Conducting Person-Oriented Research

In this topical issue, four articles present state-of-the-art scholarship concerning person-oriented research. The articles, authored by the teams of Bergman and Andersson; Schmidt, Perels, and Schmitz; Mutz and Seeling; and Strohmeier, Wagner, Spiel, and von Eye, cover an interesting range of topics and methods. Before introducing readers to these articles, we present an introduction to concepts of person-oriented research, from a conceptual and philosophy of science perspective (an overview taking a broader perspective, including the relation of person-oriented research to holistic thinking, can be found in the Bergman and Andersson article).

Psychological person-oriented research is being discussed in two forms. One is labeled idiographic psychology (Molenaar, 2004; Molenaar & Campbell, 2009; von Eye, 2004), the other person-oriented psychology (Bergman & Magnusson, 1997; Bergman, von Eye, & Magnusson, 2006; von Eye & Bergman, 2003). The fundamental tenet of both approaches states that premature aggregation of data can result in conclusions that fail to do justice to the variability in populations. By implication, the application of aggregate-level results to individuals may severely misrepresent each and every case (this is known as the ecological fallacy; see Freedman, Klein, Ostland, & Roberts, 1998; Robinson, 1950).

Sterba and Bauer (2010) noted that person orientation can be decomposed as follows:

The authors also note that, so far, the development of person-oriented theory has made more progress than the development of person-oriented methodology, and discuss options to test hypotheses that are compatible with the tenets of person-oriented psychology. In discussions of Sterba and Bauer’s (2010) work, authors note that modern statistical techniques do enable researchers to test virtually all hypotheses that can be formulated based on person-oriented theory and tenets (Molenaar, 2010) and recommend that special attention be devoted to clear-cut definitions of the units and aggregates of analysis (Mun, Bates, & Vaschillo, 2010).

It is interesting to see that, both in the theoretical and the methodological discussion, the implicit assumption is made that the scales and measures that are used to describe individuals are universally valid. Without this assumption, the comparability of individuals may be limited, and aggregate scores may not be interpretable. In the present article, we discuss implications of this assumption. We begin with characteristics of person-oriented research and, then, discuss issues of comparability.

Criteria for Person-Oriented Research

Von Eye and Bogat (2006) proposed three criteria for person-oriented research. The first criterion to be met is that a sample is analyzed under the assumption that it was drawn from more than one population (see Bergman & Andersson, 2010; Bergman, Magnusson, & El-Khouri, 2003; Mutz & Seeling, 2010). This assumption is explicit and obvious when person characteristics such as gender or personality profile are used to discriminate among individuals, or when membership in groups of individuals, that is, subpopulations, is predicted using psychological variables. Similarly, this assumption is obvious when methods such as latent class analysis, finite mixture distribution decomposition, or cluster analysis are used to identify groups or subpopulations whose existence was hypothesized but not known or had not been established before. If, however, a sample represents an aggregate of several populations and one set of parameters is estimated based on this aggregate, it is uncertain whether these parameters represent any of the populations, and the usefulness of these parameters for theory building or in guiding the planning of intervention studies is doubtful.

The second criterion is that attempts be made to establish external validity of subpopulations. In the present context, external validity comes in two forms. First, if subpopulations were identified using search or decomposition methods, researchers need to establish the existence of these groupings in the space of variables not used to identify the groupings. It is of importance that classifications be consequential. For example, in intervention contexts, groupings such as nosological classifications are of importance in particular if they correspond to treatment strategies, behavioral differences, or personality profiles. The second form involves sample specificity. To make sure clustering or decomposition results are not flukes or suggest groupings that do not exist (von Eye & Bergman, 2003), it is strongly recommended that attempts be made to replicate findings in independent samples that were drawn based on the same criteria as the original sample.

The third criterion is that groups be interpreted based on theory. Clusters can be established using statistical criteria. The application of these criteria is a necessary condition that must be fulfilled before subpopulations can be considered existing. Equally important is the attempt to interpret such groupings with reference to substantive theory (Bergman et al., 2003). In brief, subpopulations cannot be considered established if, from a theoretical perspective, they fail to make sense.

As was mentioned in the introduction, these criteria as well as the methodological discussion of Sterba and Bauer (2010), Molenaar (2010), and Mun et al. (2010) presuppose that the scales, instruments, and measures that are used to take differences in individual profiles and patterns into account are equally meaningful in all subpopulations and for all individuals. In the next section, we discuss this presupposition.

