The Rod-and-Frame Test (RFT) setup consists of a rod and a surrounding frame, both of which can independently rotate about the same pivot point. During the test, subjects are asked to align the rod with the true vertical. Nearly 80 years ago, Asch and Witkin discovered the rod-and-frame effect: when the frame is tilted, participants tend to rotate the rod past vertical, in the direction of the frame's tilt (Witkin and Asch, 1948).

The perception of verticality results from the complex integration of visual (Böhmer and Mast, 1999; Hansson et al., 2010; Dockheer et al., 2018), vestibular (Clarke et al., 2003; Pavlou et al., 2003; Kumagami et al., 2009; Hansson et al., 2010; Dockheer et al., 2018), and proprioceptive cues (Anastasopoulos et al., 1999; Trousselard et al., 2003; Barra et al., 2010). Individuals differ in how they weight these sensory inputs: field-dependent individuals, who rely more heavily on visual information to assess body orientation, tend to make larger errors. In contrast, field-independent individuals can suppress misleading visual cues and typically perform better in the RFT (Bednarek and Orzechowski, 2008).

Degeneration of vestibular hair cells and neurons is a typical manifestation of aging that can be observed as early as the fifth decade of life (Zalewski, 2015). Impairment of the otolith organs leads to increased noise in the afferent signals transmitted to the sensory integration centers of the brain. The central nervous system compensates by increasing the weighting of visual inputs in determining the vertical (Lee, 2017a,b). Such sensory reweighting (Curthoys, 2000; Peterka, 2002; Peterka and Loughlin, 2004) contributes to the heightened visual field dependence observed in older adults. Consequently, the RFT has been used not only to characterize cognitive style but also to investigate age-related changes in multisensory integration (Alberts et al., 2019).

(Witkin and Asch 1948) carried out their experiment in a darkened room using a rod and frame covered with fluorescent paint. For several decades, various RFT implementations mimicked the original setup (Erdos, 1979; Zoccolotti et al., 1993; Li and Matin, 2005; Tjernström et al., 2019). In more recent studies, the test has been projected onto a wall (Tasseel-Ponche et al., 2017), implemented using the video eye glasses (Bagust, 2005), displayed on a computer monitor (Razzak et al., 2018), or presented in virtual reality (Bringoux et al., 2009; Adamski et al., 2021; Willey and Liu, 2022; Fujimoto and Ashida, 2022).

The implementation of the Rod-and-Frame Test in a virtual reality environment (VR-RFT) offers significant methodological and practical advantages over traditional 2D or mechanical setups. By its very nature, VR-RFT is reproducible, portable, free of unwanted visual cues, and does not require controlled lighting conditions. Moreover, the immersive VR environment can mimic the visual conditions of the real world, increasing the relevance of the findings in cognitive and perceptual research (Reger et al., 2003). This is why in our research we used a VR-RFT.

In the RFT, subjects can use the edge of the frame and its imaginary diagonal as visual cues (Beh et al., 1971). Relying on one of these cues leads to the RFT illusion—an error in determining the vertical. In this study, we investigate whether individuals who can use both cues perform more accurately at tilt angles, where errors typically peak. To this end, we introduce the concept of RFT bias. Such a function, for a given frame tilt, indicates whether a subject rotates the rod toward one of the cues. We employ the variability of the bias across the set of angles (e.g., clockwise or counterclockwise), termed RFT flexibility, to assess the extent to which the subjects use both visual cues to determine vertical.

For each subject, we calculated the median alignment error across the 18 frame tilt angles θ used in the experiment. Figure 1A shows the distribution of these medians for the cohort. Although the maximum median error of 3.5° occurs at a tilt of 20° and 15°, the errors between 10° and 20° are similar. A comparable pattern is observed for negative (counterclockwise) θ, where maximum median error occurs at 20° and 10°.

The group-averaged bias is highest in the intervals [−20, −10] and [10, 20], indicating that most subjects exhibit the direct effect—rotating the rod in the direction of the tilted edges of the frame (Figure 1B). For these intervals, the median bias is 1.00. Bias values are notably lower for larger tilts: [−40, −30] and [30, 40] (0.52 and 0.40 respectively), suggesting that some participants rotate the rod toward the frame's diagonal (indirect effect).

