Why Enumerate a Permuted Block?
A permuted block is built by fixing how many assignments to each treatment will occur in a block, then randomly ordering those assignments. The possible orders are the block’s valid treatment sequences. Enumerating them makes the randomization rule explicit: it shows which sequences can occur and, under uniform selection, the probability assigned to each one.
This is different from assigning each participant independently with fixed treatment probabilities. Within a block, the number assigned to each treatment is specified in advance; randomization determines their order. For example, a block with two assignments to A and two to B can have six distinct orders, not \(2^4\) arbitrary strings.
The treatment counts must be specified before enumerating. For example, “A:B allocation ratio 2:1” does not by itself say whether a particular block contains three assignments, six assignments, or another size. Once the block’s counts are given, however, the enumeration and sequence probabilities follow directly. The calculations below concern the randomization of a single block; they do not select block sizes.
Counting Distinct Sequences
If all \(N\) assignments had different labels, there would be \(N!\) possible orders. In a block, assignments to the same treatment are indistinguishable in the resulting sequence. Permuting the \(n_j\) copies of treatment \(j\) among themselves does not create a new sequence. The number of distinct valid sequences is therefore
When the \(M\) distinct sequences are selected uniformly, each sequence has probability \(1/M\). This is a probability for a complete ordered sequence, not necessarily the probability of a particular treatment at every position. Position-specific probabilities can differ according to the counts remaining as the sequence is generated.
Worked Example: Equal Allocation in a Four-Assignment Block
Consider a hypothetical block with two assignments to A and two to B. Thus \(N=4\), \(n_A=2\), and \(n_B=2\). The number of distinct sequences is
Listing the six sequences gives AABB, ABAB, ABBA, BAAB, BABA, and BBAA. Each contains exactly two A assignments and two B assignments, and there are no duplicates.
Under uniform selection, each sequence has probability \(1/6\), or about 0.1667. As a check, the six probabilities sum to \(6(1/6)=1\). The probability that the first assignment is A is different from the probability of one particular sequence: three of the six sequences begin with A, so that position-specific probability is \(3/6=1/2\).
Unequal Allocation and Repeated Labels
Unequal allocation changes the treatment counts in the block; it does not change the counting principle. Because repeated copies of a treatment label do not distinguish sequences, the factorial denominator must include the factorial for every treatment count. Omitting that adjustment overcounts possible sequences.
Worked Example: A Two-to-One Allocation Ratio
Consider a hypothetical block containing two assignments to A and one to B. The valid sequences are AAB, ABA, and BAA. The count formula confirms that there are three:
If the three distinct sequences are selected uniformly, each has probability \(1/3\). The first assignment is A in two sequences and B in one, so the first-position probabilities are \(P(\text{A first})=2/3\) and \(P(\text{B first})=1/3\). These probabilities reflect the fixed counts in the block, not an independent draw with a constant probability at every position.
Worked Example: Enumerating All Sequences With Counts A, A, B, and C
Suppose a hypothetical four-assignment block contains two assignments to A, one to B, and one to C. Here \(N=4\), and the number of distinct valid sequences is
The twelve sequences are AABC, AACB, ABAC, ABCA, ACAB, ACBA, BAAC, BACA, BCAA, CAAB, CABA, and CBAA. Each has two A assignments, one B assignment, and one C assignment; all twelve are distinct. Under uniform selection, each sequence has probability \(1/12\), about 0.08333.
The first-position probabilities provide another check. Six sequences begin with A, three with B, and three with C. Therefore the first assignment probabilities are \(6/12=1/2\) for A and \(3/12=1/4\) each for B and C. They sum to one, and they agree with the proportions of each treatment among the four assignment slots.
Why Uniform Enumeration Gives Equal Sequence Probabilities
A common way to construct a sequence is to place all assignment slots in a container and shuffle them uniformly. To see why this gives a uniform distribution over distinct sequences, imagine temporarily labeling the copies of each treatment. For example, in the A, A, B, C block, label the A slots A₁ and A₂. There are \(4!\) equally likely orders of the four labeled slots. Each distinct treatment sequence corresponds to exactly \(2!\) labeled orders, because A₁ and A₂ can exchange positions without changing the treatment sequence. Thus each treatment sequence has probability \(2!/4!=1/12\).
The same reasoning applies to any counts: each distinct treatment sequence corresponds to \(\prod_j n_j!\) labeled-slot orders. Its probability after a uniform shuffle is therefore
An equivalent procedure is to fill the sequence one position at a time, choosing among the remaining assignment slots uniformly. For a particular valid sequence, the probability is the product of the successive probabilities of its specified labels. If treatment \(j\) appears \(n_j\) times, the factors for that treatment in the numerator multiply to \(n_j!\); the denominators multiply to \(N!\). Hence every valid sequence has probability \(\prod_j n_j!/N!\), the same value as under uniform selection from the enumerated list.