Commensurability

In the context of developmental psychology, Schmidt (1977) noted that psychological growth processes can be depicted only if the scales and the factors used to describe the processes are constant over time, that is, if the measures and scales are commensurable. Commensurability is a necessary condition of von Eye and Bergman’s (2003)dimensional identity, according to which the characteristics of and the interrelations among variables are unchanged across the levels or categories of other variables such as time.

In a context of philosophy of science, Feyerabend (1975) uses the term (in)commensurability from a broader perspective. The author discussed the following three characteristics of (in)commensurability:

1. 1.
   Any interpretation of observations is based on theoretical assumptions. It is impossible to evaluate observations in a theory-free way. This aspect of commensurability corresponds with von Eye and Bogat’s (2006) third criterion for person-oriented research, that groups be interpreted based on theory.
2. 2.
   Paradigms differ in the assumptions they make about which scientific methods result in valid scientific knowledge. This aspect of commensurability can be used to describe the relationship between aggregate-level and person-oriented research. Neither obviates the other. However, person-oriented research is needed to counter the dangers of unjustified aggregation and ecological fallacy, and aggregate-level research is needed to counter the dangers of unjustified disaggregation.
3. 3.
   Paradigms are based on specific assumptions concerning the structure of their domain, which makes it hard to compare them. The introduction of a new paradigm often implies the introduction of new terms. Based on this proposition, Feyerabend (1975) states that researchers who take the perspectives of competing theories will have problems communicating with each other because their theoretical concepts differ. From our perspective, the agendas and theoretical concepts of aggregate-level and person-oriented research do not differ to the extent that they cannot be related to each other. However, each of these two approaches to social science phenomena helps researchers avoid problems with the respective other. Specifically, generalization is still the goal of person-oriented research, and group-specific statements are an integral part of aggregate-level research. However, generalization that is based on person-specific or idiographic parameters is less risky than generalization that is based on aggregating raw data (cf. Molenaar & Campbell’s, 2009, discussion of ergodicity theorems). In turn, if the existence of subpopulations cannot be established, the estimation of group-specific parameters may lead to unnecessarily complex description of behavioral phenomena or to the interpretation of differences as important that are not even significant.

From the perspective of discussing the characteristics of person-oriented research, Feyerabend’s (1975) concept of (in)commensurability can be used to come to a more general meaning of von Eye and Bergman’s (2003) concept of dimensional identity. This concept involves scale characteristics (as originally discussed by Schmidt, 1977), structural elements such as the ones represented by factorial invariance, as well as qualitative arguments concerning, for example, the applicability of concepts in particular populations. Feyerabend’s commensurability extends these ideas to the level of theories. One can ask whether aggregate-level, variable-oriented theory building is the same as person-oriented theory building.

Interestingly, there is a clear answer to this question. Aggregate-level, variable-oriented theory building is based on the assumption that the variables used to describe behavior and its development are valid and meaningful for everybody and all possible subpopulations. In developmental research, it is also assumed that these characteristics are time-invariant. For example, the concept of developmental losses is, from a variable-oriented research perspective, meaningfully applicable to individuals from all age brackets (cf. Gestsdóttir, Bowers, von Eye, Napolitano, & Lerner, in press). In contrast, from a person-oriented perspective, variables may not be meaningful in particular subpopulations or age brackets. To continue the example, if an adolescent has not experienced any developmental losses, his or her responses to questions that address developmental losses may be random, if they are given at all.

In one word, person-oriented research does not proceed from the assumption that the validity of concepts and variables is universal (see von Eye, 2009, 2010). Person-oriented research is open to the assumption that particular concepts exist in or apply to particular populations or even individuals only.

This distinction has important implications, both in scientific and in applied contexts. In scientific contexts, theories that are based on person-oriented research strategies may use terms that are specific to populations, age groups, location, or historical time. Therefore, comparisons of individuals need to incorporate statements about the validity of concepts and variables in specific contexts. In other words, individuals cannot be compared only based on their location on particular scales, but also based on the possibly changing structure of behavior domains, and also based on the existence of behavioral domains. Person-oriented research operates thus a step beyond the research performed in differential psychology. In the latter, the basic assumption is that everybody can be assigned a location on the scales used for comparison:

In the literature, systematic application of both, the person- and the variable-oriented research strategies, has been advocated. If researchers wish to combine these strategies, the sequence of analyses proposed by Bergman (1998; cf. Spiel, 1998) may be considered:

1. 1.
   Application of exploratory, variable-oriented analyses with the goal of identifying operating factors; although this step is largely exploratory in nature, it is often based on theoretical assumptions (Feyerabend, 1975; von Eye & Bogat, 2006).
2. 2.
   Application of exploratory, person-oriented analyses with the goal of identifying possibly existing subpopulations (von Eye & Bogat, 2006).
3. 3.
   Application of confirmatory person-oriented analyses of data from independent samples with the goal of testing theoretical assumptions.
4. 4.
   Linking theories and results from the different research strategies via application of confirmatory, variable-oriented analyses (see again, Feyerabend, 1975; Molenaar & Campbell, 2009).

