What each cell measures
Each cell compares this vendor's synthetic responses against real human respondents in that demographic subgroup —
published survey ground truth, not a model's guess about the subgroup.
Higher = closer to how that real subgroup actually answered.
Demographic conditioning here is not stereotyping: every conditioned
score is checked against what real subgroup members said, not against
assumptions about them — and where the vendor's output diverges from
the real subgroup, the score drops.
- globalopinionqa — ground truth: Durmus et al. 2023, Anthropic — llm_global_opinions
- gss — ground truth: NORC at the University of Chicago — General Social Survey
- opinionsqa — ground truth: Santurkar et al., ICML 2023 — Whose Opinions Do LLMs Reflect? (derived from Pew American Trends Panel)
- subpop — ground truth: Suh et al., ACL 2025 — SubPOP: Subpopulation-Level Opinion Prediction
Low-n cells:
cells with n < 30 are shown muted and tagged low n — suggestive only instead of color-graded; they are
reported for transparency, not as findings.
CIs:
"no CI — single run" marks point estimates from a single run with no
confidence interval yet; treat the uncertainty as unknown, never zero.
Columns:
Score = distributional parity (p_dist) restricted to the subgroup ·
p_cond = conditioning strength vs the unconditioned baseline ·
N = questions answered under that conditioning · Cov. = coverage
dot derived from N (green = high, ≥100 · amber = medium, 50–99 ·
red = low, <50). Topic tables: SPS = Survey Parity Score for the
topic · p_dist = 1 − mean(JSD) · p_rank = (1 + mean(τ)) / 2 ·
p_refuse = 1 − mean(|R_model − R_human|).
Multiple comparisons:
with this many subgroup cells, a few extreme cells are expected by
chance alone — read patterns across a dimension, not single cells.
Question-type breakdown
(7 topics)
| Topic | SPS | p_dist | p_rank | p_refuse | N |
| Technology & Digital Life
low n — suggestive only
| 0.867 | 0.830 | 0.903 | 0.968 | 3 |
| Health & Science
low n — suggestive only
| 0.859 | 0.791 | 0.927 | 1.000 | 2 |
| Economy & Work
low n — suggestive only
| 0.826 | 0.813 | 0.839 | 0.967 | 5 |
| General Attitudes | 0.751 | 0.793 | 0.709 | 1.000 | 11 |
| International Relations & Security | 0.668 | 0.697 | 0.640 | 0.980 | 50 |
| Politics & Governance | 0.641 | 0.670 | 0.613 | 0.972 | 22 |
| Trust & Wellbeing
low n — suggestive only
| 0.334 | 0.320 | 0.347 | 0.980 | 7 |
Topics with N < 10 are muted and tagged
"low n — suggestive only": too few questions for a stable
3-decimal score.
Not yet measured
This vendor has no demographic-conditioned runs for age, geography,
education, or any other subgroup dimension on globalopinionqa.
No cells are fabricated — scores appear here only when a
conditioned run actually measured them.
How to submit a demographic-conditioned run →
Question-type breakdown
(7 topics)
| Topic | SPS | p_dist | p_rank | p_refuse | N |
| Health & Science
low n — suggestive only
| 0.883 | 0.859 | 0.908 | 0.967 | 2 |
| Social Values & Religion | 0.780 | 0.738 | 0.821 | 0.975 | 14 |
| Politics & Governance
low n — suggestive only
| 0.772 | 0.766 | 0.777 | 0.975 | 9 |
| Economy & Work
low n — suggestive only
| 0.690 | 0.662 | 0.718 | 0.981 | 4 |
| International Relations & Security | 0.653 | 0.697 | 0.608 | 0.974 | 33 |
| General Attitudes | 0.651 | 0.582 | 0.720 | 0.960 | 10 |
| Trust & Wellbeing
low n — suggestive only
| 0.643 | 0.678 | 0.607 | 0.951 | 3 |
Topics with N < 10 are muted and tagged
"low n — suggestive only": too few questions for a stable
3-decimal score.
Not yet measured
This vendor has no demographic-conditioned runs for age, geography,
education, or any other subgroup dimension on gss.
No cells are fabricated — scores appear here only when a
conditioned run actually measured them.
How to submit a demographic-conditioned run →