Questionnaire pre-screen

Expert review of an English-language questionnaire before pilot testing. Within two working days, receive a signed report from Dr. Milos Kankaras identifying wording and content concerns, with suggested revisions and a provenance check. The second option adds translation-risk notes for up to three target languages. This is a review of the draft, not a prediction of item statistics or evidence of cross-language equivalence.

The pre-screen reads a draft before the pilot. The audit diagnoses an instrument you already run. The development builds the one you do not have. Start wherever you are.

What you send

• The draft questionnaire in English, as docx, xlsx or csv. Item text is enough; construct labels, keying and the response format sharpen the read.
• Who it is for: the population and the setting.
• The target languages, if the instrument will be translated, up to three. Questionnaires written in Montenegrin or another language of the region are served from psihologija.me.

What comes back within 48 hours

Every item is read twice. A blind AI critique rates each item for rejection risk and names its main problem; it sees only the item text, the construct label, the keying and the response format, never any data. Milos Kankaras then reads every item himself, including the ones the AI pass let through. The AI pass decides where he reads hardest, not what he concludes. The report is his.

• One row per item to revise or cut: the critique and a suggested rewrite that keeps the construct and the keying direction.
• A provenance check: items that resemble a published instrument, with the instrument’s name, licence class and citation. No catalogued item is quoted.
• Format notes on wording: length, reading level, negations, absolutes. Notes, not verdicts.
• With target languages: a translation-risk table per language, written for the translator. Idioms, locale-bound referents, comparison groups, gender agreement, address form, terms that need one settled rendering.
• A signed report with a plain-words limitations page, so whoever reads it after you knows what it can and cannot carry.

The sample is a real pre-screen of a 75-item English work-behaviour questionnaire with three target languages, with item ids and scale codes relabelled: open the sample report (PDF, 8 pages). A client receives the same document on their own questionnaire.

Two options, one questionnaire

Prices exclude VAT; Montenegrin VAT of 21 percent is added where it applies. Longer questionnaires and more languages are quoted on request.

What the AI pass is, and is not

We tested the AI critique before offering it. On the 332 items of the OECD SSES 2018 item trial, where the real expert review had dropped 140, its cut calls were items the review had dropped in 16 of 23 cases, against a base rate of 42 percent (odds ratio 2.78, 95 percent confidence interval 1.76 to 4.40; AUC 0.67). On 162 published adult self-report items the same pass did not predict which items were statistically weakest. So the claim is exactly this: the AI pass reads the way an expert panel reads, and it is a first read only. The reviewer’s judgement is the product.

Two papers set the boundary. Maeda and Lu (2026), arXiv:2608.06609: on 52,759 items, 34 percent of them rejected from operational use, adding LLM critiques to the item text improved the prediction of rejection (fusion model AUC .80). That is why an AI first read is worth having. Lukauskas and Sarkauskaite (2026), arXiv:2608.14606: across 37 models, every one shows a strong acquiescence shift (+0.84 SD), and none is a drop-in replacement for human respondents. That is why the pre-screen never simulates a respondent.

What a pre-screen does not do

• It does not predict item statistics. On published adult scales its flags were unrelated to which items were empirically weakest.
• It does not simulate respondents and does not predict a response distribution.
• It does not replace a pilot, cognitive interviews or a psychometric analysis of real responses.
• The provenance check names instruments by name, licence class and citation. It is not a legal opinion.
• The translation-risk notes are prompts for the translator, not a translation and not a guarantee of equivalence.
• It does not review a manuscript. For a methods referee report on a paper before you submit it, see the Pre-Submission Methods Review.

Order a pre-screen

Paste the items below, or write that the file follows by email. We confirm receipt and the price by return; the 48 hours run from receipt of the file, on working days.

Name, email and the items are required.

Your name

Your email address

Organisation

Number of items

Target languages, if any

The items, with the population and the setting Paste the items here, or write that the file follows by email.

Send the order

Prefer to attach the file, or the form gives you trouble? Send it to milos@centerforpsychology.me.