Methods and data
How the numbers on this site were made
Every number on unfreeze.us comes from two things I built: the Unfrozen Research Index, a database of management and organization journal articles, and the Atlas of the Middle, an evidence map of the studies in it that are about middle managers and change. This page says how each was made, what the counts mean, and where they are weak.
The Unfrozen Research Index
Articles from 29 management and organization journals, 2007 to 2026, collected from open scholarly sources: OpenAlex, Semantic Scholar, Crossref and CORE, plus the Cohort 12 doctoral research file. Cleaning is logged.
Abstracts obtained under library or API terms are used for counting and are not republished.
The headline count
In the 21 core journals from 2007 to 2018, abstracts that mention executives outnumber abstracts that mention middle managers 786 to 66, about 12 to 1. This is a keyword count on abstracts from OpenAlex. Counting every other abstract source except the doctoral research file gives 796 to 66; every abstract-source definition tried for these journals and years gives a ratio near 12 to 1.
Keyword searches miss papers, so a recall audit coded a stratified sample of 520 abstracts against a written guide. Corrected for what keywords miss, the core-journal ratio is about 10 to 1, with a plausible range of roughly 7 to 16. The audit coding was done by AI coders and is provisional until a human second coder checks half of it.
A separate count in the Cohort 12 doctoral research file, built independently, found 853 executive to 68 middle-manager abstracts across 22 journals over the same years. Two databases, built separately, give nearly the same answer.
The Atlas of the Middle
- 196 candidate papers from 2007 to 2018 were found with a narrow middle-manager search plus a broader definition within change studies. 117 were judged to be about middle managers.
- Each study was coded for design, setting, the parts of the Unfrozen Loop it informs, whether it supports, complicates or challenges each part, and how strong its evidence is. 102 were coded from the abstract alone, 10 from the full text and 5 from the abstract plus a labeled summary.
- Coding was AI-assisted. A blind AI second coder recoded 30 papers: 97% agreement on relevance, and agreement by Loop part (Cohen's kappa) of 0.51, 0.85, 0.79, 0.76 and 1.00. An AI auditor then reviewed every link.
- In October 2026 every plain-language summary was checked twice against the study's abstract or full text by independent AI reviewers. Missing summaries were written, overstated ones were corrected, and citation years now follow the print issue. The same check flagged 18 Loop links whose fit is weak; they go to a human second coder.
What isn't settled
- No human second coder has verified the codes yet. That review is under way, and if it moves the numbers, the change will be published here.
- Agreement on the first Loop part (0.51) is only moderate.
- Most studies were coded from abstracts, which can leave out details that matter.
- Papers from 2019 to 2026 are held back and not coded, so that the next round can be preregistered before anyone codes them.
- This is an evidence map, not a systematic review or meta-analysis.
How to cite
Lopez, G. (2026). The Atlas of the Middle: An evidence map of 117 studies on middle managers and organizational change (Version 1, October 2026). Unfrozen. https://unfreeze.us/atlas/
Corrections
Found a mistake? Email info@unfreeze.us. Corrections are made and noted.