Printer Friendly

"A picture is worth a thousand words:" perceptual-personality profiling via a free-response imagery task.

The idea that patterns in writing style reveal an individual's personality structure or even hidden affect, imagery, ideation, or perception, has a checkered history in social science. For example, graphology has been controversial for more than a century but continues to be taught in some academic settings and advocated by various clinicians and personnel managers. While handwriting analysis can aid in the diagnosis and tracking of diseases of the brain and nervous system (e.g., Kopp et al., 1970; Ludewig et al., 1992), its ability to assess personality and job performance is primarily rooted in anecdotal evidence as opposed to rigorous objective testing, which has consistently shown negative results (e.g., Driver et al., 1996; King & Koehler, 2000; Neter & Ben-Shakhar, 1989; Schmidt & Hunter, 1998; for a review see Beyerstein & Beyerstein, 1992).

Content analysis of text, on the other hand, has been successful enough to have reasonable applicability. Starting in the 1940s, "projective" tests, including the classic Rorschach inkblots and word association tasks, have come into common and large-scale use (e.g., Harrower-Erickson, 1945). Projective tests have their origins in psychoanalytic psychology, which assumes that we have attitudes and motivations that are beyond or hidden from conscious awareness. Advocates of projective tests stress that ambiguous stimuli allow respondents to express thoughts that originate on a deeper level than are tapped by text-based questions. And too, there is the idea that the use of imagery-based assessment helps to avoid potential errors and response biases associated with limitations of a respondent's verbal abilities or any attempts to fake responses due to social desirability or other motivations and pressures. One of the best-known content analysis approaches is the scoring of the "picture-story" exercises of Murray's Thematic Apperception Test (TAT), which involves the "coding" or "scoring" of verbal material (or "thought sampling") for content and style. It is widely used for research into dreams, fantasies, mate selection and career choice, and to a lesser degree to assess personality or thought disorders. In a forensic context, Statement Validity Analysis (SVA) is a systematic procedure to assess the credibility of memory reports and has been used with child witnesses for decades in Germany. SVA analyzes verbal reports using nineteen criteria which reflect qualitative and quantitative differences between credible and non-credible reports (for a detailed description, see Yuille, 1988). Trained raters showed good inter-rater reliability, and SVA yielded a mean classification accuracy of 78.3% (Porter & Yuille, 1996).

Content analyses like those described above have the disadvantage of being labor intensive, because defining appropriate stimuli and scoring the verbal or written responses is extremely time consuming. Our present study aimed to overcome these disadvantages through the use of Latent Semantic Analysis (LSA), as first proposed by Deerwester et al. (1990), and we present an overview of this approach below. Here we note that LSA has proved useful in a variety of contexts, including essay grading in educational contexts (Shermis & Burstein, 2003), as well as psychological applications aimed at inferring emotion, deception, charisma, and sentiment analysis from textual information (for a review, see e.g., Luyckx & Daelemans, 2008; Mairesse et al., 2007). It does so by creating a high dimensional "semantic space" where each factor represents vocabulary words occurring in a corpus of textual items--this space is then simplified using techniques similar to standard factor analysis, to be described in the following section. The power of LSA derives from the fact that such semantic spaces can be constructed using unsupervised learning, i.e., with minimal intervention by researchers. A disadvantage of LSA that it essentially creates a black box without a clearly defined interpretation.

We offer this study merely as a first effort and case study, in which we test the effectiveness of LSA in predicting a specific set of perceptual-personality variables (Transliminality, Paranormal Belief, Intuitive Thinking, and Big Five Personality Model) that should be pertinent to an imagery-based task given previous research showing that such variables influence people's interpretation of ambiguous stimuli (e.g., Houran & Lange, 2004; Lange & Houran, 2000; Lange, Thalbourne, Houran, & Storm, 2000).

Latent Semantic Analysis

LSA provides an effective method for knowledge induction and representation by automating the process for determining the similarity of meaning of words and passages of written text (for an overview, see Deerwester et al. 1990). Based on word and context co-occurrences LSA represents the words, sentences, and paragraphs used in it as points in a very high dimensional "semantic space." Singular value decomposition (SVD), a mathematical matrix decomposition technique closely akin to factor analysis, is then used to simplify this space. While SVD itself has long been known, its application became technically feasible only after the development of efficient computer algorithms capable of dealing with large data sets.

