Laboratory experiments can show causal relations while controlling much better for disturbing factors than is possible in field studies. Those who are skeptical about external validity of experimental findings might be surprised to learn how accurately they predict the same effect in the field, and how much better they can be replicated than field studies can:
Of well-conducted experiments, the question is not if the results are externally valid, which they oftentimes are at the scale at which the results are obtained, but if the causal mechanism found at that scale, say a small group, can be generalized to different scales, e.g. national states.
Practicalities: (1) recruiting subjects
Universities with a lab usually have a website to recruit students. Another option is to recruit them online through Mechanical Turk, although its reliability is disputed:
Most experiments in Europe and the VS are done on WEIRDos: they are from Western, educated, industrialized, rich, and democratic cultures, according to
The implication is that experiments should be replicated across the planet: See Joseph Henrich's work, or a laid back introduction by:
Notice that experiments conducted in the field are not the same as field experiments, because for the latter there is no random assignment of subjects to treatments. For all experiments, researchers must obey strict protocols on how to treat subjects, for ethical reasons and to obtain valid results.
Fortunately, within Western societies, people who are recruited for experiments (usually students) do not differ from the remainder population, but Western people are not representative for the rest of the world:
(2) computer interface
Subjects often get to see the instructions for the experiment, while researchers collect their data, through a computer interface. Currently, the widely used z-tree program, which is a monster, is giving way to more user friendly software, among others the open source program oTree.
(3) data and analysis
A file of experimental data is not any different from a survey study, i.e. each subject has one row; columns are treatments, choices, and payoffs. The statistical analysis of experimental data is most of the times much simpler than of survey data, and usually consists of comparing groups with different treatments pair-wise; see this overview of options.
An early one:
An anthropological experiment in the field:
A Web experiment:
Experts at the AISSR are Theresa Kuhn, Bertjan Doosje, and Abbey Steele for field experiments.