TEST TWO: The Russell Procedure
The "Russell Procedure" (due to Dr. John Russell ), has two parts.
PART I.
Part I calculates the expected number of matches between a remote-viewing session and a target based simply on chance. This binomial
mean is found by dividing the total number of attributes for
a given target by the total possible number of attributes (93),
and then multiplying this ratio by the total number of SAM entries
for the corresponding remote-viewing session. A standard deviation
is then calculated based on the appropriate hypergeometric distribution
(see William Feller. 1968. An Introduction to Probability
Theory and Its Applications, 3rd edition. New York: John
Wiley & Sons, pp. 232-3).
Three confidence intervals are then calculated that determine if the actual number of session/target matches is different from chance. An actual match total that is outside of a given confidence interval is different from chance, which leads to the rejection of the null hypothesis.
Following this, a weighted number of matches between the session and the target is calculated. This weighted number is an alternative way of looking at this problem. Rather than simply count the number matches between a session and a target, weights are constructed for each SAM entry for the remote-viewing session based on how rare each entry occurs in general. To calculate the weights, a large pool of 240 very diverse SAM targets is used. The formula for deriving the weights is derived as follows:
Let,
Ci = the total number of times a given attribute (i) occurs in
a pool of targets
Q = the total number of targets in the pool
Thus, the probability of any attribute chosen in a remote-viewing session being represented in the pool is Ci/Q.
Since we want a weight that is large when an attribute is relatively rare in the pool, and small otherwise, we use the reciprocal of Ci/Q, times a constant of proportionality (for scaling) for the weight. Thus, our weight is,
Wi = weight for attribute i = kQ/Ci = V/Ci, where kQ=V (a constant), and k is our constant of proportionality.
We now need to determine V, which we can do by solving for it in one particular instance (since it is always a constant). We know that under conditions that all Ci equal the mean of C, then the weight for attribute i is simply V divided by the mean of C, which equals 1 by definition since all weights must be equal to 1 under such conditions. Thus, V equals the mean of C, which will be true for all distributions of Ci (again, since V is a constant). This means that our desired weight, Wi, is the mean of C divided by Ci.
The weighted mean (called the "Russell Mean") is then the summation of all of the weighted SAM entries for a given remote-viewing session. The Russell Mean is then evaluated with respect to the same confidence intervals as with the unweighted mean to determine the significance of the session's SAM entries. This test is quite rigorous (perhaps excessively so), and it evaluates a remote-viewing session based on SAM entries that are relatively rare, and thus more or less unique to a given target.
PART II.
Part II of the Russell Procedure evaluates the remote-viewing session from the perspective of how many random SAM entries would
be needed to describe the target as completely (as per the number
of session/target matches) as is done by the actual session.
To conduct this test, the SAM Program constructs pseudo sessions
composed of random SAM entries, with each entry being added one
at a time until the total number of matches with the actual target
equals that achieved by the actual remote-viewing session. The
mean and standard deviation for the total number of SAM entries
for each pseudo session are computed from a set of 1000 Monte
Carlo samples. Confidence intervals are again constructed, and
this test evaluates the efficiency of the remote viewer (as per
proportion B used in Test
One) in describing the target. When the total number of actual
session SAM entries is outside of (that is, less than) an appropriate
confidence interval, then the remote viewer's perceptive efficiency
is outside of chance, and the null hypothesis is rejected.
