PTDoes talking about errors without hurting help a team learn?
After an error, the conversation that follows can push people apart or bring them closer. The thesis asks whether nonviolent communication (observing without judging, saying what you feel and need, asking without demanding, listening) goes with a team's psychological safety and learning from errors. The author built a new scale and surveyed 308 professionals. Explained in 12 screens.
The same error can turn into two conversations
Nonviolent communication (NVC) was proposed by the psychologist Marshall Rosenberg in the 1960s. It has four steps: observing without judging, stating the feeling, naming the need and making a concrete request. The author adds a fifth pillar, empathic listening. Switch the conversation and tap a sentence:
The situation: the report sent to the client went out with the sales total swapped. The team meets to talk about the error.
Pillar definitions from Table 10 (version adjusted after expert review) and sections 2.3.1 to 2.3.5. The example sentence “you always get these reports wrong” uses one of the words that observation avoids: “always”, “never”, “every time” (section 2.3.1).
Three ideas the author set out to link in a chain
It was already known that teams with psychological safety learn more from errors. What was missing was a look at the way people talk: the author found no studies linking nonviolent communication to learning from errors at work. Tap a block or an arrow:
Model from Figure 9 (chapter 3). The thesis numbers the hypotheses differently in chapters 3 and 5; here they appear only as arrows.
None of the 21 studies linked NVC to learning from errors
In March 2023, the author searched for “nonviolent communication” in the Web of Science database and ended up with 21 studies with data. Almost all of them tested an NVC training and measured empathy, anger or self-esteem afterwards, most of them in health care. None looked at learning from errors at work.
Section 2.4; Figures 6 and 7 (counts read from the figures, adding up to 21) and countries and methods from the text. The review is a mapping: it counts and classifies the studies without pooling their results.
A new scale to measure NVC in a team's day-to-day work
Existing scales measured the effect of training, not the five pillars. The author wrote sentences for each pillar, asked three AI tools for suggestions, consulted experts and tested. Move through the stages and tap a pillar:
For psychological safety, the author compared two scales with 114 people (they measure the same thing, correlation of 0.80) and kept the one by Liang and colleagues, which was more reliable.
Height = number of sentences; hollow box = the 4th version. Tables 4, 11, 18, 24 to 26; Appendix 8. At the end, listening has two shades: compassionate and active.
Answer like a participant
Think of a conversation in your team about an error. Six real sentences from the questionnaire, all from 1 to 5. The mean of the 308 appears after your answer.
Sentences from Appendix 8 (translated by us from the Portuguese questionnaire); means from Tables 20, 21 and 22. The first four ask about frequency (1 = never, 5 = always); the last two, about agreement. This is not a diagnosis: it is only to get a feel for the questionnaire.
308 people who work in teams and talk about errors
The questionnaire circulated in August and September 2025 on the author's WhatsApp, Instagram, Facebook and LinkedIn, with a prize draw for books. 478 people answered. Only those who work in a team, depend on colleagues and discuss errors when they happen were kept:
Figure 15 and Appendix 9. To detect an effect of 0.2, the calculated minimum was 155 people.
One dot per person; counts from Appendix 10 and Table 19. The team-size table in Appendix 10 adds up to 361, more than the 308 people, so it is not shown here. The questionnaire did not offer the option “parda” (mixed race) for colour.
On a 1 to 5 scale, almost everything landed near 4
People say they practise NVC often and agree that errors teach. Psychological safety came out a little lower: on average, nobody “strongly agrees”. Before you look, a guess: which behaviour scored highest?
The floor is 1 and the hollow box goes up to 5. Tables 20, 21 and 22. Each pillar's mean is the simple mean of the sentences kept in the model (our calculation).
All three links were confirmed
Using structural equation modelling, the author tested the three arrows at once. People who say they practise NVC more also report more psychological safety and more learning from errors. The strongest link is between safety and learning. Tap an arrow or switch the model:
Standardised coefficients (β), Table 30 (without controls) and Appendix 11, model 3 (with team size and having subordinates). Thickness ∝ β. R² = how much of the differences between people the model explains.
Part of NVC's effect reaches learning through safety
NVC is linked to learning from errors by two paths: a direct one (0.269) and one that passes through psychological safety (0.307 × 0.526 ≈ 0.162). Added together they come to 0.431, a little less than the link of safety alone (0.526). Together, NVC and safety explain 43.6% of the differences in learning. Tap a band of the glass:
Tables 30 and 31 and Figure 18. The parts of the glass are coefficient × correlation (0.269 × 0.430 and 0.526 × 0.608). Our observation: the thesis calculates the path through safety as the product of the arrows but does not report a significance test of this indirect effect. Illustrative reading, at a single point in time.
Having subordinates mattered more than knowing NVC
Almost 58% of participants had already read about NVC or taken a course. The author tested whether this, gender, age, team size and having subordinates changed safety and learning. Before you look, a guess: who reported feeling safer to speak up in the team?
Our reading: 139 of the 188 people with subordinates answered thinking about the team they lead; leaders may see their own team as safer than it looks from the inside. And the effect of having subordinates on learning disappears once safety enters the model, as if it ran through it.
Table 32 (controls together, without NVC and safety) and Appendix 11, model 3 (only the significant controls, together with NVC and safety). For groups (having subordinates, gender), the value is the difference in standard deviations. Team size was measured in bands (up to 5, 6 to 10, 11 to 20, more than 20). Striped = no clear effect; light = hint (p between 0.05 and 0.10); dotted = not in that model.
The author argues for training the way teams talk
For the author, the results support training leaders and team members in communication and building safe environments together, where errors become material for learning, at work and also at school, in the family and in the community.
For leaders
- Observing the fact before judging the person, and asking instead of demanding, are behaviours that can be taught, the author says.
- Larger teams reported less safety: they call for more care with the conversation.
For HR and teams
- The new scale (20 sentences in five pillars) can serve as a diagnosis of a team's conversation skills, according to the author.
- The author links the topic to the WHO guidelines on mental health at work and to NR-01, the Brazilian standard on occupational risk management.
For society
- The thesis notes that NVC already appears in public policy, such as a 2025 recommendation on restorative practices in Brazilian prisons.
- The 58 people whose teams do not discuss errors were left out: the author suggests studying them.
What this research does not say
- 308 people, a convenience sample. People from the author's networks, with varied profiles (more women, over 40, postgraduates); the results cannot be generalised. · according to the author
- Each person spoke for themselves. It is one person's view of their own team, not the whole team measured. · according to the author
- The NVC measured is what people say they practise. Nobody observed the conversations. · our observation
- A single point in time. It shows what goes together, not that NVC causes safety or learning. · our observation
- NVC explains 9.5% of safety. The rest comes from other things, such as leadership, culture and support, the author notes. · Table 31
- Path through safety without a reported test. The indirect effect (0.162) is the product of the arrows. · our observation
Credits
BARBOSA, Lidiane Mendes. A comunicação não violenta nos processos de aprendizagem a partir do erro [Nonviolent communication in the processes of learning from errors]. Doctoral thesis (Business Administration). Mackenzie Presbyterian University, São Paulo, 2026. Advisor: Prof. Dr. Diógenes de Souza Bido.
Thesis approved on 1 June 2026, with a scholarship from CAPES (Brazil's federal agency for graduate education). The author declares using Copilot, Gemini and ChatGPT to suggest sentences for the scale, with critical review (the prompt is in Appendix 1). Data analysed in SmartPLS 4 and Jamovi. This version simplifies the work for a general audience.