Basic Terminology

A variable is anything that is measured, such as age (in years), sex (male or female), hair color (black, brown, red, blond and gray), IQ (a test result), behavior, such as taking fish oil (measured as having taken place or not, or numerically, as the number of weeks taken). A hypothesis is a prediction concerning the relationship between variables, such as "Taking fish oil increases the IQ of children." Taking fish oil is an independent variable, or the cause in the cause-and-effect model. IQ is the dependent variable, or effect. A confounding variable is an independent variable that could also influence the dependent variable even if it's not part of the hypothesis. In the example, it could be the mother's IQ, family income, school type, parents' education, family size, social class or eating school meals. A treatment is the experimental stimulus the researcher introduces and controls. In the example, the treatment would be giving fish oil to some children and giving nothing or an inert placebo to others. Experimental groups are the groups the subjects (in social science, usually people) are divided into. Many designs include a treatment group, which receives the treatment, and control groups, which receive the placebo or nothing.

True Experiment

Experimental designs have two fundamental features: a random assignment to groups and multiple groups or measurements. Random assignment means that all subjects have the same chance of ending up in the treatment or control group. The treatment and control group are roughly equivalent. Even if differences exist, they tend to cancel out. Without randomization, the design is not a true experiment. True experimental design is the gold standard for quantitative research. It maximizes internal validity, meaning within the context of the study, the researcher can be confident that any causal relationships discovered are due to treatment. Of the standard types of experimental designs, the simplest involve two randomized groups: one receives the treatment and one doesn't. Measurement takes place in each group after the treatment. An alternative two-group design has two measurements in each group: a pre-test, which establishes a baseline, and a post-test, which measures the effect of the treatment. The classic Solomon Four-Group Design combines the two in a design which has two groups with a pre-test and post-test measurement and two groups with only post-test measurement. Factorial designs include multiple variables, often on many levels.

Quasi-Experiment

Quasi-experimental designs in social sciences usually use pre-existing groups (e.g., two school classes in the same school). Quasi-experiments still a treatment, with the researcher actively introducing the stimuli, as well as a control group or multiple measurements. The designs for quasi-experiments are similar to experiments, except that there is no random assignment of subjects to the groups. For example an education psychologist testing effectiveness of two classroom teaching techniques would find it difficult to randomly assign subjects to groups as it would require splitting existing classes. Instead, she might use a new technique in one class and use another class (as similar as possible to the first) as a control.

Non-Experiment and Correlation Study

Many non-experiments aim to describe a population or its subgroups. Correlation studies attempt to discover relationships between variables which cannot be manipulated as a part of an experiment. Researchers commonly use surveys for correlation studies. Although results of correlation studies allow no causal inferences, they often cause controversy or excitement in the media. All that a correlation study can say is, for example, that children who are breastfed for longer tend to perform better at school. This doesn't mean that breastfeeding causes better school performance. Researchers who conduct such studies must account for confounding variables. In the breastfeeding example, U.S. mothers who choose to breastfeed for longer tend to be older, better educated and of higher social status and likely have different values and parenting styles than those who don't breastfeed. Only after accounting for all those confounding factors, a tentative statement of connection could be made. Still, the only way to establish a real cause-and-effect relationship would be to conduct an experiment in which different mother-child pairs would be randomly assigned to breastfeeding or non-breastfeeding groups, which is impossible for ethical and practical reasons.