Global and local perceptual style, field-independence, and central coherence: An attempt at concept validation.

Historically, the concepts of field-independence, closure flexibility, and weak central coherence have been used to denote a locally, rather globally, dominated perceptual style. To date, there has been little attempt to clarify the relationship between these constructs, or to examine the convergent validity of the various tasks purported to measure them. To address this, we administered 14 tasks that have been used to study visual perceptual styles to a group of 90 neuro-typical adults. The data were subjected to exploratory factor analysis. We found evidence for the existence of a narrowly defined weak central coherence (field-independence) factor that received loadings from only a few of the tasks used to operationalise this concept. This factor can most aptly be described as representing the ability to dis-embed a simple stimulus from a more complex array. The results suggest that future studies of perceptual styles should include tasks whose theoretical validity is empirically verified, as such validity cannot be established merely on the basis of a priori task analysis. Moreover, the use of multiple indices is required to capture the latent dimensions of perceptual styles reliably.


IntroductIon
The aim of the present study was to explore the factorial structure of visual perceptual styles. We did this by identifying tasks within the literature that are described as measuring perceptual style and carrying out exploratory factor analysis. The human visual system excels at object recognition: Objects within the visual scene are identified and perceived as wholes, even when the relevant perceptual data are incomplete. This is demonstrated by the ease with which we can identify familiar objects from incomplete line drawings (Street, 1931), or recognise faces from partially represented black and white forms (e.g., Mooney, 1957). Such abilities are said to be underpinned by a drive for perceptual closure. A related example of closure is seen in contour illusions (e.g., Kanizsa, 1974) in which, due to the organisation of local elements, boundaries and contours are perceived despite not being physically present. This tendency to group features together into a "good form" was identified as the basic law of perception by the Gestalt School of Psychology (the law of prägnanz) and highlighted that elements that are proximal to one another or that share a common property (shape, size, colour, orientation, movement in the same direction, etc.) are subject to perceptual grouping. So strong is the tendency to group visual features that it is often difficult to disambiguate constituent local features from a cluttered array. This is demonstrated by difficulty in tasks such as "spot the differences" and is especially effortful when the local features to be detected are embedded within a more complex figure, as in the Embedded Figures Test (Witkin, Oltman, Raskin, & Karp, 1971). Even when the local features are self contained and not embedded in the surrounding context, the perception of the global form still dominates (Navon, 1977).
Despite evidence suggesting a universal bias towards perceptual grouping, and a tendency to perceive the global before the local details, individual variation in the drive for global precedence is also evident.
Witkin et al. coined the terms field-dependence and field-independence (Witkin, Dyk, Faterson, Goodenough, & Karp, 1962) to classify such individual differences. An individual who is field-dependent is highly influenced by the context of the visual scene when processing features, whereas a person who is field-independent is more able to perceive an element independently from its context. The concept of field-(in) dependence was investigated with paradigms such as the Rod and Frame Test (Witkin & Asch, 1948) in which a rod is placed within a tilting frame and participants are required to adjust the rod so that it is upright. This task is accomplished most successfully by participants who can perceive the orientation of the rod independently from the angle of tilt of the frame, that is, participants who are able to resist cues from the surrounding context when making perceptual judgements.
Performance on the Rod and Frame Test correlates highly with that of the Embedded Figures Test (Witkin et al., 1962). The latter is still frequently used in research, and has a range of applications, for example to investigate perceptual style for the purposes of employee psychometric testing (Chapman & Calhoun, 2006), to investigate perceptual styles across cultures (Nisbett & Miyamoto, 2005), and in developmental research, especially in the field of atypical development and autism (Shah & Frith, 1983). Frith (1989Frith ( , 2003 coined the term weak central coherence to describe the clinical and experimental observation that individuals with autism often appear to ignore the (global) aspects of the visual scene that would be most salient to a typical observer, instead showing a tendency to focus on the smaller details, and a reduced ability to integrate material into appropriate context. Since the notion of central coherence was introduced, numerous studies have demonstrated weak central coherence in autism, although the range of tasks used to measure this is varied. For example, children with autism have been shown to succumb to the Gestalt principle of proximity significantly less than controls (Brosnan, Scott, Fox, & Pye, 2004) and to excel at the Embedded Figures Test (Shah & Frith, 1983) and the Block Design sub-test from the WAIS (Shah & Frith, 1993).
They also show superior ability to spot differences within two similar visual scenes (Teunisse, Cools, van Spaendonck, Aerts, & Berger, 2001), enhanced ability to detect targets within a visual search array (Jarrold, Gilchrist, & Bender, 2005;Plaisted, O'Riordan, & Baron-Cohen, 1998), a tendency to use a feature based, piecemeal strategy when copying the Rey-Osterrieth Complex Figure (Booth, Charlton, Hughes, & Happé, 2003;Ropar & Mitchell, 2001) and a superior ability to reproduce impossible, but not possible, figures (Mottron, Belleville, & Menard, 1999). There is also some evidence that children with autism may be less susceptible than typically developing children to geometric illusions, such as the Muller-Lyer illusion, and contour illusions such as those formed by the Kanizsa triangle (Happé, 1996, although see Mitchell, 1999, andMilne &Scope, 2008, for negative findings). The specific demands of the above tasks are wide ranging, however, and the exact nature of processes that can legitimately be subsumed under a single label of central coherence remains to be es-tablished. As the list of studies above illustrates, the term weak central coherence is often used to describe tasks that favour local over global processing styles, although this is an extension of the original concept. Witkin et al. (1962) reviewed a series of existing correlational and factor analytic studies, and concluded that field-independence was a narrow construct that refers specifically to the "ability to separate an item from its context" (p. 47). In other words, an item must be embedded within a structured context rather than merely being surrounded by amorphous material. This early research highlighted that field-independence is separate from the ability to identify an incomplete figure, as measured by Gestalt Completion tests (e.g., Street, 1931;Mooney, 1957).
Tasks requiring identification of incomplete figures were only weakly related to those that required dis-embedding, and loaded onto separate factors described as measuring "speed of closure" (Thurstone, 1944).
A unique perspective is given by Carroll (1993) who described the structure of human cognitive abilities on the basis of a comprehensive survey and re-analysis of available correlational datasets. The outcome of this analysis with respect to the visuo-spatial domain is summarized in Table 1. Theoretically, Carroll's position is consistent with Witkin's as he identified the factor of Closure Flexibility (defined as the speed of disembedding a known stimulus array from a more complex array) which broadly corresponds with Witkn's notion of field-independence. The operationalisation of the two constructs is somewhat different, however. Whereas Witkin et al. (1962)  In sum, the precise conceptual and operational definition of the construct of (weak) central coherence/field-(in)dependence, and its relationship to other dimensions of visual cognition remains unclear. While Carroll's meta-analysis confirms the existence of such a construct, it suggests a very narrow interpretation: Facility at dis-embedding a known stimulus array from a more complex array, labelled closure flexibility and measured primarily with the Embedded Figures Test. This is consistent with Witkin's definition of field-independence but narrower than the notion of weak central coherence which, in its research application, if not in Frith's original formulation, is used to describe a wide variety of tasks that represent a variety of distinct factors within Carroll's framework.
Surprisingly, despite the abundance of research on central coherence there has been little attempt to ascertain the degree to which the numerous tests that are currently used to investigate it really do measure the same construct. The research that is available finds little support for a unitary construct. For example,  carried out a principle components analysis of data collected from 70 children aged between 4 and 5 who performed the Embedded Figures Test, a test of pattern construction (similar to the Block Design subtest from the WAIS), a visuo-motor integration task that required participants to copy and maintain the configuration of a series of images, and a task that required participants to detect target shapes embedded within  (Booth, 2006 Factors whose existence was reasonably well confirmed through re-analysis of existing datasets.

