Discussion Section

The purpose of a discussion section of a research paper is to relate the results (results section) back to the initial hypotheses of the study (introduction section). The discussion section provides an interpretation of the results, presents conclusions, and supports all the conclusions with evidence from the study and generally accepted knowledge. The discussion section should describe (a) what new knowledge has been gained in the study and (b) where research should go next.

Discussion of research findings must be based on basic research concepts. When the researcher explains a phenomenon, he or she must explain mechanisms. If the researcher's results agree with the expectations, the researcher should describe the theory that the evidence supported. If the researcher's results differ from the expectations, then he or she should explain why that might have happened. If the researcher cannot make a decision with confidence, he or she should explain why that was the case, and how the study might be modified in the future. Because one study will not answer an overall question, the researcher, keeping the big picture in mind, addresses where research goes next.

Ideally, the scientific method discovers cause-and-effect relationships between variables (conceptual objects whose value may vary). An independent (exposure) variable is one that, when changed, causes a change in another variable, the dependent (outcome) variable. However, a change in a dependent variable may be due wholly or in part to a change in a third, confounding (extraneous) variable. A confounding variable is anything other than the independent variable of interest that may affect the dependent variable. It must be predictive of the outcome variable independent of its association with the exposure variable of interest, but it cannot be an intermediate in the causal chain of association between exposure and outcome. Confounding variables can be dealt with through the choice of study design and/or data analysis.

Bias is the systematic deviation of results or inferences from the truth, or processes leading to such deviation. Validity refers to the lack of bias, or the credibility of study results and the degree to [Page 375]which the results can be applied to the general population of interest. Internal validity refers to the degree to which conclusions drawn from a study correctly describe what actually transpired during the study. External validity refers to whether and to what extent the results of a study can be generalized to a larger population (the target population of the study from which the sample was drawn, and other populations across time and space).

Threats to validity include selection bias (which occurs in the design stage of a study), information bias (which occurs in the data collection stage of a study), and confounding bias (which occurs in the data analysis stage of a study). Selection bias occurs when during the selection step of the study, the participants in the groups to be compared are not comparable because they differ in extraneous variables other than the independent variable under study. In this case, it would be difficult for the researcher to determine whether the discrepancy in the groups is due to the independent variable or to the other variables. Selection bias affects internal validity. Selection bias also occurs when the characteristics of subjects selected for a study are systematically different from those of the target population. This bias affects external validity. Selection bias may be reduced when group assignment is randomized (in experiments) or selection processes are controlled for (in observational studies). Information bias occurs when the estimated effect is distorted either by an error in measurement or by misclassifying the participant for independent (exposure) and/or dependent (outcome) variables. In experiments, information bias may be reduced by improving the accuracy of measuring instruments and by training technicians. In observational studies, information bias may be reduced by pretesting questionnaires and training interviewers. Confounding bias occurs when statistical controlling techniques (stratification or mathematical modeling) are not used to adjust for the effects of confounding variables. Therefore, a distorted estimate of the exposure effect results because the exposure effect is mixed with the effects of extraneous variables. Confounding bias may be reduced by performing a “dual” analysis (with and without adjusting for extraneous variables). Although adjusting for confounders ensures unbiasedness, unnecessary adjustment for non-confounding variables always reduces the statistical power of a study. Therefore, if both results in a dual analysis are similar, then the unadjusted result is unbiased and should be reported based on power considerations. If both results are different, then the adjusted one should be reported based on validity considerations.

...