Randomize sections (A/B variants)

Randomization is how you compare options. You set up two or more Variant options under one Randomization block (shown as Variant comparison in the editor), and Strella sends each participant down one variant, or a subset, or all of them, depending on how you configure it.

When this is useful

• A/B concept tests. Show half your participants design A and the other half design B, then compare reactions.
• Order effects. Show three concepts in different orders to different people so order (and recency bias) doesn't skew the result.
• Stimulus rotation. Test which of five tagline options resonates without showing every option to every participant.

Steps

1. Add a Randomization block

   In Build, click + Add element → Randomization (under the Structure group). It can live at the top level of the guide or inside a Section.

2. Add Variant options

   Each variant is a small group of steps. Add as many as you want to compare. Drag steps from elsewhere in your guide into a variant, or add new ones inside.

   A Randomization block needs at least two Variant options.

   

3. Write each variant

   Put the variant-specific content in each option. The shared lead-in and follow-up stay outside the Randomization block.

   Example layout:

   • Statement: "I want to show you a few concepts."
   • Randomization block:
     • Variant A: Image of design A + Open question
     • Variant B: Image of design B + Open question
   • Open question: "Anything else you want to add?"
4. Optionally show only a subset of variants per participant

   If you have more variants than you want each participant to see, set the subset size. With five variants and a subset of two, each participant sees two of the five (still picked at random).

By default each participant sees one variant per Randomization block, though you can show a subset or all of them (see the step above). In Results, you can filter by variant to compare reactions across them.

How assignment works

• Random. Variant choice is automatic per participant. You can't hand-pick who sees which variant. If specific participants need specific content, use conditional sections keyed to screener answers or quotas instead.
• Sticky. If a participant leaves and comes back, they get the same variant they started with. Their place in the interview and their conditional-section eligibility stay the same too.
• Approximately balanced. Assignment is random per participant, not strictly balanced. You'll get close-to-equal counts at scale, but with small samples one variant can run ahead.

Tips

• Branches don't have to be the same length, but keeping them roughly equal helps. A 5-step branch next to a 2-step one means some participants get a much longer interview than others, which can frustrate the long-branch group.
• For more than 4 variants, use a subset: exposing every participant to all of them quickly hits diminishing returns and stretches the interview.

Related