Method
Front matter: how this dictionary is made and how to read it
wtfe records how software engineering tends to be done in fourteen countries and regions, the way a usage dictionary records how words are used. This page explains what counts as evidence, how strong each kind is, which frameworks the scale pages rely on, and where all of it runs out.
What this site is, and is not
The rule behind every page is usage, not character. A claim here says that a tendency has been reported, by whom, and how solid the report is. It does not say what any Japanese, German or Brazilian engineer is like. Most of the variation in cultural values lies inside countries rather than between them (Taras, Steel & Kirkman 2016), so a country page describes a lean within a wide spread, never a person.
The site is built like a usage dictionary, because a dictionary already solves this problem:
A dictionary does not claim that every speaker uses a word the same way, and this site does not claim that every engineer in a country works the same way.
How to read a claim
Each claim is one short statement followed by a label line: the usage label, the number of sources, the year of the newest source, and a stable id. The sources are listed underneath, each a link. The label is decided by the strongest source behind the claim, and the rule down the left edge repeats it in a form that does not depend on colour.
| Label | Sources that earn it |
|---|---|
| Well attestedLeft rule: thick, solid | Official statistics, primary documents, large surveys read on the publisher's own page, peer-reviewed quantitative studies, and an organization's own account of its own practice. |
| AttestedLeft rule: thin, solid | Frameworks such as the Culture Map, a single reputable news report, Wikipedia, industry associations, qualitative studies, and the EF English Proficiency Index. |
| ThinLeft rule: dashed | Blogs, forums, vendor and recruiter material, summary sites, and single anecdotes. Kept only when nothing better exists, and labelled so. |
Dated. When even the newest source behind a claim was published before 2016, the label line adds Dated. The claim may still hold, but workplaces change and you should look for something newer before relying on it.
Regional variants. Every country page starts with a section on variation within the country and counter-evidence (for example Japan). Read it before the themes: it is where the page argues against its own summary.
The Culture Map (2014) places Germany on the direct side of Evaluating. Meyer describes this as a tendency to deliver negative feedback without softening it.
Attested1 source2014Datedde-feedback-1
Read it from left to right: the label says one framework-level source supports it, the year says the source is from 2014, and because 2014 is before 2016 it is marked dated. The id de-feedback-1 is the same on the Japanese page. See it in context on the Germany page
The frameworks
The scale pages place countries on two published frameworks. Both are country-level averages and both have serious critics; the critiques are part of this section, not a footnote to it.
Erin Meyer's Culture Map
Erin Meyer's The Culture Map (2014) places countries on eight scales, each running between two poles (HBR 2014, Meyer's eight-scale tool):
- CommunicatingLow-context ↔ High-context
- Whether good communication is explicit and literal, or layered and read between the lines.
- EvaluatingDirect negative feedback ↔ Indirect negative feedback
- How bluntly criticism is delivered.
- PersuadingPrinciples-first ↔ Applications-first
- Whether an argument starts from theory or from examples and conclusions.
- LeadingEgalitarian ↔ Hierarchical
- How much distance is expected between a boss and the team.
- DecidingConsensual ↔ Top-down
- Whether decisions are reached by group agreement or made by the person in charge.
- TrustingTask-based ↔ Relationship-based
- Whether trust comes from reliable work or from personal relationships.
- DisagreeingConfrontational ↔ Avoids confrontation
- Whether open disagreement is seen as productive or as damaging to the relationship.
- SchedulingLinear time ↔ Flexible time
- Whether plans are fixed and sequential or adapted as things come up.
How the positions were read. The book prints no numbers. We took the positions from the charts in Meyer's own INSEAD slide deck (PDF), converted it to text, and read each country's horizontal position relative to the others. Because this is reading by eye, every position is rounded to a deliberately coarse grid of seven steps:
- −3far left pole
- −2leans left
- −1leans left
- 0middle
- +1leans right
- +2leans right
- +3far right pole
The seven steps used for Culture Map positions, read from the left pole to the right pole of each scale.
A step says "further toward this pole than that country", not a measured distance. Where only a book review or summary places a country (for example Korea on Trusting, from one review), the position is marked as secondary. Meyer treats East Asian countries as holistic rather than placing them on Persuading (Portman's review), so those cells are empty. Wherever no source places a country, the position is left empty rather than guessed.