LSA revolves around the concepts of corpus, tokens, vector space, and model. A corpus is a collection of text units (e.g., electronic documents, typed answers) that are traditionally referred to as documents. Words inside documents that represent useful aspects of the answers' semantics are said to define a series of tokens. The Vector Space Model (VSM) approach represents the tokens combined with their frequencies, while word order is ignored. For instance, a document dealing with fishery might be represented by the vector:

[("boat,3),("fisherman",1), ("weather",2)],

indicating that the word "boat" occurred three times, "fisherman" once, and "weather" twice. It seems likely that the document contained other words as well, but these were omitted. VSM assumes that documents' meaning is captured by the space defined by the token vectors, such that similar answers will have similar vectors and that vector similarity reflects similarity in meaning.

Assuming that we have d documents and t relevant terms one can construct a matrix X whose rows 1, ..., i, ..., t represent a dictionary of selected terms (or "tokens"), and whose columns 1, ..., j, ... d represent the documents containing these terms. If one considers tokens as representing items and documents as representing the unit of observation, then X is a data matrix in which the roles of the rows and columns are reversed. Each entry Xj reflects the weighted frequency with which dictionary term i occurs in document j. SVD is used to write X as the product of three matrices: X = TSD', (1) where T has size t x d and D has size d x d, both with orthonormal columns (i.e., T' T= I and D' D = I), while S is a d x d diagonal matrix. The matrices T, D, and S are referred to as the left and right singular vectors and the diagonal matrix of singular values, respectively, and together these define X's Singular Value Decomposition (SVD). By convention, the diagonal elements of S are positive and ordered in decreasing magnitude. Note that T, S, and D play roles similar to W, T, and W' as used in the principal component decomposition of the covariance matrix Q in standard factor analysis (i.e., Q = WTW').

The utility of SVD lies in the fact that keeping just the k largest singular values (i.e., setting Sk+1 = Sk+2 = ...Sd = 0) yields a matrix X* approximates X with increasing precision for greater k. Doing so produces two useful results: First, X* is a matrix of rank k which is closest to X in a least-square sense. Second, since the corresponding columns in T and D will be multiplied by 0, they can simply be deleted, yielding T* and D* with sizes t x k and d x k, respectively. The optimal number of factors will vary between applications (Landaur, 2007; Letsche & Berry, 1997;) but the use of 100-500 factors is typical. The present research uses Rehdrek's (2010) versatile Python based Gensim software for all LSA and SVD related computations.

Throughout the following we use document vectors (X) as the basic unit of analysis. This relies on the fact that a vector of tokens can be processed based on known T and S through post-multiplication by the product [TS.sup.-1]. This yields a more manageable, but approximately equivalent, representation that can be used as input to the classification algorithms. While many predictive approaches are possible, including logistic regression and discriminant analysis (e.g., Hastie et al., 2009), we have found that Support Vector Machines (SVM) often are the preferred method of analysis (for a discussion, see Bennett & Campbell, 2000). Further, prediction often benefits from first transforming the raw frequencies using a log entropy weighting function, and this is the procedure followed here (see, e.g., Landauer et al. 2007).

METHOD

Respondents

Data derived from a large convenience sample of respondents (N = 588) who were members of a survey website and subsequently compensated to complete a lengthy online questionnaire. Of the 243 respondents with known age/sex ([M.sub.age] = 32.7 yrs., SD = 9.5, range = 1953 yrs.), there were 92 women and 151 men.

Measures

A set of three illustrations was used in the imagery-interpretation task: the Mona Lisa by Leonardo da Vinci, the Sphinx at Giza, and Rodin's sculpture of The Thinker. In addition to the images we administered four perceptual-personality instruments in counterbalanced order:

The Ten Item Personality Inventory (TIPI: Gosling et al., 2003) is a brief but valid measure of the Big Five dimensions (or Five-Factor model). We used this particular measure of the Big 5 factors for its conciseness in light of the fact that test fatigue is a major concern when respondents are asked to complete a long series of questionnaires, such as in the present study. The TIPI was designed to measure the broad domains of Conscientiousness, Agreeableness, Emotional Stability, Openness, and Extraversion with only two items per dimension scored on a dichotomous scale and using items at both the positive and negative poles. Consequently, some researchers have pointed out that alphas are misleading when calculated on scales with small numbers of items (Kline, 2000; Wood & Hampson, 2005).

