Experimentation is an approach to research best suited for explanation and evaluation. Experiments are especially well suited to research projects that involve relatively well-defined concepts and propositions. Experimentation is especially appropriate for hypothesis testing. It is better suited to explanation and evaluation than to descriptive purposes. Criminal justice experiments are almost always conducted in field settings outside the laboratory.
Variables, time order, measures, and groups are the central features of the classical experiment. The most conventional type of experiment involves three major pairs of components:
1.independent and dependent variables
2.pretesting and posttesting
3.experimental and control groups
Figure 7-1 (p. 178) provides a helpful diagram of basic experimental design.
Sometimes, experimenters will prejudge results. The double-blind experiment eliminates this possibility because neither the subjects nor the experimenters know which is the experimental group and which is the control group.
In deciding who will participate in an experiment, researchers must decide on the target population and how particular members of the target population will be selected for the experiment. Once selected, the researcher randomly assigns the subjects to either the experimental or the control group. Randomization is a central feature of the classical experiment. The most important characteristic of randomization is that it produces experimental and control groups that are statistically equivalent.
Experiments potentially control for many threats to validity of causal inference, but researchers must remain aware of these threats. Reviewing again the criteria for causality:
1.that cause precedes the effect in time
2.empirical correlation between the cause-and-effect variables
3.the observed correlation between cause and effect is not due to the influence of some third variable
There are various threats to internal validity, including (see Chapter 7 for detailed discussion):
●causal time order
●diffusion or imitation of treatments
The problems of internal validity are only some of the complications faced by experimenters. They also have the problem of generalizing from experimental findings to the real world. The text considers two dimensions of generalizability: construct validity and external validity. In the language of experimentation, construct validityis the empirical text of a hypothesis and the underlying causal process that the experiment is intended to represent. Construct validity is thus concerned with generalizing from what we observe in an experiment to actual causal processes in the real world. There are three elements of enhancing construct validity:
1.linking constructs and measures to theory
2.clearly indicating what constructs are represented by specific measures
3.thinking carefully about what levels of treatment may be necessary to produce some level of change in the dependent measure
External validity represents a slightly different form of generalizability, one where the question is whether results from experiments in one setting (time and place) would be obtained in other settings, or whether a treatment found to be effective for one population will have similar effects on a different group. Statistical conclusion validity becomes an issue when findings are based on small samples of cases, and can be magnified by other difficulties of field experiments.
The basic experimental design is adapted to meet different research applications. Figure 7-3
(p. 189) provides a graphic representation of variations in the experimental design. When randomization is not possible, researchers can use different types of quasi-experimental designs. In most cases, quasi-experiments do not randomly assign subjects, and therefore they may suffer from the internal validity threats that are so well controlled in true experiments.
Quasi-experimental designs can be grouped into two categories: nonequivalent-groups designs and time-series designs. When it is not possible to create groups through randomization, a nonrandom procedure must be used; if we construct groups through a nonrandom procedure, we cannot assume that the groups are equivalent – hence the label nonequivalent-groups design. The text provides a number of interesting illustrations of this type of design. Time-series designs are common examples of longitudinal studies in criminal justice research – this type of design involves examining a series of observations on some variable over time. There are a number of variations in time-series designs, as illustrated in Figure 7-6 (p. 198).
Understanding the building blocks of research design and adapting them accordingly is preferable to trying to apply the same design to all research questions.