The group-averaged error curve ẽ(θ) shown in Figure 1A appears fairly symmetric. However, this symmetry may be misleading, as it results from averaging both symmetric (e.g., Figures 2A, C) and asymmetric (e.g., Figures 2E, G) individual error curves.

The error asymmetry index for subjects S5 (−0.13) and S8 (0.08) in Figures 2A, C is low. Subject S5 shows only the direct effect, while subject S8 also displays the indirect effect at large frame tilts.

The error curves for subjects S6 and S13 are clearly asymmetrical, as reflected in their error asymmetry index values (0.39 and −0.56). Subject S6 exhibits the indirect effect only for negative tilt angles, whereas subject S13 shows it primarily for positive tilts. These two cases shed light on how visual cue usage relates to RFT alignment error. For subject S6, the mean absolute error for positive θ is 129% higher than for negative angles (3° vs. 1.31°), with markedly lower RFT flexibility (0.03 vs. 1.35). For subject S13, the error is 256% higher for negative θ (2.88° vs. 0.81°), again with much lower flexibility (0.05 vs. 0.91).

Considering asymmetry in the case studies, we calculated E_R and E_L and the corresponding f_R and f_L (Supplementary Table S1). Such 42 pairs (two for each subject) are shown (Figure 3A). The correlation between error and flexibility was strong: ρ = −0.80 with p < 1 × 10⁻⁵. A linear fit to all data points gave E(f) = −1.82f+3.33, with 16 out of 42 points (38.1%) within the 95% confidence band. To further verify that the flexibility is related to the error in the RFT, we divided the cases into two groups using the median of f = 0.32 (Supplementary Table S2). The inset in Figure 3A shows the distribution of E for such groups. The median of E is equal to 2.84° and 1.56° for low (f < 0.32) and high (f≥0.32) flexibility, respectively. The percent change of 82% is statistically significant (p = 6 × 10⁻³). Figure 3B illustrates the relationship between error asymmetry α_e and flexibility asymmetry α_f. The Spearman correlation coefficient was ρ = −0.85 with p = 3 × 10⁻⁵. The linear fit yielded α_e = −0.44α_f−0.03, with 11 out of 21 points (52.4%) within the 95% confidence band.

To determine whether subjects who use the frame's diagonal while searching for the vertical are more accurate, we calculated the bias function b(θ) at θ = 35° where such a cue is visually salient, and the error E(θ) at θ = 15° where the group-averaged error curve reaches maximum. Similarly, the calculations were repeated for the corresponding negative tilt angles. We collect results in Supplementary Table S3 and present the spread of these points in Figure 4A. We can see that the higher the bias, the higher the error. The correlation between them is strong ρ = 0.75 with p < 1 × 10⁻⁴. The best linear model was E(b) = −1.80b+2.95. In general, 15 of 42 data points (35.7%) were within the 95% confidence band. Figure 4B shows the distribution of errors for low-bias cases (b ≤ −0.5, indicating reliance on the diagonal cue) and high-bias cases (b ≥0.5, indicating reliance on the edge cue). The median E in the low-bias group was almost four times smaller than that in the high-bias group (1.25° vs. 4.50°, p = 0.003). As seen in Figure 4C there was no statistically significant difference in error E made by both groups at θ = ±35° (1.50° vs. 2.00°, p = 1).

The RFT is a classic paradigm for studying visual field dependence (Witkin and Asch, 1948). Historically, performance on this test has been interpreted primarily in terms of susceptibility to the frame-induced illusion, assumed to reflect a stable individual trait—visual field dependence (Witkin and Asch, 1948; Nair et al., 2018; Abdul Razzak et al., 2020; Abdul Razzak and Bagust, 2022). However, this view may oversimplify the perceptual and cognitive strategies involved. Specifically, participants are not limited to a single visual cue; both the edges of the tilted frame and its diagonals can serve as reference axes (Beh et al., 1971). The diagonals of a tilted square frame can approximate the gravitational vertical, particularly at larger tilt angles. Thus, the selection and weighting of visual cues may influence RFT performance beyond general field dependence.