In addition, the following additional questionnaires were answered by a subset of 260 respondents. Tobacyk's (1988) Revised Paranormal Belief Scale (RPBS) consists of 26 statements that are to be rated on seven-point Likert-type scales. Lange, Irwin et al. (2000) found that the RPBS comprised only two, moderately correlated Rasch-scaled belief subscales, representing different control issues. New Age Philosophy appears beneficial to individuals by instilling a greater sense of control over interpersonal and external events (e.g., belief in psi), whereas Traditional Paranormal Beliefs seem to be more culturally-transmitted and beneficial in maintaining societal control via a belief in magic, determinism, and a mechanistic view of the world. Scores on New Age Philosophy are more consistently and strongly related to the experience and report of anomalous experiences than Traditional Paranormal Belief scores. Several studies support the internal reliabilities and construct validities of these subscales (Houran, Thalbourne, & Ashe, 2000; Houran & Lange, 2001; Houran, Irwin, & Lange, 2001).

The Revised Transliminality Scale (RTS: Lange, Thalbourne et al., 2000) is a Rasch version of Thalbourne's (1998) original 29-item, true/false scale (Form B). Twelve items from the original scale are excluded from the scoring due to age and gender biases. However, the remaining seventeen test items constitute a unidimensional Rasch (1960/1980) scale. These 17 test items, which share a common underlying dimension, span seven domains: Hyperesthesia, (fleeting) Hypomanic or Manic Experience, Fantasy-Proneness, Absorption, Positive (and perhaps obsessional) Attitude Towards Dream Interpretation, Mystical Experience and Magical Thinking. The Rasch reliability of the RTS is .82 (Lange, Thalbourne et al., 2000) and the testretest reliability is .82 (p < .001) over seven weeks (Houran et al., 2003).

Finally, we administered a 17-item, Rasch scaled questionnaire on self-reported experiences of intuitions in workplace settings called the Business Intuitions Inventory (BII: Lange & Houran, 2010). Questions were designed by a panel consisting of social and industrialorganizational psychologists and professionals who claimed to have had frequent and accurate intuitions in their careers. Items cover content related to the phenomenology of intuitions identified by Sinclair and Ashkanasy (2005). The response format was a four-point Likert Scale anchored by "Disagree Completely" and "Agree Completely." Lange and Houran (2010) reported a Rasch reliability of .83, as well as initial validity data for the scale.

Rasch Scaling

Each of the instruments referred to above was subjected to separate Rasch scaling runs. Using Linacre's (2012) versatile Winsteps software, rating scale models were estimated in which the probability of each answer is modeled based on [P.sub.ijk], the probability that respondent i on item j gives rating k. According to this formulation, ln([P.sub.ij(k-1)]/[P.sub.ijk]) = [R.sub.i] - [T.sub.i] - [F.sub.k] (1) where [R.sub.i] denotes the intensity of respondent's i trait level, [T.sub.i] reflects the trait level addressed by item j, and [F.sub.k] indicates the point where rating k has the same likelihood as rating k-1, given R and T. In Eq. 1, R, T, and F are all expressed in the same metric, i.e., the log-odds (or, logit) of observing rating k rather than k-1. We note that binary answers, as used for the transliminality items, are a special case of this model.

The extent to which respondents' answers fit the above model is quantified by their Outfit. For each person i and item j, Eq. 1 allows the computation of a standardized residual [z.sub.ij], whose distribution is approximately unit normal. Assuming there are n items, the person (respondent) Outfit is defined as: Outfiti = S [z.sub.ij.sup.2] / (n-1) (2).

In other words, the Outfit index is a scaled [|.sup.2] statistic with a theoretical optimal value of 1.0. Values below 1.0 reflect answer patters that over-fit the model (i.e., a lack of noise), whereas noisy (idiosyncratic) answers result in Outfit values greater than 1.0. For details we refer to Wright and Masters (1982), but for our purposes it is sufficient to note that Outfit values allows us to order respondents in terms of the degree of noise in their answer patterns.

