Quasi-experiment

from Wikipedia, the free encyclopedia

A quasi-experiment is a research design and belongs to the four main types of experiments in the systematics of the experimental plans . The term experiment describes investigations that enable a statement to be made about a causal relationship between two variables. A quasi-experiment compares natural groups without randomized assignment of test persons. The research design determines the value of the causal conclusions that are determined on the basis of the empirical findings.

The quasi-experiment contains essential components of sufficient quality criteria, but does not allow a complete control of all experimental components, since among other things there is no randomized sample selection.

Cause and effect in the experiment

The aim of the experiment is the valid assessment of the dependency relationship of cause and effect, or the answer to the question "Causes the causative attribute A of the independent and (by treatment English treatment ) random variable variable X the measured value W of the dependent response variable Y?". The variation of the independent variable precedes the measurement of the dependent variable in time, so only the independent variable can act on the dependent variable. In quasi-experiments, variables that lead to alternative explanations of the causal relationship due to their influence on the dependent variable are referred to as disruptive factors and must be checked or eliminated.

Differentiation from other experimental research designs

In contrast to the experiment, there is no randomized assignment of test persons to the experimental and control groups in the quasi-experiment . Test objects are selected according to existing properties, e.g. B. according to socio-demographic characteristics or group membership. The quasi-experiment compares natural groups and the rigorous experiment compares randomly selected groups.

The lack of randomization leads to confounding factors, since differences between experimental and control groups cannot be clearly attributed to the independent variable and thus impair the internal validity (validity). Internal validity, as the most important quality criterion of experiments, describes the causal attribution of a change in behavior or characteristics of the dependent variable to the change in the independent variable. The internal validity is guaranteed by the randomization method. The randomized assignment minimizes or neutralizes existing differences between groups as a source of disruptive factors and thus functions as a statistical error compensation.

Disturbing factors can occur in the form of a wide variety of artifacts, e.g. B. due to an ambiguous time sequence, selection distortion due to the selection, statistical regression, etc. The internal validity decreases more sharply if several disruptive factors act simultaneously. An impairment of the external validity (inductive generalizability of the test results) is not to be expected in quasi-experiments. The demarcation from the ex-post-facto experiment is based on the so-called self-selection, in which test subjects assign themselves to test groups.

Criteria for the indication of the quasi-experiment

A quasi-experiment exists if:

  • no randomized allocation of test subjects to the study groups is possible.
  • External changes are interpreted as manipulation of the independent variable, even though the change was caused by researchers.
  • Interventions are made that influence the dependent variable of the investigation.

Types of quasi-experimental designs

Quasi-experimental design without a control group

Experiments without control groups are often used in field research. With regard to internal validity, these are subject to a variety of impairments. It is therefore only possible to draw conclusions by excluding alternative declarations.

One-group pre-test-post-test design with non-equivalent dependent variables

In addition to dependent variables for which changes are claimed as a result of the intervention, dependent variables for which no corresponding changes are assumed are recorded. This design assumes that confounding factors affect the values ​​of the two groups of dependent variables in the same way. For this reason, many possible impairments of the internal validity can be controlled, so this test plan can be used in many ways and allows causal interpretations of the results.

One-group, repeated intervention design

This type of design is useful when treatments are used, suspended, and resumed. At least four measurement times are realized at which the treatments are temporarily suspended. This design allows the covariation of treatments and dependent variables to be analyzed over time. The test plan is suitable for research questions that focus on transient effects or do not allow a longer study period.

Quasi-experimental test plans with control group

The foundation of a quasi-experimental design is an investigation design in which at least two samples are taken. These two samples can, for example, each represent a so-called experimental or control group. Experimental and control groups are generally surveyed in order to establish a cause-effect relationship, which, however, only works when comparing two identical groups. The group comparability in a quasi-experimental test plan cannot be assumed, however, since no procedures for controlling confounding factors are used when assigning individual persons to the two groups, such as B. Randomization or parallelization processes.

In order to differentiate between the respective quasi-experimental test plans with control groups, different measurement times are recorded. There is the so-called post test, in which only one point in time is measured, and the pre-test post test, in which two points in time are measured.

Quasi-experimental test plans with control group and pretest-posttest

Quasi-experimental test plans with a control group and pre-test post-test are carried out, as the pre-test comparison between the experimental and control group can be used to exclude possible differences that already existed before the systematic variation of the independent variable in the experimental group. Differences that already occur during this first measurement prevent later differences between the experimental and control group from being clearly attributed to the causal relationship to be examined. Even with a pre-test post-test, however, it cannot be ruled out that there are further independent variables in which the two groups differed beforehand, as the test does not necessarily guarantee that all relevant variables were taken into account.

Time series designs

A time series is a sequence of measurements of one (or more) variables at successive points in time, both before and after the intervention (treatment). At least three before and after measurements are required. The external validity is very high, as the survey takes place in the everyday life of the study participants. Time series test plans are divided into single and multiple time series test plans.

