SPURIOUS CORRELATION AND IT'S DETECTION
RANDOMIZATION
The reason statisticians design experiments and then
randomly execute them is to try and eliminate the effect of
concomitant or "lurking variables". If all the experiments
at the "low level" of a variable are performed before the
experiments at the "high level", there is a chance that a
background or unknown variable may be the real cause of the
significant difference between two means. The social
scientist unable to design and randomize and replicate must
be ever careful of the "lurking variables".
Lurking variables, like the time frame example in the
Demand For Gas study are identified when the augmented model
using Intervention Series causes the conditional impact of
the heretofore significant X variable to be non-significant.
Another example of exposing a Lurking Variable is the
Australian Wine Study. As soon as the model was augmented
with the ARIMA structure, a proxy for population,
the statistical significance of the Italian car registration vanishes.
CLICK HERE:Home Page For AUTOBOX