assignment04-week4

Assignment 04: Spatial autocorrelation and spatial regression

Use a data from PAAVO (postcode area data) and download data from the server of Statistics Finland (help: see lecture notes).

In this exercise set, you will use PAAVO postcode area data provided by Statistics Finland. The data should be downloaded directly from the Statistics Finland server (for guidance, see the lecture notes).

More information about the PAAVO data can be found here: https://stat.fi/en/services/statistical-data-services/statistical-databases/paavo

1. Global spatial autocorrelation

Investigate the global spatial autocorrelation of household median income in 2025 (variable tr_mtu) across Finnish postcode areas.

Construct three spatial weight matrices based on k nearest neighbours:
6 nearest neighbours
10 nearest neighbours
15 nearest neighbours
For each weight matrix, calculate Moran’s I.

Tasks:

Compare the values of Moran’s I across the different weight matrices.
Explain how and why the level of spatial autocorrelation changes when the number of neighbours increases.

2. Local spatial autocorrelation

Calculate the local spatial autocorrelation of household median income in 2025 using Local Moran’s I.

Use the local_moran() function.
Visualize the results using a map (see lecture notes for instructions on mapping Local Moran’s I results).

Tasks: - Interpret the resulting clusters (e.g. high–high, low–low). - Explain what these clusters reveal about the spatial distribution of household income.

3. Linear regression and spatial dependence

Estimate a linear regression model (OLS) where household median income in 2025 is explained by:

the proportion of highly educated people
the unemployment rate

Construct the explanatory variables as relative proportions as demonstrated in the lecture notes:

p25_data$tyottom <- (p25_data$pt_tyott / p25_data$pt_tyoll) * 100
p25_data$korkk   <- (p25_data$ko_yl_kork / p25_data$ko_koul) * 100

(Unemployment rate = number of unemployed / number of employed) (Share of highly educated = number with academic degrees / number of educated people)

Tasks: - Interpret the regression coefficients. - Evaluate whether the results are consistent with economic intuition. - Investigate the spatial autocorrelation of the model residuals (e.g. using Moran’s I). - Explain what the residual spatial autocorrelation indicates about the adequacy of the OLS model.

4. Spatial regression model (continuation of Exercise 3)

Extend Exercise 3 by estimating an appropriate spatial regression model (e.g. spatial lag or spatial error model).