Answer Preprocessing

Standard to LSA applications, the text material to be analyzed needs "purification" (e.g., Berry & Browne, 2005) to determine which tokens should be used to best capture their semantic contents. Here, tokens are defined as series of letters and "|", encoded in all lower case letters, together with end-of-sentence markers (i.e., "!", "?", or ".") and word combinations (e.g., "well-defined"). Thus, a sentence like "The fisher fishes for crab." has tokens "the", "fisher"' "fishes", "for", "crab", and "#', where "#" denotes the end of the sentence. Some concepts are captured by a sequence of two or more words, rather than a single word (e.g., New York, morning-after pill). Therefore, all possible pairs of two adjacent tokens in the sentence, called bigrams, were added as well. Thus, in "The fisher fishes for crab." we add: "the|fisher", "fisher|fishes", "fishes|for", "for|crab", "crab|#'--where the non-letter "|" is used to ensure that any two words will have a unique bigram. The present corpus was deemed too small to also include sequences of more than two words.

To eliminate spelling errors, all answers were corrected manually using the dictionary in MSOffice. Also, words deemed to have limited impact in discriminating among concepts (so-called "stopwords") were removed, as were words that occurred either very frequently or very infrequently. To further reduce the number of different tokens, words were normalized down to their essential roots by a process also known as stemming (Landauer, 2005). For example, and depending on the exact stemming method being used, the words "fishing," "fished," and "fisher" would reduce to the root "fish." Stemming does not intend to find the morphological roots of words, but rather it assumes that words with a common stem will usually have strongly related meanings. Such stems need not be part of the language to be useful. For instance, the words "argue," "argued," and "argues" might reduce to the stem "argu." In this research we use the well-known Porter stemming algorithm as included in Rehdrek's (2010) Gensim software.

Classification using Support Vector Machines (SVM).

The SVM approach finds the optimal separating hyperplane between two classes of vectors by maximizing the margin between the classes' closest points (Hastie et al., 2009). Points on the wrong side of this hyperplane are weighted down to reduce their influence. When a linear separator cannot be found, transformations can be applied to create new spaces to increase separability. This approach can be formulated as a quadratic optimization problem that can be solved by standard techniques (Vapnik, 1998). We use the R interface to libsvm in package e1071, which implements the radial basis function kernel method to achieve optimal linear separation (Meyer, 2012).

The e1071 software also performs <<-fold cross-classification on the training data (e.g., Geisser, 1993). In this approach, (1/n) th of the data is randomly omitted during model-fitting, and the model is then applied to the remaining data to determine its' generalizability. Our experience indicates that using << = 7 for smaller samples is close to optimal. The software computes the accuracy for each of the n predictions separately, as well as the overall accuracy. In this context, accuracy is defined as the proportion of low respondents categorized as low and high respondents categorized as high. Thus, when the top and bottom half of the respondents (on some variable) are compared, successful generalizability requires proportions over 0.5. Although the 7 runs are not independent, an approximate test of statistical significance is obtained via the binomial distribution. We note that the Type I error falls below 5% whenever the accuracy in 6 or 7 out of 7 runs exceeds 0.5.

LSA Analyses

All LSA analyses reported here rely on Rehurek's (2010) Gensim software, which can be downloaded from http://radimrehurek. com/gensim/install.html. In addition to implementing the LSA computations described in the introduction, Gensim contains a general library for topic modeling, document indexing and similarity retrieval for large corpora of texts, as well as the Porter stemming algorithm.

RESULTS

Preliminaries

Combining the data for the Sphinx at Giza, Mona Lisa, and the Thinker, yielded a total of 2,352 typed answers. The purification rules outlined earlier identified 5,390 different tokens, among those were 3,818 bigrams, which defined 86,731 token occurrences. The Gensim software (Rehurek, 2010) was used to approximate the Singular Value Decomposition of the 2,352 x 5,390 sparse data matrix, by a 2,352 x 200 array of vectors. These vectors are extracted such that each subsequent vector explains less variation than the preceding one, as quantified by their associated singular values.

SVD essentially creates more compact and less noisy data, but--analogous to factor analysis without any rotation--without regard to structure. Hence, no serious attempts were made to interpret tokens' loadings across the 200 vectors. However, as is usual, the first few factors (about 10) reflect syntactical features, whereas later ones start to address the answers' semantics. For instance, the first five largest loadings on Factor 14 reflect that "happy" and "serious" (and all other words that reduce to these stems) are seen as opposites because their loadings have different signs: 0.27*happy--0.25*serious|#--0.24*serious + 0.21*#|happy + 0.19*happy|#.

