Person-Centered Research Research Paper

Person-centered research does not need to be quantitative and can in certain situations be carried out by case-oriented research and by using a qualitative approach. Obviously, the study of the single individual, studied quantitatively using the p-technique, is in one way person-centered, but is not normally so according to the definition given above (since the focus is usually then on relationships between variables within the individual). P-technique and time-series analysis is also discussed elsewhere in the encyclopedia. The perspective given in this research paper is of carrying out quantitative person-centered research on a sample of persons aiming at explaining and understanding inter-individual and/or intra-individual differences.

A very short historical overview is given of the emergence of person-centered research, followed by a presentation of the theoretical foundation of the modern person-approach. Finally an overview of common methods for carrying out person-centered research is presented.

1. The Emergence Of Person-Centered Research

A major type of person-centered research is the typological approach by which individuals are categorized into different types. It has ancient roots and is already reflected in the classical categorization of individuals into four basic temperaments: sanguine, phlegmatic, melancholic, and choleric. The typological view can be regarded as a reflection of a view that people are functioning wholes and that there are only a limited number of typical ways in which a functioning system can be organized. From this perspective, finding and understanding these types are important scientific goals. Creating typologies is also a reflection of a basic tendency in man of categorizing encountered phenomena.

The typological approach has been, and is still, stronger in the life sciences (especially taxonomy in biology and diagnosis in medicine) than it is in the social and behavioral sciences (for an historical overview, see Misiak and Sexton 1966). Gangestad and Snyder (1985) discussed the importance of a categorical approach in personality research and pointed to the emergence of types, each sharing a common influence (like an infectious disease caused by a specific germ). The typologies presented by Jung, Kretschmer, Sheldon, and others are still of interest today and psychiatric diagnoses in the Kraepelin tradition are influential in clinical practice. There is also a concern with typological research in sociology, including, for instance, the search for ‘ideal types’ (Bailey 1994).

The word ‘typology’ has many meanings—see Cattell 1957 for an overview of 45 different meanings of the word. It has lead to much confusion and even to resentment. This resentment was often directed against the connotation of ‘innate’ implied in many earlier typologies and against their subjectivity. Waller and Meehl (1998) make a number of thoughtful distinctions with regard to type and typology, preferring the use of the terms taxon for, roughly speaking, a meaningful type, and taxometrics as a generic name for (their) procedures for finding taxa. It should be pointed out that neither of these limitations apply to modern typological research within the personapproach described.

One of the first proponents of a comparatively modern person-centered research strategy was William Stern who already in the beginning of the twentieth century discussed an approach in which individual patterns in many traits were the units of analysis. Other early proponents of the importance of considering persons as functioning wholes were Kurt Lewin and Gordon Allport. A systematic approach to the person-centered study of personality development was undertaken by Jack Block based on longitudinal data. He presented an empirically based typology of longitudinal personality types and was, as far as we know, the first one to use the term ‘person-approach’ (Block 1971). Examples of research looking for basic categories are given by the work of Lars Bergman and David Magnusson studying stability and change in patterns of extrinsic adjustment problems, by Sigrid Gustafson studying a psychopathy-linked pattern called aberrant self-promotion, by Lea Pulkkinen, studying personality styles in a developmental perspective, by Richard Robins, Oliver Johan and Avshalom Caspi. searching for personality types, and by Ed Seidman relating typical experiential neighborhood profiles to antisocial behavior.

Perhaps the most forceful modern development of person-centered research has taken place within the new developmental science (Cairns et al. 1996). There, a holistic-interactionistic paradigm has grown strong and a new type of person-centered research has emerged (Magnusson 1999). It is focused on the study of individual development and has been called ‘personapproach’ (for an overview, see Magnusson 1998). Its theoretical and research strategic fundaments are described in Sect. 3.