What is unknown. Meyer's maps come from her observation of, and interviews with, participants in her management courses; no sample size or error is published, and the HBR article says only that the scales draw on her research and that of other experts (HBR 2014). Treat the Culture Map as a well-informed framework, which is why claims that rest on it are labelled Attested at most.
Hofstede's six dimensions
Geert Hofstede's model scores countries on six dimensions, roughly from 0 to 100 (Hofstede 2011, DOI). Long-term orientation was added in 1991 and indulgence in the 2000s, the latter by Michael Minkov from World Values Survey data (global-iq, Hofstede 2011):
- Power distanceLow power distance ↔ High power distance
- How far less powerful members accept that power is distributed unequally.
- IndividualismCollectivist ↔ Individualist
- Whether people define themselves by personal goals or by the groups they belong to.
- Motivation towards achievement and successConsensus-oriented ↔ Decisive
- Formerly masculinity/femininity: whether competition and achievement or care and quality of life drive people.
- Uncertainty avoidanceTolerant of ambiguity ↔ Avoids ambiguity
- How threatened people feel by ambiguous or unknown situations. Not the same as risk avoidance.
- Long-term orientationShort-term ↔ Long-term
- Whether a society favours tradition and quick results or pragmatism and future rewards.
- IndulgenceRestrained ↔ Indulgent
- How freely people may act on their desires and enjoy life.
Which values. The scores shown are The Culture Factor's current values (country comparison tool, read in October 2026). They differ from the tables in older books and articles, mostly in individualism and long-term orientation. The change appears to date from around 2023; we have not found a public announcement of the revision, so mixing old and new tables is the thing to avoid.
| Country | Dimension | Older table | Current | Older value cited by |
|---|---|---|---|---|
| South Korea | Individualism | 18 | 58 | Cyborlink |
| Germany | Individualism | 67 | 79 | Simply Psychology |
| Germany | Long-term orientation | 83 | 57 | Simply Psychology |
Korea is the clearest case: the familiar picture of Korea as strongly collectivist rests on the old individualism score of 18, while the current value is close to the middle.
The critiques
- Ecological fallacy. Correlations between country averages do not hold for individuals, so using a country score to predict a person is an error (Brewer & Venaik 2014). Hofstede himself warns that the dimensions depend on the level of aggregation and that country scores must not be read as individual values (Hofstede 2011).
- McSweeney 2002. Brendan McSweeney rejected the assumptions that national culture is uniform and that it causally determines behaviour, and pointed out that some country samples were small and drawn mainly from IBM's sales and service staff (Human Relations 55(1), Royal Holloway). Williamson replied in the same journal (Williamson 2002).
- The IBM sample, 1967–73. The original data are about 116,000 questionnaires from IBM employees, collected between 1967 and 1973 (sources give 53 to 72 countries). It may partly measure IBM's own corporate culture, and it is old (global-iq, Simply Psychology).
- Nation is not culture. A meta-analysis of 558 studies that used Hofstede's framework found about 80% of the variation in cultural values inside countries; occupation and socio-economic status are sometimes better units than nationality (Taras, Steel & Kirkman 2016).
GLOBE, and why it is not used
The GLOBE study (House et al. 2004) surveyed about 17,000 middle managers in 62 societies, in three industries: finance, food processing and telecommunications. Its nine dimensions are measured twice, as current practices ("as is") and as values ("should be") (global-iq, Andrews University review). We do not use it. None of the three industries is software; on many dimensions practices and values correlate negatively, and how to interpret that is still disputed; and its country averages face the same ecological-fallacy critique as Hofstede's (Brewer & Venaik 2014). A third set of national averages would add apparent precision without adding evidence about engineers. That last point is our judgement.
Evidence from software engineering
Studies that measure national culture in software work directly are few. These are the ones the site leans on.
Rastogi et al. 2018: pull requests and the submitter's country
Rastogi, Nagappan, Gousios and van der Hoek analysed 70,740 pull requests from 17 countries in 1,069 of the busiest GitHub projects, using logistic regression (ESEM 2018, Microsoft Research, TU Delft). The odds of acceptance differed by the submitter's country:
| Submitter's country | Odds | Difference from the United States, drawn as a bar |
|---|---|---|
| Switzerland | 146 | |
| Netherlands | 130 | |
| Japan | 128 | |
| United Kingdom | 114 | |
| Canada | 113 | |
| India | 102 | |
| United States(baseline) | 100 | |
| Russia | 94 | |
| Sweden | 81 | |
| Germany | 78 | |
| Brazil | 76 | |
| Italy | 73 | |
| China | 68 |
Solid bars to the right of the centre line: higher odds than the United States. Outlined bars to the left: lower odds. Full bar width is a difference of 50.
