The P300 is a positive change in the brain's electrical activity that appears roughly 300 to 600 milliseconds after a person sees or hears something rare and meaningful to them. It is recorded with EEG sensors on the scalp and is largest over the parietal midline, at the position called Pz.

How many words are shown in a CIT block?

A standard block has seven different items: one probe, one target and five irrelevants. Each is repeated 20 to 40 times in shuffled order, so the person sees between 140 and 280 items in a block, at roughly one every two seconds.

Why does the person have to press a key?

The key press makes sure every item is read, because the target cannot be found without looking. It also lets the examiner confirm that the person is attending and can tell the target apart. The key press is not treated as a truthful or untruthful answer.

How long does a CIT take?

Each detail is tested in a block of five to ten minutes. With set-up, explanation, two to four blocks and rests between them, an appointment usually lasts 60 to 90 minutes.

What is the bootstrap test?

It is a way of checking how stable the difference between probe and irrelevants is. The software redraws random samples of the recorded responses 1,000 times and counts how often the probe response is the larger. We report an item as recognised when that happens in at least 90 per cent of the resamples.

What happens if the recording is poor?

The block is reported as unusable. That happens when the target fails to produce a clear response, when too many showings are lost to movement, or when the key presses show the person was not doing the task. An unusable block is neither a pass nor a fail.

Can the cap read my thoughts?

No. EEG records the timing and size of electrical activity at the scalp. It cannot recover words, images or memories. In a CIT the only thing examined is whether one item on a list produced a larger response than the others.

Do you keep the raw data?

Yes. The complete unfiltered recording and the exact timing of every item are retained with the result, so the analysis can be repeated independently if the result is ever challenged.

How CIT Testing Works | P300 Concealed Information Test Explained

The Principle: Recognition Leaves a Trace

Everything in a CIT follows from one observation about how the brain handles things that matter.

Imagine reading a list of first names at a steady pace: Daniel, Marcus, Stephen, Oliver, Graham. If one of them is your own name, something happens when you reach it that does not happen for the others. You do not choose it and you cannot easily stop it. Your attention is caught for a fraction of a second, because your brain has spent your whole life treating that word as important.

That catch of attention has an electrical signature. Between about 300 and 600 milliseconds after the word appears, the summed activity of large groups of brain cells swings positive, most strongly over the back of the top of the head. Researchers named it the P300: P for positive, 300 for the earliest time in milliseconds at which it tends to peak. It was first described in the 1960s and has been measured in thousands of studies since.

Two conditions make a P300 large. The item must be rare within the sequence, and it must be meaningful to the person. A Concealed Information Test arranges both. The real detail is one item among six or seven, so it is rare. And it is meaningful only to someone who knows it is the real one. For anyone else, it is just another word in the list.

This is the whole logic of the test. We do not look for signs of stress, guilt or effort. We look for one item being treated differently from its neighbours by a system that responds before the person has had time to decide anything.

Probe, Target and Irrelevants

The item list is the test. A well-built list can be scored honestly; a badly built one cannot be rescued by any amount of analysis.

The probe

The probe is the true detail. If a sum of money went missing from a blue cash tin, "blue cash tin" might be the probe in a block about the container. It has to be something an involved person would have noticed and remembered. Details that were central to the event are remembered better than background ones: the object that was handled is a better probe than the colour of the wall behind it.

The irrelevants

The irrelevants are the alternatives: "red folder", "grey envelope", "black wallet", "white box", "green pouch". They must be the same kind of thing as the probe, about the same length, equally plausible, and equally ordinary. A person who knows nothing should have no way to guess which is real. If the probe is the only two-word item, or the only unusual word, or the only one with any personal meaning to the tested person for unrelated reasons, the block is compromised before it starts.

We normally use five irrelevants to one probe. With that ratio, a person with no knowledge has no reason to respond to the probe more than to any of the others, and the comparison has five times as much data on the irrelevant side.

The target

The target is one more item of the same kind, and it does two jobs. First, it gives the tested person a task. They are told before the block: "press Y when you see white box, press N for everything else". That forces them to read every item, because they cannot find the target without looking. Second, it is a built-in check of the recording. The target is rare and, because of the instruction, meaningful to everyone. So everyone should produce a P300 to it. If the target does not produce one, something is wrong with the recording or with how the person is attending, and we treat the block as unusable.

This arrangement is called the three-stimulus protocol. It was the standard design in the P300 CIT literature for many years, and our procedure is based on Rosenfeld's published P300 CIT protocols.