The Articles in This Topical Issue

The present topical issue includes four articles. Each of these is implicitly based on the assumption of dimensional identity, and each of these covers new and innovative aspects of person-oriented research. The first of these articles, contributed by Bergman and Andersson, presents a well-balanced state-of-the-art overview of theoretical, methodological, and empirical aspects of the person-oriented approach to research in psychology. There are three highlights that make this treatise worth reading in particular. The first is the presentation of the person-oriented theoretical thinking itself. In contrast to other renderings of theoretical elements of person orientation (e.g., von Eye & Bogat, 2006), this presentation establishes the connection to holistic thinking and the holistic aspects of person orientation (see also Bergman, von Eye, & Magnusson, 2006). The second highlight is the discussion of theoretical elements of the variable-oriented approach. This discussion is well balanced in that it involves a review of the pros and cons of both the person- and the variable-oriented approaches instead of a priori favoring one approach over the other. The third highlight is the discussion of considerations that need to be performed when deciding whether one approach or the other is to be pursued (or both). This discussion of considerations carries over to the empirical example in which the attention control – activity level in early adolescence is studied in its implications for the development of criminal behavior in adolescence.

The second article in this topical issue was contributed by Schmidt, Perels, and Schmitz. The authors, well known for their work in time series analysis, ask questions concerning the combination of person-oriented and nomothetic approaches with time series analysis. The authors present analyses in which they apply both approaches in the evaluation of self-regulating training that had been taken by a sample of adolescents and their parents. The results of these analyses show two important aspects of the differences between person- and variable-oriented research. First, similar results can be achieved with both approaches. This can be predicted when ergodicity exists (Molenaar & Campbell, 2009). However, as soon as ergodicity is violated, one can expect that person-oriented analyses show that group-level trends do not apply to everybody. This is also shown very clearly in this article. One aspect that makes this article particularly interesting is that the authors answer both variable- and person-oriented questions using the same methods of analysis.

The third article in this topical issue, contributed by Mutz and Seeling, is novel in two important respects. First, based on the Brunswick Lens Model, which is known to have been predominantly applied in variable-oriented research, the authors propose combining variable- and person-oriented data analysis. In fact, the authors propose a new aspect to the variable- versus person-oriented approach debate. They suggest that, under certain conditions, the two approaches cannot be separated. The second novel aspect in this article is that person-oriented research is applied, for the first time, to judgment processing. Students and experts in forestry are compared in their ratings of timber quality. Comparing these subgroups of raters, the authors identify two judgment strategies that are differentially applied by the individuals from their two target groups.

Reiterating the idea that person- and variable-oriented strategies of data analysis can be applied to the same data set, Strohmeier, Wagner, Spiel, and von Eye study the cross-time and cross-setting stability of bullying. The authors ask questions concerning the effects of entering a new social peer group and the effects of an interruption in the contact with peers on the incidence rates of bullying in two adolescent samples. Like Schmidt and collaborators, Strohmeier and colleagues find plausible and interpretable results at the aggregate level. However, the authors also find that change patterns are not universal. Instead, they identify four groups of adolescents who show specific and unique patterns of change, thus qualifying the aggregate-level results. This article is unique in its rich application of variable- and person-oriented methods of analysis. The authors conclude that results from the two approaches can complement each other.

In sum, this topical issue both reviews concepts, methodology, and application of the person- versus variable-oriented approaches to psychological research and carries them further by adding new facets. Directions for future work are out lined. Scholars will need to discuss in more detail, from philosophy of science perspectives, the ramifications of person-oriented research strategies, from conceptual perspectives, the implications of (in)commensurability and the risk of committing an ecological fallacy, from methodological perspectives, the testability of hypotheses that are compatible with the person- and the variable-oriented viewpoints, and, from an application-oriented perspective, the implications of person-oriented research for causal statements and the derivation of intervention strategies.