|
|
|
|
|
Quasi-Experimental Research ANOVA (Analysis of Variance) One-Way Analysis of Variance Two-Way Analysis of Variance Key Terms to Understand
Module Objectives: At the end of this module the student should be able to:
*This course is based on the assumption that students already have had the appropriate statistics and research methods courses from their undergraduate/graduate studies. However, it may have been a few years since some of the students have had an opportunity to review some of the material covered in this course. If you need to refresh your memory concerning statistics/biostatistics please review some of the sources presented on the Background page for this module as well as Module One Background. It is recommended that you bookmark those reference sources you find as valuable. Also be sure to be very conversant with all Key Terms presented in each Module of the course.
This course is intended to discuss the various research designs used in quantitative research for the health sciences. We began with descriptive research designs before examining some of the issues regarding correlational research. Now we will examine quasi-experimental and pre-experimental research designs. Pre-Experimental Research Designs Pre-experimental research designs are those designs for studies that the researcher cannot adequately control or chooses not to control the selection of the study's subjects. Therefore, generalization of the study's findings are limited only to those subjects tested or exposed to the study's treatment. Research designs generally considered to be classified as pre-experimental are the One-Group Post-test-Only Design, the Post-test-Only Design with Nonequivalent Groups, and the One-Group Pretest-Post-test Design. All of these designs have serious weaknesses and numerous threats to internal and external validity. Can you identify several serious weaknesses with these designs? The question that is often asked by students is "Why are these designs used when they have such serious weaknesses associated with them that the findings often cannot be interpreted with any confidence? The answer is that researchers may have to conduct a study without adequate controls because of the circumstances surrounding the study environment. In this case limitations involving the study environment could mean that there are concerns with the treatment, the experimental group, and problems measuring norms associated with the dependent variable(s). See the characteristics of Pre-Experimental Designs Quasi-Experimental Research Designs Quasi-experimental research designs represent an improvement over the pre-experimental designs due to their greater control of threats to internal validity. Quasi-experimental designs attempt to examine the causal relationships between the independent and dependent variables but encounter difficulties doing so because of the threats to validity which are inherent in these designs. Quasi-experimental designs are generally classified as either Nonequivalent Control Group Designs or Interrupted Time-Series Designs. Some of the research designs by classification are described in the table found on this link: Quasi-experimental Designs Experimental Research Designs Experimental research designs differ from other research designs discussed in the earlier modules since they are designed to provide the researcher with the greatest amount of control in order to study causality. True experimental designs eliminate all factors that may affect the dependent variable except the cause or the independent variable. In most studies in health sciences, those factors which cannot be completely eliminated are controlled. Besides controlling or eliminating factors that may effect the outcome of the study, experimental designs have two other goals: A) to eliminate bias; and, B) to ensure the study has adequate precision or rigor. The problem of bias, especially researcher bias, is usually controlled, in part, through randomization. However, establishing proper experimental control (use of a control group) is the most confident way of eliminating a source of bias. Understanding the "concept of control" is critical if the researcher is to control extraneous variables in the study. A "control" group allows all extraneous variables (factors) to equally effect a study's experimental and control group. Since both groups are equally effected by the extraneous variables, any differences between the dependent variables in the study can be entirely attributed to the effect of the independent variable. In other words, we have held the extraneous variables in the study "constant". Control groups and randomization alone cannot control for all study factors. Often, a study participant's expectations or participant bias can effect a study's outcome. In such studies the use of "blindness" can control for participant bias and even researcher bias. Use of a "single-blind or double blind" technique prevents the participant or the observer from knowing which participant was exposed to the independent variable. Blinding protects a study from the effects of confounding. Knowledge of the content area to be studied by the researcher is also important in providing adequate experimental controls. This is one aspect of research that cannot be overlooked since often it is the knowledge and skill of the researcher that allows the identification and control of unique extraneous variables. Some of the more commonly used experimental research designs are briefly described in the following Table: Description of Experimental Research Designs Controlling Internal Validity Threats There are many other designs which could be presented in the above table; however, it would be pointless unless you understand this very critical point -- As the researcher you control the research study and must adjust and fine tune YOUR design to account for the validity issues that interfere with the ability of the research design to accurately answer the research question. The threats to the validity of a study which must be controlled by the researcher are internal validity, external validity, statistical conclusion validity, and construct validity. Internal validity threats include:
External validity refers to the ability or capacity of the study's finding to generalize from the study sample to the study population. Statistical validity refers to the power to draw sound statistical conclusions from the study's findings. Construct validity refers to the compatibility of the constructs which are the focus of the study to the constructs once they are operationalized. For instance, if the study is to measure program effectiveness then the operationalized construct for measuring program effectiveness in the study must actually measure program effectiveness. If it does not actually measure what it is intended to measure then we have a problem with construct validity. Read the relevant documents for this module in the Background Information page before proceeding to the Case Study. |
|
|
