Planning statistical research

Most scientific work starts with a question about the world we live in. For
a statistician, these questions can be classified into a few different kinds.

  1. We can start with questions about a single Unit, like a biological
     organism, a manufactured product, a plot of ground, a city block or any
     of many different things we want to understand. In this type of study,
     we are often concerned with the dynamics of the Unit; how does it vary
     in time? Can we identify what aspects seem to control others? (See
     research subject.)
  2. Sometimes, our interest in a Unit may be purely instrumental, that is,
     we are interested in the Unit because of a population to which it
     belongs or an environment in which it resides. Our desire to describe a
     population may be satisfied by a statistical report on a sample from
     that population. The methodology is to select a sample and observe just
     the Units which are selected and then to summarize our results and
     interpret their meaning for the population. When we deliberately
     manipulate something in the process of observation, we call it an
     experiment. When we attempt to observe the Units without affecting
     them, we call it a survey.
  3. Sometimes, we want to understand the internal workings of a Unit so we
     look at its components and how they relate to each other. Physiology is
     an example of a discipline in which this kind of focus is common.

In much research, we use all three modes at different stages. A useful
synthesis of this kind of thinking was provided by Ludwig von Bertalanffy.

In every type of research, we must be concerned with managing the
observational error that is inherent in all empirical research. We can
increase the precision of our research by

  1. using a more precise instrument (eg., a more powerful telescope)
  2. increasing the number of observations so the constant we try to measure
     stands out better against the noise.
  3. changing the design of our research. This last approach can become very
     technical, so we will postpone its discussion until optimum
     experimental design.