Experimental Discovery, Data Models, and Mechanisms in Biology: An Example from Mendel’s Work - Mechanism and Causality in Biology and Economics - page 110

obtained by de-idealizing principled models and be used to explain phenomena
revealed by data models.
By and large, a data model represents the structure of data from observations,
measurements, or experiments. Different philosophers have different character-
izations of data models. 11 There is no need to develop a general conception of
data model here; I intend to give only an account of models of experimental data.
A model of experimental data can be produced, constructed, and extracted from a
specific experimental design, arrangement, and results (i.e., the whole process),
with some conditions for control. 12 Because the experimental process can be
represented by a symbolic or conceptual diagram (for instance, the diagrams
shown as Figs. 6.1 , 6.2 , 6.3a, b ), the diagram in turn can be seen as a data model.
One can also call those diagrams diagrammatical models, each of which represents
some specific pattern, structure, or regularity. The diagrammatical model in Fig. 6.1
represents the pattern of effects produced by injecting two different strains (R and S)
of bacteria into mice. The diagrammatical model in Fig. 6.3a represents the pattern of
effects produced by crossing certain forms of plants possessing pairs of contrastive
traits. Therefore, a data model can reveal a significant phenomenon (in Bogen and
Woodward's sense; see next section).
Why were the data models of the formation of hybrids in effect a key to the
research about heredity? What allowed the models and generalizations to be
reinterpreted into the so-called fundamental laws of heredity by later geneticists?
From the view of genetics, G1, G2, and G3 can be interpreted as generalizations of
trait transmission, a view that presupposes that there are bearers of traits that can
move from parents to offspring. In addition, these generalizations also show that
dominating traits appear in the offspring of every generation, whereas recessive
traits hide in the second generation and reappear in the third generation. One can
infer from the phenomenon that the bearers of these traits would not be blended, and
thus one can derive the law of segregation (of the trait bearers). G3 and G4 describe
the occurrence or nonoccurrence of paired traits and the fixed ratios in the number
of offspring. From this one can infer the “law” of dominance—that is, the existence
of dominating and recessive bearers—though this is not a general case for most
traits. G5 corresponds to the law of independent assortment; it is an empirical
expression of the independent distribution of trait bearers. According to such a line
of thinking, we should note that “Mendelian” laws of heredity, two or three, are
really theoretical rather than empirical, for they imply the notion of trait bearer ,

11

Recently, there have been several waves of debate over the relation between data and phenom-
ena, stemming from Jim Bogen and James Woodward's 1988 paper (Bogen and Woodward 1988 ).
For review articles, see Harris ( 2003 ), Bogen ( 2010 ), Woodward ( 2010 ), McAllister ( 2010 ), Teller
( 2010 ), and Brading ( 2010 ).
12 Mendel set up three control conditions for his experimentation with plants: (1) The experimental
plants must necessarily possess constant differing traits, (2) their hybrids must be protected from
the influence of all foreign pollen during the flowering period, and (3) there should be no marked
disturbances in the fertility of the hybrids and their offspring in successive generations (Mendel
1966 [1866], p. 3).