We ran a separate model for each country to preserve national differences. But that made comparison harder (see 2.2). To compare the results, we grouped 138 country-specific topics into 20 shared categories. Here you can inspect those groupings, compare the three collections and see how apparently similar issues are framed differently in each country.
Cross-country comparison is constructed, not simply found by the models. TRACE fitted a separate topic model for each country, producing 75 topics for England, 30 for Scotland and 33 for the Republic of Ireland. One researcher then mapped those 138 topics by hand into 20 shared analytical categories.
This mapping makes comparison possible, but it does not establish that the underlying policies, institutions or experiences are equivalent. Grouping England’s SEND topics, Scotland’s additional-support-needs topics and the Republic of Ireland’s special-education topics together is an interpretive decision. Another researcher, with different domain or institutional knowledge, could reasonably draw the boundaries differently.
Model-generated alternative mappings were used as sensitivity tests. They provide evidence about how dependent particular results are on the production mapping, but they are not independent human validation and may reproduce the same semantic assumptions.
Different national terms can point towards a shared theme. Here, England’s SEND, Scotland’s ASN and Ireland’s special needs topics are grouped under Special and Inclusive Education.
A pooled model was also tested as an alternative specification. TRACE retained separate country models for the production analysis and constructed the cross-country comparison afterwards. The pooled-model result is reported as a sensitivity test, not as proof that the production specification is uniquely correct.
Real excerpts from the source documents, shown to make the reading inspectable. TRACE does not alter the text.
The decision about how topics should be grouped together requires someone who understands the evidence. No amount of technical checking substitutes for that.
How the three records compare once every country topic has been placed in a shared category, and which topics ended up where.
The shares come from our mapping. They tell you what these records contain under that mapping, not that a category means the same thing in all three countries.
Whether a shared category carries the same meaning in each country. We tried to test that, and the measure could not tell a real result from none, so it was withdrawn.
Every comparison on this site rests on a mapping we made by hand. Use it with that in mind, and challenge it if you know these systems better than we do.