Worked Example: Checking a Sequence by Sequential Probabilities
Use the block with two A assignments, two B assignments, and one C assignment. Its sequence count is
For the particular valid sequence ABABC, the successive probabilities when drawing uniformly from the remaining slots are \(2/5\), \(2/4\), \(1/3\), \(1/2\), and \(1/1\). Therefore,
The product matches the probability obtained by counting all 30 distinct sequences. This calculation also shows why the probability changes from one position to the next: after each assignment, the available treatment counts change.
A Practical Enumeration and Verification Process
For a small block, a hand list is a useful check. For a larger block, software can generate the multiset permutations, but the same checks apply. A reliable construction follows these steps.
Write the number of assignments required for each treatment and verify that the counts sum to the block size.
Use \(M=N!/\prod_j n_j!\). This gives the number that the enumeration must contain.
Permute the treatment labels while treating copies of the same label as indistinguishable. Do not count exchanges of identical copies as new sequences.
Confirm every sequence has the required treatment counts, that no sequence is repeated, and that the number listed is \(M\). If selection is uniform, assign probability \(1/M\) to each and check that all probabilities sum to one.
If the list has fewer than \(M\) entries, a sequence has been omitted or duplicates have been mistaken for one another. If it has more than \(M\), it likely contains a repeated sequence or counts identical treatment copies as distinct. A second useful check is to count how many sequences begin with each treatment. When \(n_j>0\), that number is \((N-1)!/[(n_j-1)!\prod_{\ell\ne j}n_\ell!]\); dividing by \(M\) gives \(n_j/N\), the probability that treatment \(j\) is first under uniform sequence selection.
How This Differs From Other Randomization Rules
| Randomization rule | What is selected at random? | Implication for a block |
|---|---|---|
| Uniform permuted-block randomization | One distinct ordering of the block’s specified treatment counts | Only sequences with those counts are valid; each valid sequence has probability \(1/M\). |
| Complete randomization | A treatment assignment from the specified set of trial assignments | Unless block counts are fixed as part of the design, the treatment totals within a given group of participants need not equal the permuted-block counts. |
| Sequential selection from remaining slots | One remaining assignment slot at each position | When each remaining slot is equally likely to be selected, it produces the same sequence probabilities as uniform selection from the distinct valid sequences. |
The comparison with complete randomization is about the assignment rule, not the clinical merits of either design. The earlier material on complete randomization describes uniform sampling from a set of assignments. In a permuted block, the set is instead restricted to the permutations of the block’s specified treatment counts. A sequential draw from remaining slots is not a different probability model when implemented as described above; it is an alternative way to generate the same distribution.
Common Mistakes and Limitations
- Counting labeled copies as different sequences. With two A assignments, exchanging the first A and the second A does not change the treatment sequence. Divide \(N!\) by the factorials of the repeated counts.
- Assuming every position has the same treatment probability. In an AAB block, the probability of A at the first position is \(2/3\), but after an A has occurred, the remaining counts have changed. The probability must be based on the assignments still available.
- Confusing sequence probability with a position probability. In a four-assignment AABB block, one sequence has probability \(1/6\), whereas the first assignment is A with probability \(1/2\).
- Assuming that the formula specifies the randomization procedure by itself. The formula counts valid sequences. To conclude that each has probability \(1/M\), the procedure must select uniformly from those distinct sequences (or use an equivalent uniform shuffle of assignment slots).
Enumeration is transparent and particularly useful for small blocks, but listing every sequence becomes cumbersome as the block grows. The count formula still gives the number of possibilities, while software can generate or sample valid sequences. In either case, implementation should be checked against the intended treatment counts and probability rule; an enumeration that is correct on paper does not by itself verify that a randomization system uses it correctly.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions.
- A block contains three A assignments and one B assignment. How many distinct sequences are valid, and what is the probability of each under uniform selection?
- A block contains one A, one B, and one C assignment. List all valid sequences and give the probability of each.
- For a block with two A assignments, one B, and one C, how many sequences begin with B? What is the probability that the first assignment is B?
- In an AABB block, why is the probability of one specified sequence \(1/6\), while the probability of A in the first position is \(1/2\)?
- A proposed enumeration for an AABC block lists 12 entries but includes AABC twice. Is the list valid just because its length equals the formula’s count? Explain.
Solutions: (1) The count is \(4!/(3!1!)=4\). The sequences are AAAB, AABA, ABAA, and BAAA, so each has probability \(1/4\). (2) There are \(3!/(1!1!1!)=6\) sequences: ABC, ACB, BAC, BCA, CAB, and CBA. Each has probability \(1/6\). (3) With B first, the remaining labels are A, A, and C, which can be arranged in \(3!/2!=3\) distinct ways. Thus \(P(\text{B first})=3/12=1/4\). (4) Uniform selection gives each of the six distinct sequences probability \(1/6\). Three sequences begin with A, so \(P(\text{A first})=3(1/6)=1/2\). (5) No. The count of 12 is necessary but not sufficient: a duplicated entry means at least one other distinct valid sequence is missing. The list must contain 12 unique sequences, each with two A assignments, one B, and one C.