Building a Sample That Looks Like California
A score is only as trustworthy as the people behind the numbers, so the first job was to make sure our respondents resembled California's actual online casino players rather than whoever happened to be easiest to reach. We used stratified sampling, which sets quotas across meaningful groups, in our case region within the state, age band above 21, and level of playing experience, so that no single slice of the population could dominate the results. Every respondent was age-verified before their answers counted.
Random Sampling: The Backbone
The core of the sample was drawn randomly. In a random sample, every eligible person has a known and roughly equal chance of being selected, which is what allows the results to be generalized beyond the specific people who answered. Random selection is the single best defense against selection bias, the quiet distortion that creeps in when a survey mostly hears from enthusiasts, complainers, or a provider's own loyal customers. It is slower and more expensive than convenience polling, but it is the reason the headline figures hold up.
Snowball Sampling: Reaching Players Others Miss
Random methods struggle to reach one group in particular: active offshore casino players, who are a minority of the public and are not always eager to identify themselves in a cold survey. To reach them we added snowball sampling, where verified respondents refer other eligible players they know, and those referrals refer more, so the sample grows through real-world networks. This surfaces experienced players that a purely random approach would rarely find.
Snowball sampling has a known weakness: because it travels through social ties, it can over-represent tightly connected groups. We limited that effect by capping how far any single referral chain could extend and by blending the snowball responses with the random core rather than letting them stand alone. The two methods together give both breadth and depth.
How to Read the Sample
No survey recruited online is a perfect mirror of a population, and we do not claim ours is. What stratified, random, and snowball sampling buy together is a sample broad enough to trust and deep enough to be interesting. The next chapter shows how we tested the data for significance so that differences between casinos reflect real gaps rather than noise, and you can see how it all rolls up on the CAC Score methodology page.
Testing for Significance: T-Tests, Variance and Confidence Behind the CAC Score
An average can lie. We ran the California survey data through SPSS with t-tests, variance analysis and confidence intervals so that every number in the score earns its place.
By Derek LoomisMay 28, 2026
From Survey to Score: Weighting the Eight CAC Pillars Out of 100
Survey data on one side, hands-on testing on the other. Here is exactly how the two become eight weighted pillars and one number in a green poker chip.
By Ryan CarriganJune 4, 2026
The CAC team weighs in
The analysts behind the study talk through what this chapter means for a California player.
Four thousand two hundred and seventeen verified California players — roughly one percent of the active audience. That's a genuine sample, not a vibe.
And 21-plus is verified by government ID. We don't accept self-declared age; the population definition and the ethics both demand it.
Regional quotas matched the real player distribution — Southern California 58%, the Bay Area 20%, and so on — so no region is over-represented.
The piece people miss is funded-account screening. A casino only earns a survey score from respondents who actually played there, minimum 100 verified players.
That screening is the reason a per-casino number like Ignition's 98 actually means something — it's not strangers guessing, it's funded players.
Keep reading
Our full 67-page methodology and dataset: a stratified survey of 4,217 verified California players aged 21+, the eight-pillar weighting model, complete data tables and statistical analysis behind every score on this site.
Download the full study (PDF, 67pp) →