Further Evidence that Probability Density Shape is a Proxy for Correlation
Bibliographic record
Abstract
Previous work demonstrated a discrimination performance cost for selecting “target” correlation populations among irrelevant “distractor” populations in two-class scatterplots (Elliott, 2016). This cost cannot be eliminated by increasing featural differences, e.g., color, between the dots in the two populations (Elliott & Rensink, VSS 2018). These findings do not agree with predictions from feature-based attention models (Wolfe, 1994), motivating us to investigate whether feature information can in fact be used to select target correlation populations. Observers performed a correlation discrimination task for scatterplots containing a target and a distractor population. Both populations had the same mean, standard deviation, color, and number of dots; the resulting two-class plots were distinguished by the correlation of the target population only. In the first of two counterbalanced conditions, targets were more correlated than the distractors; in the second, they were less. Results showed that observers can successfully discriminate two-class plots based on the correlation of their target populations. Increased JNDs were found when targets had higher correlations than distractors, replicating the results of Elliott (2016); however, there was no cost for targets with lower correlations. This asymmetry supports the proposal (Rensink, 2017) that estimation of correlation in scatterplots is based on the width of the probability density function corresponding to the dot cloud; for a two-class plot this appears to be a single density function dominated by the width of the lower-correlation (and thus wider) population. In addition, there is a resistance to feature selection: performance is the same regardless of the difference in features between target and distractor populations. This suggests that a two-class scatterplot is coded as a single ensemble, with observers unable to select items based on the value of their features because ensemble structure is prioritized over item-level feature information (Brady & Alvarez, 2011).
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.
How this classification was reachedexpand
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 | 0.003 | 0.005 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| 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 itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".