The Difficulty of Effect Modification

Interaction terms in regression models can be very attractive, but the effects they measure can be complicated to explain. In the previous example, the relationship between the explainer variables age and lvef is straightforward in the absence of effect modification. As we have seen earlier in this chapter, older patients tend to have lower ejection fractions than younger patients. However, the inclusion of the product term suggests that this relationship between age and lvef is a function of...

Observational Studies

Commonly the relationship between an exposure and the disease cannot be elucidated by an experiment. For example it is not feasible randomly to allocate subjects to either use cell phones and others to avoid them in order to determine if there is a causal relationship between these devises and brain cancer.* In these circumstances, the research must be observational. In observational studies, the investigators do not control the exposure. Instead, they observe the exposure-disease relationship...

The Role of Discovery Versus Confirmation

Certainly, the production of hypoglycemia and seizure disorder was not part of the research protocol for Dr. Janbon and colleagues, and their unanticipated findings fell into the category of exploratory research. Janbon and colleagues would not be criticized for pursuing the surprise findings of their research efforts. The discovery that the antihypertensive agent minoxidil could unexpectedly reduce hair loss, and the finding that the antihypertensive, anti-anginal compound sidenafil can...

The US Carvedilol Program

In the late 1980s interest in the heart failure research community focused on the use of beta blockers. Considered anathema for the treatment of chronic CHF, reinvestigation of the issue suggested that beta blockade could be useful in relieving the symptoms of heart failure. The US Carvedilol program 11 evaluated the medication carvedilol (previously approved for the treatment of essential hypertension) for the treatment of CHF. In these research efforts 1,094 patients were selected for entry...

The Normal Distribution and Its Estimators

Bell Shape Curve Equation

Most of traditional hypothesis testing is based on the assertion that the variable of measurement is governed by the normal distribution. Although there are many probability distributions that have been discovered over the past 300 years, the one most frequently is the bell-shaped curve (Figure 3.1). Fig. 3.1. The bell-shaped curve. The formula for the curve is complicated, but probability is measured as area under the curve, and provided in tables (Appendix A). The shaded area is the...

The Rubric of Hypothesis Testing

There are of course important similarities between scientific and statistical hypothesis evaluations that allow statistical hypothesis testing to be very naturally and easily folded into the scientific method. Each procedure begins with a hypothesis, the veracity of which the investigator seeks to determine. In each case, the data must be the product of a well-designed and well-conducted research effort. Since the data are most commonly quantitative, both the scientific method and statistical...