4 Developing an Index

This chapter runs through the process of developing a whole index for a single nation. Its purpose is to demonstrate the top-level workflow and ideas, before later chapters dive into the technical, nitty-girtty details of the build process.

For demonstration purposes, let’s run through the workflow to build a Health Inequalities Index for Scotland.

4.1 End goal

The end goal is to compute a resilience score, which is a combination of vulnerability and capacity:

The process to achieve the end goal is as follows:

1. Add indicators
2. Combine indicators to form domains
3. Combine domains and compute vulnerability and capacity scores
4. Combine vulnerability and capacity scores to make an overall resilience score

4.2 Add indicators

In the Resilience Index GitHub repo, navigate to the correct domain/strategic-cause/nation subfolder. To add vulnerability indicators to the Scottish Heath Inequalities index, navigate to R/vulnerability/health-inequalities/scotland:

If the model has subdomains, like in the case above, navigate into the subsequent folder:

To add an indicator, add a new .R file. Each file should contain a stand-alone modular script, that should compute the statistic for a single indicator:

The output of this script is then saved in a .rds format in the data/ folder in the root of the repo, in a structure which mirrors that of the file path in the R/ folder:

4.3 Combine indicators

Once indicators have been added, they are joined in a build-index.R script, located in the root of the domain/strategic-cause/nation subfolder:

The process to combine domains varies by the strategic-cause and model of interest. The OECD handbook on constructing composite indicators should be used as a reference text throughout. To combine vulnerability indicators for Scottish Health Inequalities, the following steps needs to be taken:

1. Load indicators into the R session:

# Load indicators
healthy_lives_indicators <-
  load_indicators(
    path = "data/vulnerability/health-inequalities/scotland/healthy-lives",
    key = "lad_code"
  )

2. Scale and align the indicators so that higher values equate to higher vulnerability scores: multiply misaligned indicators by -1:

healthy_lives_scaled <-
  healthy_lives_indicators %>%
  mutate(
    gcse_qualifications_percent = gcse_qualifications_percent * -1,
    healthy_eating_percent = healthy_eating_percent * -1,
    job_related_training_percent = job_related_training_percent * -1,
    activity_levels_met_percent = activity_levels_met_percent * -1,
    vaccine_rate_mean = vaccine_rate_mean * -1,
    young_people_training_percent = young_people_training_percent * -1,
    cancer_screening_percent = cancer_screening_percent * -1
  )

2. Apply functional transformations (e.g., log) to address non-normal distributions if necessary

3. Normalise indicators to a mean of 0 and SD +-1:

healthy_lives_weighted <-
  healthy_lives_scaled %>%
  normalise_indicators()

4. Weight indicators using PCA or MLFA if appropriate

5. Calculate domain scores by ranking and quantising the (weighted) normalised scores:

  healthy_lives_weighted %>%
  calculate_domain_scores(
    domain_name = "healthy_lives",
    num_quantiles = 10
  )

4.4 Combine domains

Combine domains with equal weighting and rank and quantise the output to produce a composite score:

health_inequalities_scores <-
  healthy_lives_scores %>%
  left_join(
    healthy_people_scores,
    by = "lad_code"
  ) %>%
  left_join(
    healthy_places_scores,
    by = "lad_code"
  ) %>%
  select(
    lad_code,
    ends_with("domain_score")
  ) %>%
  calculate_composite_score(
    index_name = "health_inequalities",
    num_quantiles = 10
  )

4.5 Combine vulnerability and capacity

Currently, vulnerability and capacity composite scores are not computed on to produce a resilience score. Rather, the two scores are plotted using a bivariate choropleth map, which examines the intersection of the two variables. For an example of how these plots are run, see here.