Classification

Respondents were classified as either high or low on each of the four questionnaire-based variables (Big 5 personality profile, Transliminality, Intuitive Thinking, and Paranormal Belief), where low and high scores are defined in relation to these variables' respective median values. Respondents were also categorized as either high or low based on the Outfit of their responses to the Rasch model. Analogously, respondents' age (i.e., above/below the median) and sex were used as classifiers as well. Except for sex, in case of ties, the tied respondents were randomly classified as either high or low so as to obtain approximately equal group sizes. Using random sub-samples of data, we noticed that the major analyses described below did not show improvements when more than 40 LSA coordinates were employed. For this reason, all following analyses rely on a semantic space as defined by the first 40 dimensions.

Respondents' sex and age, as well as the classification of their questionnaire scores (low vs. high) were generated using the SVM method as mentioned above. To establish whether the fitted model generalizes, <<-fold cross-classification was used, with n = 7 runs. In Table 1, classification results are marked as statistically significant whenever at least 6 of the 7 runs yielded over 50% correct predictions. The base rate for sex is different because 63% of respondents are women.

The first column in Table 1 shows the success of SVM-based classifications of low vs. high levels of the variables listed in the margin using 40 LSA coordinates. It can be seen that respondents' impressions of the Mona Lisa, the Giza Sphinx, and Rodin's Thinker can be used to predict whether they score low vs. high on Age, Agreeableness, and Conscientiousness (p < .05). Similarly, when the Outfit of respondents' answers is used to define low vs. high categories, the 40 coordinates of their textual answers successfully predicted age, Extraversion, and Transliminality. Although the percentage of correct classifications does not exceed 65%, the percentages gain in importance by the fact that this reflects the findings' generalizability, rather than the values reached upon convergence on the training set.

DISCUSSION

This research explored the usefulness of computerized Latent Semantic Analysis of respondents' open-ended text in predicting those respondents' trait levels of several personality and perceptual-oriented variables. Specifically, we found that respondents' open-ended, written impressions of images of the Mona Lisa, the Giza Sphinx, and Rodin's Thinker could predict whether they scored low or high on Age, Agreeableness, and Conscientiousness (p < .05). Similarly, the Outfit of respondents' text significantly predicted the variables of respondents' Age, Extraversion, and Transliminality. Nevertheless, several constraints limit the generalizability of these modest results. For instance, the sample was not representative, and we only considered a selected few respondent traits that we expected to be salient in the present context given their established role in the interpretation of ambiguous stimuli (Houran & Lange, 2004; Lange & Houran, 2000; Lange et al., 2000). Most important to note was our use of "stemming" to reduce the size of the vocabulary, given the relative small sample on which to base the semantic space. Landauer and Dumais (http://www.scholarpedia. org/article/Latent_semantic_analysis) noted that "stemming often confabulates meaning, e.g., flower becomes flow." However, we believe that such instances are rare in our study as inspection of 10% of the data indicted no clear confabulations, while combining the different words into a single token was unobjectionable. Moreover, we argue that our stemming process benefitted the statistical quality of the token frequency estimations.

Notwithstanding these caveats, we argue that the present findings are reasonably promising for the most part and provide additional support for content analysis approaches in general, and more specifically, the core premise that free-response text to sets of standardized imagery can efficiently and significantly predict trait levels of at least some key variables relevant to studies of perceptual-personality and individual differences. That said, we obtained only weak to moderate effects, and obviously considerable work needs to be done in order to produce results of strong, practical application. Indeed, on a methodological level, refinement and validation of LSA for perceptual-personality profiling could offer many potential advantages over standard questionnaire approaches, including: Enhancing the internal validity of data by reducing any temptation to fake responses; Reducing reliance on a respondent's verbal abilities; Tapping both conscious and unconscious motivations; Reducing test fatigue for studies that require the measurement of many traits and characteristics that normally would require a multitude of separate assessments.

Of course, future work needs to build on these preliminary results in order to explain the nuances we observed in how certain features of the free-response text predicted the variables of Age, Agreeableness, Extroversion, Conscientiousness and Transliminality. This latter variable is especially curious to us, as Transliminality reflects the tendency for psychological material to cross (trans) thresholds (limines) into and out of consciousness. Reviews show that, in addition to paranormal belief and sometimes self-reported and laboratory-based paranormal experience (Thalbourne & Houran, 2003), the major correlates of transliminality are syncretic cognitions (Houran et al., 2006; Lange et al., 2000)--i.e., the fusion of perceptual qualities in subjective experience such as: physiognomic perception (the fusion of perception and feeling); synesthesia (the fusion of sensory modalities) and eidetic imagery (the fusion of imagery and perception). Accordingly, transliminality is conceptualized as enhanced interconnectedness between brain hemispheres, as well as among frontal cortical loops, temporal-limbic structures and primary or secondary sensory areas or sensory association cortices (Houran et al., 2006; Thalbourne et al., 2001; Thalbourne, Crawley & Houran, 2003).