Definition
Tasks loading highly on the factor Visualisation The ability to comprehend imaginary movements in a 3-dimensional space or the ability to manipulate objects in imagination.
Block Design and Object Assembly (WAIS) Block counting tasks Block rotation tasks Visuo-spatial perspective tasks Spatial Relation The ability to perceive spatial patterns or to maintain orientation with respect to objects in space.

Card Rotation Task Flags and Figure Rotation Closure Speed
The ability to combine disconnected, vague visual stimuli into a meaningful whole; to unify an apparently disparate perceptual field into a single concept.

Gestalt Completion Test Street Pictures Closure Test Incomplete Pictures Closure Flexibility
The manipulation of two configurations simultaneously or in succession. Speed of detecting and dis-embedding a known stimulus array from a more complex array.

Embedded Figures Test Hidden Patterns Test Copying Test Perceptual Speed
The ability to locate a unique item in a group of identical items. Finding, in a mass of distracting material, a given configuration which is borne in mind during the search.

Cancellation tests Finding "A"s Test Comparison tests
Factors whose existence and/or cognitive interpretation was less well confirmed Serial Perceptual Integration The ability to apprehend and identify a visual pattern when parts of the pattern are presented serially or successively at a high rate.  were also administered, however these did not correlate with each other and were not entered into the factor analysis (see Booth, 2006).
Other authors have reported the strength of correlations between different tasks used to measure perceptual styles, although these correlations are often presented as secondary to the primary hypothesis of the research, which is usually aimed at comparing performance of a developmentally delayed group against a control group. Table 2 presents a summary of these findings. Most, although not all, of this work stems from the field of developmental disorders, specifically autism research, therefore the table specifies whether the data are collected from a clinical or a neuro-typical population.
In general these studies suggest a modest degree of correlation between different tasks purported to measure central coherence, but they do not provide evidence for a single factorial structure. Furthermore, the scope of the analyses was limited as each study typically included only a small subset of tasks.
Since the aim of the present study was to explore the factorial structure of many tasks that have been used to measure either central coherence, field-(in)dependence or global-local perceptual style, we reviewed the literature and identified 14 tasks that are employed for this purpose. Based on this literature review and speculative task analysis, we defined these tasks as measuring the following constructs:

Participants
Ninety participants, 49 females and 41 males, were recruited to the study via posters displayed on the university campus and an e-mail that was sent to a list of registered volunteers. All participants were students: 60 undergraduates, 7 MPhil students, and 23 PhD students.
The average age of the participants was 21 years and 3 months with a standard deviation of 3 years. The exclusion criteria were: speaking a language other than English as a first language and/or being older than 30 or younger than 18. We recruited participants from a range of faculties across the university, the percentage of participants from each faculty was Arts, 21%; Engineering, 14.4%; Medicine, 9%; Law, 7.8%; Pure Science, 25.7%; and Social Science, 20.7%. Participants provided a history of any developmental disorder or existing condition that may affect their performance on the tasks (e.g., uncorrected visual impairment, motor problems, etc.). Four participants disclosed a diagnosis of dyslexia, one was red/green colour blind, and three reported having a lazy eye. These were noted in all cases but as they represent a cross section of the typical population were not considered grounds for exclusion from the study. 1

Experimental tasks: "Pen and Paper"
The Stimuli were line drawings of geometric patterns. Some of the patterns contained the embedded target configuration. Participants were required to mark, for each item, whether or not the target configuration occurred (see Figure 1). Following an untimed practise session of 10 stimuli, two parts of the test were given. In each part, participants were allowed 60 s to mark whether the target was present or absent in as many patterns as possible. The dependent variable was the number of correct responses given in both parts, out of a possible 200.

GesTalT compleTion TesT (From The educaTional TesTinG services kiT; eksTrom eT al., 1976)
Drawings composed of black patches representing parts of objects were presented and participants were asked to write down what each drawing depicted (see Figure 1). The experiment was presented in two parts. Following an un-timed practise session participants were given 120 seconds to identify as many objects as they could in each part. The dependent variable was the number of objects identified correctly in both parts, out of a possible 20.

vosp-silhoueTTes(WarrinGTon & James, 1991)
The stimuli of this task were drawings of 30 objects presented as silhouettes. Each participant had unlimited time to identify each object. The dependent variable was the number of objects correctly identified, out of a possible 30.

spoT The diFFerences TesT
This was a traditional spot the difference puzzle adapted specifically for this study. The stimuli originally appeared in a pre-1990 edition of a Polish popular weekly magazine called Przekrój. These images were chosen as they were considered to be sufficiently challenging for adults and were highly unlikely to have been seen previously by any of the participants. Two versions were given: a kitchen scene and a fishing scene (see Figure 1). Each scene was presented as a black and white line drawing reproduced twice on one piece of A4 paper. The participant was informed that the two pictures differed in a number of small details, and were asked to mark any differences they detected by putting a cross in the appropriate place on the right-hand side image. A 60 s time limit was given for each picture. The differences could concern placement of features, size of features, number of clustered features, orientation of features, and addition/subtraction of features. The order of scene presentation (kitchen or fishing) was counterbalanced between participants. The dependent variable was the total number of differences detected, out of a possible 39 and 34 for the kitchen and fishing scenes, respectively.

rey-osTerrieTh complex FiGure (rey FiGure)
Participants were presented with a laminated card showing the Rey figure (Osterrieth, 1944), given a blank piece of paper and a pencil, and asked to reproduce the figure. Following a break of 5 min during which they engaged in another, unrelated task, they were given a surprise recall test and asked to re-draw the figure on a new piece of paper from memory. Participants were filmed in both conditions.
The copies were scored for accuracy and strategy, recall was scored for accuracy only. Accuracy was scored according to Osterrieth's (1944) criteria, adapted by Taylor (1959; reproduced  A high score on this system therefore indicates a globally biased drawing style, whereas a low score indicates a more locally biased, piecemeal drawing style. The scoring system template is illustrated in Figure 1. In total, three dependent variables were obtained from this test: copy accuracy score, recall accuracy score, and strategy score.

Experimental tasks -computerised
The following computerised tasks were presented on a Viglen laptop computer, the screen of which was 1024 pixels wide (285 mm) and 768 pixels high (215 mm

hierarchical FiGures TesT (based on navon, 1977) stimuli
Hierarchical stimuli consisted of large "global" letters composed of smaller "local" letters. Target stimuli were either "H" or "S" and neutral letters were "X". The stimuli were compatible, neutral, or incompatible depending on the pairing of target and distractor stimuli and are detailed in Figure 2. All stimuli were black and were presented on a grey background (Michelson contrast = 76%). The global outline of the stimuli subtended 3.66° x 4.87°.

design and procedure
A selective attention design was used, whereby participants were instructed to indicate via a two-alternative choice key press whether the letter at the designated level was "H" or "S". A total of 144 trials were presented in 12 blocks. In half of the blocks the participant was instructed to identify the letter at the global level and in the other half the letter at the local level. Within each block the three different stimulus

ReseARch ARticle
Hidden Patterns Test 1 (fragment): Participants must indicate, underneath each stimulus, whether or not it contains the target (model).
Spot the Differences Test: Participants must detect as many differences between the two images as possible in 60 s.
Copying Test 1 (fragment): Participants must reproduce as many figures on the grid as possible in 90 s. VOSP Silhouettes Test 3: Participants must identify as many items from their silhouette as possible (maximum = 30, no time limit).
Example below is a bicycle.
Gestalt Completion Test 1 : Participants must identify as many types: compatible, neutral, or incompatible, and were presented randomly but equiprobably. To discourage the participant from looking at a fixed point on the screen where "local" letters always appeared, the stimulus appeared randomly either below or above a fixation point that remained on the screen for 500 ms. Each stimulus stayed on the screen for 150 ms, and was replaced by a pattern mask that remained until the participant made a response, following which the screen went blank for 500 ms before presentation of the next fixation point. Response time and accuracy were recorded.

