What Our Own Test Data Shows About Online IQ Test-Takers

652 Completions. Here Is What the Data Says.

We queried our own database. This article is the result.

Most IQ content online cites the same handful of external studies. The same Mensa threshold. The same Flynn Effect graph. Almost nobody publishes data from their own users. We are in the position of having it, so we are going to use it.

One caveat first. This is not a random population sample. People who seek out an IQ test online are self-selected. Curiosity about your own cognitive performance correlates modestly but consistently with above-average intelligence. That bias shapes everything that follows, and we will come back to it.

The Headline Numbers

652 completions, filtered to remove incomplete and invalid submissions. Mean IQ: 103.1. Median: 107. Standard deviation: 15.6. Lowest score in the dataset: 64. Highest: 139.

The standard deviation is worth pausing on. The theoretical population SD for IQ is 15. Ours is 15.6. In our previous analysis at 622 completions it was 16.1. It has tightened as the dataset has grown, which is exactly what you would expect from a calibrated scoring model. We take it as a good sign.

The mean-median gap tells the more interesting story. When the median (107) sits above the mean (103.1), the distribution is left-skewed. A tail of lower scores pulls the average down while most of the data sits higher. The band breakdown shows exactly where.

Where Scores Actually Land

Full band breakdown across 652 completions:

• Under 70: 0.9%
• 70-79: 8.9%
• 80-89: 13.5%
• 90-99: 14.3%
• 100-109: 17.2%
• 110-119: 33.0%
• 120-129: 10.9%
• 130+: 1.4%

The 110-119 band contains 33.0% of all results. In a true normal distribution centred at 100 with SD 15, you would expect roughly 15-16% in that range. We see double. This is the clearest statistical signature of self-selection.

It is not a calibration error. People who find the site, sit through 36 timed questions, and reach the results page are not a random cross-section of the population. They skew above average. The data is consistent with this and consistent with what academic literature reports about online cognitive testing populations. For context on what each band means in practice, our guide to IQ score ranges breaks it down in detail.

The Finding That Will Not Move

The IQScore assessment covers four domains: logical reasoning, pattern recognition, spatial reasoning, and abstract reasoning. We looked at average performance as a percentage of maximum points available in each domain.

Three domains cluster together:

• Pattern recognition: 28.6%
• Abstract reasoning: 28.6%
• Spatial reasoning: 27.3%

Then there is logical reasoning. At 11.8%.

Less than half the rate of every other domain. We first flagged this at 622 completions. The numbers looked unusual enough that we ran the query three times. Now we have 652 completions and every decimal point is identical. The finding is not noise. It is stable.

Two explanations are worth considering, and we genuinely do not know which accounts for more of the gap.

The first is question structure. Logical reasoning tasks here are formal and deductive: syllogisms, conditional logic chains, argument validity. These require suppressing an intuitive response and applying a strict rule instead. Research on dual-process theories of reasoning consistently shows that the intuitive system generates plausible-but-wrong answers to formal logic problems. Most people trust that intuition. It costs them points.

The second explanation is familiarity. Pattern recognition, abstract, and spatial tasks resemble puzzle formats people encounter through games and everyday problem-solving. Formal logical reasoning is less commonly practiced outside academic or professional contexts. Someone who has worked through deductive proofs carries a structural advantage that has nothing to do with raw intelligence.

We checked whether the gap narrowed among higher scorers. It does not. People in the 120-129 band still underperform on logical reasoning relative to their scores elsewhere. That rules out the simple explanation that lower overall scores are dragging the domain average down. Something about formal logic is harder for this population at every level. Our article on pattern recognition in IQ tests explains why some reasoning formats feel more natural than others.

If your IQScore result was lower than you expected, logical reasoning is the most likely culprit. It is also the domain most responsive to targeted practice, precisely because the gap seems to reflect familiarity as much as raw ability.

What This Data Does Not Tell Us

Three limitations, stated plainly.

This is not a population study. 652 self-selected online test-takers are not representative of any national or global population. Conclusions about how intelligence distributes across humanity belong in large, stratified academic research, not here.

Score variance is real. A standard deviation of 15.6 means the same person could reasonably score several points differently on a retest. One result is informative. It is not a fixed ceiling. Our article on whether IQ can change covers the evidence on score stability and what actually shifts it.

Online tests estimate cognitive ability. They do not diagnose it. The WAIS-V, administered by a licensed psychologist, is the appropriate tool for clinical or educational assessment. IQScore is a calibrated screening instrument, more rigorous than most online tests and less rigorous than a full clinical battery.

What We Are Looking at Next

The logical reasoning gap is the finding we want to dig into further. The next question is whether it narrows among users who complete the test more than once, and whether it varies between score bands at the extremes. We will revisit the full distribution once the dataset passes 1,000 completions. By that point, organic search traffic will have diversified the user base enough to check whether the self-selection effect has shifted.

The data is from real people. That makes it worth publishing honestly, including the parts we cannot yet fully explain.

If you have not yet taken the assessment, the full test takes under 25 minutes and results are available immediately with no sign-up required.