Automatic Coding of Text Answers to Open-Ended Questions: Should You Double Code the Training Data?
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
Open-ended questions in surveys are often manually coded into one of several classes (or categories). When the data are too large to manually code all texts, a statistical (or machine) learning model must be trained on a manually coded subset of texts. Uncoded texts are then coded automatically using the trained model. The quality of automatic coding depends on the trained statistical model, and the model relies on manually coded data on which it is trained. While survey scientists are acutely aware that the manual coding is not always accurate, it is not clear how double coding affects the classification errors of the statistical learning model. We investigate several budget allocation strategies when there is a limited budget for manual classification: single coding versus various options for double coding where the number of training texts is reduced to maintain the fixed budget. Under fixed budget, double coding improved prediction of the learning algorithm when the coding error is greater than about 20–35%, depending on the data. Among double-coding strategies, paying for an expert to resolve differences performed best. When no expert is available, removing differences from the training data outperformed other double-coding strategies. When there is no budget constraint and the texts have already been double coded, all double-coding strategies generally outperformed single coding. As under fixed budget, having an expert to solve disagreement in training texts improves accuracy most, followed by removing differences.
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 | 0.071 | 0.003 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.002 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.004 | 0.001 |
| Research integrity | 0.000 | 0.000 |
| 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