Validating the Online Circle Test (OL-CT): cross-format equivalence, one-week reliability, and contextual stability in Japanese undergraduates

Abstract

Background

The Circle Test is a drawing-based task for assessing time perspective, but conventional paper-and-pencil administration requires substantial human resources for scoring and quantification, limiting its scalability for large or remote surveys to be conducted. We developed an online web-based circle test with automated scoring and evaluated its validity and reliability.

Methods

Thirty-six Japanese undergraduates completed both the Online Circle Test (OL-CT) and paper-and-pencil Circle Test (PP-CT) in a counterbalanced crossover design (Visit 1) and repeated the OL-CT after 7–9days either in the laboratory or at home (Visit 2). Continuous indices included log-transformed absolute and proportional areas, and categorical indices included temporal dominance and temporal relatedness. Analyses evaluated (a) cross-format equivalence between PP-CT and OL-CT, (b) one-week test–retest reliability of the OL-CT, and (c) contextual stability across laboratory versus home administration.

Results

Cross-format equivalence was moderate for log-transformed absolute areas (ICC = 0.73–0.77) but excellent for proportional areas (ICC = 0.89–0.90). The categorical agreement was high (88.9%), with no systematic category shift. One-week test-retest reliability was moderate-to-excellent for absolute areas (ICC = 0.69–0.89) and excellent for proportional areas (ICC = 0.93–0.95). Categorical stability was also high (temporal dominance: 91.7%, κ = 0.88; temporal relatedness: 86.0%, κ = 0.76). The OL-CT indices did not differ between the laboratory and home retesting contexts.

Conclusion

In this Japanese undergraduate sample, the OL-CT showed promising preliminary psychometric properties as a digital adaptation of the Circle Test, particularly when proportional indices are treated as the primary continuous outcomes. The findings support its potential use as a scalable tool for assessing time perspective across formats, over a one-week interval, and across testing contexts.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40359-026-04646-8.

Introduction

Time perspectives describe individuals’ views on the past, present, and future, which are relatively stable individual characteristics [1]. Recent reviews suggest that a balanced time perspective is consistently associated with higher subjective well-being, adaptive health behaviors, academic achievement, and lower psychopathology vulnerability [2, 3]. Therefore, a variety of methods have been developed to measure time perspective.

Time perspective has traditionally been assessed using questionnaire-based (self-report) and performance-based (projective) measures. Self-report instruments such as the Zimbardo Time Perspective Inventory (ZTPI) [4] and the Temporal Focus Scale [5] are easy to administer and psychometrically established. However, questionnaire methods are influenced by introspective ability and social desirability, whereas performance-based measures may capture aspects of temporal representation in a more behaviorally expressed format [6]. Performance-based measures have the potential to address these limitations; however, they are time-consuming to administer and are not well-suited for large-scale surveys.

A widely used performance-based measure is Cottle’s circle test [7], in which participants were asked to draw three circles representing the past, present, and future respectively. Traditionally, there are two indexes based on drawings namely temporal dominance, and temporal relatedness. Temporal dominance refers to the relative preeminence of one temporal zone, as indicated by the circle size. Temporal relatedness refers to the degree of perceived linkage among the three temporal zones. Because the Circle Test explores implicit aspects of time perspective [8], its utility has been demonstrated across diverse populations and outcomes [9, 10]. However, because the original scoring methods require considerable effort, they are unsuitable for large-scale surveys.

The use of digital technologies can be a useful way to address these challenges. Web-based online versions of the circle test allow participants to complete the task and automatically record geometric information, such as circle centers, radii, and final layouts [11], while other versions can also compute the absolute area [12]. However, these studies rely on manual judgment for scoring of temporal dominance and temporal relatedness, therefore scoring still requires considerable effort. In addition, these studies have not sufficiently established the equivalence between the original paper-based and online versions, the short-term test–retest reliability of the digital format, or its robustness across different testing contexts.

To overcome these challenges, we newly developed a web-based online circle test (OL-CT) that automatically scores temporal dominance and temporal relatedness. In addition, the system enables automated capture of both end-state geometric features of the three circles and process-level drawing data (e.g., latencies and kinematic traces); however, the present study focuses on the psychometric evaluation of the end-state indices. The OL-CT has at least three advantages. First, the administration and scoring can be conducted automatically. It requires no rater training and can relieve heavy burdens on human raters. Second, uniform instructions are provided by the browser and therefore reduce the variability caused by inter-rater differences and environmental inconsistencies. Third, OL-CT scores can be provided and used directly for the interpretation of results and further data analyses. Thus, the OL-CT is intended primarily as a digital version of Cottle’s Circle Test for researchers studying time perspective, especially in settings that require standardized administration across paper-and-pencil and online formats. However, the utility of the OL-CT remains constrained by the need to establish its psychometric properties. Drawing inspiration from contemporary digital validation frameworks [13], we prespecified two objectives with a strict analysis–reporting correspondence: Objective 1 (Validity)—to evaluate cross-format equivalence between the paper-and-pencil Circle Test (PP-CT) and the OL-CT for both continuous and categorical indices; Objective 2 (Reliability)—to estimate the one-week test–retest reliability of the OL-CT and assess its contextual stability across laboratory versus home administration.

