) och för varje region skattas en multipel linjär regressionsmodell och två varianter av Durbinmodellen. Durbinmodellerna är skilda i valet av vikt-matris, det vill säga definitionen av de rumsliga vikterna. Varje modelltyp skattas med två olika variabeluppsättningar: 1) en fullständig uppsättning med tolv förklarande variabler valda utifrån tidigare studier, och 2) en reducerad uppsättning framtagen från den fullständiga genom minimering av det bayesianska informationskriteriet i den linjära modellen. Modellerna utvärderas genom att jämföra förklaringsgrader och genomföra likelihood-kvottest.
Studiens resultat tyder på att det finns en rumslig komponent i den kommunala arbetslösheten som inte till fullo kan förklaras med de förklarande variabler som används. För de tre landsdelarna (Götaland, Norrland, Svealand) visar likelihood-kvottesten på signifikant fördel för Durbinmodellerna med full variabeluppsättning. För storstadsregionerna är resultaten mindre klara, men förklaringsgraden tyder på att det kan finnas fördelar med rumsliga modeller. Vidare pekar resultaten på bety-dande skillnader mellan regioner, och på att valet av viktmatris inte har stor påverkan på förklaringsgraden. (Less) - Abstract
- The study presents an analysis of municipal unemployment in Sweden. Its aim is to examine whether unemployment can be modeled using linear regression or re-quires an additional spatial component. To do this, a Durbin model – a model that extends multiple linear regression using spatially weighted variables – is employed.
The country is divided into five regions (Götaland, Norrland, Svealand, Stockholm and Malmö-Gothenburg) and for each region a multiple linear regression model and two variations of the Durbin model are estimated. The two Durbin models differ in the choice of the weight matrix, i.e. the definition of the spatial weights. Each model is estimated using two different variable sets: 1) a full set with twelve explanatory... (More) - The study presents an analysis of municipal unemployment in Sweden. Its aim is to examine whether unemployment can be modeled using linear regression or re-quires an additional spatial component. To do this, a Durbin model – a model that extends multiple linear regression using spatially weighted variables – is employed.
The country is divided into five regions (Götaland, Norrland, Svealand, Stockholm and Malmö-Gothenburg) and for each region a multiple linear regression model and two variations of the Durbin model are estimated. The two Durbin models differ in the choice of the weight matrix, i.e. the definition of the spatial weights. Each model is estimated using two different variable sets: 1) a full set with twelve explanatory variables which are chosen based on previous studies and 2) a reduced set obtained by eliminating variables from the full set by means of minimizing the Bayesian information criterion in the linear model. The models are evaluated by comparing the coefficient of determination, and by implementing likelihood ratio tests.
The results of the study suggest that there is a spatial component in the local un-employment rate which cannot be fully explained by the explanatory variables used. For the three regions Götaland, Norrland and Svealand the likelihood ratio tests show significant results in favor of the Durbin models with the full variable set. In the metropolitan areas, the results are less clear, but the coefficient of determination indicates possible advantages for spatial models. Furthermore, the study shows large differences between the five regions, and that the choice of weight matrix has little impact on the coefficient of determination. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/4467304
- author
- Flöhr, Adam LU
- supervisor
- organization
- course
- STAK01 20102
- year
- 2014
- type
- M2 - Bachelor Degree
- subject
- keywords
- Spatial econometrics, regional unemployment, Durbin model
- language
- Swedish
- id
- 4467304
- date added to LUP
- 2014-06-17 11:34:08
- date last changed
- 2014-06-17 11:34:08
@misc{4467304,
abstract = {{The study presents an analysis of municipal unemployment in Sweden. Its aim is to examine whether unemployment can be modeled using linear regression or re-quires an additional spatial component. To do this, a Durbin model – a model that extends multiple linear regression using spatially weighted variables – is employed.
The country is divided into five regions (Götaland, Norrland, Svealand, Stockholm and Malmö-Gothenburg) and for each region a multiple linear regression model and two variations of the Durbin model are estimated. The two Durbin models differ in the choice of the weight matrix, i.e. the definition of the spatial weights. Each model is estimated using two different variable sets: 1) a full set with twelve explanatory variables which are chosen based on previous studies and 2) a reduced set obtained by eliminating variables from the full set by means of minimizing the Bayesian information criterion in the linear model. The models are evaluated by comparing the coefficient of determination, and by implementing likelihood ratio tests.
The results of the study suggest that there is a spatial component in the local un-employment rate which cannot be fully explained by the explanatory variables used. For the three regions Götaland, Norrland and Svealand the likelihood ratio tests show significant results in favor of the Durbin models with the full variable set. In the metropolitan areas, the results are less clear, but the coefficient of determination indicates possible advantages for spatial models. Furthermore, the study shows large differences between the five regions, and that the choice of weight matrix has little impact on the coefficient of determination.}},
author = {{Flöhr, Adam}},
language = {{swe}},
note = {{Student Paper}},
title = {{Rumslig autokorrelation i kommunal arbetslöshet}},
year = {{2014}},
}