Newest! Random block sizes
Newer! Specify initial subject ID number
New! How Randomization Plans Are Generated
Up to 20 treatments can be specified. The randomization plan is not affected by the order in which the treatments are entered or the particular boxes left blank if not all are needed. The program begins by sorting treatment names internally. The sorting is case sensitive, however, so the same capitalization should be used when recreating an earlier plan.
A randomization can be carried out as many blocks to insure against serious imbalance should the study be terminated prematurely. For example, instead of having 100 subjects randomized to one of two treatments in a single block, they might be randomized as 5 blocks of 20 subjects. This will insure that the number of subjects on each treatment will be equal every time the total number entered is a multiple of 20. See Fleiss (1986, sec. 3.1). The number of subjects per block must be a multiple of the number of treatments.
The seed for the random number generator (Wichmann and Hill, 1982, as modified by McLeod, 1985) is obtained from the clock of the local computer and is printed at the bottom of the randomization plan. If a seed is included in the request, it overrides the value obtained from the clock and can be used to reproduce or verify a particular plan.
Initial subject ID number: By default, the
randomization plan will number treatments starting with 1. The
starting value can be changed by placing the chosen value in the box
labeled Initial subject ID number. A starting value different from
1 might be entered when plans call for a particular set of consecutive
subject IDs to be used. The starting value would be the first ID number
of the set. It is also helpful when a study is extended and an additional
set of random assignments is needed. The additional assignments can now
be numbered starting where the original set left off.
Some Special Situations To randomize subjects with respect to several experimental factors,
use a set of labels composed of all possible combinations of the original
factor levels. For example, if subjects are to be randomly assigned to
one of two levels of dietary supplement (LOW/HIGH) and one of two types
of oil (PEANUT/RICE), use a set of 4 treatment labels: "Low/Peanut",
"High/Peanut", "Low/Rice", "High/Rice". To generate random permutations of treatments, use the random
permutation generator.
To generate a plan with unequal numbers of subjects on each treatment,
enter the labels in the worksheet in proportion to how they are to compose
the plan. For example, to have twice as many subjects receive treatment
A as treatment B, enter the label A twice and B once. Randomization.com uses the pseudo-random number generator of Wichmann
and Hill (1982) as modified by McLeod (1985). The generator uses three
seeds. The first seed is always 12345. The second seed is always 23456.
The third seed may be specified by the user, but is typically obtained
from the local computer's clock as
Each block is generated in turn. With randomly permuted blocks, subjects are assigned to treatment in blocks
to insure that equal number of subjects have been assigned to each treatment
each time the number of subjects is a multiple of the block size.
Some concern has been voiced that a study could be compromized, especially
an open label study, if the block size becomes known. Someone keeping careful
track would know the treatment that would be given to the remaining members of
a block once there is only one treatment left to be assigned. In order to
counter this, it has been suggested that more than one block size be used so
that no one could be sure of when a particular block of assignments would be
finished.
The generator now makes it possible to assign treatments with random block
sizes. The user merely specifies the different numbers of subjects per block
and the number of each type of block desired. The generator randomizes the
order in which blocks of various sizes appear in addition to randomizing
treatment within each block.
If everything has gone as planned, random block size has been implemented
in a way that gives the same results as the previous version of the generator
when there is only one block size. That is, this new version should be able to
recreate randomization plans generated by the previous version if the labels,
block size, and seed are provided. The previous
version of the generator is still available. Please let me know
immediately if ANY discrepancies are seen.
Notice! I could spend a HUGE amount of time
writing code to trap errors. I trap the most likely ones. However, someone
hell-bent and determined could probably come up with a set of conditions that
would cause the program to fail ignominiously. However, it should fail rather
than generate nonsense. If in practice a set of conditions causes the program
to fail, let me know and I'll write the commands to trap it and generate the
appropriate alert.
Notice! There may be small changes to the program to add
comments or trap errors. These WILL NOT be announced or otherwise documented.
Anything that affects the way calculations are performed WILL be
documented.
References Fleiss, JL (1986). The Design and Analysis of Clinical
Experiments. New York: John Wiley and Sons. [back to Randomization Plan generator]
To demonstrate that this method produces a random
permutation, notice that this is equivalent to selecting numbers out
of a hat without replacement. At the first step, a value is selected at
random and placed into slot kt. At the second step, a value is
selected from those that remain and is placed in slot kt-1, and so
on.
McLeod, A. Ian (1985), "Remark AS R58. A remark on algorithm AS 183.
An efficient and portable pseudo-random number generator," Applied
Statistics, 34, 198-200.
Wichmann BA and Hill ID (1982), "Algorithm AS 183. An efficient and
portable pseudo-random number generator," Applied Statistics, 31,
188-190.