Paperonium

← Back to the plain-language case

Abstract places the statistical-regularity bias reduction at the low-probability location; Results show the opposite

Hum Brain Mapp (2023) · doi:10.1002/hbm.26520 · PMC10681649

Summary

The abstract states that statistical regularities "reduced this attention bias when the target was presented at a low probability location". The Results report the opposite direction. In BA17 and BA19 the spatial bias vector lies further from the target for the high-probability target location (HTPL > LTPL in distance, t(68) = 4.93 and 2.85). The Methods define a larger distance as less bias towards the target, and target-quadrant (ROI_IN) activity is lower for HTPL (LTPL > HTPL, t(206) = 3.04). Every reported effect size recomputes from its t statistic (d = t/√n), and the Discussion and Conclusions describe the direction as the Results do, so this is an error in the abstract's wording, not in the analysis. It still matters because the direction is what separates the paper's two competing hypotheses: the abstract, read on its own, points to the account the paper's data argue against.

What the paper reports

Abstract:

Salience acted primarily by reducing the attention bias towards the target location when associated with irrelevant distractors, while statistical regularities reduced this attention bias when the target was presented at a low probability location, particularly at the lower levels of the visual hierarchy.

Introduction, last paragraph (the two hypotheses):

In the priority maps framework, we expected increased activation for targets presented in high versus low probability location (HTPL > LTPL), with sensory salience that should further boost these spatially specific responses when the target is also the salient item (Awh et al., 2012; Zhang et al., 2021). As an alternative hypothesis, if mechanisms of “expectation suppression” apply also to recurring targets (see Zhang et al., 2021, for the case of recurring distractors), one may predict a reduction of the spatial bias towards the representation of the target location, for targets presented at the high versus low probability locations (HTPL < LTPL).

Materials and Methods, "Group-level analyses of spatially specific occipital responses" (definition of the outcome):

Any increase of distance between this point and the subject-specific vector expresses a modification of the spatial bias away from the target location (i.e., a decrease of the vector's magnitude and/or a deviation of the vector's sense away from the target location, see Figure 5).

Results, "Attention priority signals modulate activity in the occipital cortex" (ROI analysis):

Specifically, the ROI_IN activity was larger in LTPL compared with HTPL (“LTPL > HTPL”, t(206) = 3.04, p < 0.005, Cohen's d = 0.21), while target-probability did not affect activity in ROI_DIAG (“LTPL > HTPL”, p > 0.9).

This revealed a significant effect of target-probability location in BA17 (“LTPL > HTPL”, t(68) = 3.42, p < 0.003, Cohen's d = 0.41), while this effect did not reach significance in BA18 (p > 0.9) and BA19 (p > 0.2).

Same section (bias-vector analysis):

The comparisons between HTPL and LTPL, separately for the three BAs, revealed a significant effect of statistical regularities in BA17 (HTPL > LTPL, t(68) = 4.93, p < 0.001, Cohen's d = 0.59) and BA19 (HTPL > LTPL, t(68) = 2.85, p = 0.017, Cohen's d = 0.34), but not in BA18 (p > 0.9). In BA17 and BA19, the spatial bias vector was more distant from the target location in HTPL than in LTPL, which explains the target-probability location × BA interaction.

Same section, summary paragraph:

In sum, we found evidence that statistical regularities modulate the spatial bias in area BA17 and BA19, with a loss of the spatial specificity towards the target location in the HTPL condition.

Discussion, "Occipital spatial representations are modulated according to statistical regularities and salience":

In both regions, these analyses showed a loss of the attentional bias for targets presented at the high- compared to the low target-probability location (Figure 5).

The finding of a reduced bias towards the target location in the high-target probability condition is consistent with the “expectation suppression” account.

Conclusions:

Statistical regularities decreased the spatial bias towards the target representation primarily in BA17, and this was independent of parietal-occipital interactions.

What the paper's own numbers show

Method. I took each reported test of target-probability location (HTPL vs LTPL) and fixed its direction from the printed contrast and the Methods definition. For the bias-vector distance, "HTPL > LTPL" means the HTPL vector is further from the target point (−11, 11), that is, less bias towards the target at HTPL. For ROI_IN activity, "LTPL > HTPL" means a weaker response in the target quadrant at HTPL. To check that the printed statistics are internally consistent (so that the direction is not a typo in the test itself), I recomputed Cohen's d for a paired test as d = t/√n. Here n = df + 1. The degrees of freedom correspond to 23 participants × 3 salience conditions (n = 69, df 68) and 23 × 3 salience conditions × 3 Brodmann areas (n = 207, df 206). I also recomputed two-sided p values from t and df. recompute.py.txt prints every number below.