2. Theoretical And Research Strategic Fundaments Of The Person-Approach

The person-approach refers to a holistic view on individual functioning and development. Magnusson and Allen (1983) summarized the essence of a person- approach in the following way: ‘The person oriented approach to research (in contrast to the variable centered approach) takes a holistic and dynamic view; the person is conceptualized as an integrated totality rather than as a summation of variables’ (p. 372). A basic proposition of a holistic perspective, with consequences for the application of a person-approach, is that individual functioning and development can be described in terms of dynamic, complex, and adaptive processes in which mental, biological, and behavioral factors in the individual, and social, cultural, and physical factors in the environment are involved.

The person-approach is equally applicable to research on the current functioning of individuals and to research on individual development. Thus, it has implications for research in any field of psychological inquiry. Hitherto it has mainly been used in developmental and personality research.

For a long time the holistic view was considered to be too vague and lacking of specific content to serve as the basis for strong theories for understanding and explaining individual functioning and development. However, since 1970 the holistic model has been filled with content to the extent that it now forms a scientific basis for planning, implementation, and interpretation of empirical studies on specific aspects of individual functioning and development.

Substantively, contributions come from psychological research on mental factors and behavior and from biological and medical research on the functioning of the brain and the role of physiological factors in the total functioning and development of individuals. New findings in these areas have helped to enrich the ‘black box’ in S-R models with substantive contents and have helped in closing the gap between different explanations of behavior in terms of mental, biological, and environmental factors.

Contributions to the effectiveness of a holistic model as a theoretical framework for empirical research on specific phenomena, also derive from the presentation of modern models for dynamic, complex processes, particularly the general systems theory. Three aspects of these models are important for the discussion here of a person-approach.

(a) From a holistic perspective, mental, behavioral and biological aspects of individual functioning, and social, cultural, and physical factors of the environment are incorporated into one integrated theoretical framework. Thus, the role of a single variable cannot be finally investigated and understood in isolation from its context.

(b) A central principle in dynamic, complex processes is the principle of dynamic interaction (continuously ongoing reciprocal influences) as contrasted to statistical interactions in data. The new models for dynamic, complex processes provide a theoretical framework for investigating and understanding the dynamic processes of interaction of operating factors within the individual, and the continuous reciprocal interaction between the individual and the environment in the person-environment system.

(c) For the relations among operating factors at the level of the individual, mutual dependencies characterized by dynamic interactions and non-linearities can be a characteristic feature. The same holds true for the interaction of a single individual with his/her environment. For instance, individuals’ psychological and physiological stress reactions to increasing stimulation from the environment are often nonlinear. In fact, most common variable-oriented research methods are not suited for handling nonlinear relations and dynamic interactions. This is exemplified by the fact that many such methods use the correlation matrix as the data to be analyzed, a matrix which mainly reflects linear relations, not nonlinear relations and interactions.

The process of organization of current and developmental processes which takes its start at conception is guided by the principle of self-organization. Self-organizing ability is a characteristic of open systems and refers to a process by which structures emerge without ‘prescriptions from the outside.’ Within subsystems, the operating components organize themselves to maximize the functioning of that subsystem with respect to its purpose in the total system. At a higher level subsystems organize themselves in order to fulfill their role in the functioning of the totality. Within a specific system, say the cardiovascular system, each of the operating factors (e.g., systolic blood pressure, diastolic blood pressure, and heart rate) do not function and develop independently of the others. The specific role of each operating factor is determined by the role it plays in the system. The operating factors are organized and function in terms of functional configurations, in what will be referred to in the following text as patterns. From this perspective, the important individual differences are not to be found in differences in any single variable taken out of its context of other, simultaneously operating variables. Instead, they are to be found in differences in the patterning of operating variables in the system under investigation. This applies to all levels of organization.

For the discussion of the implications of the person approach for an effective measurement model and methodological approach, two aspects of the organization of mental, biological, and behavioral structures and processes are fundamental:

(a) Within subsystems individuals differ to some extent in the way in which operational factors are organized and function.

