Evaluating CloudResearch’s Approved Group as a solution for problematic data quality on MTurk
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Maintaining data quality on Amazon Mechanical Turk (MTurk) has always been a concern for researchers. These concerns have grown recently due to the bot crisis of 2018 and observations that past safeguards of data quality (e.g., approval ratings of 95%) no longer work. To address data quality concerns, CloudResearch, a third-party website that interfaces with MTurk, has assessed ~165,000 MTurkers and categorized them into those that provide high- (~100,000, Approved) and low- (~65,000, Blocked) quality data. Here, we examined the predictive validity of CloudResearch's vetting. In a pre-registered study, participants (N = 900) from the Approved and Blocked groups, along with a Standard MTurk sample (95% HIT acceptance ratio, 100+ completed HITs), completed an array of data-quality measures. Across several indices, Approved participants (i) identified the content of images more accurately, (ii) answered more reading comprehension questions correctly, (iii) responded to reversed coded items more consistently, (iv) passed a greater number of attention checks, (v) self-reported less cheating and actually left the survey window less often on easily Googleable questions, (vi) replicated classic psychology experimental effects more reliably, and (vii) answered AI-stumping questions more accurately than Blocked participants, who performed at chance on multiple outcomes. Data quality of the Standard sample was generally in between the Approved and Blocked groups. We discuss how MTurk's Approval Rating system is no longer an effective data-quality control, and we discuss the advantages afforded by using the Approved group for scientific studies on MTurk.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.104 | 0.008 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.003 | 0.000 |
| Scholarly communication | 0.001 | 0.000 |
| Open science | 0.004 | 0.005 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it