Internal vs external validity
When testing cause-and-effect relationships, validity can be split up into two types: internal and external validity.
Internal validity refers to the degree of confidence that the causal relationship being tested is trustworthy and not influenced by other factors or variables.
External validity refers to the extent to which results from a study can be applied (generalized) to other situations, groups or events.
Trade-off between internal and external validity
Better internal validity often comes at the expense of external validity (and vice versa). The type of study you choose reflects the priorities of your research.
A causal relationship can be tested in an artificial lab setting or in the ‘real world’. A lab setting ensures higher internal validity because external influences can be minimized. However, the external validity diminishes because a lab environment is different than the ‘outside world’ (that does have external influencing factors).
A solution to this trade-off is to conduct the research first in a controlled (artificial) environment to establish the existence of a causal relationship, followed by a field experiment to analyze if the results hold in the real world.
Threats to internal validity
There are seven factors that can threaten the internal validity of your research. They are explained below using the following example:
The management of company X wants to know if flexible working hours will improve job satisfaction among employees. They set up an experiment with two groups: 1) control group of employees with fixed working hours 2) experiment group with employees with flexible working hours. The experiment will run for six months. All employees fill in a survey measuring their job satisfaction before the experiment (pre-test) and after the experiment (post-test).
|Confounding factors||Unexpected factor or variable that influences the causal relationship tested in the study.||A new (better) manager starts during the duration of the study which may improve job satisfaction.|
|Maturation||The passage of time influences the dependent variable (job satisfaction).||During the six-month experiment, employees become more experienced and better at their jobs. Therefore, job satisfaction may improve.|
|Testing||The pre-test (used to establish a baseline) affects the results of the post-test.||Employees feel the need to be consistent in their answers in the pre-test and post-test.|
|Participant selection||Participants in the control and experimental group differ substantially and can thus not be compared.||Instead of a randomly assigning employees to one of two groups, employees can volunteer to participate in an experiment to improve job satisfaction. The experimental group now consists of more engaged (more satisfied) employees to begin with.|
|Mortality||Over the course of a (longer) study, participants may drop out. If the drop out is caused by the experimental treatment (as opposed to coincidence) it can threaten the internal validity.||Really dissatisfied employees quit their job during the study. The average job satisfaction will now improve, not because the “treatment” worked, but because the dissatisfied employees are not included in the post-test.|
|Regression towards mean||Extreme scores tend to be closer to the average on a second measurement.||Employees who score extremely low in the first job satisfaction survey probably show greater gain in job satisfaction than employees who scored average.|
|Instrumentation||There is a change in how the dependent variable is measured during the study.||The questionnaire in the post test contains extra questions compared to the one used for the pre-test.|
Threats to external validity
There are two factors that can threaten the external validity of a study. Both are explained using the example introduced earlier.
|Testing||Participation in the pre-test influences the reaction to the ‘treatment’.||The questionnaire about job satisfaction used in the pre-test triggers employees to start thinking more consciously about their job satisfaction.|
|Participant selection||Participants of the study differ substantially from the population.||Employees participating in the experiment are significantly younger than employees in other departments/ companies. So, the results can’t be generalized.|