Method

Study design

The study consisted of two visits. In visit 1, participants completed both the paper-and-pencil Circle Test (PP-CT) and the online Circle Test (OL-CT). They were randomly allocated to either the OL-CT-first or PP-CT-first order. Participants returned for Visit 2, 7–9 days after the first visit, to repeat the OL-CT for test–retest reliability analyses. To examine contextual stability, participants were assigned to the Visit 2 laboratory or home retest context after completing Visit (1) Specifically, after Visit 1, participants were stratified by sex and age (≤ 20 vs. >20 years) and then block-randomized to the two retest contexts. Block randomization was performed with a block size of 2 to keep the groups as balanced as possible within each stratum. The randomization sequence was computer-generated prior to data collection. Because the study was conducted by a single researcher, allocation concealment by an independent party was not feasible; however, to minimize selection bias, the pre-generated allocation sequence was stored in a separate file and accessed only when scheduling Visit (2) All other procedures (instructions, timing, and scoring pipelines) were identical across visits and order groups.

Study population

Participants were recruited over a two-week period in April 2025 through in-class announcements in psychology and general education courses, as well as the distribution of flyers on campus (convenience sampling). Because this study was designed as an initial psychometric validation study, the sample size was determined pragmatically based on feasible recruitment within the fixed study period rather than on a formal a priori power analysis. Accordingly, the findings should be interpreted as preliminary, particularly with respect to the precision of the ICC estimates. To contextualize sample-size adequacy for the ICC-based reliability analyses, we referred to the planning framework proposed by Walter and colleagues [14], which specifies the minimum acceptable ICC, expected ICC, significance level, statistical power, and number of repeated observations per participant. Under assumptions of two observations per participant, a minimum acceptable ICC of 0.70, an expected ICC of 0.88, a two-tailed α of 0.05, and 80% power, the required sample size would be 32 participants. The final sample of 36 therefore exceeded this benchmark for hypothesis testing of relatively large ICCs. However, this does not imply high precision for all ICC estimates, and the confidence intervals for some ICCs remained relatively wide. Interested individuals scheduled an appointment via an online registration form or email for Visit 1 and a second session approximately 7–9 days later (Visit 2). The second session was conducted either in the laboratory or at home, depending on the assigned context. After completing Visit 1, each participant received a snack valued at approximately ¥500 as an incentive. Eligibility criteria included: (1) being 18 years or older, (2) providing written informed consent, (3) understanding Japanese and being able to follow task instructions, and (4) no motor or visual limitations (corrected vision permitted) that would preclude drawing the circles, operating a mouse/trackpad, or viewing the on-screen materials.

Data collection

PP-CT task (Visit 1 only)

The PP-CT followed Cottle’s procedure [7] and administered on A4 landscape paper and the participants were asked to draw three circles representing past, present, and future. The left half of the sheet printed the Japanese instruction text identical in content to the online version, and, beneath it, a demographic panel with fields for sex, age, birth month/day (MMDD), and the last three digits of the mobile number (XXX) was provided. From these entries, an anonymous participant ID (MMDD-XXX) was formed and written on the sheet. The right half contained a 120*120 mm blank drawing frame printed in thin black lines, centered vertically with margins > 10 mm to the page edges.

Completed sheets were digitized using a flatbed scanner at 600 dpi, and images were processed with ImageJ (v1.54) [15] to extract calibrated radii, areas, centroids, and pairwise spatial indices (center-to-center distances and overlap areas) of the three circles. From these geometric features, we derived (a) continuous indices: absolute areas (mm²) and proportional areas (each circle’s area divided by the sum of the three), and (b) two qualitative typologies followed Cottle’s procedure [7]: temporal dominance and temporal relatedness.

Temporal dominance was assigned by comparing proportional areas: cases were labeled Past-, Present-, or Future-dominant when one time zone’s proportional area clearly exceeded the other two by a pre-specified margin. Drawings were labeled Balanced when all three shares were within the pre-specified tolerance of an equal one-third split (i.e., each circle’s proportional area deviated by less than 10% from the mean proportional area of the three circles).