(b) Only a limited number of states are functional for each subsystem and for the totality.

The number of ways is restricted in which operating factors in a certain subsystem can be organized in patterns, in order to allow the subsystem to play its functional role in the totality, and the number of ways is also restricted in which subsystems can be organized to form the total pattern for the total organism (cf. Bergman and Magnusson 1997). This view implies, among other things, that the studied complex systems have inherent restrictions which lead to (a) certain states being frequent or typical and (b) other states not occurring. The first aspect has been much studied and provides a motivation for the search for types. The second aspect has been much less studied but can also be of importance: What does not occur but, in principle, could occur contains information about how the system under study functions (cf. the concept of ‘white spots’ as discussed by Bergman and Magnusson 1997).

As emphasized above, the person-approach is a theoretical perspective. As such it forms the theoretical framework for the planning, implementation, and interpretation of specific substantive issues. This implies, among other things, that it has to be distinguished from the methods applied for treatment of data within the perspective of a person-approach. However, in most cases, pattern-based methods emerge as the most natural method choices. A brief overview of a selection of such methods is given in Sect. 4.

3. A Selection Of Methods For Carrying Out Person-Centered Research

3.1 Classification And Cluster Analysis

It was mentioned in the introductory section that classification is a basic approach for carrying out person-centered research. Before discussing different types of classification analysis a few basic properties of such an analysis should be mentioned. In most forms of classification analysis the basic data are contained in the similarity or dissimilarity matrix between all pairs of subjects. Similar subjects are then grouped together in the classification analysis. The (dis)similarity between each pair of subjects could be a subjective rating but is most commonly calculated according to a formula that takes into account the (dis)similarity of the profiles of values in the variables for the two subjects. One common formula that takes into account differences in both level and form of the value profiles is the averaged squared Euclidean distance. It should be pointed out that the choice of (dis)similarity coefficient usually has important consequences for the results obtained in the classification analysis and that this choice should always be made on the basis of considerations in the specific case. The measurement characteristics of the different variables must be considered since the results of many methods are not invariant under linear transformations of the involved variables.

A classification can be achieved in diferent ways:

(a) It can be strictly theoretically derived as in the construction of ideal types.

(b) It can be totally empirically driven as in cluster analysis where subjects are classified together in clusters on the basis of their profile similarity. This type of classification approach is the most common one and is described.

(c) Some kind of model-based quantitative approach can also be used.

Examples of (c) are latent structure analysis and latent class analysis, originally proposed by Paul Lazarsfeld, where a number of latent classes is assumed and then the fit to the empirical data is tested and parameters are estimated (Goodman 1974). If the model holds, all relationships between variables within a latent class should disappear (the assumption of local independence).

3.1.1 Cluster Analysis. In cluster analysis, a large number of methods are available for classifying objects on the basis of their (dis)similarities. Major types of cluster analysis are hierarchical methods (agglomerative or divisive), partitioning methods, and methods that allow overlapping clusters. Within each type of methods a variety of specific methods and algorithms exist. Perhaps the most common form of analysis is the agglomerative hierarchical cluster analysis. This group of methods starts with each of the n subjects being its own cluster. In Step 1 the two most similar subjects are joined to form one cluster giving in all n-1 clusters. In Step 2 the two most similar clusters are joined to form one cluster, giving in all n-2 clusters. The process is repeated until every subject is in one and the same cluster that occurs at Step n-1. The result is a hierarchical classification tree. Often the researcher concentrates on one cluster solution with a fairly small number of clusters and each cluster is described by its profile of means in the variables studied (called the cluster centroid). All members in that cluster should have profiles similar to the centroid. Different specific methods of hierarchical agglomerative cluster analysis have different rules for how to decide which two clusters are most similar. For instance, in the single linkage ( = nearest neighbor) method the similarity between two clusters is given by the (dis)similarity of the two subjects, one from each of the two clusters, that are most similar. What method of cluster analysis is most appropriate of course, depends on the specific situation. Evaluations of the sensitivity of different clustering algorithms to the effects of errors of measurement and of the ability to recover a known cluster structure indicate that, as expected, no method appears to be generally superior to the others. Methods that often see to perform well include Ward’s minimum variance method and average linkage cluster analysis (two hierarchical methods), and k-means relocation analysis based on a reasonable start classification (Morey et al. 1983). More recently, methods based on so called Beta- flexible clustering have been suggested. For an overview of methods for cluster analysis see Gordon 1981.

