untitled

(ff) #1

15 Combining Information


Meta-analysisis the integration of data from disparate sources. While this
can encompass a wide variety of phenomena, it is most commonly applied
to data obtained from sensors that are observing the same (or at least over-
lapping) environments. The sensors can be different sensors or they can be
the same sensor observing at different times. Scientific experimentation is
also a form of sensing. In this case one is observing natural phenomena.
Such observations are generally subject to uncertainty due to the lack of full
knowledge about the phenomena being observed as well as the limitations
of the measuring devices being used. One can reduce these uncertainties by
making a series of independent observations. Meta-analysis is the process of
combining the evidence afforded by the observations.
Meta-analysis goes by many names. The generic name iscombining infor-
mation(CI). In the social and behavioral sciences, it is also known asquan-
titative research synthesis. In medicine, it is often calledpooling of resultsor
creating an overview. Physicists refer to CI asviewing the results of research. For
chemists, CI is used for determining physical constants based on empirical
studies, and is calledcritical evaluation(NRC 1992). When applied to sensors,
CI is most commonly calleddata fusion. Sensor data fusion has the most elab-
orate forms of CI, and the field of multi-sensor data fusion has a standard for
data fusion, called theJDL Modelthat divides it into 4 levels (Steinberg et al.
1999).
Because of the uncertainty inherent in most forms of empirical knowledge,
it is normally stated as a probability distribution (PD) on a set of possi-
ble states. For example, when one speaks of a temperature measurement
of 30.5◦±0.4◦C, one is asserting that the measurement has a normal distri-
bution whose mean was 30.5◦C and whose standard error was 0.4◦C. Now
suppose that one performs a second, independent, measurement of the same
Free download pdf