Contrasting the levels of an independent variable can be accomplished either between-subjects or within-subjects. That is, subjects can be randomly assigned on one of several groups. Subjects in each group are treated to or observed under the single condition that has been assigned to that treatment group. In this situation, the levels of the independent variable are contrasted between subject, as each single subject is exposed to only one treatment condition.

Alternatively, the levels of an independent variable can be contrasted within-subjects by randomly selecting a sample of participants from an identified population and observing all subjects under each level of the treatment (independent variable).

Choosing between these methods of contrast warrants careful consideration, for each approach has advantages and disadvantages. In true experiments it is of utmost importance that the choice be made in view of possible sources of invalidity, which will confound the results.

The within-subjects method of contrast is very attractive, as this procedure is designed to reduce between-subjects sources of variation. Consequently, in most cases, this method offers greater statistical power. Nevertheless, the within-subjects designs can be affected by statistical difficulties when the data do not meet the assumptions of sphericity or additivity. Modern computer programs provide tests of these assumptions but the researcher must be prepared for further analyses should either of these assumptions not be met.

The within-subjects model can also introduce sources of bias that are known as range effects and order effects. A range effect can occur when subjects are tested across various levels of a quantitative independent variable; if present, a range effect results in a bias toward the mean of the range of stimuli or the range of responses that are required. An industrial psychology study by Kennedy and Landesman (1963) confirmed the presence of a range effect in a simple manipulation task; Poulton (1975) has written on the presence of these effects in a variety of experiments.

The order in which the levels of the independent variable are presented to the subjects can cause a carry-over bias; that is, the results of experiencing a previous level of the independent variable can affect the subjects' response on a subsequent level. While counter-balancing is often suggested as a procedure to reduce or eliminate order effects, there is considerable evidence that counter-balancing can introduce an asymmetric transfer effect, a bias that cannot be removed via statistical manipulation (Poulton & Freeman, 1996).

Alternatively, a between-subjects contrast will not introduce either range or order effects; however, this method usually requires a considerably larger number of subjects to achieve the same level of statistical power as the within-subjects approach. Without a careful estimation of the number of subjects required to attain an adequate level of statistical power, the researcher can obtain results that are not statistically significant, thereby raising the probability of making a Type II error.

Various approaches have been developed to compare the methods of contrast, i.e., that allow a researcher to determine if the experimental results are different or if a between-subjects or a within-subjects contrast is used. A method developed by Erlebacher (1972) provides an efficient albeit complex process to test effects of these models on the data collected in empirical studies.

When examining the literature on a specific topic, if it is found that one cluster of studies render significant results while another does not, consider the possibility that the method of contrasting the levels of the independent variable might be a confounding variable. Examine whether some of the studies used a between-subjects approach while others used a within-subjects contrast. Then, consider the possibility of developing a study that allows a comparison of the method of contrast of the levels of the independent variable using Erlebacher's (1972) technique. Perhaps such an experiment should be one of the initial studies in all lines of inquiry. technique. Perhaps such an experiment should be one of the initial studies in all lines of inquiry.