muller-lyer line lenGTh illusion Task stimuli
Six stimulus pairs were created. Each consisted of one horizon-

design and procedure
Each stimulus pair was presented eight times (N = 48 trials) in random order. Prior to stimulus presentation a central fixation cross appeared for 500 ms. The stimulus remained on the screen until the participant made a response, following which the screen went blank for 500 ms before presentation of the next fixation cross. Response time and accuracy were recorded.

kanizsa illusory conTour Task (based on rinGach & shapley, 1996) stimuli
Stimuli were illusory rectangles induced by white "pac-man" figures presented on a black background. The dimensions of the stimuli were governed by the angle of pac-man rotation. In half of the images the pac-man figures were rotated to create the perception of a "fat" rectangle while in the other half they were rotated to create the perception of a "thin" rectangle. The degree of rotation was either 5º, 10º, or 15º from the horizontal midline, this resulted in a percept with varied degrees of "fatness" or "thinness". Participants were instructed to identify whether the shape was fat or thin via a two-alternative choice key press. The images were presented at different orientations: straight, rotated 45º to the left, or 45º to the right to prevent any participant using a strategy of ascertaining the shape of the induced rectangle by looking at the angle of orientation of one inducer only. Each stimulus subtended 5.48º × 8.52º. Control, non-illusory stimuli were created that were identical to the illusory stimuli apart from white line contours (2 pixels wide) that were drawn to highlight the rectangle. Michelson contrast of the stimuli was 87%.

design and procedure
Two separate blocks of trials were administered: the illusory block and the control block. The order of block presentation was counterbalanced between participants. In each block, the six stimuli were presented nine times giving a total of 54 trials per block. Each stimulus was equally likely to be presented upright, oriented to the left, or oriented to the right. Prior to stimulus presentation a central fixation cross appeared for 500 ms. The stimulus remained on the screen until the participant made a response, following which the screen went blank for 500 ms before presentation of the next fixation cross. Response time and accuracy were recorded.

visual search For a conJuncTive TarGeT (based on plaisTed eT al., 1998) stimuli
The stimuli consisted of an array of letters in a virtual grid, from which participants were instructed to search for a target amidst distractors. The target was a red "X", the distractors were red letters "T" and green letters "X". Each letter measured approximately 0.5° x 0.5° and the virtual grid subtended 20.3° x 20.3°.

design and procedure
Participants were instructed to press one of two keys to indicate whether the target was present or not. From a total of 60 trials, 30 contained the target. In each trial 5, 15, or 25 distractors were presented with equal probability but random selection. Prior to the array presentation, a fixation cross appeared on the screen for 500 ms, and disappeared once the stimulus appeared. The stimulus remained on the screen until the participant made a response, following which the screen went blank for 500 ms before presentation of the next fixation cross. Response time and accuracy were recorded.

impossible-possible FiGures TesT stimuli
The stimuli were figures used by von Karolyi, Winner, Gray, and Sherman (2003), and were adapted from the set of possible and impossible images developed by Schacter, Cooper, and Delany (1990).
These were geometrically possible (n = 12) or impossible (n = 11) black line drawings presented on a white background (Michelson contrast = 87%). Participants indicated via a two-alternative choice key press whether the presented stimulus was geometrically possible or impossible. Each stimulus subtended approximately 5.1° x 5.5°.

ReseARch ARticle design and procedure
Prior to the presentation of each figure, a fixation cross appeared on screen for 500 ms. One of the 23 figures then appeared at random and remained on the screen until a response was made, following which the screen went blank for 500 ms before the reappearance of the fixation cross. Response time and accuracy were recorded.

Good Form Task (adapTed From Williams & boloGna, 1985) stimuli
"Good form" experimental stimuli were pairs of vertical brackets that were designed to elicit perceptual closure. "Poor form" control stimuli were horizontal and vertical brackets that did not elicit perceptual closure (see Figure 2). In each block participants were instructed to sort the stimuli into two groups (arbitrarily classified as left or right) via a two-alternative choice key press. Theoretically this task could be successfully completed by selectively attending to the right bracket of each pair only. The experimental images subtended 1.2° x 1.8°, the control images subtended 3.65° x 1.8°.

design and procedure
In each trial, one bracket pair appeared on the screen; 144 trials were organised into two conditions (experimental good form and control poor form) and six separate blocks. Four out of six blocks were defined as "simple" as only one stimulus pair was associated with each response. In the simple blocks the irrelevant (left) bracket was predictable, that is it always faced the same way. The other two blocks were defined as "orthogonal" as two stimulus pairs were associated with each response. In these blocks the irrelevant left bracket was unpredictable, in other words, it could face either direction. Williams and Bologna (1985) found that reaction time to classify the stimuli was significantly longer in the orthogonal experimental blocks than in the simple experimental blocks or any control blocks. They interpreted this as a result of perceptual grouping, that is the greater the tendency for perceptual grouping the harder it was to selectively attend to the relevant right bracket (and ignore the irrelevant left) to the detriment of performance. Since the aim of this task was to test the drive for perceptual grouping rather than memory, visible prompts were available at all times to remind participants which stimulus pairs were associated with which response. The order of block administration was counterbalanced across participants. After instruction, a fixation cross appeared on the screen for 500 ms, this was replaced by the stimulus which remained on the screen until a response was made, following which, the screen went blank for 500 ms before presentation of the next stimulus. Response time and accuracy were recorded.

