Sunday, March 15, 2009

Taxonomies of Experiment III: Silva, Bickle and Landreth

A third taxonomy of experiment can be derived from an article by Alcino Silva (UCLA) published in Journal of Physiology - Paris 101 (2007) 203–213 and work that Silva and I are doing along with John Bickle, who is at the University of Cincinnati. (Bickle and Silva have a related article that will soon be published in the Oxford Handbook of Philosophy and Neuroscience. Bickle is the editor of that volume.) This taxonomy is a work in progress.

The proposed taxonomy of experiment covers some of the same considerations that Craver and Sweatt considered. But it holds that there are 3 broad classes of experiment that are distinguished by their goals. The goals are: 1) description of phenomena, 2) assessment of causal relations among phenomena, and 3) development of tools to facilitate 1 and 2. Let's call experiments of class 1 Descriptive Experiments, those of class 2 Connective Experiments, and those of class 3 Validation Experiments.

Descriptive experiments focus on the dissection and description of phenomena without regard for the evaluation of causal hypotheses, per se. Causal considerations will of course affect the interpretations of one's measurements in these experiments, e.g. in the use of an imaging technique. But the goal of these experiments is not to assess the causal relations among the phenomena that constitute the subject matter. For example, one can dissect the hippocampus and describe its parts without testing hypotheses about the interactions of those parts.

Connective Experiments attempt to determine whether states of phenomena depend on each other. These assessments are made on the basis of manipulations (intervetions) and measurements of the phenomena of interest. There are 3 forms of connective experiment: 1) positive manipulations, which increase the value of an independent variable; 2) negative manipulations, which decrease the value of an independent variable; and 3) neutral measurements, which measure correlation between an independent and dependent variable under normal test conditions (roughly equivalent to Craver's activation experiments).

Validation Experiments validate the use of a tool, demonstrating that it is a reliable means of manipulating or measuring phenomena of interest. For example, the demonstration that knockout mice can be used to reveal the role a protein (e.g. CamKII) plays in both spatial learning and long-term potentiation validated the use of knockouts in the neuroscience of learning and memory. These experiments did not invent the knockout technique of course, but they did adapt a tool for use in neuroscience and led to a swarm of innovative transgenic approaches.

These forms of experiment are not entirely distinct. Validation experiments draw more attention when they simultaneously introduce a tool and reveal undiscovered phenomena or undiscovered causal dependencies. Descriptive experiments are often performed in such a way as to reveal causal information, e.g. that glutamate receptors can be found in pyramidal cells. The three different goals of experiment are mutually dependent, but any one of them can be performed with little regard for the others.