Show the full module text
Module 5 — Where Screeners Fit
What a screening questionnaire is for, what it cannot be for, and which of the two courses to open next.
Self-Identification · Part Two — The evidence
The big idea
Short on capacity today? The big idea: You have read the criteria and started on the history. The obvious next move is to take a test, and there are two courses next door that are nothing but tests, each taught with its limits shown. This module is the sixty seconds before you open one. It exists because the most common way to waste a screener is to ask it a question it was never built to answer.
Step 1 — The lesson
A. Two lamps, and there is no third
A screening instrument is a sorting device. It was built to split a large group cheaply into a smaller group worth looking at closely and a larger group that probably is not, so that the expensive procedure gets spent where it is most likely to find something. It is an allocation tool that happens to arrive in the shape of a questionnaire about you.
Which is why every screener in both of the sister courses returns one of exactly two results:
- low or no indication, or
- possible or elevated indication.
There is no third lamp. Nothing in either course says you have it, and no arrangement of answers will produce that, because the instrument has no mechanism for producing it. Read literally, “possible or elevated indication” means: your answers resemble the answers of the group this questionnaire was built to flag. That resemblance is real and it is worth something. It is not a finding about you.
Diagram — A · Two lamps, and there is no third. A screening instrument is a sorting device: it was built to split a large group cheaply into a smaller group worth looking at closely and a larger group that probably is not, so that the expensive procedure gets spent where it is most likely to find something. Which is why every screener in both of the sister courses returns one of exactly two results: low or no indication, or possible or elevated indication. There is no third lamp. Nothing in either course says you have it, and no arrangement of answers will produce that, because the instrument has no mechanism for producing it. Read literally, possible or elevated indication means that your answers resemble the answers of the group this questionnaire was built to flag. That resemblance is real and it is worth something, but it is not a finding about you. The corollary catches almost everybody: a score is not a dose. Thirty-eight is not more autistic than thirty-one; it is further past a line somebody drew in a validation sample. The number tells you which side you came down on. It does not rank you.
The corollary catches almost everybody: a score is not a dose. Thirty-eight is not more autistic than thirty-one, it is further past a line somebody drew in a validation sample. The number tells you which side you came down on. It does not rank you.
B. The negative lamp is usually the more trustworthy one
Here is the asymmetry, in one study.
In 2018, researchers gave the ASRS-v1.1 — the most widely used adult ADHD screener there is — to 40 adults being treated for major depression and 55 healthy controls, then established who actually had ADHD by structured clinical interview. Full-syndrome ADHD was present in 12.5 per cent of the depressed patients.
Against that reference standard the screener returned:
Sensitivity — 60 per cent
Specificity — 68.6 per cent
Positive predictive value — 21.4 per cent
Negative predictive value — 92.3 per cent
Translated: of the people it flagged, roughly one in five turned out to have ADHD. Of the people it did not flag, about nine in ten indeed did not. The authors’ own conclusion is the sentence to keep — a negative screen strongly suggests the absence of ADHD, while a positive result requires careful evaluation to work out whether the symptoms come from the depression or from a second condition underneath it.
Diagram — B · It beeps at the bottle tops. In 2018, researchers gave the ASRS-v1.1 — the most widely used adult ADHD screener there is — to 40 adults being treated for major depression and 55 healthy controls, then established who actually had ADHD by structured clinical interview. Full-syndrome ADHD was present in 12.5 per cent of the depressed patients. Against that reference standard the screener returned a sensitivity of 60 per cent, a specificity of 68.6 per cent, a positive predictive value of 21.4 per cent and a negative predictive value of 92.3 per cent. Translated: of the people it flagged, roughly one in five turned out to have ADHD; of the people it did not flag, about nine in ten indeed did not. The authors' own conclusion is that a negative screen strongly suggests the absence of ADHD, while a positive result requires careful evaluation to work out whether the symptoms come from the depression or from a second condition underneath it. That shape — high negative predictive value, low positive predictive value — is the normal shape for adult screening, not one unlucky study.
