Research Design Basics

Before thinking about statistics, it helps to understand the fundamental design choices in survey research. These choices determine what you can conclude from your data.

Between-subject vs. within-subject designs

How you expose respondents to conditions shapes what you can measure and how many twins you need.

Between-Subject

Different twins see different conditions. Each twin sees only one version, which avoids carry-over effects.

• Best for: concept tests, pricing experiments
• Needs larger samples (split across conditions)
• Cleaner causal inference

Within-Subject

The same twins see all conditions. Each twin evaluates every version, giving direct comparisons with fewer twins.

• Best for: preference ranking, A/B comparison
• Needs smaller samples
• Order effects possible — use randomisation

Rule of thumb

“Which is better?” = within-subject. “How does each land on its own?” = between-subject. When in doubt, within-subject with randomised order is the safer default on Elaiia.

Response bias in survey research

Understanding common biases helps you design better questions — whether for humans or AI twins.

Social Desirability

Respondents answer in ways they think are socially acceptable rather than truthful. Especially strong on sensitive topics (health, finance, politics).

Acquiescence Bias

The tendency to agree with statements regardless of content (“yea-saying”). Mitigate by mixing positively and negatively worded items.

Primacy & Recency

First and last answer options get chosen more often. Mitigate by randomising option order.

Variance in AI twin data

AI twins tend to cluster more tightly around the mean than humans — they reproduce less individual spread. The twins’ mean stays unbiased, but variance is typically lower. Delta Labs accounts for this by design: regular validation studies ensure that reduced variance doesn’t distort means, medians, or correlations in practice.