Studies of perception, imagery and memory all provide evidence for a threshold that mediates unconscious-conscious awareness, and findings from several experiments are consistent with the neurological interconnectedness model of transliminality in particular (Crawley, French & Yesson, 2002; Fleck et al., 2008; Houran et al., 2006). The prospect that free-response text from imagery tasks may prove to be a reliable indicator of the level of such brain-based functionality is conceptually consistent with and may extend current uses of handwriting analysis in the diagnosis and tracking of diseases of the brain and nervous system (Kopp et al., 1970; Ludewig et al., 1992).

At the theoretical level, the results give further credence to an interactionist framework that emphasizes the links between the person and the environment (cf. Buss, 1987; Rentfrow & Zilca, 2010; Swann et al., 2002). According to this approach, personality and individual differences inherently encompass a mixture of self-attribution and stimulus-attribution variables, such that personality structure can be inferred indirectly from one's external (physical) activities and appearance and vice versa. For instance, Rentfrow and Gosling (2003) found that music preferences are related to basic personality characteristics, values, self-esteem, intelligence level, and even political orientation. Personality structure can also be inferred reliably from leisure activities like movie and reading preferences (Rentfrow, Goldberg, & Zilca, 2010). Interestingly, even people without formal training can form reasonably accurate impressions of people based solely on such cues. Our preliminary findings suggest that the interactionist framework can be extended and adapted in new and promising ways that push the boundaries of psychological assessment.

REFERENCES

Bayne, R., & O'Neill, F. (1988). Handwriting and personality: a test of some expert graphologists' judgments. Guidance and Assessment Review, 4, 1-3.

Bennett, K. P. & Campbell, C. (2000). Support vector machines: hype or hallelujah? SIGKDD Explorations, 2(2). http://www.acm.org/sigs/sigkdd/ explorations/issue2-2/bennett.pdf.

Ben-Shakar, G., Bar-Hillel, M., Blum, Y., Ben-Abba, E., & Flug, A. (1986). Can graphology predict occupational success? Two empirical studies and some methodological ruminations. Journal of Applied Psychology, 71, 645-653.

Berry, M. W., & Browne, M. (2005). Understanding search engines. Philadelphia, PA: SIAM Publications.

Beyerstein, B. L., & Beyerstein, D. F. (Eds) (1992). The write stuff: Evaluations of graphology, the study of handwriting analysis. Amherst, NY, US: Prometheus Books.

Deerwester, S., Dumais, S. T., Furnas, G. W., Landauer, T. K., & Harshman, R. (1990). Indexing By Latent Semantic Analysis. Journal of the American Society for Information Science, 41, 391-407.

Driver, R. H., Buckley, M. R., & Frink, D. D. (1996). Should we write off graphology? International Journal of Selection and Assessment, 4, 78-86.

John, O. P., Donahue, E. M., & Kentle, R. L. (1991). The Big Five InventoryVersions 4a and 54. Berkeley, CA: University of California, Berkeley, Institute of Personality and Social Research.

John, O. P., Naumann, L. P., & Soto, C. J. (2008). Paradigm shift to the integrative Big Five trait taxonomy: History, measurement, and conceptual issues. In O. P. John, R. W. Robins, & L. A. Pervin (Eds.), Handbook of personality: Theory and research (pp. 114-158). New York, NY: Guilford Press.

Foltz, P. W., & Dumais, S. T. (1992). Personalized information delivery: an analysis of information filtering methods. Communications of the ACM, 34(12), 51-60.

Furnham, A., & Gunter, B. (1987). Graphology and personality: another failure to validate graphological analysis. Personality and Individual Differences, 8, 433-435.

Geisser, S. (1993). Predictive Inference. New York Chapman and Hall.

Gosling, S. D., Rentfrow, P. J., & Swann, W. B., Jr. (2003). A very brief measure of the Big Five personality domains. Journal of Research in Personality, 37, 504-528.

Harrower-Erickson, M. (1945). Large scale Rorschach techniques: a manual for the group Rorschach and multiple choice test. Springfield, Illinois: Charles C Thomas Publisher, Ltd.