That shape — high negative predictive value, low positive predictive value — is the normal shape for adult screening, not one unlucky study.
C. Why the base rate does that, and how it flips
The reason is arithmetic, and you do not need any of the arithmetic to see it.
If the thing being screened for is uncommon in the group taking the screen, most of the people it flags will be people it got wrong — not because the instrument is bad, but because the group it is wrong about is so much bigger. A screen that misfires on a modest fraction of a very large unaffected group will still produce more false alarms than there are true cases in the small affected one. Nothing is broken. That is what catching a rare thing costs.
You can watch it happen at full size. In 2021, the same ADHD screener was run over two general population samples — 642 people in the UK and 579 in the USA. It indicated probable ADHD in 26.0 per cent and 17.3 per cent of participants, against an expected prevalence of 2.5 per cent. The estimated positive predictive value was around 11.5 per cent, and the authors put the over-identification at seven to ten times.
Now watch it run the other way, which is the part nobody tells you. In 2016, researchers followed 476 adults seen consecutively at an autism diagnostic referral service — a room where 73 per cent went on to be diagnosed. There the autism-spectrum quotient had a positive predictive value of 0.76 and sensitivity of 0.77, but a specificity of 0.29 and a negative predictive value of 0.36. Which means 64 per cent of those who scored below the cut-off were autistic anyway.
Same class of instrument, opposite asymmetry. Nothing changed except who was in the room.
Diagram — C · The same rail, two rooms. If the thing being screened for is uncommon in the group taking the screen, most of the people it flags will be people it got wrong — not because the instrument is bad, but because the group it is wrong about is so much bigger. In 2021 the same ADHD screener was run over two general population samples, 642 people in the UK and 579 in the USA. It indicated probable ADHD in 26.0 per cent and 17.3 per cent of participants, against an expected prevalence of 2.5 per cent; the estimated positive predictive value was around 11.5 per cent, and the authors put the over-identification at seven to ten times. Now the other way. In 2016, researchers followed 476 adults seen consecutively at an autism diagnostic referral service, a room where 73 per cent went on to be diagnosed. There the autism-spectrum quotient had a positive predictive value of 0.76 and sensitivity of 0.77, but a specificity of 0.29 and a negative predictive value of 0.36 — which means 64 per cent of those who scored below the cut-off were autistic anyway. Same class of instrument, opposite asymmetry, and nothing changed except who was in the room. So the rule is not that negatives are reliable and positives are not. It is that which lamp you can lean on depends on how common the thing is among people like you, and nobody has measured your in-between.
So the rule is not negatives are reliable and positives are not. It is that which lamp you can lean on depends on how common the thing is among people like you — and you, four modules into a course you sought out because something clicked, are neither a general population nor a referral clinic. You are in between, and nobody has measured your in-between. Which is exactly why neither lamp settles it.
One finding sits in both studies, and it is Module 3’s differential gate showing up as arithmetic: anxiety inflates these scores. The 2016 study named generalised anxiety as a condition that can mimic autism and push the questionnaire up; in the 2018 study, the number of items endorsed tracked the patients’ anxiety levels. A high score is a real signal — just not a signal about only one thing.
Diagram — D · One needle, everything on it. One finding sits in both studies, and it is Module 3's differential gate showing up as arithmetic: anxiety inflates these scores. The 2016 study named generalised anxiety as a condition that can mimic autism and push the questionnaire up; in the 2018 study, the number of items endorsed tracked the patients' anxiety levels. A high score is a real signal — just not a signal about only one thing. Before you reorganise your self-understanding around a questionnaire, ask what else produces this pattern: the differential list from Module 3, and the honest one — sleep, grief, a job that would break anyone. Finding something is not disqualifying, because co-occurrence is the norm. But a high score with no recognition behind it is the weakest single piece of evidence in this course. And a low score when you do recognise yourself is not an answer either: in the one setting where the question was live for everybody in the room, most people who scored under the line were autistic regardless.
