Research Design: Qualitative, Quantitative, and Mixed Methods - Essay Example

Clever Hans, among other obvious examples such as the Tuskegee syphilis study, Mengele's research, and Stanley Milgram's experiments on conformity and obedience, demonstrated that the claims we make and research we do can always be confounded by oversights, biases, and expectations, and can have unethical or disastrous implications when conducted incorrectly. The issues with designing research problems, methods, and studies can be put into a few categories: Choice of experiment design such as the usage of qualitative or quantitative data, the nature of the hypothesis in question, moral and ethical concerns, countering for bias and experimental error, and designing models that allow factors such as causation to be confidently discussed. Creswell argues that variables should be defined in three categories: Independent, dependent, and control (2009, 151). Independent variables are those that the researcher can control, dependent variables are ones that the researcher expects to be changed, and “dependent” on the independent variable. It's important to note that, when research examines correlations about data, the choice of the independent and dependent variable can become arbitrary: A causal model has to be established by more than just declaring which is independent and which is dependent. If I am saying that height and weight are correlated, I can have height be the independent variable that leads to changes in the dependent variable of weight or I can do it vice versa: The results and the correlations will be identical. Where the independent and dependent variable matters for experiments or for situations where causality can be argued due to a complex model. In the case of an experiment, the independent variable is what is being administered, and the dependent variable is what is being observed. In this case, it is impossible to swap them: Offering people medication and seeing if it cures their migraines with a placebo study is rather different from offering people migraines and seeing if it produces a cure. Control variables are, in any complex model, by far the most important. This is because the only way to control for spuriousness is to control for all possible confounding variables. There is undoubtedly a correlation between the number of homeless people and the number of doctors in a city, but neither cause the other: The city's size controls both. Experimental designs control by making sure groups are identical, whereas data analysis controls by making sure that data points are compared that are identical in every way aside from the difference being analyzed. A simple study that makes a correlation between education and eventual income has to control for, among other factors, race, socioeconomic status of parents, a wealth of social networks, perceived quality of the college, the actual quality of the college, gender, sexual orientation, and dozens of other factors to be meaningful. The design of the study must also be determined to be quantitative, qualitative, or mixed. Creswell (2009) defines qualitative procedures as “rely[ing] on text and image data” and argues that qualitative procedures cannot be easily enumerated. Note that asking someone, “How does this make you feel?” and writing down their response is qualitative while asking “How does this make you feel, from 1 to 9, one being worst and nine being best”, is quantitative even though both are asking about the same phenomenon because the latter can be mathematically charted. The choice of the hypothesis in question is quite important. “Does Drug X make people healthier” is meaningless: What is health, philosophically, and what elements are more important for health than others? “Does Drug X reduce the frequency of migraines” is a testable hypothesis that has no obvious philosophical problems that are insurmountable. Moral, ethical, and legal issues are incredibly complex and include privacy, lack of coercion, compliance with institutional ethics and standards, compliance with professional norms and ethics, and compliance with general cultural norms. In general, experiments should not deceive people or leave them much worse off for participating, and certain thresholds of harm are intolerable. My question then becomes: How can I construct causal models that are meaningful? In any complex real-world environment, dozens of factors are playing upon each other. How can I test and control for these models? And how can we ever really be sure that bias is countered? Two equally prestigious universities can publish two studies using similar methods that disagree diametrically on an issue, and it seems that oftentimes this is directly traceable to the partisan background or bias of the researchers Read More