Additional psychometric variables choice rT
As most of the experimental tasks administered above required participants to make an alternative choice by pressing one of two keys as quickly as possible, this control task provided a baseline measure of choice reaction time.

stimuli
The stimuli were red and blue circles with a diameter of 5.36°.
They appeared in the centre of the computer screen against a white background (Michelson contrast = 58% and 52%, respectively).

design and procedure
The circles were presented in random order. Participants were instructed to indicate the colour of the stimulus via a two-alternative choice key press. Prior to the presentation of each stimulus a fixation cross appeared on the screen for 500 ms, the stimulus then appeared and remained until a response had been made, after which the screen went blank for 500 ms before the reappearance of the fixation cross. In both the motion and form coherence tasks participants were asked to identify the patch that contained the coherent signal via two alternative forced choice key press. Auditory feedback was given after each trial. Three sets of trials in each task were administered. The final threshold for each task was calculated as the average threshold of the three sets of trials. Signal coherence was varied by modifying the number of coherent elements within the target patch. Each set of trials started with signal at 75% coherence. Following a correct response, coherence decreased by 1.5 dB and following an incorrect response coherence increased by 0.5 dB (Kaernbach, 1991). Each set was terminated after

General inTellecTual abiliTy
IQ was assessed with the Wechsler Abbreviated Scales of Intelligence (WASI; Wechsler, 1999). This consists of four subtests: two verbal (Vocabulary and Similarities) and two performance (Block Design and Matrix Reasoning).The tasks were administered according to the manual. As the Block Design subtest is also an index of perceptual styles this was included as one of the experimental variables and IQ was derived from two subtests only (Vocabulary and Matrix Reasoning), using norms provided in the test manual.

General procedure
The study received ethical approval from university ethics sub-committees. All participants provided informed consent before taking part and received £10 for their participation. The tasks were administered during two sessions (at least one day apart) that lasted approximately 2 hours each. Four different schedules counterbalanced the order of task administration. Three participants did not return for the second session.

Preliminary analysis
Descriptive statistics for all tasks are presented in Appendix A. In order to establish whether the tasks used in this study produced within-task patterns of results that were consistent with those reported in published studies, preliminary analyses were carried out and are reported in Appendix B. In all cases median response times based on correct responses only are reported and analysed. There were some missing data points for some variables (reflected in the varying degrees of freedom). All of the tasks showed the expected pattern of results based on previously published studies (see Appendix B).

Selection of tasks for correlation and factor analyses
The following tasks were selected for factor analyses: (a) Block

Reliability analyses
Indices of reliability were computed for all measures entered into correlation and factor analyses, with the exception of the WASI IQ variables whose psychometric properties are well described in the literature.
Measures of split-half, parallel test, and internal consistency reliability were computed, as appropriate. For the Rey figure copying strategy, two indices were obtained: internal consistency (based on data from all participants, scored and agreed jointly by both authors) and inter-rater reliability (using data from 30 participants, scored independently by a person who was blind to the authors' scores). The results are presented in Table 3.

Relationships between tasks daTa preparaTion
Some data points were missing due to equipment failure, administrators' errors, or participants' failure to attend one of the assessment sessions. For most variables one to four data points were missing, which constituted 1.1 -4.4% of potentially available data. The only exceptions were Rey figure strategy and Visual Search (accuracy and reaction time) with 15 (16.7%) and 9 (10%) data points missing, respectively. The missing data points were replaced using expectation maximization (EM) method (Tabachnick & Fidell, 2001). The EM procedure included all cognitive variables and WASI raw scores. Little's MCAR test was carried out on all variables and was not significant (χ 2 ReseARch ARticle = 445.724, df = 452, p = .574) indicating that data can be assumed to be missing at random.
All variables that were entered into correlation and factor analyses were winsorised for outliers: All scores that were more than 2.33 standard deviations away from the mean (which, under normal distribution, corresponds to the top and bottom 1% of cases) were set to the value of 2.34 standard deviation away from the mean. After this treatment, no variables showed extreme departures from normality (defined as the absolute value skewness greater than 2 and/or the absolute value of kurtosis greater than 7; West, Finch, & Curran, 1995, as cited in Fabrigar, Maccallum, Wegener, & Strahan, 1999 Table 4 reports zero-order and partial correlations, controlling for estimated full scale IQ based on two subtests from the WASI, and choice RT, for all variables. It is apparent that the significant correlations appear mostly between non-computerised tasks. Also, the differences between zero-order and partial correlations are mostly negligible, suggesting the relationships between variables in the study are not mediated by general cognitive ability or choice reaction time.