D. What to do when the result disagrees with you
A low score when you recognise yourself. Do not file this as an answer. The 2016 study is the reason: in the one setting where the question was live for everybody in the room, most people who scored under the line were autistic regardless. Screeners miss adults for reasons unrelated to whether the description fits — a presentation trained smooth over thirty years answers the way a non-autistic person would, and the requirements that actually decide it are the four the questionnaire never asks about. A low score is a reason to keep going with better evidence, which is the document you started in Module 4.
A high score when you do not recognise yourself. Take it seriously, and slowly. A quarter of an ordinary population sample screened positive on that ADHD instrument. Before you reorganise your self-understanding around a questionnaire, ask what else produces this pattern — the differential list from Module 3, and the honest one: sleep, grief, a job that would break anyone. Finding something is not disqualifying; co-occurrence is the norm. But a high score with no recognition behind it is the weakest single piece of evidence in this course.
In both directions the instruction is the same. A screener is a question generator, not an answer. Its best use is to change what you go and find out next.
Diagram — E · Three bearings, one fix. A screener is a question generator, not an answer, and its best use is to change what you go and find out next. Before: write your prediction down before you answer a single item — one line, I think this will come back elevated, and here is why — because the gap between your prediction and the result is the informative part, and it is destroyed the moment you see the score. Take three: Modules 2, 3 and 4 of each course are the three broad screeners, and one result is a coin. Together: read them against each other rather than one at a time, because three that agree, or three that split two-to-one, tell you something a coin cannot. After: convert whatever you get into one next question rather than a verdict — possible or elevated indication becomes which requirement is my evidence weakest on, and who still alive can speak to it; low or no indication becomes which items did I answer for the version of me that has been practising for thirty years. Never re-take the same screener hoping for a different lamp: nothing has changed except your mood and the fact that you now know what the items are looking for, and the answer to a result that bothers you is a different kind of evidence.
Where to start
Two courses, and they are the next thing you read.
- Autistic Self-Discovery — nineteen modules, at /self-discovery-autism
- ADHD Self-Discovery — ten modules, at /self-discovery-adhd
Each opens with a module called How to Read a Screener, and each carries the same instruction, which is worth following literally:
Take the three broad screeners first — Modules 2, 3 and 4. Read them together rather than one at a time. Then come back with a question rather than a number.
Three instruments disagreeing teaches you more in an afternoon than one instrument agreeing does in a month, because the disagreement is where the thing you want to know is hiding.
If you think both may apply to you, start at Module 1 of each. Not interleaved — run one course through its three broad screeners, then the other. Both were built for people who arrive suspecting both, which is a great many people, and neither assumes you have ruled the other out.
→ Those two opening modules, the ones to read before you answer a single item, are Autistic Self-Discovery, Module 1 and ADHD Self-Discovery, Module 1.
F. What helps
1. Write your prediction down before you answer a single item.
One line: I think this will come back elevated, and here is why. The gap between your prediction and the result is the informative part, and it is destroyed the moment you see the score — nobody can reliably remember what they expected.
2. Take three, and read them against each other.
That is what Modules 2, 3 and 4 of each course are for. One result is a coin. Three that agree, or three that split two-to-one, tell you something a coin cannot.
3. Convert whatever you get into one next question, not a verdict.
Possible or elevated indication becomes: which requirement is my evidence weakest on, and who still alive can speak to it? Low or no indication becomes: which items did I answer for the version of me that has been practising for thirty years? Write it at the bottom of the Module 4 document, where you will see it again.
4. Do not re-take the same screener hoping for a different lamp.
Nothing has changed except your mood and the fact that you now know what the items are looking for. If a result is genuinely bothering you, the answer is a different kind of evidence — the history, an informant, a fortnight of consequences logged — not the same questionnaire on a better day.
Up next