Cross-sectional research designs are commonly vilified, but are they as limited as generally believed? In this 2019 paper, published in Journal of Business and Psychology I explore how to optimize cross-sectional research designs, and argue that the frequently touted alternative, the longitudinal design, cannot generally accomplish what people assume.
Limitations of Cross-Sectional Designs
The cross-sectional research design, where all variables are measured at the same time, is frequently criticized for leaving the issue of causality ambiguous. This criticism is particularly likely with survey research where all variables are assessed through self-report. It is widely believed that the longitudinal design, with variables assessed at 2 or more time points, offers advantages because it allows for the determination of whether variables assessed at one time can predict variables assessed at a later time. It is believed that being able to link variables over time can provide evidence for causality.
Are Longitudinal Designs Really Better than Cross-Sectional Designs?
To draw an inference that variable X is a cause of variable Y, one must be able to show that X occurred before Y. With the typical longitudinal design, X is measured before Y is measured, but this does not mean that X occurred before Y occurred. To determine that we need to know when X and Y occurred so we can measure X before Y happens. If we are interested in whether employee feelings of fair treatment drive their job satisfaction, we would need to know when their feelings of fairness occurred relative to the feelings of satisfaction. If both feelings have been formed prior to the study, measuring one before the other is not telling us what we want to know.
How to Optimize Cross-Sectional Research Designs
The cross-sectional design is not as limited as often assumed. There is much that we can do with such designs.
- Establish that two or more variables are related. Before we can show that X causes Y, we need to show that they are related.
- Rule out alternative explanations for why X and Y are related. Two variables might be related due to the action of a third variable that is producing a spurious relationship. By including expected causes of X and Y, we can test their effects statistically.
- Ask people explicitly about when X and Y happened. There are a number of ways to do this that can shed light on the sequence of events.
- Use qualitative methods that ask people to explain an event, what led up to it, and what were the consequences. You can look for consensus among people to enhance confidence in conclusions.