Temporal relatedness was coded based on the pattern and magnitude of pairwise overlaps: Atomicity (no meaningful overlap among any pair), Continuity (exactly one substantive pairwise overlap forming a “chain” with the third circle separate), Integration/Projection (two or more substantive overlaps, including triple-overlap configurations or instances where one circle is substantially projected into another), and Other for atypical or ambiguous arrangements (e.g., tangencies, severe distortions, or labeling anomalies). Two independent raters, both trained in the study-specific scoring protocol before formal coding, coded all PP-CT drawings for both typologies while blinded to participant identity and any online outputs. Training consisted of reviewing the Cottle-based classification rules and the study’s operational definitions for temporal dominance and temporal relatedness, followed by calibration on practice cases to standardize coding decisions. Inter-rater agreement was 88.9% for dominance (κ = 0.84) and 100% for relatedness (κ = 1.00). When initial ratings differed, the two raters reviewed the case together with reference to the predefined scoring rules and reached a final consensus code through discussion.

OL-CT task (Visit 1 and Visit 2)

Participants completed the OL-CT independently in a browser-based application on laboratory-standardized laptops (Lenovo IdeaPad 16 Pro) with external mice. The program (https://reurl.cc/4LZNyK) proceeded as follows: (1) a full-screen 800*800px canvas with high-precision event logging; (2) demographic entry and automatic anonymous ID (MMDD-XXX); (3) two-point screen calibration (200/250px) using a ruler to compute pixel-to-mm, with > ± 5% ratio error triggering repeat; (4) brief guided practice (create/move/resize/toggle/delete); and (5) asked to draw three circles for Past, Present, and Future and labels each. The system records geometry, overlaps/proportions, dominance/relatedness, and kinematics (Fig. 1). (6) automatic export (static CSV, dynamic CSV, JPEG) and secure upload with ZIP fallback (A step-by-step description of the OL-CT program flow is provided in Appendix (Appendix 1).

Fig. 1.

User interface of the Online Circle Test (OL-CT) showing the drawing canvas and Japanese instructions

The OL-CT platform generated the same geometric feature set (radii, areas, centroids, and pairwise overlap areas), from which identical dominance and relatedness rules were algorithmically applied to assign categorical labels. For quality control, one trained human rater—blinded to the algorithm’s output—independently coded each OL-CT image using the same codebook. Human–algorithm concordance was perfect (κ = 1.00 for both dominance and relatedness), confirming the functional equivalence of the automated scoring to expert judgment.

Statistical methods

Analyses were organized to mirror the prespecified objectives: Objective 1 (Validity)—to evaluate cross-format equivalence between the paper-and-pencil Circle Test (PP-CT) and the Online Circle Test (OL-CT); Objective 2 (Reliability)—to examine one-week test–retest reliability of the OL-CT and assess contextual stability across laboratory versus home administration.

For all analyses, a significance level of α = 0.05 (two-tailed) was applied, and 95% confidence intervals (CIs) were reported where appropriate. In regard to quantitative analysis of the circle test, data preparation followed a two-pronged approach. First, the absolute areas (mm²) of the Past, Present, and Future circles, as well as their total area were extracted. Natural-log transformations were then applied to these area variables for inferential analyses in order to reduce positive skewness, stabilize variance, and better approximate the distributional assumptions of the parametric models used in this study [16, 17]. Accordingly, the absolute-area results reported in this manuscript should be interpreted on the variance-stabilized log scale, that is, as relative (multiplicative) rather than raw additive differences in drawing magnitude. The log-transformed total area was also used to index each participant’s overall drawing scale. Second, proportional areas for Past, Present, and Future were calculated dividing each circle’s area by the participant’s total area of the three circles within the same administration. Thus, the denominator was not the physical size of the paper frame or computer display, but the within-drawing total area, and the resulting values were treated as compositional shares of relative temporal salience. In regard to qualitative analysis of the circle test, based on Cottle’s original distinction [7]; temporal dominance profiles and temporal-relatedness configurations of the three circles were coded as categorical indices based on the spatial arrangement of the circles.

The statistical analyses in this study encompassed both continuous and categorical dimensions. For quantitative indices, including log-transformed absolute and proportional areas, intraclass correlation coefficients (ICC 3,1) were employed to evaluate cross-format equivalence and one-week test–retest reliability. Categorical indices—temporal dominance and relatedness—were summarized using frequency distributions and assessed for within-person stability across formats and time using McNemar–Bowker tests. Additionally, independent-samples t-tests served as randomization checks for the counterbalanced administration order, while contextual stability between the laboratory and home settings was evaluated through mean comparisons with 90% confidence intervals and chi-square tests of independence. To evaluate the validity and reliability of the Circle Test, we conducted the following analyses:

• Objective 1(a): To examine the validity of cross format equivalence between the paper and online version for quantitative outcomes (absolute and proportional areas; within subjects), data obtained from Visit 1 were utilized. Intraclass correlation coefficients (ICCs) were calculated using ICC (3,1) (two-way mixed effects, absolute agreement) with 95% confidence intervals. For categorical comparability (temporal dominance and relatedness), PP-CT classifications were cross-tabulated with OL-CT classifications and evaluated using the McNemar–Bowker test of symmetry for paired multi-category data.