Hastie, T., Tibshirani, R., & Friedman, J. (2009). The elements of statistical learning: data mining, inference, and prediction. New York: Springer.

Houran, J., Hughes, L. F., Thalbourne, M. A., & Delin, P. S. (2006). Quasiexperimental study of transliminality, vibrotactile thresholds and performance speed. Australian Journal of Parapsychology, 6, 54-80.

Houran, J., Irwin, H. J., & Lange, R. (2001). Clinical relevance of the two-factor Rasch version of the Revised Paranormal Belief Scale. Personality and Individual Differences, 31, 371-382.

Houran, J., & Lange, R. (2001). Support for the construct validity of the twofactor conceptualization of paranormal belief: a complement to Thalbourne. European Journal of Parapsychology, 16, 52-61.

Houran, J., & Lange, R. (2004). Redefining delusion based on studies of subjective paranormal ideation. Psychological Reports, 94, 501-513.

Houran, J., Thalbourne, M. A., & Ashe, D. D. (2000). Testing a psychoanthropological view of paranormal belief and experience. North American Journal of Psychology, 2, 127-138.

Houran, J., Thalbourne, M. A., & Lange, R. (2003). Methodological note: erratum and comment on the use of the Revised Transliminality Scale. Consciousness and Cognition, 12, 140-144.

Jennings, D. L., Amabile, T. M., & Ross, L. (1982), Informal covariation assessment: Data-based versus theory-based judgments. In D. Kahneman, P. Slovic, & A. Tversky (Eds.), Judgment under uncertainty: heuristics and biases (pp. 211-238). Cambridge, England: Cambridge University Press.

King, R. N., & Koehler D. J. (2000), Illusory correlations in graphological inference. Journal of Experimental Psychology: Applied, 6, 336-348.

Kline, P. (2000). Handbook of psychological testing. London, Routledge.

Kopp, W; Paulson, G; Allen, J; Smeltzer, F; Brown, F.; & Koss, W. (1970). Parkinson's disease: L-dopa treatment and handwriting areas, Current Therapeutic Research, 12, 15-124.

Landauer, T. K. (2007). LSA as a theory of meaning. In T. Landauer, D. McNamara, D. Simon, & W. Kintsch (Eds.) Handbook of latent semantic analysis (pp. 3-34). Mahwah, NJ Erlbaum Associates.

Landauer, T. K., McNamara, D., Simon, D., & Kintsch, W. (Eds.). Handbook of latent semantic analysis. Mahwah, NJ: Lawrence Erlbaum Associates.

Lange, R., & Houran, J. (2000). Modeling Maher's attribution theory of delusions as a cusp catastrophe. Nonlinear Dynamics, Psychology, and Life Sciences, 4, 235-254.

Lange, R., & Houran, J. (2010). A transliminal view of intuitions in the workplace. North American Journal of Psychology, 12, 501-516.

Lange, R., Irwin, H. J., & Houran, J. (2000). Top-down purification of Tobacyk's Revised Paranormal Belief Scale. Personality and Individual Differences, 29, 131-156.

Lange, R., Thalbourne, M.A., Houran, J., & Storm, L. (2000). The Revised Transliminality Scale: reliability and validity data from a top-down purification procedure. Consciousness and Cognition, 9, 591-617.

Letsche, T.A., & Berry, M. (1997). Large scale information retrieval with latent semantic indexing. Information Sciences--Applications, 100, 105-137.

Linacre, J. M. (2012). Facets computer program for many-facet Rasch measurement, version 3.70.0. Beaverton, Oregon: Winsteps.com.

Ludewig, R., Lewinson, T. S., & Lewinson, T. S. (1992). Profiles and limits of medical graphology: determination of current status and perspectives. Zeitschrift fur die gesamte innere Medizin und ihre Grenzgebiete, 47, 549-557.

Luyckx, K., & Daelemans, W. (2008). Personae: a corpus for author and personality prediction from text. In Proceedings of the 6th International Conference on Language Resources and Evaluation. Marrakech, Morocco: European Language Resources Association.

Mairesse, F., Walker, M. A., Mehl, M. R., & Moore, R. K. (2007). Using linguistic cues for the automatic recognition of personality in conversation and text. Journal of Artificial Intelligence Research, 30, 457-500.

Meyer, D. (2012). Support vector machines. The Interface to libsvm in package e1071. e1071 Vignette.

