Abstract
This article examines and appraises a novel approach for generating shared group constructs through aggregative analysis: the application of Honey’s aggregation procedure to repertory grid technique (RGT) data. Revisiting personal construct theory’s underlying premises and adopting a social constructivist epistemology, we argue that, while “implicit theories” of the world, elicited via RGT, are unique to individuals, the constructs on which they are founded may be shared collectively. Drawing on a study of workplace performance, we outline a protocol for this novel use of Honey’s approach, demonstrating how it can be utilized to generate shared constructs inductively to facilitate theory building. We argue that, unlike other grid aggregation processes, the approach does not compromise data granularity, offering a useful augmentation to traditional idiographic approaches examining individual-level constructs only. This approach appears especially suited to addressing complex and implicit topics, where individuals struggle to convey thoughts and ideas.
Introduction
Studying complex social or behavioral phenomena presents researchers with a dilemma. While individual participants’ data offer depth and granularity, the multilevel nature of associated issues often necessitates aggregation compromising individual-level detail (see Hodgkinson 1997b, 2002). In this article, we offer a novel use for Honey’s (1979a, 1979b) aggregation approach, examining and appraising its utility for generating shared group constructs while preserving individual-level granularity of repertory grid technique (RGT; Kelly 1955, 1963) data.
Since its inception, RGT has been used widely to elicit individual psychological constructs. Aggregative analyses of these data at group level have typically adopted a nomothetic perspective, relying on methods such as principal components or cluster analysis. Such methods risk losing the inherent complexity and individual perceptual richness in elicited data. In contrast, Honey (1979a, 1979b) offers a potential “hybrid” approach for data aggregation, capitalizing on strengths of both nomothetic and idiographic approaches. We commence by reappraising the theoretical foundations and epistemological assumptions of RGT and aggregation. Within this, we outline how Honey’s approach can be reconciled with Kelly’s original personal construct theory (PCT) through adopting a social constructivist epistemological position. Building on this, we offer a protocol for using RGT and Honey’s inductive grid aggregation to elicit shared constructs, illustrated with worked examples from a study of workplace performance. We appraise this novel use of Honey’s aggregation approach, with particular consideration of prioritizing depth versus data aggregation.
Reappraising RGT and Aggregation
RGT is grounded in Kelly’s (1955, 1963) PCT. PCT’s fundamental postulate states an individual’s processes are channeled psychologically by the way they anticipate events, their interpretation of associated information varying according to 11 underlying corollaries. Of these, the individuality and commonality corollaries are particularly pertinent to aggregative analysis. The former notes that persons differ in their construction of events; the latter that where one person’s construction is similar to another’s, underlying psychological processes are similar. Kelly (1963) argued individuals’ constructions of the world are abstract and personal, being subject to revision or replacement when tested against everyday reality using notions of similarity and difference. Constructs therefore develop and change as a consequence of individuals’ reflection on past and anticipation of future experiences. Formed by one relationship of similarity and one of difference, they are expressed through bipolar anchors such as unhappy–cheerful (Kelly 1955, 1963). The ways individuals speak about these anchors reveal the meanings they attach to a construct.
To enable construct elicitation, Kelly developed RGT, an idiographic technique allowing individuals to express their own subjective understandings of their social practices (Daniels et al. 1995). Originally developed within clinical psychology settings (Slater 1977), RGT is now used more widely, a recent bibliometric review noting 46% of empirical articles were from outside psychology, disciplines including health, computer science, marketing, and business administration (Saúl et al. 2012). RGT uses broad questions focusing on elements such as people, objectives, activities, or events (Jankowicz 2004) to help participants to formulate their own constructs relevant to the topic being explored.
Collection, analysis, and interpretation of RGT data can adopt a variety of approaches. These usually focus on the individual level, emphasizing the unique nature of constructs, highlighting the technique’s utility in enabling insights into an individual’s construct system, and often adopting an interpretivist epistemology (see Table 1: Individual level). Once communicated, such constructs may be shared across individuals with varying idiosyncrasy (Arnold et al. 2010; Grice 2004; Kelly 1963; Simpson and Wilson 1999), offering epistemological justification for group-level analysis. Within an interpretivist epistemology, this is likely to involve manual content analysis to pinpoint similarities and differences within and between individuals’ constructs.
Comparison of Individual- and Group-level Analytical Approaches.
Note: In both within individual-level and group-level approaches, data collection usually follows the traditional pattern of conducting interviews using the RGT (see “Procedure” section).
a The group-level analytical approaches presented here are not exhaustive.
