Goal

Converting the ONS multiple deprivation index onto Dutch stats

The variables included are a subset of what might be available with a bit more work

The variables included need to be reshuffled a bit probably as well

knitr::opts_chunk$set(warning = F, echo=F, message = F)

## Reading layer `wijk_2018_v2' from data source `C:\Users\richa\OneDrive\DATA\wijk_2018_v2.shp' using driver `ESRI Shapefile'
## Simple feature collection with 3174 features and 137 fields
## geometry type:  MULTIPOLYGON
## dimension:      XY
## bbox:           xmin: 10425.16 ymin: 306846.2 xmax: 278026.1 ymax: 621876.3
## epsg (SRID):    NA
## proj4string:    +proj=sterea +lat_0=52.15616055555555 +lon_0=5.38763888888889 +k=0.9999079 +x_0=155000 +y_0=463000 +ellps=bessel +units=m +no_defs

Comparison

ONS MDI

7 domains, 39 indicators, broadly speaking

Income Welfare assistance, measures taken by family / children. Includes asylum-seekers on assistance.
Labour market Focusing on working age population: Welfare aimed at work, disability benefits, and carers allowance
Education and skills Two subdomains: Children and education levels (flow), and adults skills and language proficiency (stock)
Health Years of life lost, illness disable ratio, morbidity, and mental health
Crime Violent crime, burglary, theft, destruction of property
Barriers to living Distance to post office, primary school, shop/supermarket, GP, small dwelling, homelesness
Residential quality Central heating, maintenance, air pollution, traffic accidents

Maximum likelihood factor analysis, with shrinkage

Setup for the Netherlands

Three spatial scales: Neighbourhoods, larger scale neighbourhoods, municipality.

Most data is available for larger areas. Smaller areas suffer from data-attrition because of privacy concerns mostly. Larger scale neighbourhoods is the smallest spatial unit that has health data avilable (“wijken”)

Overview of results

The following results are all: Higher = less deprivation

Income

Contains the StatNL variables:

Percentage of people in lower two quintiles of income nationally.

Percentage of people recieving an income around or under the social minimum (definition from StatNL)

For both measures income is

Employment

Welfare, disability

Health

Mortality

Crime

Burglary, destruction of property

Living

Average price of housing, percentage of properties > 20 yrs old, occupation rates

Residential quality

Distance to supermarkt

Unweighted overall average

Running a PCA using ranked data

For the ONS MDI they use a factor analysis with shrinkage terms. In our case, we have a number of regions with missing values.

One way around the problem is the use of a NIPALS PCA, which allows for estimation of PCA loadings if some values are missing.

The underlying assumptions are quite different from the ones used for the ONS MDI.

First step: Determine the number of dimensions to retain. The following barplot contains the eigenvalues for each number of dimensions retained.

Looks like two might be a good guess.

Using scaled data

Looks like two might be a good guess here too.

GWPCA

Requires complete cases, dropping 759 cases.

The main variable in the first component

Glyphplots

For Zwolle there is a disontinuity in the lead item for the first component. Other than that, some local differences.

Getting Started on an MDI for the Netherlands