Why experiments matter in organisational behaviour
Organisational behaviour (OB) asks causal questions: does job insecurity reduce performance, do intrinsic rewards increase effort more than pay, does ethical leadership change how employees behave? Most doctoral studies answer such questions with cross-sectional surveys, which show associations but struggle to rule out reverse causation, omitted variables and common method bias (Antonakis et al., 2010; Podsakoff et al., 2003). In an experiment the researcher manipulates the proposed cause, holds other conditions constant and, with random assignment, makes groups comparable on everything else.
Experiments do not replace other designs. They answer a narrow causal question well, often at the cost of realism. A thesis that uses one should say why the causal question matters and how the design balances internal validity against relevance to real UAE workplaces.

How UAE-based OB research has been designed
A review of verified studies by researchers at UAE institutions shows a field dominated by surveys. We did not find a published OB experiment by UAE-affiliated authors among the sources we could verify, which is itself a signal: an experimental design can be a methodological contribution in the UAE setting.
| Study | Topic | Design | What an experiment could add |
|---|---|---|---|
| Singh and Shaurya (2021) | AI and HR practices in UAE firms | 8 interviews and a survey of 248; PLS-SEM | Test whether a specific AI-supported HR practice changes employee responses |
| Almazrouei and Alnahhal (2026) | Transformational leadership and performance, public sector | Survey of 233 employees; PLS-SEM with mediators | Manipulate leadership messages in a vignette to test the causal path |
| Sakka (2026) | Leadership, flexibility and digital transformation | Survey of 600 managers; covariance-based SEM | Separate cause from outcome where both are self-reported by one manager |
| Aw et al. (2011) | Field research challenges in the UAE | Methodological discussion of surveys and interviews | Plan for a mobile, multilingual and diverse population |
Almazrouei and Alnahhal (2026) recommend longitudinal designs and private-sector samples, and Singh et al. (2021) note that few UAE firms had integrated AI into HR, limiting generalisation. Aw et al. (2011) describe features of UAE fieldwork that also shape experiments: a large expatriate share of the population, high mobility, incomplete sampling frames and the need for several languages.
Choosing an experimental design
| Design | What is manipulated and where | Strength | Main risk |
|---|---|---|---|
| Laboratory experiment | A task or message in a controlled session | High internal validity | Artificial setting; student samples |
| Vignette or scenario experiment | Descriptions of situations shown to employees online | Feasible with working adults | Intentions rather than behaviour |
| Field experiment | A real practice, such as feedback or rewards, inside an organisation | Realism and behavioural outcomes | Access, contamination between groups |
| Quasi-experiment | A change introduced by the organisation, without random assignment | Studies real interventions | Groups may differ before treatment |
Online panels make vignette experiments with working adults feasible, but panel samples have their own attention and representativeness problems that need screening and reporting (Aguinis et al., 2021). In UAE workplaces, field experiments depend on gatekeepers; agree the design, the randomisation unit (individual, team or branch) and what happens to the control group before you start.
Manipulations, controls and hypotheses
Start from a hypothesis derived from theory, for example that recognition framed as intrinsic (meaningful work) increases discretionary effort more than recognition framed as extrinsic (a bonus). Then operationalise each part: what exactly differs between conditions, what stays the same, and how the outcome is measured, preferably as behaviour rather than self-report.
Manipulations must be valid as well as effective. Chester and Lasko (2021) examined 348 experimental manipulations in a leading social psychology journal and found that validity evidence was often missing or weak. Pilot your manipulation, show that it changes the intended construct and not others, and include a manipulation check in the main study. Use a control condition that differs only in the active ingredient, and keep instructions, timing and language identical across conditions.
Sample size, ethics and data protection in the UAE
Justify the sample from the smallest effect that would matter and the planned analysis, not from what is available (Lakens, 2022); G*Power covers the t-tests, ANOVA and regression models typical of OB experiments (Faul et al., 2007). Between-subjects designs with several conditions need large samples, so consider within-subject or factorial designs where carry-over effects are unlikely.
UAE universities require ethics approval before recruitment. At UAE University, for example, projects involving humans need Human Research Ethics Committee approval (United Arab Emirates University, 2025). Experiments raise particular issues: deception must be justified and followed by debriefing, employees must be free to decline without consequences from their employer, and results must not be used to evaluate individuals. Identifiable data fall under the federal Personal Data Protection Law, Federal Decree-Law No. 45 of 2021, with separate regimes in some free zones (United Arab Emirates, 2021). See the explainers on the UAE Personal Data Protection Law, UAE University research policies and, for studies in health settings, Abu Dhabi's human subject research standard.

Analysis, validity and reliability
Match the analysis to the design. Two or more independent conditions call for t-tests or ANOVA with effect sizes and confidence intervals; factorial designs test interactions; regression or structural models are useful when the experiment also measures mediators. Randomisation protects the comparison between conditions, but a mediator measured after the manipulation is not randomised, so causal claims about the mediating path need the same caution as in survey research (Antonakis et al., 2010).
Report the reliability of multi-item outcome scales in each language version, and check that measures mean the same thing across the nationality and language groups you compare. Where outcomes are self-reported, separate them in time or source from the manipulation to reduce method bias (Podsakoff et al., 2003). Short interviews after the experiment can show how participants understood the materials, which helps interpret both significant and null results.
Worked example (illustrative)
A hypothetical doctoral researcher studies whether recognition framing affects voluntary extra effort among customer service employees in Dubai. Working with two companies, she designs an online vignette experiment with three conditions (intrinsic recognition, extrinsic recognition, neutral control), translated into Arabic and English and piloted with 30 employees to validate the manipulation (Chester & Lasko, 2021). Participants are randomly assigned within each company. The outcome is the number of optional practice tasks completed after the vignette, a behavioural measure, with a self-reported intention measure as secondary.
A power analysis for a one-way ANOVA sets the target sample (Faul et al., 2007). Ethics approval covers consent, voluntary participation independent of managers, debriefing and pseudonymised storage under the PDPL (United Arab Emirates, 2021). Short post-experiment interviews with a subsample explore how employees interpreted the messages, which helps explain any null result.

What to report, and common mistakes
| Element | What to state |
|---|---|
| Hypotheses | Derived from theory, stated before data collection |
| Conditions | Exact materials for each condition, in all languages used |
| Assignment | Unit and method of randomisation |
| Manipulation validity | Pilot evidence and manipulation check results |
| Sample | Power analysis, recruitment route, exclusions with reasons |
| Analysis | Tests, effect sizes, confidence intervals |
| Ethics | Approving committee, consent, debriefing, data protection |
- Calling a correlational field survey an experiment.
- Manipulating several things at once, so the cause of any effect is unclear.
- Relying on one manipulation check as proof of validity.
- Recruiting through managers in a way that makes participation feel compulsory.
- Generalising from one company or nationality group to the UAE workforce.

For related design guidance, see designing experimental research in psychology and primary data collection in the UAE.
Key takeaways
- Experiments support causal claims that cross-sectional surveys cannot (Antonakis et al., 2010; Podsakoff et al., 2003).
- UAE-based OB research is dominated by surveys, so a well-designed experiment can be a contribution (Almazrouei & Alnahhal, 2026; Sakka, 2026; Singh et al., 2021).
- Validate manipulations with pilots and checks, not a single significance test (Chester & Lasko, 2021).
- Justify sample size from the expected effect and planned analysis (Lakens, 2022; Faul et al., 2007).
- Obtain ethics approval and plan PDPL-compliant data handling before recruiting (United Arab Emirates University, 2025; United Arab Emirates, 2021).