3.1.2 Should All Subjects Be Classified? It has long been recognized that multivariate outliers may disturb the results of cluster analysis and it has been suggested that in some situations the coverage must be less than 100%, i.e., not everybody can be classified. Bergman (1988) drew attention to the fact that there often exist a small number of ‘unique’ individuals, not similar to any other subjects that should not be forced into a cluster. He presented a procedure, RESIDAN, for a priori identifying and analyzing separately what he calls a residue of unclassified subjects. In a subsequent residue analysis, rare or non-existent patterns are studied which may be of theoretical significance (cf. the discussion about ‘white spots’ in Sect. 3).

3.1.3 Classification In The Study Of Individual Development. Important and intriguing methodological problems for person-centered research occur when studying individual development. A variety of approaches are available.

(a) Directly analyzing longitudinal patterns of values in variable profiles have often proved disappointing. If only profile form is studied, a direct longitudinal classification might work better than otherwise, since the amount of information that has to be summarized by the cluster membership variable is drastically reduced. An example of the usefulness of a longitudinal classification strategy in this situation is given in Block’s (1971) study of longitudinal personality types.

(b) The analysis of cross-sectional patterns followed by linking over time is simpler and also often more robust than most other methods. One standard method of this type is called Linking of Clusters after Removal of a Residue (LICUR, Bergman 1998). LICUR is suitable in situations where both form and level of the profile is considered to be of importance. It includes procedures for removing a residue before the cluster analyses that are performed at each age separately and procedures for deciding the number of clusters. The results of the cluster analyses are linked between adjoining ages by cross-tabulating the classifications obtained at the two ages and it is tested for over frequented and under frequented cells (i.e., cluster membership combinations occurring more or less frequently than expected by chance). LICUR is applicable in a large number of settings, for instance when different variables are measured at the different measurement occasions and during periods of dramatic developmental shifts. Model-based alternatives to LICUR are provided by longitudinal extensions of latent class analysis, for instance latent transition analysis developed by Collins and Wugalter (1992).

3.2 Some Other Approaches For Person-Centered Analyses

3.2.1 Identifying A Set Of Typical Classes. Sometimes the interest is not in a complete classification but rather in identifying homogenous subgroups of subjects (‘dense points’ in a multivariate space) believed to indicate system states that are in some way ‘important.’ It is then believed that many subjects have entered a stable ‘functioning’ system state with a characteristic value pattern and that there are only a limited number of such states. Frequent typical patterns are regarded as indicators of such optimal states. One may also be interested in recognizing patterns that a priori are considered as ‘important.’ For overviews of various techniques relating to these issues see Young and Fu (1986).

3.2.2 Studying All Possible Patterns Using Configural Frequency Analysis. Configural Frequency Analysis (CFA) is a set of methods for studying all possible value patterns for a set of studied variables. The involved variables have to be discrete and are often dichotomized to make the number of value patterns to be examined manageable. CFA was originally suggested by Gustav Lienert. Lienert and his coworkers have developed CFA in a number of ways and in Germany it has become a research tradition. For a basic introduction see Krauth and Lienert (1982). A more recent overview is given by von Eye (1990) who also has contributed to various newer developments of CFA. The idea of types in a psychiatric setting was elaborated by Joachim Krauth who pointed to the importance of identifying combinations of variable values that emerge above chance level. His line of reasoning relates to the discussion about optimal systems states and designs as being more frequent than other states and designs. For further information on CFA, see von Eye (1990).