For group-level aggregation approaches operationalized within other, often implicitly ascribed, epistemologies (see Table 1: Group level), this frequently involves using similarity matching/rating or variable reduction techniques to develop constructs that can be described or manipulated statistically, risking losing the depth of individual-level data. In contrast, grid aggregation approaches relying on either data or theory-driven content analysis offer for the former greater flexibility and closeness to data and for the latter greater transparency (Green 2004). Yet Honey’s (1979a, 1979b) use of inductive content analysis has rarely been mentioned as an approach, despite potential for revealing similarities and differences between individuals’ constructs while preserving the inherent complexity and individual perceptual richness in elicited data. Rather, the dominant view is that all such group-level aggregation approaches are epistemologically incompatible with Kelly’s original ideas and likely to result in substantial distortions (Easterby-Smith et al. 1996; Marsden and Littler 2000).
Contrary to viewing Honey’s (1979a, 1979b) approach as incompatible, Hill (1995) contends that within a social constructivist epistemology grid, aggregation can facilitate accurate expression of common or shared constructs, embodying all participants’ categorized views, while conserving idiosyncrasy and richness though maximum participant-specific information (Gergen 2015). From this epistemological stance, grid aggregation satisfies the core tenets of RGT and maintains granularity by not reducing elicited constructs to themes, reference concepts, or components. Compared to other group-level approaches where participants rate similarity between their own constructs and reference concepts derived from prior research (see Table 1: Group-level), the influence of existing concepts is also minimal, suggesting aggregative RGT data analysis is possible (Jankowicz 2004).
A literature search (of Business Source Complete and PsycINFO databases) revealed the novelty of Honey’s grid aggregation, the approach being referenced in only six peer-reviewed studies and five unpublished doctoral theses or conference proceedings since its 1979 inception. 1 We contend this is likely to be for three reasons: First, Honey’s (1979b) original article was published in a journal with a predominantly practitioner readership (Industrial and Commercial Training 2018). As such, it is unlikely to have come to the attention of many scholars. Second, scholars aware of the approach might be unconvinced that advantages outweigh the disadvantages. Finally, despite Kelly’s (1955, 1963) commonality corollary recognizing the potential for shared abstractions, and grid aggregation being considered compatible with RGT within a social constructivist epistemology as previously outlined, some researchers (Easterby-Smith et al. 1996; Marsden and Littler 2000) still deem shared constructs and cross-grid analysis epistemologically incompatible, maintaining “the grid is par excellence a technique for measuring individual perceptions” (Easterby-Smith et al. 1996:26).
Guided by a social constructivist epistemology, we now consider how these concerns can be addressed, thereby allowing us to offer a novel use of Honey’s (1979a, 1979b) aggregation approach for generating shared group constructs from RGT data, maintaining PCT’s individuality corollary as well as congruence with the commonality corollary (Kelly 1955, 1963).
Using RGT and Grid Aggregation to Elicit Shared Constructs
Our exemplar study focuses on conceptualizations of individual workplace performance behaviors, a widely researched phenomenon in management and organization studies (Campbell 2010), yet one where controversy remains, particularly regarding conceptualization (Griffin et al. 2007). For example, while Borman and Motowidlo’s (1997) distinction between task and contextual performance has been supported empirically (Oh and Berry 2009), it is criticized for being broadly defined. Conversely, empirical scrutiny of Campbell and colleagues’ (1993) widely cited eight-factor model offers sparse support (Varela and Landis 2010). Given these, our study’s objective was to identify both idiosyncratic constructs pertaining to one person and constructs where there was communality across persons (Kelly 1955). These would be used to refine our understanding of the phenomenon.
Participants
A heterogeneous sample of 25 managers and professionals with at least three years’ experience was selected purposefully from public, private, and third-sector organizations across various sectors, on the basis that common ground across their constructs would indicate commonality (see Daniels et al. 1995; Hodgkinson 1997a).
Procedure
Data on individual workplace performance constructs based on participants’ day-to-day experience of interacting with others in their own working environments were elicited using traditional RGT structured, semistandardized interviews (detailed in Jankowicz 2004), each lasting on average 45 minutes. At the start of their interview, each participant is asked to provide nine elements in our study comprising “three high, three medium, and three low (workplace) performers, with whom they had interacted in their current or former work environment.” Although elements can be introduced in various ways, we asked participants to select the persons whose behaviors they were going to discuss during the interview to ensure familiarity (Curtis et al. 2008). Participants needed to have observed their chosen nine elements’ work behaviors sufficiently to make statements about their performance. During each interview, elements serve as referent points of comparison, providing the participant with an interaction with the environment when thinking about their constructs. Participants record their elements on separate cards and at the top of an interview grid (see Figure 1: Individual elements) to aid construct elicitation.