• Objective 2(a) Test–retest reliability (within subjects) between Visit 1 and Visit 2 for quantitative indices was calculated using ICC(3,1) with 95% confidence intervals, and test–retest stability for categorical dominance and relatedness classifications was examined using McNemar–Bowker tests.

• Objective 2(b): To assess the contextual stability of the OL-CT (between subjects), we evaluated whether the administration environment influenced the outcomes by comparing the mean values and 90% confidence intervals (CIs) of continuous indices (absolute and proportional areas) between the laboratory and home settings at Visit 2. In accordance with equivalence testing standards (α = 0.05), overlapping 90% CIs and the absence of substantial mean differences were interpreted as evidence that the physical environment did not exert a systematic effect on the metrics. For categorical indices (dominance and relatedness), we employed chi-square tests of independence to compare the frequency distributions of classifications across the two contexts.

Results

Sample description

A total of 36 Japanese undergraduates participated (23 women, 13 men; M age = 20.78, SD = 2.15) and completed both visits without attrition. In Visit 1 (laboratory), participants were randomized to two order groups: OL-CT-first (PP-CT-second) (n = 18; 9 women, 9 men; M age = 21.56, SD = 2.55) and PP-CT-first (OL-CT-second) (n = 18; 14 women, 4 men; M age = 20.00, SD = 1.33). In Visit 2 (7–9 days later), the same participants were block-randomized by sex and age into laboratory (n = 18; 11 women; M age = 20.61, SD = 2.17) and home (n = 18; 12 women; M age = 20.94, SD = 2.18) retest contexts. All datasets met the prespecified quality-control criteria.

Objective 1a — cross-format equivalence

Order randomization checks

To evaluate whether the counterbalanced administration order introduced systematic differences in OL-CT performance, independent-samples t tests compared Visit 1 OL-CT indices between participants who completed PP-CT first (PP-first; n = 18) and those who completed OL-CT first (OL-first; n = 18). For log-transformed absolute areas, there were no significant order effects for the Past or Future circles, nor for the total OL-CT area (Past: t(34) = − 0.28, p = .781; Future: t(34) = − 0.40, p = .692; total: t(34) = − 1.25, p = .220). The log-transformed Present circle area was larger when OL-CT was administered first than when it followed PP-CT, t(34) = − 2.28, p = .029. For proportional indices, no order differences reached conventional significance (Past: t(34) = 0.75, p = .458; Present: t(34) = − 1.71, p = .096; Future: t(34) = 0.45, p = .654). Importantly, no significant order effects were observed for any proportional index, suggesting that the main conclusions regarding the more robust proportional outcomes were not materially affected by administration order.

To further examine the order effect for the Present circle, we computed within-person format-difference scores (OL-CT minus PP-CT) for the log-transformed Present area. An independent-samples t test on these change scores was not significant, t(34) = − 0.66, p = .515, indicating that the magnitude of the cross-format difference in Present area did not vary significantly by administration order. However, this result does not rule out the possibility that the order groups differed on unmeasured baseline characteristics that influenced Present circle size.

Cross-format agreement for continuous indices

Results for cross-format agreement are summarized in Table 1. Across log-transformed absolute areas, the ICCs (0.73–0.77) indicate moderate cross-version agreement: each significantly exceeded zero yet remained below the 0.75 benchmark typically considered “good” for single observations. When areas were expressed as proportions of total drawing space, agreement rose sharply; ICCs for Past (0.89), Present (0.90), and Future (0.89) all surpassed the “good” threshold and approached the 0.90 criterion for excellence, with narrow confidence intervals entirely above 0.75. Hence, while log-transformed absolute area metrics are reasonably consistent, proportional scores showed stronger cross-format agreement and appear to be the more robust continuous indices in the present sample. Therefore, subsequent analyses relied on proportional indices as the primary continuous outcomes.

Table 1.

