Attention
Why two attention tests give you different answers
You can take our attention test and another one on the same afternoon, get a good result on one and a poor result on the other, and nothing has gone wrong with either of them. This page is why.
The short version
"Attention" is not one thing, and the two tests are not measuring it in the same units. Ours counts how many trials you got right at a fixed pace. The best-known laboratory measure counts how many milliseconds slower you were when a distraction was present. One is a count of correct answers; the other is a subtraction between two timings. There is no reason for those to move together, and quite a few reasons for them not to.
The four reasons, in order of how much they matter
1. The canonical test reports a difference. Ours refuses to.
The Attention Network Test — the standard against which browser attention tests are usually justified — divides attention into three networks and reports the efficiency of each one as a change in reaction time: from a warning signal, from a spatial cue, and from conflicting flankers.
We do not publish a difference score, anywhere, as the output of any domain. Two noisy measurements subtracted from one another produce a third that is noisier than either, and it arrives looking more precise than either. On a browser the effects in question are tens of milliseconds, and we have published our own measurement of a touchscreen adding four times more latency than the response it was supposed to be measuring. A number that is mostly hardware should not be reported as though it were mostly you.
So the two tests are not disagreeing about your attention. They are answering different questions, and only one of them is a question a browser can answer honestly.
2. Our deadline is a number we chose, and a different deadline is a different test.
This is the weakest joint in our instrument and it belongs near the top rather than buried.
An interference task with no time limit measures almost nothing: congruency effects on accuracy "are found only in trials with fast responses (< 600 ms)", with accuracy "attaining near-perfect levels at about 600 ms". Give people as long as they want and nearly everyone gets nearly everything right.
So there has to be a deadline, and somebody has to pick it. We picked 650 milliseconds. Push it to 900 and the task drifts toward the ceiling that source describes; pull it to 450 and it drifts toward chance. Neither version would be wrong — they would be different tests wearing the same name, and any other attention test you take has made this same choice, usually without telling you what it was.
3. A test can measure a different part of "attention" entirely.
Sustained attention over twenty minutes, noticing something at the edge of vision, switching between two rules, resisting a distraction for one second — these come apart from each other, which is why the canonical test reports three separate numbers rather than one.
Ours measures one of them: holding a rule while an automatic, unasked-for answer competes with it, for about ninety seconds. If another test measures how long you can stay with something dull, it is entitled to disagree with us, because it is not measuring what we measured.
4. Your device is inside the score, and we cannot get it out.
The 650 milliseconds is counted from the moment the trial is drawn. Drawing it takes time, registering your keypress takes time, and we have not measured either on your hardware and cannot from inside a web page. A slower screen makes this task harder in a way that has nothing to do with you. [S] That is an inference from how browsers work, not a measurement of your device — and it is the reason we do not put a decimal point anywhere near this score.
How ours differs from the one it sits next to on this site
A fair question, since symbol match also involves paying attention and also has a clock on it. If two of our own tests measured the same thing, the composite would silently count that thing twice.
| Symbol match | Ignore the Word | |
|---|---|---|
| What you do | Look up a symbol in a key and enter its digit, over and over | Answer where an arrow points while a word says otherwise |
| Is there conflict? | No. Nothing is pulling you toward a wrong answer | Yes, and it is the entire point |
| What is fixed | The time. The number of items is up to you | The number of items. The time per item is fixed and short |
| The score is | A rate — how many in the window | An accuracy — how many of 48 you got right |
| Nearest classical factor | Approximately P, Perceptual Speed | None in that reference. See below |
Different output types, not just different pictures. One asks how much you can get through; the other asks how often you can hold a rule. You can be good at one and ordinary at the other.
The uncomfortable part: our usual anchor does not cover this
Four of our instruments are anchored to the ETS Kit of Factor-Referenced Cognitive Tests, which demonstrates 72 tests as markers for 23 cognitive factors. There is no attention factor among them. The nearest is P, Perceptual Speed — which is roughly what symbol match already does. "Concentration" appears only in the companion temperament guide, listed with no marker scales at all.
So this instrument does not have the lineage the other four have, and we would rather say that here than let the family it sits in imply one. It is anchored instead to the interference literature directly, and its single load-bearing source is named on the instrument page.
What to do with two disagreeing results
- Do not average them. They are not two readings of one quantity.
- Read the units before the number. If the other test reports milliseconds, it is measuring speed and calling it attention. That is a legitimate thing to measure and it is not this.
- Ask what its deadline was. If it does not say, its score cannot be compared with anything, including its own results from last month.
- Neither one is a diagnosis. Not ours, and not one that produces a longer report.
What next
- Take the test, if you have not.
- The instrument register — every test here, what it measures and what it refuses to claim.
- How everything on this site is scored.