In the case of a simple time series, several data collections are carried out in a test group before and after an intervention at fixed time intervals. Disruptive factors (maturation, test effects, changes in the measuring instruments, statistical regression) can be controlled by the multiple measurements. Only the disruptive factor of time influences (events in the meantime) cannot be controlled here and limits the internal validity .

In the case of multiple time series test plans, the data is collected in at least two groups. In the case of two groups, the simple time series test plan is usually supplemented by a control group. Basically, the survey is carried out in both groups in parallel, whereby the control group is not exposed to any intervention. It is only influenced by what has happened in the meantime. In this way, this disruptive factor can also be checked and the internal validity ensured. The multiple time series design is one of the most meaningful quasi-experimental designs.

Regression discontinuity analysis

The basic idea of Regression Discontinuity Analysis (RDA) is to find a variable that affects whether or not a person has been treated. It is therefore tried a fully controlled assignment on the basis of an assignment variable ( English variable assignment , or continuous variable assignment given) to be generated. This variable serves as a control variable, must be observable and have a discontinuity. An assignment variable X can be any variable that is normally distributed and continuous, was recorded before the treatment and does not correlate with the treatment. It provides additional information that the investigator can use to divide the sample. A threshold value is defined on this assignment variable X (e.g. the mean value ). The assignment of the individuals to the control group or treatment group then takes place on the basis of this threshold value. The threshold value should be as close as possible to the mean value of X in order to ensure optimal selectivity . The allocation of persons is carried out strictly and clearly along the threshold value. In order to be considered a treatment effect, the discontinuity must occur exactly at the threshold value. Any alternative explanation for a discontinuity precisely at that threshold value would have to be plausible in order to be able to trace the change back to something other than the treatment.

Solution approaches for problem areas of the quasi-experiment

The endangered internal validity of the quasi-experiment can reduce the value of the research results. Nevertheless, it is possible to carry out quasi-experimental investigations with meaningful conclusions. Quasi-experiments are about control techniques such as B. to expand the use of several dependent variables so that they become more meaningful. The expression of disruptive factors and their influence on the dependent variable must be calculated from the measurement results. For example, one tries to counter the problem of the inequality of the comparison groups in the effect variable before treatment by statistically homogenizing the groups. This can be done by identifying outliers in the pre-test group comparison and excluding them from the further calculation. This again ensures that the samples to be compared do not differ beforehand.

literature

  • Bortz, Jürgen; Döring, Nicola (2006): Research methods and evaluation for human and social scientists , 4th edition, Heidelberg.
  • Hertel, Silke; Clever, Julia; Schmitz, Bernhard (2010): Quasi-experimental experimental plans , 1st edition, Göttingen, published in: Holding, Heinz; Schmitz, Bernhard (Ed.): Handbook Statistics, Methods and Evaluation.
  • Sarris, Viktor; Reiss, Siegbert (2005): Brief Guide to Experimental Psychology , 1st edition, Munich.
  • Sedlmeier, Peter; Renkewitz, Frank (2008): Research methods and statistics in psychology , 1st edition, Munich.
  • Shadish, WR; Cook, TD; & Campbell, DT (2002): Experimental and Quasi-Experimental Designs for Generalized Causal Inference , Boston: Houghton Mifflin.

Web links

supporting documents

  1. See Sarris, V .; Reiss, S. (2005): page 61.
  2. See Sedlmeier, P .; Renkewitz, F. (2008): page 124.
  3. See Bortz, J .; Döring, N. (2006): page 54.
  4. See Sarris, V .; Reiss, S. (2005): page 20 and page 39.
  5. See Sedlmeier, P .; Renkewitz, F. (2008): page 132.
  6. See Sedlmeier, P .; Renkewitz, F. (2008): page 124.
  7. See Bortz, J .; Döring, N. (2006): page 54.
  8. See Sedlmeier, P .; Renkewitz, F. (2008): page 132.
  9. See Sarris, V .; Reiss, S. (2005): page 41.
  10. See Bortz, J .; Döring, N. (2006): page 5.4
  11. See Sarris, V .; Reiss, S. (2005): page 73 f.
  12. See Bortz, J .; Döring, N. (2006): page 53.
  13. See Hertel, S .; Klug, J .; Schmitz, B. (2010): page 52.
  14. See Hertel, S .; Klug, J .; Schmitz, B. (2010): page 49.
  15. See Hertel, S .; Klug, J .; Schmitz, B. (2010): page 53 ff.
  16. See Hertel, S .; Klug, J .; Schmitz, B. (2010): page 53 ff.
  17. See Hertel, S .; Klug, J .; Schmitz, B. (2010): page 53 ff.
  18. See Hertel, S .; Klug, J .; Schmitz, B. (2010): page 55 ff.
  19. See Hertel, S .; Klug, J .; Schmitz, B. (2010): page 55 ff.
  20. See Hertel, S .; Klug, J .; Schmitz, B. (2010).
  21. Shadish, WR; Cook, TD; Campbell, DT (2002).
  22. See Hertel, S .; Klug, J .; Schmitz, B. (2010): page 49.
  23. See Bortz, J .; Döring, N. (2006): page 56.
  24. See Sedlmeier, P .; Renkewitz, F. (2008): page 178.