Cross-version intraclass correlations for log-transformed absolute area and proportion scores

Measure	Time-zone	ICC(3,1)	95% CI	F(35, 35)	p
Log-transformed absolute area	Past	0.77	0.60 – 0.88	7.88	< 0.001
	Present	0.76	0.57 – 0.87	7.17	< 0.001
	Future	0.73	0.53 – 0.85	6.39	< 0.001
Proportion	Past	0.89	0.80 – 0.94	17.43	< 0.001
	Present	0.90	0.81 – 0.95	17.73	< 0.001
	Future	0.89	0.80 – 0.94	17.45	< 0.001

ICC intraclass correlation coefficient (single-measure, absolute agreement, two-way mixed); all dfs = 35

To complement the ICC-based evaluation of cross-format equivalence, Bland–Altman plots were examined for the continuous indices (Fig. 2). Panels A–C show agreement for the log-transformed absolute areas of the Past, Present, and Future circles, respectively. Panels D–F show agreement for the corresponding proportional indices. For each panel, the x-axis represents the mean of the PP-CT and OL-CT values, and the y-axis represents the difference between methods (OL-CT − PP-CT). Solid lines indicate the mean bias, and dashed lines indicate the 95% limits of agreement.

Fig. 2.

Bland–Altman plots for cross-format agreement between the paper-and-pencil Circle Test (PP-CT) and the Online Circle Test (OL-CT) for log-transformed absolute areas (A–C) and proportional areas (D–F)

For the log-transformed absolute areas, the OL-CT showed positive mean bias relative to the PP-CT for the Past (bias = 0.495), Present (bias = 0.473), and Future (bias = 0.427) circles, with wider limits of agreement overall. In contrast, the proportional indices showed minimal mean bias for the Past (bias = 0.011), Present (bias = − 0.002), and Future (bias = − 0.008) dimensions, together with narrower limits of agreement. Visual inspection did not suggest pronounced heteroscedasticity, but dispersion was clearly greater for the log-transformed absolute area measures than for the proportional indices. These plots support the interpretation that proportional indices are more robust to format-related variation than absolute area measures.

Cross-format agreement for categorical indices

A four-by-four McNemar–Bowker test showed no systematic shift in dominance categories between versions (χ²(3) = 2.00, p = .572). Agreement statistics are summarized in Table 2. Thirty-two of the 36 participants (88.9%) received the same label in PP-CT and OL-CT; the four discrepancies were evenly distributed across off-diagonal cells. The same pattern was observed for temporal-relatedness classifications (χ²(3) = 4.00, p = .261). Again, 88.9% of the cases were concordant, and the few mismatches showed no directional bias. The absence of significant asymmetry and the high exact match rate confirm that the OL-CT faithfully reproduces the qualitative profile structure of the PP-CT, extending the metric equivalence of continuous indices to categorical outcomes.

Table 2.

Cross-format agreement for categorical indices

Categorical Index	Agreement	McNemar–Bowker	df	p	Mismatches
Temporal Dominance	88.9%	2.00	3	0.572	4
Temporal Relatedness	88.9%	4.00	3	0.261	4

N = 36. Agreement (%) refers to the percentage of participants whose classification labels (e.g., Past Dominance, Balanced) were identical across both formats. The McNemar–Bowker test evaluates the symmetry of the disagreement; a non-significant p-value indicates no systematic bias in classification changes between formats

Objective 2a — one-week test–retest reliability

Test–retest reliability of continuous indices

Table 3 reports the pre-specified ICC(3,1) estimates (95% CIs) indexing one-week test–retest reliability for OL-CT continuous indices, separately for (a) log-transformed absolute circle areas and (b) proportional scores.

Table 3.

Week-to-week intraclass correlations for log-transformed absolute area and proportional scores

Measure	Time-zone	ICC(3,1)	95% CI	F	p
Log-transformed absolute area	Past	0.79	0.62 – 0.89	8.47	< 0.001
	Present	0.69	0.47 – 0.83	5.38	< 0.001
	Future	0.89	0.79 – 0.94	16.69	< 0.001
Proportion	Past	0.95	0.91 – 0.98	40.67	< 0.001
	Present	0.93	0.87 – 0.96	27.99	< 0.001
	Future	0.94	0.89 – 0.97	34.71	< 0.001

All values refer to the stability between Time 1 and Time 2

A clear pattern emerged in that the log-transformed absolute circle areas showed only moderate consistency, whereas the proportional areas displayed excellent reliability. More specifically, ICCs for the log-transformed absolute metrics ranged from 0.69 to 0.89, indicating that participants tended to redraw their circles at a somewhat different absolute scale after 7–9 days. However, the precision of these estimates was not uniform, and the 95% confidence interval for the Present absolute-area index was relatively wide (0.47–0.83), with a lower bound below 0.50, indicating limited precision for this particular estimate. In contrast, all proportional indices exceeded 0.90, with correspondingly large F ratios, indicating strong short-term consistency in the relative psychological space allotted to the past, present, and future over the 7–9-day retest interval. This dissociation replicates the cross-format findings from Visit 1 and suggests that proportional scores are the most stable continuous outcomes of the OL-CT within this short retest window.

