The country changes how people answer the same engagement survey

Not every rule in a society is written down

Institutional theory sees three layers of rules. On top, laws and regulations that anyone can read. Below, norms and traditions nobody wrote down but everyone follows. At the bottom, what is taken for granted and hardest to change. The thesis calls the two lower layers informality rules. Tap each layer.

Since no database of informality rules exists, the thesis uses each country’s culture as a proxy: the scores of the GLOBE project, an international study that measures nine cultural traits in dozens of countries. Section 2.2 and Figure 1 (Scott’s pillars).

Every year, the same 12 statements reach every site, in any country

The company uses Gallup’s Q12 questionnaire, widely used around the world to measure engagement. Each person says how much they agree with each statement, from 1 (strongly disagree) to 5 (strongly agree). There is also an overall satisfaction question (Q00). Answer as if it were about your own job.

According to the thesis, the survey is the same in every country and does not consider local differences; results come out by site, but the analysis tends to treat them as a whole. Your answers are not stored.

Out of almost 20,000 rows of answers, 3,226 made it through the cleaning

The company handed over the 2022 survey in a spreadsheet, with no data on age, gender or tenure. The author removed faulty answers, repeated rows and countries that were too small. Tap each step (widths not to scale).

Before comparing, the thesis discounts how high each country tends to answer

In some places people mark 4 and 5 for almost everything; in others they are more reserved. If that were all, countries would just be “more optimistic” or “more critical”. So, besides the original scores, the author subtracts the country’s overall average from each answer and looks at what is left: which questions sit above or below that place’s way of answering.

Illustration with made-up numbers. Two imaginary countries and three questions. In the thesis, the tests run in both versions: the “direct” one and the “modified” one, with the average removed.

On all 12 questions, the countries answered differently

An analysis of variance (ANOVA) compares the averages of the 18 countries on each question. On every one, the difference is too large to be chance (p < 0.001), both with the original scores and with each country’s average removed. In other words, it is not just one country being more optimistic than another.

Even with optimism removed, Brazil differs from the other countries on different questions

The thesis compared Brazil with each of the other 17 countries, question by question. Each dark square is a statistically significant difference (p < 0.05). Switch the scale and tap a country.

Engagement follows how society acts, not how it says it should act

GLOBE measures each cultural trait in two ways: practices (“how things are”) and values (“how things should be”). The thesis matched the 18 countries’ engagement averages with each trait. The rule: a trait only counts if it links strongly to at least 3 of the 12 questions (dashed line). Tap a column.

“Strongly” = p < 0.05 on each question. Two practice traits pass: assertiveness (11 questions) and institutional collectivism (3). No value trait passes. With only 18 countries, the thesis accepts 90% and 80% confidence as medium and weak hints, shown in the panel.

More assertiveness goes with less engagement; more collectivism, with more

Assertiveness measures how tough and confrontational people in a society are in their relationships. Institutional collectivism measures how much institutions encourage collective action and reward. Each bar shows how a trait links to one question (t statistic): to the left, reverse; to the right, direct.

With 12 questions, one hit by luck is easy; three at once is rare

Each test at 95% confidence has a 5% chance of flagging a link that does not exist. With 12 questions per cultural trait, that happens often. Draw a made-up trait, with no real link at all, and count the false “hits”.