Meyer, G. J., & Kurtz, J. E. (2006). Advancing personality assessment terminology: time to retire "objective" and "projective" as personality test descriptors. Journal of Personality Assessment, 87, 223-225.

Neter, E., & Ben-Shakhar, G. (1989). The predictive validity of graphological influences: a meta-analytic approach. Personality and Individual Differences, 10, 737-745.

Porter, S., & Yuille, J. C. (1996). The language of deceit: an investigation of the verbal clues in the interrogation context. Law and Human Behavior, 20, 443-458.

Rasch, G. (1960/1980). Probabilistic models for some intelligence and attainment tests. Chicago, IL: MESA Press.

Rehurek, R. (2010). Software framework for topic modelling with large corpora. Proceedings of the LREC 2010 Workshop on new challenges for NLP frameworks, pp. 45-50, Valletta, Malta: ELRA

Rentfrow, P. J., Goldberg, L. R., & Zilca, R. (2010). Listening, watching, and reading: the structure and correlates of entertainment preferences. Journal of Personality, 79, 223-257.

Rentfrow, P. J., & Gosling, S. D. (2003). The do re mi's of everyday life: the structure and personality correlates of music preferences. Journal of Personality and Social Psychology, 84, 1236-1256.

Schmidt, F. L., & Hunter, J. E. (1998). The validity and utility of selection methods in personnel psychology: practical and theoretical implications of 85 years of research findings. Psychological Bulletin, 124, 262-274.

Shermis, M. D., & Burstein, J. (2003). Automated essay scoring: a cross disciplinary perspective. Mahwah, NJ: Lawrence Erlbaum Associates.

Sinclair, M., & Ashkanasy, N. M. (2005). Intuition: myth or decision-making tool? Management Learning, 36, 353-370.

Thalbourne, M. A. (1998). Transliminality: further correlates and a short measure. Journal of the American Society for Psychical Research, 92, 402419.

Thalbourne, M. A. (2000). Transliminality and creativity. Journal of Creative Behavior, 34, 193-202.

Thalbourne, M.A., & Houran, J. (2003). Transliminality as an index of the sheepgoat variable. European Journal of Parapsychology, 18, 3-14.

Tobacyk, J. J. (1988). A Revised Paranormal Belief Scale. Unpublished manuscript, Louisiana Tech University, Ruston, LA.

Vapnik, V. (1998). Statistical learning theory. New York: Wiley.

Wood, S. A., & Hampson, S. E. (2005). Measuring the Big Five with single items using a bipolar response scale. European Journal of Personality, 19, 373-390.

Wright, B. D., & Masters, G. N. (1982). Rating scale analysis. Chicago: MESA Press.

Yuille, J. C. (1988). The systematic assessment of children's testimony. Canadian Psychology, 29, 247-262.

Acknowledgements: This study was supported by a bursary (#38/12) awarded to the second author, for which we express our gratitude.

Rense Lange

ISLA--Instituto Politecnico de Gestao e Tecnologia

James Houran

Integrated Knowledge Systems

Author info: Correspondence should be sent to: James Houran, Ph.D., 2833 Duval Dr., Dallas, Texas 75211 USA Jim_houran@yahoo.com

TABLE 1: Prediction of Demographics and
Personality Variables

                        Trait     Person
                        Levels    Outfit
Demographics

Age                     60.8 *    60.1 *
Sex (63%=female)         62.8      62.1

Personality--Big 5     (n= 588)

Openness                 53.1      53.5
Neuroticism              52.4      50.1
Agreeableness           57.3 *     52.8
Conscientiousness       58.7 *     54.5
Extraversion             52.8     57.9 *

Additional Variables   (n= 260)

Transliminality         56.3 *    64.2 *
Intuitive Thinking       54.6      53.8
Paranormal Belief        49.5      47.3
COPYRIGHT 2015 North American Journal of Psychology
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2015 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Lange, Rense; Houran, James
Publication:North American Journal of Psychology
Article Type:Report
Geographic Code:1USA
Date:Dec 1, 2015
Words:5746
Previous Article:Self-blame attributions and cardiac symptom experiences in cardiac rehabilitation patients: a preliminary study.
Next Article:Sequential love priming as a compliance-gaining technique.
Topics:

Terms of use | Privacy policy | Copyright © 2021 Farlex, Inc. | Feedback | For webmasters |