To complement the ICC-based evaluation of one-week test–retest reliability, Bland–Altman plots were also examined for the continuous OL-CT indices (Fig. 2). Panels A–C show agreement for the log-transformed absolute areas of the Past, Present, and Future circles, respectively. Panels D–F show agreement for the corresponding proportional indices. For each panel, the x-axis represents the mean of the T1 and T2 values, and the y-axis represents the difference between time points (T2 − T1). Solid lines indicate the mean bias, and dashed lines indicate the 95% limits of agreement. For the log-transformed absolute areas, the mean differences were negative across the Past, Present, and Future circles (biases = − 0.334, − 0.348, and − 0.297, respectively), indicating that T2 values were, on average, slightly smaller than T1 values, with relatively broad limits of agreement. By contrast, the proportional indices showed minimal mean bias overall (Past: −0.001; Present: −0.003; Future: 0.003) and comparatively narrower limits of agreement. Visual inspection did not suggest marked heteroscedasticity, although dispersion was greater for the log-transformed absolute area measures than for the proportional indices. Overall, the Bland–Altman plots were consistent with the ICC results in indicating stronger week-to-week stability for the proportional indices than for the absolute area measures.

Fig. 3.

Bland–Altman plots for one-week test–retest agreement of the Online Circle Test (OL-CT) for log-transformed absolute areas (A–C) and proportional areas (D–F)

Test–retest stability of categorical indices

To assess whether participants retained the same qualitative time-orientation pattern after one week, we cross-tabulated temporal dominance and temporal relatedness labels obtained at Visit 1 with those obtained at Visit 2. As in Visit 1, symmetry was tested using a four-by-four McNemar–Bowker procedure, and overall agreement was indexed using weighted Cohen’s κ.

For temporal dominance, 33 of the 36 cases (91.7%) fell on the principal diagonal, indicating perfect label replication. Agreement was excellent (κ = 0.88, p < .001), and no systematic reclassification occurred (χ²(3) = 0.33, p = .56). In other words, whether a participant was Past, Present, Future, or Balanced-dominant at baseline, the same classification was overwhelmingly reproduced at retest.

A similarly robust pattern emerged for temporal relatedness. 86% of the sample received identical labels at both time points (κ = 0.76, p < .001). The McNemar–Bowker statistic again indicated symmetry (χ²(3) = 4.00, p = .41), signifying that no particular profile was disproportionately likely to change. Integration/Projection remained the modal category, while Atomicity, Continuity, and Other patterns were comparatively rare and stable.

Objective 2b — contextual stability across laboratory and home testing

Contextual effects on continuous indices

To examine whether the environment in which the task was performed influenced the continuous indices, we compared the results of participants who completed the second session (T2) at home (n = 18) with those who completed it in the laboratory (n = 18). Table 4 presents the descriptive statistics and 90% confidence intervals (CIs) for both the log-transformed absolute areas and proportions across the two environments.

Table 4.

Descriptive statistics and 90% confidence intervals for continuous indices by environment at T2

Variable	Home Mean (SD)	Home (90% CI)	Lab Mean (SD)	Lab (90% CI)
Log-transformed absolute area
Past	7.63 (0.55)	(7.41, 7.85)	7.71 (0.41)	(7.54, 7.88)
Present	7.64 (0.79)	(7.32, 7.96)	7.71 (0.48)	(7.51, 7.91)
Future	7.61 (0.89)	(7.24, 7.98)	7.61 (0.77)	(7.29, 7.93)
Total	8.86 (0.51)	(8.65, 9.07)	8.88 (0.33)	(8.75, 9.01)
Proportion
Past	0.342 (0.147)	(0.267,0.419)	0.343 (0.184)	(0.282,0.402)
Present	0.330 (0.116)	(0.267,0.391)	0.329 (0.150)	(0.282,0.378)
Future	0.329 (0.185)	(0.256,0.402)	0.329 (0.177)	(0.253,0.405)

All values refer to the stability between Time 1 and Time 2

The analysis revealed that the 90% CIs for all log-transformed absolute area indices (Past, Present, Future, and Total) showed substantial overlap between the home and laboratory settings, with remarkably similar mean values. For instance, the log-transformed total area at home was 8.86 (90% CI 8.61, 9.12), whereas it was 8.88 (90% CI 8.72, 9.05) in the laboratory. More importantly, the proportional areas of the three circles were almost identical between the two groups, with their 90% CIs being nearly indistinguishable. Specifically, the mean proportion for the Future circle was 0.329 (90% CI 0.241, 0.417) in the home environment and 0.329 (90% CI 0.237, 0.421) in the laboratory. These results demonstrate that the physical environment did not exert any systematic effect on the continuous metrics of the OL-CT, further supporting the tool’s contextual stability and robustness for its remote administration.