FacTor analyses
The factor analyses reported below used the Alpha Factoring method of factor extraction. This method was chosen because it "considers the variables in the analysis to be a sample from the universe of potential variables" (SPSS, 2005). This was appropriate in our case, the "universe" being all potential variables measuring perceptual style.
Alpha factoring also maximises the reliability (internal consistency) of the extracted factors. This results in a conservative estimate of the proportion of total variance explained by the latent factors. Factors were rotated using the Direct Oblimin method.    Preliminary analyses examined the suitability of the data for factor analysis which followed the recommendations of Field (2005).

Reliability of the Measures Used in the study
Bartlett's test of sphericity was highly significant (χ2 = 514.4, df = 210, p < .001), indicating some relationships existed between variables, which makes the correlation matrix suitable for factor analysis.
Determinant of correlation matrix was .002 (well above the recommended minimum value of .00001) indicating that multicollinearity was not a problem. On the other hand the Kaiser-Meyer-Olkin measure of sampling adequacy was .653, which is above the minimum recommended value of .500 yet "mediocre" (Kaiser, 1974, as cited in Field, 2005. This indicates that the pattern of correlations was rela- Seven factors were extracted using Kaiser's criterion of eigenvalues greater than one. Together they explained 44.8% of total variance. Most communality values were lower than .50; that is the seven latent factors could typically account for less than half of the variance of individual variables. Only Block Design showed high communality values (above .80, see Table 5). The pattern matrix, representing coefficients of regression of variables on factors, is displayed in Table 6, while the structure matrix, representing factor loadings (coefficients of correlations between variables and factors), is displayed in Table 7.
We interpreted the seven factors as follows: Factor 1: Dis-embedding. This factor received six substantial (greater than .40) factor loadings, three of which (Block Design, Embedded Figures accuracy, and RT) were significant according to Stevens' (1992, p. 394) Carroll's (1993) Closure Flexibility factor, defined as "speed of detecting and dis-embedding a known stimulus array from a more complex array" (p. 341). The fifth variable, Spot the Differences, was not included in Carroll's analyses, but also appears to require dis-embedding. Only the sixth variable, Block Design (which was the highest loading Factor 1 variable), represents a different factor in Carroll's analysis, namely, Visualisation. Overall, we decided that the term disembedding offers the best description of the demands shared by the tasks loading on Factor 1, but we understand it as broadly equivalent to the concept of Closure Flexibility (Carroll, 1993), as well as weak central coherence (Frith, 1989), and field-independence (Witkin et al., 1962).
Of the seven factors identified in our analysis, the Dis-embedding factor was the only one approaching Stevens' (1992, p. 395) criteria for a reliable factor (four or more loadings higher than .60). The remaining six factors are not considered reliable, thus their interpretation must remain tentative.   Carroll (1993) identified tasks that require speed in searching for and finding or correctly comparing stimuli which can be arranged by pairs, columns, rows, or at random, as representing the factor of Perceptual Speed. This description seems to apply to our Visual Search task (where the target stimulus must be found quickly amongst the array of distractors), and Muller-Lyer task (where the rapid comparison of the length of two lines is required). Although the Hidden Patterns Test has been identified by Carroll, and in our own analysis, as representing Closure Flexibility (Dis-embedding), it also requires speeded search and comparison of stimuli, which may explain why it also loads equally strongly on Factor 7. All three variables mentioned above also appear to favour a local processing style.
While the last correlation is negative it represents a positive relationship: Good ability to dis-embed (Factor 1) scores are associated with good (accurate and/or fast) performance on the variables that load onto Factor 4 (primarily Kanizsa and Spot the Differences).