The figures are from the Ehime University WWS2019 slides summarising the paper. The study also found that a pull request is more likely to be accepted when the submitter and the person who merges it are in the same country. This held for 16 of the 17 countries (China +13.4%, United States +2.4%); India was the exception at −7.4%. The cause is not known (WWS2019). Acceptance is a judgement made by reviewers, so these numbers describe reviewers as much as submitters.
Nadri et al. 2021: perceived race
Comparing developers by the race a reader would infer from their name, pull requests from developers perceived as white had 6–10% higher odds of acceptance (arXiv 2021).
Herbsleb & Mockus 2003: distance
Work that spanned sites took about two and a half times as long as similar work within one site, mainly because more people were involved (IEEE TSE 2003). In a 2025 retrospective the authors note how much the environment has changed since (TSE 2025).
Neumann et al. 2024: teams that did not match their country
In four case studies, two in Japan and two in Germany, the cultural traits of agile team members differed markedly from what the countries' Hofstede scores predict. The authors argue for analysing each case rather than applying a national label to a team (arXiv, EASE 2024).
Lambiase et al. 2024: LLM adoption
In a survey of 188 software engineers analysed with PLS-SEM, adoption of LLM tools was driven by habit and expected performance; Hofstede's cultural values did not significantly moderate it (arXiv 2024).
How not to read this site
- Not as a prediction about a person. About 80% of the variation in cultural values is within countries (Taras, Steel & Kirkman 2016). The person you work with may sit anywhere in that spread.
- Not by stepping from the average to the individual. Applying a country score to one person is the ecological fallacy, and Hofstede warns against it himself (Brewer & Venaik 2014, Hofstede 2011).
- Not over the company. Company and team culture often outweigh national culture: the Japanese and German teams in Neumann et al. 2024 did not match their national profiles, and whether code review is adopted depends on the organization, the product and the team (modern code review literature review). Engineering is itself an occupational culture that can narrow national differences (Taras et al. 2016).
- Not as fixed. Cohorts born after 1958 score higher on individualism and indulgence and lower on power distance, but countries keep roughly the same order relative to each other: the gaps shrink while their direction stays (Beugelsdijk et al. 2015, PDF).
- Not from nowhere. A position only means something relative to where you stand. Seen from Japan the United States is direct; seen from the Netherlands it is diplomatic (HBR 2014). The vantage strip on each scale page re-centres the scale on the country you choose.
- Not without the reader's bias. Differences in pull request acceptance and same-country preference are partly differences in reviewers (Rastogi et al. 2018). A 2026 study of 48 engineers in four countries found European colleagues systematically underrated East African engineers whose performance matched their own (Khosravi & Payberah 2026). Whoever reads this site brings their own position too.
Limits of the data
- Stack Overflow survey coverage. The 2024 Developer Survey had 286 responses from Japan (0.5%) and 144 from South Korea (2024 developer profile), far too few to describe either country's developers. The 2025 survey breaks results down only for its five largest countries: the United States, Germany, the United Kingdom, India and France (2025 work).
- Octoverse uses self-reported location. GitHub assigns developers to countries by the location on their profile, and the counts include student and hobby accounts (Octoverse 2025). Countries where many developers use other platforms, China for example, may be undercounted; that is our inference, not GitHub's statement.
- Ukraine has no Culture Map data. Meyer's slides do not place Ukraine on any scale (Meyer deck), so its scale positions read "no source". Its Hofstede scores do exist (Ukraine page).
- The Nordics are a region. Denmark, Sweden, Finland and Norway share an entry, and Meyer's charts do not place all four on every scale. Where a position comes from one country, the note says which; Hofstede scores are given per country (Nordics page).
Corrections
If a claim is wrong, out of date or mislabelled, please open an issue on GitHub with two things: the claim's id (the small code at the end of its label line, such as de-feedback-1) and a link to the source that shows the problem. A better source is the most useful correction of all.
Ids are the same in English and Japanese, so a report filed against either language fixes both. Every id is also an anchor: /countries/de/#de-feedback-1 links straight to the entry.