Contextual effects on categorical indices

To examine whether the testing environment influenced the categorical outcomes of the OL-CT, we compared the frequency distributions of the Temporal Dominance and Temporal Relatedness types between the home (n = 18) and laboratory (n = 18) settings. Chi-square tests of independence were conducted to assess the association between the testing environment and categorical indices. For Temporal Dominance, the distribution of dominance types (Past, Present, Future, and Balanced) did not differ significantly between the home and laboratory conditions (χ²(3) = 0.34, p = .95). Similarly, for Temporal Relatedness, no significant difference was found in the distribution of relatedness patterns (Atomicity, Continuity, Integration/Projection, and Others) across the two environments (χ²(3) = 0.67, p = .88). The contingency tables showed highly consistent classification patterns across contexts. These results suggest that the categorical classifications yielded by OL-CT are robust to contextual variations and remain stable, whether the test is administered in a controlled laboratory setting or remotely at home.

Discussion

The present study provides preliminary psychometric evidence that the online Circle Test (OL-CT) may function as a promising digital adaptation of Cottle’s Circle Test [7], with the strongest and most consistent support observed for proportional indices in this Japanese undergraduate sample. Across two visits, the OL-CT showed moderate-to-excellent cross-format equivalence with the paper-and-pencil Circle Test (PP-CT), strong one-week stability, and no clear contextual differences between laboratory and home administrations. Taken together, these findings support that the OL-CT has potential as a scalable tool for time-perspective research, although replication in larger and more diverse samples is needed before broader generalization is warranted.

Psychometric properties: reliability, stability, and format equivalence

A central finding is that continuous indices derived from proportional (relative) circle areas show excellent psychometric performance, whereas absolute areas (even after log transformation) are comparatively less interchangeable across measurement occasions and formats. Specifically, the cross-format agreement between PP-CT and OL-CT for log-transformed absolute areas was moderate (ICCs = 0.73–0.77), whereas the proportional indices showed excellent agreement (ICCs = 0.89–0.90). Consistent with established guidelines for interpreting ICCs [18], this pattern suggests that the OL-CT preserves the relative allocation of psychological “time space” (past/present/future) particularly well, even if the absolute drawing scale varies across modalities.

The same dissociation emerged for one-week test–retest stability: log-transformed absolute areas showed moderate-to-good short-term reliability (ICCs = 0.69–0.89), whereas proportional indices showed excellentt short-term consistency over the 7–9-day interval (ICCs = 0.93–0.95). These findings indicate that proportional indices are especially robust across a one-week retest interval; however, longer-term stability remains to be established. Taken together, these results support a practical methodological recommendation: when the goal is comparability across formats, testing contexts, or short-term repeated assessments, proportional indices should be treated as the primary continuous outcomes, and absolute area metrics may remain useful when the research question explicitly concerns overall drawing magnitude or when measurement conditions are tightly standardized.

From a measurement perspective, the weaker stability/interchangeability of the absolute areas is not unexpected. The absolute area is inherently sensitive to motor constraints (mouse vs. pen), device ergonomics, and implicit scaling strategies, all of which can shift without necessarily altering the underlying construct (time perspective). In the present study, these absolute-area indices were analyzed on the log scale to reduce skewness and stabilize variance; therefore, the corresponding reliability and equivalence estimates should be interpreted as applying to variance-stabilized absolute-area measures rather than to raw additive mm² differences. More generally, similar geometric patterns may also be multiply realizable, in the sense that they can reflect different psychological meanings depending on a person’s cultural background, life stage, value system, or situational context [19–21]. In contrast, proportional indices normalize much of this variance, functioning as a within-person compositional representation of the temporal salience. This interpretation is reinforced by the observation that order effects were limited and primarily apparent in absolute indices, whereas proportional indices were comparatively unaffected by the administration order.

Categorical indices: preservation of qualitative profiles

Beyond continuous scores, the OL-CT also reproduced the qualitative profile structure of the Circle Test. Cross-format agreement for Temporal Dominance and Temporal Relatedness classifications was high (both 88.9%), and no systematic asymmetry was detected. These findings are particularly important because categorical classifications are often used to describe “types” of time perspectives in applied settings [7]. The high concordance suggests that OL-CT preserves the interpretive framework of the classical task, extending equivalence from metric indices to qualitatively interpretable profiles.

Test–retest results converged with this conclusion: Temporal Dominance showed 91.7% exact agreement and excellent weighted kappa, again without evidence of systematic reclassification over time. In practical terms, OL-CT appears capable of capturing stable individual differences in qualitative time-orientation patterns, supporting its use in longitudinal designs and individual-difference research.