ReseARch ARticle
The first of these two correlations is negative, that is, higher perceptual speed/local bias is associated with reduced global bias.
Correlations between factor scores and the baseline/psychophysical variables (form and motion coherence thresholds, IQ, and choice RT) were generally weak (see Table 9). The only moderate (.30 or above) correlations were observed between choice RT and Factors 2 (Global Bias) and 4 (Kanizsa); high scores on those factors are associated with slower choice reaction times. IQ correlated weakly with Factors 1 (Dis-embedding) and 3 (unnamed). The Motion and Form Coherence tests were threshold tests, therefore a high score represents poor performance and a low score represents good performance. The negative correlations between Factor 1 (Dis-embedding) and these tests indicate that the individuals who are good at dis-embedding tend to be good at detecting both the coherent motion and coherent form signals. The positive correlation between Factor 4 (Kanizsa) and form coherence thresholds indicates that good performance on the tasks that load onto this factor is related to good sensitivity to coherent form. The positive correlation between motion coherence thresholds and Factor 5 (Integration) indicates that good integration is related to good sensitivity to coherent motion. Finally, the negative correlation between Factor 6 (Cognitive Flexibility) and Form Coherence indicates that high cognitive flexibility is related to high sensitivity to detect coherent form. Overall, however, there is no evidence that any of the seven factors identified in our analysis reflect primarily the low level efficacy of visual perception, speed of choice reaction, or general intelligence.
The preceding analyses suggested that the extracted factors represent mainly specific dimensions of visual perception, and are only weakly loaded with more general aspects of cognition (namely general intelligence or general speed of processing). However, in order to obtain the factorial structure of visual cognition that is independent from any influence of those general factors, another factor analysis was run to control for individual differences in IQ and choice RT. The second analysis was based on the matrix of standardised residuals, remaining after the scores of visual perception tests were regressed on general intelligence (WASI IQ based on two subtests) and general speed of processing (Choice RT) scores. The results were not substantially different to the factor analysis reported above and are presented in Appendix C.
Following the suggestion of one of the reviewers, we also analysed the data using the confirmatory factor analysis (CFA). Two models were tested against the data: (a) a single factor model, testing the prediction that all variables represent a single continuum of global-local perceptual bias; (b) a two factor model, testing the prediction of distinct "global" and "local" dimensions of visual perception. Individual variables were allocated to either "global" or "local" factors, depending on our analysis of the task demands. We classified the following Local accuracy and RT, Muller-Lyer accuracy and RT, Visual Search accuracy and RT, and Good Form RT) were classified as measuring local perceptual styles. The "global" and "local" factors were assumed to be correlated. The analyses were carried out using AMOS software.
For the first, single factor analysis, the CFA algorithm failed to converge at all; no solution was obtained. The second, two factor model provided a poor fit to the data (according to Blunch's, 2008, andByrne's, 2001, interpretation): The parameter estimates were not statistically significant, and the fit indices were unsatisfying, χ 2 (169) = 317.1, p < .001; CFI = .561; RMSEA = .099). Thus, it can be concluded that the model is considerably different from the data. While both models could be modified to improve their fit to the data (by removing certain variables from the analysis, and adding or deleting parameters), the basic fact remains: Neither a single factor nor a two factor model represent the data well -a conclusion consistent with the results of our exploratory factor analysis.

dIscussIon
The aim of this study was to investigate the relationships within a set of tasks that are commonly described in the literature as measuring (weak) central coherence/field-(in)dependence (Frith, 1989(Frith, , 2003Witkin et al., 1962) as well as related, but typically poorly defined, constructs of global and local perception. Many studies have investigated these constructs, primarily in the context of autism (e.g., see Happé & Frith, 2006), but also in dyslexia (e.g., Brosnan et al., 2002;von Karolyi et al., 2003) and typical adult cognition (see Carroll, 1993). However, there have been very few attempts to clarify the relationship between these constructs, or to validate the tasks purported to operationalise them. We investigated this issue by surveying the literature on (weak) central coherence/field-(in)dependence, global and local perception; identifying a set of visual only tasks that are used to measure these constructs, and, finally, measuring the strength and direction of the relationship between them in a group of typically functioning adults.
Our search for the relevant tasks was made harder by the conceptual and terminological inconsistencies apparent in the literature. We identified the following predominant (if sometimes implicit) assumptions. The terms (weak) central coherence and field-(in)dependence are synonymous and represent the tendency to dis-embed elements from the surrounding context, and to segment local details from the global configuration. Broadly speaking, individuals who show weak central coherence/field-independence could be considered as having a locally biased perceptual style. This is in contrast to those with a more globally biased perceptual style, that is people who are strongly influenced by the surrounding context and would be described as having strong central coherence/being field-dependent.
Despite the implicit assumption within the literature that weak central coherence/field-independence is equivalent to a locally biased perceptual style and strong central coherence/field-dependence is equivalent to a globally biased perceptual style, the direct relationship between these constructs has not been examined systematically. We hypothesised that three potential outcomes were possible from our exploratory factor analyses: robi ReseARch ARticle 1. All tasks would load on a single factor, representing a continuum of weak to strong central coherence (field-independent to field-dependent; local to global perceptual style). This factor would receive positive loading from tasks that are easier to complete for those who have a locally biased perceptual style, and negatively loaded with tasks that are easier to complete for those who have a globally biased perceptual style.
2. Local (field-independent) and global (field-dependent) tasks would load onto two separate and uncorrelated factors, indicating that local or global bias do not occur on a continuum but in fact represent independent dimensions of visual cognition.
3. The tasks would share little variance. Several different factors would emerge; they would represent some very narrow aspects of visual cognition, or be merely task-specific.
The first hypothesis is consistent with the assumptions we identified in the literature. However the outcome of our analyses was largely consistent with the third hypothesis. It revealed that the tasks share relatively little variance -contrary to what would be expected if they measured a single construct. The factor analyses identified as many as seven factors, only one of which could be considered reliable (Stevens, 1992). This reliable first factor, which we labelled "Dis-embedding", Factor 1 (Dis-embedding) also broadly replicated the Closure Flexibility factor identified in Carroll's survey, defined as "speed of detecting and dis-embedding a known stimulus array from a more complex array" (Carroll, 1993, p. 341 analysis it belonged to a separate factor of Visualisation, defined as "the ability to comprehend imaginary movements in a 3D space, or the ability to manipulate objects in imagination" (Carroll, 1993, pp. 315-316). However, since our battery included no other tasks, apart from Block Design, that met the definition of visualisation, and since the Visualisation and Closure Flexibility factors are hard to distinguish empirically (Carroll, 1993, pp. 338-339) our outcome is not necessarily at odds with Carroll's. Furthermore, the Block Design task, which was described very well by our seven factors (communality of over 80%), appears to have a multifactorial structure, as it loaded substantially and significantly onto two factors, and moderately on a further two.
While local perception is reasonably well represented by the tasks that load on Factor 1 (Dis-embedding) and to some degree Factor 7  Duncan, 1995) which suggests that Visual Search requires efficient integration of features. However, this is integration in a broader sense than outlined in the introduction. Tasks that we initially identified as requiring the integration or grouping of discrete elements (e.g., Good Form Task, Gestalt Completion Test, and Kanizsa task) did not load onto Factor 5, nor onto any discrete factor which could represent global grouping or Gestalt perception.
It is important to note that the pattern of factor loadings reported here is unlikely to reflect individual differences in either IQ or general speed of making choice reactions. These two variables showed generally weak correlations with performance individual perceptual style tasks (see Table 4) and the extracted factors (see Table   9). Moreover, the second factor analysis which specifically controlled for the effects of IQ and choice RT produced results very similar to the first. Whilst we cannot be sure we eliminated Spearman's g factor completely from our analysis, since only two tasks, Matrix Reasoning and Vocabulary, were used to measure it, we can be certain that this was not the main source of variance that was captured.
What we captured was much more specific to visual perception.
The pattern of correlations and factor loadings obtained in our analyses speaks against the idea of a single continuum from global to local bias, synonymous with the continuum of central coherence or that of field dependence-independence. Our data suggest that, instead, central coherence and field dependence-independence should be understood more narrowly, as the capacity for dis-embedding only, which is not related to capacity for integration, gestalt grouping, or global perception. This outcome is consistent with some autism studies, which also demonstrate that in autism, one's ability to dis-embed has relatively little bearing on performance on tasks that measure global perception (see Mottron, Dawson, Souliéres, Hubert, & Burack, 2006, who raise the point that enhanced local perception in autism does not necessarily imply reduced global perception in autism).  Table 9). The correlations between Motion or Form