Why the key press is not a truth question. People sometimes ask why they press N for the probe, as if that were a lie being tested. It is not. The keys only confirm the person is reading and can tell the target from the rest. The measurement is the brain response, not the answer.

Building the Lists Before the Appointment

This is the part of the work the tested person never sees, and it decides whether the result means anything.

We start from the account given by the person arranging the test. From that account we pull out candidate details and ask three questions of each one.

Would an involved person definitely know it? If they could have been involved without noticing this detail, a flat response tells us nothing.
Could an uninvolved person know it? Has it been said out loud, written in a message, posted, shown in a photograph, or mentioned in an accusation? If so, it is out.
Can we write five alternatives that are just as believable? Some details have no natural alternatives, or the real one is so much more likely than any other that people would guess it.

Details that pass all three become blocks. For each block we draft the probe, the target and the irrelevants, and check the set for give-aways: length, familiarity, spelling, anything that makes one item look different on a screen. Where possible we ask someone with no knowledge of the case to look at the list and say which item they think is the real one. If they can pick it, the list is rewritten.

The person arranging the test sees and approves the lists. The person being tested does not see them until the words appear on screen, and is never told which item is the probe.

The Equipment

What is actually on the head, and what it records.

We record with a BrainBit Flex 4, a four-channel EEG amplifier with a fabric cap. Four sensors sit along the midline and back of the head, at positions named Fz, Cz, Pz and Oz in the standard electrode map, with reference and ground clips on the ears. The amplifier samples each sensor 250 times a second and sends the readings to a laptop over Bluetooth.

The channel that matters for scoring is Pz, over the parietal midline, towards the back of the top of the head. That is where the P300 is largest in most people. The other channels help the examiner judge the quality of the recording and spot interference such as blinks, which show up most at the front.

The signals are tiny. A P300 is a few millionths of a volt. A blink, a clenched jaw or a loose sensor can produce a hundred times that. So most of the examiner's effort at the start goes into contact: parting hair under each sensor, adding a small amount of conductive gel, and watching the live traces until they are steady and quiet. If contact cannot be made good enough, we do not start the test.

The sensors are passive. They measure voltage at the skin. Nothing is transmitted into the head, and the cap cannot read thoughts, images or words. All it can show is the timing and size of electrical activity after each item appears.

What the Tested Person Sees and Does

The task is deliberately dull. Dull is good: it keeps the recording clean.

Each block opens with an instruction screen that names the target and the keys. When the person is ready, items begin to appear one at a time in the centre of the screen. Each stays for 300 milliseconds, about a third of a second, then the screen clears. The person presses Y for the target or N for anything else. After the key press there is a pause of between 1.4 and 2 seconds, varied at random so the next item cannot be anticipated, and the next item appears.

The order is shuffled so that the person cannot predict what is coming, with a rule that the same item never appears twice in a row. Over a block, every item is shown the same number of times. With one probe, one target and five irrelevants each shown thirty times, a block has 210 items and takes around eight minutes including a short rest in the middle.

During the block the person is asked to do four things: keep their eyes on the centre of the screen, blink between items if they can, keep the jaw and forehead relaxed, and answer every item. They do not need to speak. They are not asked whether any item is true, and they are not told how they are doing.

Why so many repeats

The brain is electrically busy all the time. The response to any single word is far smaller than the background activity it sits in, so a single showing is unreadable. But the background is unrelated to the moment the word appeared, while the response is locked to it. If you cut out the second or so of recording after each showing of an item and average those snippets together, the unrelated background tends towards zero and the locked response remains. Twenty showings give a rough picture. Thirty or forty give a much steadier one. This is why the test cannot be shortened to a handful of words.

From Recording to Result

The analysis in the order it happens, without the mathematics.

The whole session is recorded raw. Every sample from every channel is stored, with the exact moment each item appeared. Nothing is discarded at this stage.
The recording is filtered. Slow drift below 0.3 cycles per second and fast activity above 30 are removed, along with mains hum at 50. The P300 lives comfortably inside that band.
It is cut into epochs. For each item shown, a window is taken from 200 milliseconds before the item to 1,300 milliseconds after. The 200 milliseconds before gives a baseline to measure from.
Contaminated epochs are rejected. Any window in which the signal swings by more than 100 microvolts is dropped, because a swing that size is movement or a blink, not brain activity.
Epochs are averaged by item type. One average for the probe, one for the target, one for the irrelevants pooled.
The P300 is measured. We find the most positive 100-millisecond stretch between 300 and 900 milliseconds, then the most negative stretch after it, and take the difference. This peak-to-peak measure is the one recommended in the published P300 CIT literature because it is less sensitive to slow drift than a simple peak.
The target is checked. If the target's P300 is not clearly larger than the irrelevants', the block is unusable and scoring stops there.
The probe is tested against the irrelevants. This is the step that produces the result, described next.