Contextual stability: robustness across testing contexts

A further contribution of this study is the direct evidence that the testing environment did not systematically influence OL-CT outcomes. Neither continuous indices nor categorical classifications differed between the laboratory and home administrations. This supports the contextual stability of the OL-CT and strengthens the case for remote assessment, an especially relevant advantage for large-scale data collection and contexts where in-person testing is impractical. These results align with broader evidence that well-designed web-based cognitive and behavioral tasks can yield reliable data, provided that the sampling and procedural controls are adequate [22].

Broader implications for digital assessment

The OL-CT’s automation and scalability matter not only for convenience but also for measurement innovation. Digital assessment can enable standardized scoring pipelines, reduce examiner burden, and make large-scale and remote data collection feasible. In addition, digital formats create opportunities to capture process-level data (e.g., drawing order, latencies, corrections, and cursor trajectories), which may provide information beyond final geometric outcomes. Although the present study focused on classical Circle Test indices (areas and categorical profiles), future work can examine whether these process indicators offer additional insight into how individuals construct and revise temporal representations during task performance.

Another practical advantage is that a performance-based measure, such as the Circle Test, may be comparatively less vulnerable to certain response-style distortions that can affect self-reports, including social desirability pressure [6]. This does not imply immunity to bias, but it supports the rationale for using the OL-CT as a complementary tool in time-perspective research.

Limitations and future directions

Several limitations should guide the interpretation and motivate follow-up research. First, the sample size was modest and restricted to Japanese undergraduates, which limits both the precision and the generalizability of the present findings. Although several ICC estimates were encouraging, some confidence intervals were relatively wide, indicating limited precision and suggesting that the study was better suited to detecting larger effects than modest but potentially meaningful differences. Accordingly, the present findings should be interpreted as preliminary and should be replicated in larger and more diverse samples, including different age groups, cultural contexts, and clinical populations. In addition, the psychological meaning of a given circle configuration may not be one-to-one; similar temporal layouts may arise from different underlying value systems, cultural backgrounds, religious worldviews, developmental stages, or situational contexts. Accordingly, the interpretation of OL-CT profiles should remain context-sensitive, and future studies should examine the measure in more age-and culturally diverse samples. Second, while one-week stability was strong, especially for proportional indices, these estimates should be interpreted as evidence of short-term consistency over a 7–9-day interval rather than as evidence of trait-like stability. In this sense, the present test–retest findings represent a minimum estimate of temporal stability. Longer retest intervals are needed to determine whether the observed reliability is maintained over periods such as 4–6 weeks. Third, despite the null findings for environmental effects, future studies should systematically examine device variability (e.g., trackpad vs. mouse; screen size differences) and its impact on absolute versus proportional indices. Future studies should also compare raw-scale, transformed-scale, and robust analytic approaches in order to evaluate the sensitivity of absolute-area conclusions to distributional assumptions. Another limitation is that the present study focused on cross-format equivalence and short-term reliability, but did not examine external validity of the OL-CT. Accordingly, the validity of the OL-CT as a measure of time perspective in the present study remains primarily inferred from its similarity to the paper-and-pencil Circle Test. Future research should evaluate criterion-related validity using established time-perspective measures, as well as known-groups and outcome-based validity in more diverse and clinical samples. Finally, the present findings should be regarded as preliminary because the sample was modest and restricted to Japanese undergraduates. Replication in larger and more diverse samples will be necessary to establish the generalizability of the OL-CT across age groups, cultural contexts, and testing conditions.

Conclusion

Overall, the OL-CT showed promising preliminary psychometric properties as a digital adaptation of the classical Circle Test in this Japanese undergraduate sample, particularly for proportional indices. Proportional scores showed the strongest cross-format agreement and the strongest short-term reliability over one week, whereas the absolute area indices were comparatively less robust and should be interpreted more cautiously. Longer-term stability of the OL-CT remains to be established in future studies with extended retest intervals. These findings support the OL-CT as a scalable digital version of Cottle’s Circle Test for time-perspective research, while indicating that proportional indices may be the preferable continuous outcomes in future use. Replication in larger and more diverse samples will be necessary to establish the generalizability and precision of these findings.

Abbreviations

OL-CT

Online Circle Test

PP-CT

Paper-and-pencil Circle Test

Authors' contributions

Author contributions HH conducted the experiments, performed data collection and data analysis, and drafted the manuscript. AM supervised the overall revision of the manuscript, including its structure, data analysis, and the preparation of tables, and was responsible for the writing and checking of all sections. All authors read and approved the final manuscript.

Funding

Not applicable.

Data availability

The Online Circle Test（OL-CT）and datasets analysed during the current study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

Ethical approval for this study was obtained from the Chuo University Research Ethics Committee (approval number 2025-020; approved on 16 April 2025). All procedures involving human participants were performed in accordance with the ethical standards of the institutional research committee and with the Declaration of Helsinki and its later amendments or comparable ethical standards. Written informed consent was obtained from all participants prior to participation.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.