Norms, Beliefs, and Determinants of Harsh Discipline in Schools: Experimental Evidence from Zambia

2024-09-01

2026-06-30

For each vignette shown to a respondent, we measure:
1. Personal acceptability of the teacher's disciplinary response
2. Social expectations about the disciplinary response (normative and empirical expectations)
3. Personal behavioural attitude
4. Beliefs about the consequences of the disciplinary response for the child
The randomization of vignette attributes allows each of these outcomes to be estimated as a function of the type of offence, the severity and nature of the disciplinary response, and the child's gender — providing experimental identification of the determinants of adult attitudes, norms, and beliefs about harsh discipline.

The study is a factorial vignette experiment embedded in the endline survey of a cluster-randomised trial evaluating a comprehensive school health program in 225 primary schools in Zambia. The vignette experiment is administered to teachers in these schools.
Each respondent is presented with one vignette describing a scenario in which a teacher reacts to a learner's misbehaviour. Vignettes take place "today, in a community similar to yours" and involve a 10-year-old learner. The attributes of each vignette are randomly assigned to the respondent from a factorial design:
(a) the learner's gender.
(b) the type of misbehaviour ( accidentally spills water on their book in class", " makes noise and disturbs other learners in class", or " beats another learner in class")
(c) the disciplinary response ("whips the learner with a stick", "makes the learner kneel in front of the class for 10 minutes”, "sits the learner down and talks to them, explaining why their behaviour was wrong", "makes the learner sweep the classroom alone after school").
This yields 2 × 3 × 4 = 24unique vignettes.
For each vignette, respondents report: (A) whether they personally find the adult's behaviour acceptable; (B) how many members of their community (fellow teachers) — out of five — would find the behaviour acceptable, and how many would react in the same way; (C) how likely they themselves would be to act in the same way; and (D) their beliefs about the likely consequences for the learner.
The random assignment of vignette attributes allows experimental identification of how respondents' attitudes, social expectations, self-projected behaviour, and beliefs about consequences vary with the type of offence, the severity and nature of the disciplinary response, and the child's gender. Content and wording of the scenarios were developed on the basis of focus group discussions and key informant interviews with teachers in Ndola.

Not available

Randomization done in office by a computer.

The randomization of vignettes was conducted at the individual level.

No

Using baseline sample sizes as an upper bound:
- Teachers: 1,714 (planned; will be lower due to turnover)

Using baseline sample sizes as an upper bound: - Teachers: 1,714 (planned; will be lower due to turnover)

Vignette experiment (unit = respondent):
The vignette experiment uses within-respondent factorial randomisation, so the "arms" are the cells of the factorial design. Each respondent is randomly assigned to one cell. Expected sample sizes per level (based on baseline sample of 1,714 teachers):
(2 × 3 × 4 = 24cells; ≈1,714 teachers):
- By disciplinary response (4 levels — talk, kneel, whip with stick, sweep classroom): ≈429 teachers per level
- By misbehaviour (3 levels — accidental, disturbs class, beats another learner): ≈571 per level
- By learner gender (2 levels — boy, girl): ≈857 per level
- Per full cell: ≈71 teachers