Annotated completed interview grid with analysis.
Next, each participant assigns an overall rating to each element ranging from 1 (very low) to 5 (very high), noting it in the first grid line, directly underneath each of the nine element names. Subsequently, bipolar constructs are elicited using the difference method (Epting et al. 1971): Presenting a triad of three name cards, each participant was instructed to “pair up two of the persons (elements) that have something in common regarding their (performance-related) behaviors, that differentiates them from the third person (single element).” Participants were then asked to “elaborate on these behaviors” (the constructs), the “Pair” description being noted on the left and the “Single” description on the right of the grid (see Figure 1). 2 The attribution of a description to pair or single depends therefore on the triad presented. Further bipolar constructs are elicited until a participant can think of no more. Finally, for each element, the participant assigns a rating to each construct using a five-point Likert-type scale (5 = very much like the pair, 1 = very much like the single; Palmer 1978). In our study, 317 bipolar constructs were generated (Figure 1: Example…), ranging from 6 to 18 (SD = 3.06) per participant.
Data Analysis
Individual-level analysis
Initial analysis focuses on individual participants’ construct systems. Similarity scores (“importance scores” in Jankowicz 2004) are calculated for all elicited constructs, indicating the likeness between each construct and the participant’s overall rating; in our example, those most similar (Honey 1979a) to the overall performance rating (Figure 1, right: Similarity). To do this, first, each participant’s overall ratings for elements (Figure 1: first line of grid) are compared against their ratings for each construct, absolute differences across each construct being cumulated. For the construct “Do not think about their work at home; job stops at 5pm”—“Engrossed in her work, never stops thinking about her work,” comparison of overall ratings with this construct’s specific ratings resulted in absolute differences of 1,0,0,1,0,1,0,0,1; cumulated to a “Comp1” value of 4 (Figure 1, right); the “pair” description being associated with high performance and the “single” description with low performance. Given attribution of “pair” and “single” construct descriptions depends on the triad presented, an alternative triad might have resulted in these being reversed. Rather than reverse ratings for each construct, next overall ratings are reversed (Figure 1, bottom: “Reversed…”), and the comparison and cumulating process is undertaken again, absolute differences being recorded (Figure 1, right) as a “Comp2” value: 24 for the first construct. The absolute difference between these two sets of comparison values represents each construct’s similarity score. A relatively high construct score, in our example 20, indicates great similarity to the overall (performance) rating. A relatively low construct score indicates difference. These scores are also used for subsequent aggregative analyses across grids.
Aggregative (group-level) analysis: A novel use of Honey
For aggregative analysis, each participant’s grid constructs are first ranked separately according to their similarity scores and then divided equally into top, medium, and tail terciles (Honey 1979b; Jankowicz 2004); the number of constructs in each tercile being dependent on the number of constructs in the grid. Top constructs are those associated most, and tail constructs are those associated least with the topic being investigated. Constructs for all participants are then examined together. Within each tercile, constructs are sorted into categories and subcategories according to commonalities represented by narrower, more specific aspects or subordinate components of the higher-level categories; noncategorizable constructs being placed in a miscellaneous category. Next, the constructs forming categories, as well as the ratio of top, medium, and tail data in each, are examined. Categories comprising predominantly of top- and medium-level constructs are retained forming an initial model, given individual participants consider these to be important. Categories with more tail than top constructs are discarded as participants do not associate these strongly with the topic.
Following Honey (1979b), our entire categorization process was undertaken independently by two researchers to reduce unwitting data distortion. The first categorization comprised nine categories and 10 subcategories, and the second comprised 12 categories and two subcategories. These were compared and contrasted taking into account respective subcategories, 55% of constructs being categorized into nine conceptually identical categories. Given partial overlap, an expert panel (Honey 1979b) was used to categorize the remaining 45% of constructs. Five management and organization studies and industrial/organizational psychology experts, split into two groups, were asked to sort the uncategorized constructs. Where constructs could not be placed in the existing categories, we requested they sort them into either a new or a “miscellaneous” category. A final facilitated discussion comprising all experts was undertaken to resolve sorting disagreements. Their resulting categorization had 11 categories and 57 subcategories, constructs for each category having been elicited from, on average, 15 participants (Table 2). This final model was discussed with all participants to check understanding and conclusions drawn (Hill 1995). Subsequent comparison with Borman and Motowidlo’s (1997) task/contextual performance distinction and Campbell’s (2010) and Campbell et al.’s (1993) eight-factor model (Table 2) revealed aspects where each had neglected to capture the complexity of performance, neither incorporating constructs categorized as “displaying self-confidence” or “balancing work and life.”
