Chapter 7. Quantitative Research Designs
7.1 Design follows the question
The design defines when and how measurements are made, how comparison groups are formed, and what conclusions are defensible. Select the design before data collection and describe it in enough detail for a reader to understand the inferential limits.
| Design | Basic structure | Can support | Key threats/limits |
|---|---|---|---|
| Cross-sectional survey | Variables measured at one time | Prevalence and associations | Temporality, confounding, nonresponse, common method |
| Longitudinal panel | Same units measured over time | Change and temporal ordering | Attrition, time-varying confounding, measurement consistency |
| Repeated cross-section | Different samples from same population over time | Population-level trend | Changes in sample composition |
| Randomised experiment | Random assignment to conditions | Strong causal inference under sound implementation | Noncompliance, attrition, spillover, limited generalisability |
| Quasi-experiment | Intervention without full randomisation | Causal evidence if assumptions/design are credible | Selection bias and unmeasured confounding |
| Case-control | Select by outcome and compare prior exposures | Efficient study of rare outcomes | Recall/selection bias; not direct incidence estimation |
| Cohort | Follow exposed/unexposed units over time | Incidence and temporal relationships | Confounding, attrition, cost |
| Secondary-data analysis | Analyse existing records/database | Efficient analysis of large or historical data | Fit-for-purpose, missingness, coding and provenance |
| Time-series/panel modelling | Repeated observations over time or units | Dynamics and within-unit patterns | Autocorrelation, nonstationarity, dependence |
7.2 Cross-sectional research
Cross-sectional studies are common in business dissertations because they can be feasible within limited time. Their major limitation is that exposure and outcome are measured at roughly the same time, so temporal ordering may be unknown. Results should usually be described as associations, not causal effects. The discussion should acknowledge plausible reverse direction and unmeasured confounding.
7.3 Experimental and quasi-experimental designs
Random assignment aims to make groups comparable on average at baseline. A pre-registered protocol, allocation concealment where applicable, consistent intervention delivery, outcome blinding where feasible, sample-size planning, intention-to-treat analysis and transparent reporting strengthen an experiment. Quasi-experiments may use difference-in-differences, interrupted time series, regression discontinuity or matched comparisons, but each method depends on assumptions that must be justified and checked where possible.
7.4 Longitudinal and panel designs
Repeated measurements can clarify change and temporal sequence but require consistent measures, appropriate intervals and an attrition plan. The analysis must account for dependence among observations from the same person, team, organisation or country. Ordinary regression that treats repeated observations as independent may underestimate uncertainty.
7.5 Secondary data
Before using secondary data, inspect the data dictionary, sampling design, collection purpose, variable definitions, timing, consent and licensing terms, missingness, revisions, weights and linkage quality. Data collected for operational purposes may not measure research constructs in a theoretically valid way. Document transformations and avoid treating administrative fields as self-explanatory.