Appendix A — Data and methodology

This appendix explains how chapter outputs are produced from the raw survey answers. It is meant to be read together with the rendered book.

A.1 Respondent inclusion

Analyses in this book use only respondents who submitted the survey. A row is included when its submitdate_0 field is non-empty. Partial responses that were saved without a submission timestamp are excluded, because they typically contain only early answers and would distort question-level percentages.

Within that submitted sample, the configured country filter (FILTER) is applied via socio1_0. Question-level N can still differ when submitted respondents skipped a question or it did not apply to them.

A.2 What you are looking at

Every chapter analyses one survey question. Question types are handled differently:

Single-select options are summarised as counts and percentages.
Pre-defined, multi-select options (the respondent ticks one or more boxes) are shown as a selection heatmap.
Likert-scale items are shown as ordered response distributions, sometimes split by country or age group.
Time-allocation items compare actual and desired percentages by activity.
Free-text answers (the respondent types their own text, typically under “other, please specify”) are cleaned, grouped into categories, and shown as a summary table.

A question can combine types: for example, a heatmap for ticked options and a table for typed-in “other” answers.

Note

All figures and tables describe the country or set of countries selected in the configuration (FILTER). Unless stated otherwise in a chapter, the numbers are for that scope only, not the whole survey.

A.3 The selection heatmap

A.3.1 What it shows

The heatmap visualises who selected what for a multi-select question. It has two panels side by side:

Left panel (the heatmap): a grid where every column is one respondent and every row is one answer option (category). A dark blue tile means that respondent selected that option; a white tile means they did not.
Right panel (the bar chart): for each option, the total number of respondents who selected it, with the count printed at the end of the bar.

A.3.2 How it is constructed

Only the pre-defined selections are used (the typed-in “other” answers are excluded from the heatmap).
Each respondent-by-option combination is marked as selected (1) or not selected (0), so every respondent is represented even if they selected nothing for a given option.
Respondents (columns) are ordered by how many options they selected, and options (rows) are ordered by how often they were selected. This sorting is what gives the plot its characteristic “staircase” shape.

A.3.3 How to read it

A row that is mostly dark is a popular option; this is confirmed by a long bar on the right.
A column that is mostly dark is a respondent who selected many options; columns are sorted so these accumulate toward one side.
The shape of the filled area tells you whether selections are concentrated in a few options (a few dark rows) or spread out.

A.4 Free-text answers: from raw text to a tidy table

Typed-in answers are inconsistent: the same idea appears with different spellings, capitalisation, abbreviations, and several answers are often put in one box. The summary table you see in a chapter is the result of normalising and grouping these raw responses.

A.4.1 Step 1: Normalisation (tokenisation)

Each raw answer is:

Split into separate entries wherever it contains a comma (so “Python, R” becomes two entries).
Lowercased and stripped of extra whitespace.
De-duplicated (identical raw answers are collapsed) and emptied entries are dropped.

The result is a clean list of individual response entries ready to be categorised.

A.4.2 Step 2: Recoding into categories

Each chapter defines a recode map: an ordered list of rules, where each rule has a pattern (a regular expression) and the category it should map to. Every response entry is compared against the rules and assigned to the matching category. Rules are applied in order, so earlier rules take precedence.

Some rules instead mark responses to be excluded (for example blank, nonsensical, or “not applicable” answers). Excluded responses do not appear in the summary table.

A.4.3 Step 3: The summary table

The cleaned, categorised entries are counted per category and presented as a “pretty” table with three columns:

Response Category: the category an answer was mapped to.
Frequency (n): how many response entries fall into that category.
Percentage (%): that category’s share of all counted entries.

The table title carries a suffix that tells you which part of the question it summarises:

[Others]: the typed-in “other, please specify” free text.
[Main]: the main pre-defined answers (used when those were also free text rather than tick boxes).

Tip

Because identical raw answers are de-duplicated and multi-answer boxes are split before counting, n counts normalised response entries, not unique respondents. Read the frequencies as the relative weight of each theme, not as a headcount of people.

A.5 Where to find the recoding tables

To keep the grouping of free text into categories transparent and reproducible, every chapter records exactly how each raw answer was handled, and these are collected in the Recoding appendix (Appendix B).

For each question, the allocation (codebook) table lists, one row per normalised raw answer:

Raw response: the normalised text as typed by a respondent.
Assigned category: the category it was mapped to (empty if excluded).
Allocation type: one of:
- Unchanged: the raw answer was already a clean category label.
- Recoded: the raw answer was mapped to a different category by a rule.
- Excluded: the answer was removed from the analysis.
Frequency (n): how often that raw answer occurred.

So if you see a category in a chapter’s summary table and want to know which raw answers it contains, look up that question in Appendix B.