Quantity Location Reported Recomputed from the paper's t and df
Bias-vector distance, BA17 Results, bias-vector analysis HTPL > LTPL, t(68) = 4.93, p < 0.001, d = 0.59 d = 4.93/√69 = 0.594; two-sided p = 5.6 × 10⁻⁶. Less bias at HTPL
Bias-vector distance, BA19 Results, bias-vector analysis HTPL > LTPL, t(68) = 2.85, p = 0.017, d = 0.34 d = 0.343; two-sided p = 0.0058 (× 3 = 0.0173, matching the reported 0.017 if adjusted for three BAs; the paper does not name the correction). Less bias at HTPL
Bias-vector distance, BA18 Results, bias-vector analysis p > 0.9 No effect in either direction
ROI_IN activity, BA17 Results, ROI analysis LTPL > HTPL, t(68) = 3.42, p < 0.003, d = 0.41 d = 0.412; two-sided p = 0.0011. Weaker target response at HTPL
ROI_IN activity, all BAs Results, ROI analysis LTPL > HTPL, t(206) = 3.04, p < 0.005, d = 0.21 d = 3.04/√207 = 0.211; two-sided p = 0.0027. Weaker target response at HTPL
Consistency check: ROI_IN > ROI_DIAG, BA17 / BA18 / BA19 Results, ROI analysis t(137) = 5.46 / 13.48 / 15.79; d = 0.46 / 1.15 / 1.34 d = 0.465 / 1.147 / 1.344 (n = 138)
Consistency check: SDD > STD in ROI_DIAG Results, ROI analysis t(137) = 6.69, d = 0.57 d = 0.569
Direction of the statistical-regularity effect Abstract Bias reduced at the low-probability location All four significant tests: reduced at the high-probability location; none at the low-probability location

All reported effect sizes agree with their t statistics to the printed precision, so the signs and contrasts are internally consistent. Every significant test of target-probability location points one way: towards a smaller target bias at HTPL. The abstract assigns the reduction to the other location.

Figure

Abstract and Results excerpts that place the reduction of the target bias at opposite target-probability locations

Figure 1. The left panel quotes the abstract sentence on statistical regularities. The right panel gives the bias-vector and ROI_IN contrasts and a Results sentence, quoted from section 3.2 of the paper. The abstract puts the reduced bias towards the target at the low-probability location (LTPL). The Results put it at the high-probability location (HTPL). The footer notes the Methods definition that links a larger bias-vector distance to less bias, and that the Discussion and Conclusions follow the Results.

Signed Cohen's d for the four significant target-probability tests, reported and recomputed, all on the side of a smaller target bias at HTPL

Figure 2. Each significant test of target-probability location in the Results, shown as a signed Cohen's d (positive = smaller bias towards the target at HTPL). Orange circles show the d printed in the paper; blue diamonds show d = t/√n recomputed from the paper's t and df. The two agree for every test, and every test lies on the HTPL side of zero. The abstract's wording corresponds to the left side, where no reported test lies. The BA18 bias-vector and ROI_IN tests (p > 0.9) and the BA19 ROI_IN test (p > 0.2) are not drawn, because no t or d is given for them.

Why it matters

Explanations considered

A separate observation, which does not bear on the direction: the degrees of freedom show that the paired tests pool the three salience conditions (df 68 = 23 × 3 − 1) or the salience conditions and BAs (df 206 = 23 × 9 − 1) as if they were independent observations, so the p values may be smaller than a test on 23 participant-level means would give. This does not change which location shows the reduced bias.

Status

As of 2026-10-24:

Suggested correction

In the Abstract, replace

statistical regularities reduced this attention bias when the target was presented at a low probability location

with

statistical regularities reduced this attention bias when the target was presented at a high probability location

This leaves the rest of the sentence ("particularly at the lower levels of the visual hierarchy") unchanged. It agrees with the bias-vector results in BA17 and BA19, the ROI_IN results, Supplementary Figure S2, the Discussion and the Conclusions. If the PubMed and indexing records are generated from the article, they should be updated as well.