3.2.3 Analyzing Specific Properties Of Patterns. Instead of considering the complete value patterns in a pattern analysis, specific aspects of the patterns can be highlighted. Two such analyses are the following:

(a) In some cases the maximum and minimum scores of the subject’s profile of scores are at focus. The variable taking the maximum value and the variable taking the minimum value may be seen as the essential features of the profile with the other scores providing the background. Of course, this implies that the different variables are scaled in a way that allows for comparisons between variables.

(b) The degree to which a subject’s profile is even or uneven can be studied in different ways and an obvious measure of profile scatter is its variance around the mean of the scores in all variables constituting the profile.

3.2.4 Abstracting Group Memberships From Qualitative Information. A good example of this approach is the work done by Singer et al. in linking life histories to mental health outcomes. Richly detailed descriptions of individual lives form the input data for a procedure for discerning generalizable features of aggregates of multiple lives.

4. Some Final Comments

We believe that the theoretical perspective included in the person-approach presented above provides person-centered research with a useful metatheoretical framework. But to reiterate: a personapproach should not be confused with a methodological approach using pattern analysis although such methods often are natural within a holistic orientation.

Sometimes reality may not be continuous but may rather operate to produce more or less discrete types. Or put differently: only certain configurations of system states are in some way optimal and become stable and often observed. A discussion of this issue was given in Sect. 3. It can then be argued that in person-centered research, methods for patternbased analysis are often more useful than standard variable-based methods. Many common variablebased methods do not handle interactions well and in many situations pattern-based methods are more naturally used for this purpose. In fact, taking interactions seriously, such an approach, focusing on variables, tends to become very complicated. Lee Cronbach even made the metaphor of entering a hall of mirrors when one pursues such a goal.

It has sometimes been claimed that results from many methods used in person-centered research, for instance cluster analysis, are untrustworthy. In order to evaluate the validity of results from such studies two aspects should be discussed separately: (a) the technical aspects of the methods applied and (b) the appropriateness of the application of these methods in different settings. There is nothing wrong technically with any of the major methods used to carry out person-centered analysis. The problem arises when a method is inappropriately applied. Inconsistent results are often caused by the use of a clustering procedure that does not match the combined requirements of the problem under study and the data available. Important considerations for obtaining useful results from a cluster analysis are: (a) That the variable profile under study adequately summarizes the information Gestalt of interest; (b) that the values in the different variables are comparable and an appropriate (dis)similarity coefficient has been chosen; (c) that a sound clustering algorithm has been chosen; (d) that only variables are included in the value profile which have a reasonably high reliability and, if finding homogenous clusters is the focus of interest, the profile is constituted by only a limited number of variables; and (e) that in many cases not all subjects should be classified. A number of validation procedures are available for studying the results of a classification analysis.

Two issues that are sometimes confused in person centered research are the question about identifying generic classes (‘natural clusters’) and the question about ascribing the subjects in the sample to the appropriate class. It is a general experience that the second purpose tends to be more difficult to achieve. If the purpose is to identify typical value profiles that frequently emerge in different settings, one way of validating these typical profiles is to compare centroids between different samples split-halves. Those that replicate might be regarded as types. These typical profiles need not, of course, together comprise a total classificatory system that encompasses everybody in a specific sample. On the contrary, the usefulness of a partial typology should be stressed since it can often be a more realistic goal. For instance, methods for determining how well the (dis)similarity matrix is represented by the classification solution are sometimes erroneously used for evaluating to what extent the first purpose has been achieved. Such methods are mainly relevant when the study of individual class membership is at focus.