The bootstrap test

A single comparison of two averages does not say how much to trust the difference. The probe average might be larger by luck: a few noisy epochs can shift it. To find out how stable the difference is, the analysis repeats the comparison 1,000 times. Each time, it draws a fresh random sample of probe epochs and an equal-sized random sample of irrelevant epochs, with replacement, averages each, and measures the P300 in both. It then counts how many of the 1,000 rounds had the probe larger.

If the person does not recognise the probe, the probe is just another irrelevant, and it should come out larger about half the time. If it comes out larger in 90 per cent of rounds or more, we report the item as recognised. Below that, we report it as not distinguished. The figure on the report, for example "larger in 94.2 per cent of 1,000 resamples", is a statement about this recording: how consistently it separated the probe from the alternatives. It is not a probability that the person did something.

The Checks That Protect the Result

A test with no way of saying "this reading is no good" should not be trusted when it says anything else.

Target check. Described above. A missing target response means the person was not attending, the contact was poor, or the timing was off. Whatever the cause, the block is not scored.
Clean-epoch count. If too many showings of the probe are rejected for movement, the average rests on too little data. The report gives the count for every item so this is visible.
Key-press accuracy. Someone who presses N for everything, or misses the target repeatedly, is not doing the task. That is recorded and may make the block unusable.
Reaction times. Unusually slow or uneven responses to particular items can indicate that the person is doing something extra on those items. The pattern is noted.
Timing calibration. A Bluetooth link adds a small, variable delay. The software estimates the true timing from the least-delayed packets so that item onsets line up with the recording.
Raw data kept. The unfiltered recording and the item timings are retained, so every step above can be repeated independently.

What Can Go Wrong

Stated plainly, because a method is only as trustworthy as its account of its own failures.

The detail had leaked

The most common way for a CIT to mislead is for an uninvolved person to know the probe. They recognise it because they were told, and the test reports recognition, correctly, with no way of knowing the source. This is why so much weight is placed on what has and has not been disclosed before the appointment.

The detail was never noticed, or was forgotten

An involved person who did not take in a particular detail will not respond to it. Peripheral details fade faster than central ones, and memory weakens with time. A flat response to one probe is weak evidence of non-involvement for that reason; flat responses across several well-chosen probes are stronger.

The probe meant something for another reason

If the probe happens to be the tested person's mother's name, their home town or the make of their own car, it will draw a response regardless. The list-building stage tries to catch this, and the person arranging the test is asked about it directly.

Countermeasures

In 2004 Rosenfeld and colleagues reported that participants taught to make a small covert response to irrelevant items could shrink the gap between probe and irrelevants in the three-stimulus protocol. A later design, the Complex Trial Protocol, was developed partly to resist that. We use the three-stimulus protocol and do not claim it is immune. Signs such as slowed, uneven reaction times and a distorted target response can point to an attempt, but they are not proof.

Individual differences

Some people produce small P300s to everything, including the target. In those cases the target check fails and no conclusion is offered. Very poor scalp contact, heavy movement, or strong tiredness have the same effect.

The research is mostly from laboratories

Laboratory studies commonly report detection of 80 to 95 per cent of people who hold the tested information. Those figures come from controlled experiments and are not results from our own testing. Real cases differ: the detail may be older, less rehearsed, or partly known. Field validation of the P300 CIT is limited, and a result should be weighed as one piece of information.

Known-Answer and Searching Tests

The test described on this page assumes somebody knows the true detail. There is a second form for when nobody does.

In the standard, known-answer CIT, the examiner knows which item is the probe, and the question is whether the tested person recognises it. In a searching CIT, the true answer is unknown. The list contains candidate answers, and the question is whether any one of them stands out from the rest. That is a harder statistical problem, because whichever item happens to come out largest will look special unless the analysis accounts for having picked it afterwards. Our searching test uses a permutation test for that reason and requires agreement across separate blocks. It is a weaker instrument than the known-answer test and is reported more cautiously.

Arranging a Test

If the method fits your case, the next step is a conversation about details.

Read who qualifies for a CIT test to check your situation against the requirements, or go straight to booking. We will ask what happened, what the other person has been told, and who else knows. If a sound list cannot be built, we will tell you before any appointment is made.

Book a CIT Test

How CIT Testing Works

On this page