At small values of θ, the frame's edges play a critical role in the rod alignment strategy. They cause the subject to rotate the rod past the true vertical, as illustrated in Figure 2A for subject S5. The alignment error remains much smaller than the frame's tilt angle (i.e., the maximum possible RFT error), reflecting the multisensory nature of vertical perception. As θ increases, so does the sensory conflict. Due to the frame's symmetry, at θ = 22.5°, neither the edges nor the diagonals are reliable cues for verticality. The saturation and eventual decrease of alignment error at larger tilt angles may reflect a shift in reliance from visual to non-visual (vestibular and proprioceptive) inputs, though this interpretation requires further investigation.

On average, subject S5 exhibits the classic direct effect—rotating the rod in the direction of the frame's tilt—as indicated by a high positive bias value (Figure 2B). In contrast, at θ = 40° and θ = −35°, subject S8 (Figure 2D) rotates the rod opposite to the frame's tilt, toward one of the diagonals, consistent with the indirect effect. The errors E(θ) for S8 are substantially smaller than those of S5 (Supplementary Table S2). This raises the question of whether utilizing the diagonal as a visual cue might be advantageous for reducing alignment error. At first glance, this hypothesis may appear controversial, since both the direct and indirect effects contribute to RFT error, as illustrated by subject S6 at θ = ±40° (Figures 2E, F). Nevertheless, Figures 2E, G reveal that RFT errors tend to be smaller on the side of the tilt spectrum where the indirect effect is observed. Similar effect was observed by (Abdul Razzak and Bagust 2022) in a study that used two frame tilts (±18°). A plausible explanation is that subjects' ability to visualize the frame's diagonals reflects their capacity to perceive the visual context in a way that mitigates the RFT illusion at intermediate tilt angles, where alignment errors are typically largest.

Figure 3A shows strong negative correlation ρ = −0.80 with p < 1 × 10⁻⁵ between mean absolute error and RFT flexibility. Moreover, the error E for the low-flexibility group is 84% higher than that of the high-flexibility group. One can see in Figure 3B that the asymmetry in RFT flexibility mirrors the asymmetry in alignment error (the correlation coefficient ρ = −0.85 with p = 1 × 10⁻⁵), further supporting the notion that an individual's cue-use strategy is closely tied to task performance. The most compelling evidence that the ability to use the reconstructed frame's diagonal to determine the vertical underlies good performance at tilt angles where the RFT error is highest is provided by data presented in Figure 4. We can see that subjects who relied on the diagonal to determine the vertical at ±35° (for such angles, this cue is visually salient), at ±15° made the error almost four times smaller than those who relied on the edge. It should be emphasized that the error that both groups made at ±35° was not statistically different.

The presented findings suggest that RFT errors—arising from multisensory (visual, vestibular, proprioceptive) integration—also reflect an individual's propensity for cue integration, which attenuates the sensitivity to the edge cue.

Accurate perception of verticality at intermediate tilt angles may be particularly important for maintaining postural control and balance in natural environments, especially for populations with impaired vestibular function or in aging, where reliance on visual cues may increase (Agathos et al., 2015).

A significant question is whether RFT flexibility in visual cue use is a fixed individual characteristic or a malleable skill. Future studies could investigate whether individuals can be trained to improve their flexibility—for example, through explicit instruction or targeted practice—and whether such training translates to better RFT performance and, potentially, to improvements in other real-world spatial orientation tasks (Willey and Jackson, 2014) and balance control. The review of (Cortés-Pérez et al. 2021) shows the examples of such line of research that involve both immersive and non-immersive VR.

Several limitations are inherent to the present study.

We revisited the classic Rod and Frame Test (RFT) using a virtual reality implementation. Our findings suggest a need to reinterpret the significance of RFT alignment error: it reflects not only visual field dependence but also individual differences in how visual context is processed. Participants who used both visual cues—the frame's edges and diagonals—were better able to reduce or even overcome the RFT illusion, particularly at tilt angles most relevant to balance control in everyday environments.