Execute a systematic sample and a two-factor experiment

Try

A club has a numbered list of 96 members and wants 12 interviews. “Take every eighth member” does not yet tell another person how to begin. If someone always starts with their friend at position 1, the procedure has not used a random start. Write the possible start positions and decide how each will be given the same chance.

The worked systematic plan chooses one integer uniformly from 1 through 8, then selects that position and every eighth position afterward. With start 5, the selected IDs are 5, 13, 21, 29, 37, 45, 53, 61, 69, 77, 85 and 93. The first selection counts toward the 12. A physical draw from eight identical numbered slips, mixed well, is an executable random-start method; a trusted uniform integer generator is another.

Predict

Suppose the list alternates junior and senior members. Predict the composition of a sample with interval 8. Because adding an even number preserves odd/even position, every selected person comes from the same alternating category. Could a uniform random start still give each person the same inclusion probability? Yes: each position belongs to exactly one of the eight possible start sequences. Equal individual probabilities do not guarantee a balanced realized sample or make the sample a simple random sample.

Experiment

Carry out the worked plan once with start 2 and once with start 7. Both select 12 distinct people, but each follows one residue class. Only eight sample sets are possible. By contrast, a simple random sample of 12 gives every 12-person subset the same probability. For example, the subset containing positions 1 through 12 is impossible under the systematic plan.

Procedure	Random action	Resulting structure
Systematic	Uniform start 1 through 8	Every eighth position afterward
Simple random sample	Uniformly shuffle all 96 distinct IDs, take 12	Every 12-ID subset has equal probability
Stratified	Separate junior/senior frames, sample within each	Guarantees the prespecified group counts
Cluster	Randomly select existing groups, observe members in those groups	Whole groups enter together
Census	Obtain measurements for all target members	No subset is selected; missing responses still matter

Repair the periodic problem by randomly shuffling the whole list before the systematic procedure, or by using the SRS procedure. A random start alone does not break the original repeating pattern. With a uniform start and complete responses, the sample mean can remain unbiased over all starts even though individual samples are unrepresentative and variability is large. Calling every periodic alignment “bias” without specifying the estimator and randomization distribution obscures this distinction. A convenient nonrandom start introduces a separate selection problem.

Now shift from choosing people to assigning treatments. Thirty-two consenting adult volunteers compare two formats for a harmless puzzle-training activity. Factor A is presentation (paper or screen); factor B is example order (worked example first or practice first). There are four treatment combinations, not two. Predict what is lost if every paper participant gets examples first while every screen participant gets practice first.

Understand

Assign unique IDs, shuffle all 32 uniformly, and use the four blocks of shuffled positions shown above. This is random assignment, not evidence that the volunteers represent all adults. Eight people per treatment provide replication across participants; eight repeated measurements on one person would not replace it. Choose the outcome before starting, such as the number of correctly solved new puzzles after a fixed delay.

Keep puzzle content, total time, room conditions and test delay comparable. Participants can see whether they use paper or a screen, so promising double blinding would be misleading. Coded responses can keep the scorer unaware of treatment. A no-training comparison is useful if the question concerns training versus none, but it is not automatically necessary for the relative comparisons among these four training packages. A placebo is not a universal requirement for every experiment.

If prior puzzle experience is an important source of variability, form experience blocks first and randomize within each, allocating equal counts to the four combinations when block sizes permit. Do not assign all novices to one treatment and claim that statistical analysis later guarantees removal of confounding. Participation must be voluntary, with withdrawal without penalty, minimal personal data and a safe procedure.

Suppose illustrative cell means are 7 and 8 for paper under example-first and practice-first, and 11 and 9 for screen. The screen-minus-paper contrast is 4 with examples first but 1 with practice first. This difference of contrasts, 3, describes interaction. It does not by itself establish a statistically reliable interaction: sample sizes and individual variation are needed. Nor does a group mean show that every participant benefited.

Apply

Try a fresh allocation problem: 40 consenting volunteers test a memory activity with music absent/present and instructions spoken/written. Twenty report prior experience and twenty do not. Write all four combinations, an allocation algorithm that balances experience, a measurable outcome and a feasible blinding decision. Explain whether random assignment establishes an effect for every adult.

Compare an executable plan

Within each experience block, uniformly shuffle the 20 unique IDs and allocate consecutive groups of five to the four prespecified combinations. Each treatment contains ten people, five from each block. Use the same word list and a fixed delayed recall test; keep total time and instructions’ substantive content constant. A scorer of coded responses can be blinded, while participants hear the music and the spoken instructions. Assignment supports causal comparisons of these packages for participants under the protocol, with uncertainty. Volunteers and the particular activity limit population and setting generalization; the result does not prove an effect in every individual.

Continue

For each practice design, separate the selection procedure, treatment assignment and measurement procedure. State the random mechanism, available IDs, group sizes and handling of nonresponse or withdrawal. Then give population and causal scope as two separate sentences. The goal is an auditable plan, not the unexplained label “randomized study.”