Applying variable-oriented methods it is often assumed that (the same) linear relations approximately hold over individuals and that the essential features of a multivariate data set are captured by, for instance, the correlation matrix. This enables the researcher to use modern powerful statistical methods to construct models of data that are testable within the confinements of these basic assumptions. However, as was pointed out in Sect. 3, in settings when these assumptions cannot be assumed to hold a person-centered approach emerges as an alternative. It is natural that the recognition of multivariate complexity and higherorder interactions that follows with this perspective also makes it extremely difficult to formulate a coherent testable model of the data. Of course, this does not mean that person-centered research needs to be explorative in its methods. Theoretical considerations will lead to expectations about, for instance, types and antitypes and about typical developmental streams which are testable using various methods.

Finally, during the last decade person-centered research has received increased attention. For this we believe there are sound motives, as explicated in the section about the person-approach. In some situations one can trace a disappointment in the meager outcome in the understanding of a studied process given by even sophisticated variable-oriented methods. Applied with good judgment, person-centered research may then offer a deeper insight in how a system works. This applies to systems at the level of the individual as well as at other levels.

Bibliography:

1. Bailey K D 1994 Typologies and Taxonomies. Sage, New York
2. Bergman L R 1988 You can’t classify all of the people all of the time. Multivariate Behavioral Research 23: 425–41
3. Bergman L R 1998 A pattern-oriented approach to studying individual development: Snapshots and processes. In: Cairns R B, Bergman L R, Kagan J (eds.) Methods and Models for Studying the Individual. Sage, Thousand Oaks, CA, pp. 83–121
4. Bergman L R, Magnusson D 1997 A person-oriented approach in research on developmental psychopathology. Development and Psychopathology 9: 291–319
5. Block J 1971 Lives Through Time. Bancroft Books, Berkeley, CA
6. Cairns R B, Elder G H Jr, Costello E J 1996 Developmental Science. Cambridge University Press, Cambridge
7. Cattell R B 1957 Personality and Motivation Structure and Measurement. World Book, New York
8. Collins L M, Wugalter S E 1992 Latent class models for stage sequential dynamic latent variables. Multivariate Behavioral Research 27: 131–57
9. Eye A von 1990 Introduction to Configural Frequency Analysis. The Search for Types and Antitypes in Cross-Classifications. Cambridge University Press, Cambridge, UK
10. Gangestad S, Snyder M 1985 To carve nature at its joints: On the existence of discrete classes in personality. Psychological Review 92: 317–49
11. Goodman L A 1974 Exploratory latent structure analysis using both identifiable and unidentifiable models. Biometrika 61: 215–31
12. Gordon A D 1981 Classification: Methods for the Exploratory Analysis of Multivariate Data. Chapman and Hall, London
13. Krauth J, Lienert G A 1982 Fundamentals and modifications of configural frequency analysis (CFA). Interdisciplinaria 3, Issue 1
14. Magnusson D 1998 The logic and implications of a person approach. In: Cairns R B, Bergman L R, Kagan J (eds.) Methods and Models for Studying the Individual. Sage, Thousand Oaks, CA, pp. 33–63
15. Magnusson D 1999 Holistic interactionism—a perspective for research on personality development. In: Pervin L A, John O P (eds.) Handbook of Personality, 2nd edn. Guilford Press, New York, pp. 219–47
16. Magnusson D, Allen V L 1983 Implications and applications of an interactional perspective for human development. In: Magnusson D, Allen V L (eds.) Human Development: An Interactional Perspective. Academic Press, New York, pp. 369–87
17. Misiak H, Sexton V 1966 History of Psychology. Grune & Stratton, New York
18. Morey L C, Blashfield R K, Skinner H A 1983 A comparison of cluster analysis techniques within a sequential validation framework. Multivariate Behavioral Research 18: 309–29
19. Waller N G, Meehl P E 1998 Multivariate Taxometric Procedures. Distinguishing Types from Continua. Sage, Thousand Oaks, CA
20. Young T Y, Fu K S 1986 Handbook of Pattern Recognition and Image Processing. Academic Press, Orlando, FL