Categories and Their Features.
Note: N = number; ✓ included; (✓) = partially included × = not included.
Appraisal
Taking a social constructivist framework, the methodological procedure outlined here demonstrates that it is possible to retain individual richness when aggregating personal constructs to the group level. Through such extension of the idiographic usage of Kelly’s RGT to a nomothetic application, we demonstrate that, as individuals’ elicited constructs are a product of interactions through social relationships, they can be aggregated using Honey’s (1979a, 1979b) approach, offering a flexible, multipurpose methodology. The social constructivist position remains true to Kelly’s PCT, countering aforementioned criticism that such aggregation is epistemologically not defensible for this method. Moreover, this procedure appears suitable for participant-generated rather than researcher-supplied elements, emphasizing utility for maintaining the data’s inherent richness.
We propose that the approach and associated protocol discussed here makes two methodological contributions: First, unlike many scholars who, following Kelly’s Individuality corollary, have analyzed data within single grids at the individual level, we have illustrated how data may be aggregated across grids without compromising individual-level detail. Our novel use of Honey’s (1979a, 1979b) approach and associated protocol allows consideration of the content of participants’ constructs and their associated ratings. In drawing on all information provided and comparing individual thoughts and ideas, we address concerns regarding loss of data richness when aggregating information across several grids (Easterby-Smith et al. 1996). Such comprehensive aggregation enables insight into the prevalence of constructs, offering a structured, replicable alternative to techniques for eliciting shared understandings such as focus groups. Second, we highlight how this inductive approach can facilitate new understandings, even for comparatively well-researched topics. As such, we address Hibbert and colleagues’ (2014) call for methodologies and practices that can offer new, contextualized theoretical insights. Our research reveals Honey’s grid aggregation approach can allow a heterogeneous group of individuals’ constructs to be used as a basis for new theoretical understanding from which, although each participant has had different experiences, aggregative analysis can reveal commonalities regarding behaviors.
Applying this to individual grid data showed the potential to reveal new aspects considered important but previously not included in existing frameworks in a comparatively well-researched topic. We recognize that our research has only established the utility of aggregating one group’s grid data in one context, which may not be useful or possible where data are highly idiosyncratic. Further work is therefore needed to evaluate the extent to which this new use can be applied with other groups and alternative contexts.
We note using Honey’s (1979a, 1979b) aggregative approach to analyzing RGT data requires substantial time investment. While this is an associated cost, it enables the researcher to remain immersed in the minutiae of participants’ actual data (Patton 2015). Acknowledging immersion is a standard component regarding the analysis of qualitative data, we note its pertinence for a research context where use of specific (e.g., Idiogrid) or generic software (e.g., SPSS) to analyze data elicited via the RGT (Scheer 2016), at the individual or group level, is the norm.
Conclusion
We examined and appraised a novel approach and offered a protocol to enable complimentary idiographic and nomothetic approaches to preserve individual granularity while undertaking aggregative analysis of data elicited via the RGT. Using Honey’s (1979a, 1979b) approach within a social constructivist epistemology offers a novel alternative not only to traditional solely idiographic approaches within the RGT but also to other group-level data elicitation methods such as focus groups.
Offering a single exemplar our research is invariably constrained, and we recognize further application of this use of Honey’s aggregation approach within our protocol would allow boundary conditions to be examined (Dubin 1976; Sackett and Larson 1990), providing a better understanding of where the RGT and such subsequent aggregation might best be used.
Finally, our exemplar study reveals Honey’s approach to grid aggregation using content analysis can provide the basis for new theoretical insights even into well-researched topics. Our protocol regarding how to aggregate individual-level data to the group level while retaining idiosyncratic complexities provides an epistemologically consistent guide for fellow researchers. We therefore propose scholars, when faced with the dilemma of whether to focus on depth or aggregation, now consider utilizing RGT combined with Honey’s aggregative approach within a social constructivist epistemology.
Footnotes
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.