Coherence Thresholds and Group Embedded Figures Test accuracy
were also negative (see Table 4). That is, the more field-independent the individual, the more able they were to integrate the target elements of either the motion or form signal.
In conclusion, the results of our exploratory factor analysis indicate that the 14 tasks we selected based on their use in the literature for measuring (weak) central coherence/field-(in)dependence or global/ local perceptual style do not measure a unitary construct. However, we did find evidence in favour of the existence of a relatively narrow factor that represents individual differences in the ability to dis-embed relevant visual stimuli -the construct that largely corresponds to the notion of weak central coherence/field-independence, and partly also to the concept of Closure Flexibility (Carroll, 1993). In contrast, global grouping as defined by the ability to pull detached elements into a coherent whole was not represented by a single factor. Indeed, both task analysis and the outcome of the factor analysis suggests that multiple processes are involved in perceptual integration. Given that there was no significant relationship between Factor 1 and Factors 2 and 5, our results suggest that dis-embedding (or weak central coherence or field-independence) does not predict reduced global perception.
While the conclusions reached here apply directly to the adult neuro-typical population only, they may have implications for the studies of cognitive development and developmental disorders, especially autism. This is definitely so if we assume that central coherence/field dependence is a general characteristic of human cognition, and that individuals with autism represent the tail end of the normal distribution of that characteristic; that is they are different from neuro-typical individuals in degree rather than kind. If this is the case, then studies of the autism population would be expected to reveal a similar pattern of correlations to the one observed here. An alternative possibility is that all individuals with autism (or a subgroup of individuals) are qualitatively different in their cognitive skills either because of some specific deficit, or enhancement (e.g., Caron et al. 2006 Correlation between copy and recall accuracy r(88) = .23, p < .05 Correlation between copying strategy and recall accuracy r(88) = .42, p < .05 Note. a Accuracy analyses are not presented due to the majority of participants performing at ceiling in these tasks.

ReseARch ARticle
Preliminary analyses established that the data were suitable for the fac- As in the previous analysis, Kaiser's criterion of eigenvalues greater than 1 resulted in extraction of 7 factors. Together they explained 44.6% of total variance. Most communality values were again lower than .50.
Only Block Design showed a high communality value (above .80; see Table C1). Coefficients of regression of variables on factors (pattern matrix) and factor loadings, that is coefficients of correlations between variables and factors (structure matrix) are presented in Tables C2 and   C3, respectively.
The factorial solution obtained for the correlation matrix of standardised residuals was very similar to that obtained previously for the correlation matrix of raw scores. Correlations between corresponding factors scores from both analyses (estimated using regression methods) were very high (r = .90 -.99). Note, however, that Factor 6 (Cognitive Flexibility) obtained in the previous analysis now appears as Factor 7 (and vice versa). Also the vector of Factor 7 (previously 6) is reversed, that is, high scores on this factor now represent low cognitive flexibility.