The Interpersonal Principle of Complementarity:
A Meta-Analysis

G. Scott Acton
Northwestern University

Author Note

According to the interpersonal principle of complementarity, dominance invites submission, submission invites dominance, hostility invites hostility, and friendliness invites friendliness. Narrative reviews of the literature on the principle of complementarity in interpersonal theory have turned up contradictory and inconclusive results. The present meta-analysis is meant to determine the overall effect size of the principle of complementarity and to find those factors, if any, that mediate this effect size. The effect size of complementarity was found to be large, but homogeneity was achieved only after eliminating half of the studies as "outliers." Possible mediating factors examined include global or summative ratings versus sequence of behaviors, and document source (published versus unpublished). Effect sizes differed significantly based on these mediating variables.

Why do people tend to elicit specific types of responses from others with whom they interact--often the same types of responses across many relationships? Carson (1969) has dealt extensively with the security-enhancing properties of interpersonal relationships, and proffers three related explanations. First, certain classes of behavior from the "other" tend to co-occur with things reinforcing to the individual, and thus become secondarily reinforcing. For example, feeding an infant tends to co-occur with caring and attending behaviors emitted by the parents. Second, each individual has a "plan" or strategy for interacting with the "other." This plan may vary somewhat from relationship to relationship. For example, a student may have a somewhat different strategy for impressing a professor and for wooing a romantic partner. If a strategy is unexpectedly unsuccessful, anxiety results. Such anxiety is avoided in future interactions by the use of strategies that are less vulnerable to disruption. The individual sets out to narrow the range of possible responses of the "other" so that a rebuff is unlikely. Third is the homeostatic or congruence hypothesis, according to which individuals have a need to keep their behavior in line with their self-perceptions and their perceptions of others. If the behavior of others disconfirms one's self-perception, this produces anxiety, which one will attempt to reduce.

Sullivan (1953) develops the concept of a "self-system," which is the individual's collection of self-perceptions. The self-system actively protects one from information that would cause one to reevaluate all pre-existing self-perceptions. It does this through a process of "selective inattention." Part of this process is taking evasive maneuvers that allow one to maintain congruence between one's interpersonal world and one's self-perceptions. These evasive maneuvers can be defined on the Interpersonal Circle, a two-dimensional model of the domain of interpersonal behavioral tendencies having dimensions of power (dominance versus submission) and affiliation (friendliness versus hostility; Leary,1957; Carson, 1969). Such evasive maneuvers include avoiding certain segments of the Interpersonal Circle that usually result in behaviors in others that disconfirm one's self-system, and consequently forcing others to give way to one's own interpersonal strategy, regardless of their own wishes.

The interpersonal principle of complementarity specifies ways in which a person's interpersonal behavior evokes restricted classes of behavior from an interactional partner, leading to a self-sustaining and reinforcing system. The principle of complementarity is defined on the Interpersonal Circle, such that reciprocity tends (probabilistically) to occur on the power axis (dominance elicits submission, and submission elicits dominance), and correspondence tends (probabilistically) to occur on the affiliation axis (friendliness elicits friendliness, and hostility elicits hostility). Carson (1969) defines interpersonal complementarity using quadrants of the Interpersonal Circle, while Kiesler (1983) defines interpersonal complementarity using sixteenths of the Interpersonal Circle (see Figure 1 [note: this figure is not yet available on the internet]). However, the ideas of reciprocity on the power axis and correspondence on the affiliation apply in each case--the only difference is the degree of resolution on the Interpersonal Circle to which they are applied.

At least two important narrative reviews of the literature on interpersonal complementarity have appeared. Kiesler (1983), speaking of the twin processes of reciprocity and correspondence, concludes, "A growing body of empirical research confirms these two rules of complementarity [eighteen studies cited] with some mixed results [three studies cited]" (p. 201).

Orford (1986), however, after reviewing fourteen studies, concludes that the empirical research is far from confirming interpersonal complementarity as a general rule. He reports, "Not so clearly predicted by the theories, however, is that friendly-dominance is a not uncommon response both to friendly-dominance . . . and to hostile-submission" (Orford, 1986, p. 374). Theoretically, dominance should elicit submission, and hostility should elicit hostility. Orford also reports, "In particular, the theories have missed the fact that hostile-dominance is most likely to meet with hostile-dominance in return, and that this is particularly likely in behavior between equals (fathers and mothers, husbands and wives, fellow students) . . . . Hostile-submission may produce hostile-dominance, as the Leary and Carson/Kiesler theories predict, but reciprocated hostile-submission is as common, if not more so, and friendly-dominance . . . seems to be most common" (p. 375). Thus, hostile submission in fact elicits friendly dominance, which is a difference on the affiliation axis where similarity (i.e., correspondence) is predicted. Thus, the theoretical predictions of the principle of complementarity seem to have been disconfirmed. Orford concludes that the evidence on complementarity is mixed--there are a few verifications and a few disconfirmations. Orford suggests several possible mediating variables, including status differences, setting, and time in relationship. However, the narrative method of review is not adequate to test these hypotheses.

The present study represents an attempt to determine not only the overall effect size for the principle of complementarity, but also the role of mediating variables such as global or summative ratings versus sequence of behaviors, and document source (published versus unpublished).


Retrieval of Studies

Studies were located by searching PsychINFO (a database that covers psychology-related articles, books, chapters, and dissertations going back to 1967) using the search term "complementarity." Thereafter, additional studies were located using the ancestry method (looking in the references section of obtained studies). All dissertations that could be obtained for free (including almost all of them) were ordered using interlibrary loan.

Criteria for Selection of Studies

The first criterion for a retrieved study to be used in the meta-analysis was that an effect size for complementarity could be calculated from the reported results. Second, studies that did not pertain to the Interpersonal Circle (a model of interpersonal behavioral tendencies having dimensions of power and affiliation) were excluded (e.g., studies on the Structural Analysis of Social Behavior were excluded).

Studies Used in the Meta-Analysis

One hundred four studies were candidates for the meta-analysis. These studies are listed in Appendix A. However, only a sample of those studies was used. Of the studies actually retrieved from the library or through interlibrary loan (including dissertations), the first twenty-four in alphabetical order were coded. Effect sizes were calculated for nine of these studies. Seven studies were excluded because they did not include suitable information to calculate an effect size. Eight studies were excluded because the author was uncertain how to calculate an effect size for them. From the nine studies used, twenty-one effect sizes for complementarity were calculated.

Calculation of Effect Sizes

Effect sizes (d) were obtained in the standard way. Because sample sizes were sometimes small, correction procedures described in Hedges and Olkin (1985) were routinely applied to effect sizes to yield unbiased estimates of their population values. (When effect sizes are large, corrected and uncorrected effect sizes are nearly identical.)

Meta-Analysis of Effect Sizes

The meta-analysis utilized the methods developed by Hedges and Olkin (1985). All analyses were done using the program DSTAT. First, weighted mean effect sizes and their 95% confidence interval (CI) were calculated. A mean effect size is statistically significant when the CI does not include .00. Second, the homogeneity of within-category effect sizes was determined for each category. Third, moderator variable analysis was conducted to determine whether effect sizes varied significantly with study characteristics.


Description of Database

Table 1 lists the included studies of complementarity and the study characteristics [note: this table is not yet available on the internet]. The overall effect size for complementarity is d = .7404. Cohen has suggested guidelines for interpreting the magnitude of d, so that a small effect size is .20, a medium effect size is .50, and a large effect size is .80. A medium effect size is conceived as one large enough to be visible to the naked eye. The overall effect size for complementarity in the present meta-analysis is rather large.

These findings are complicated by the fact that the hypothesis of homogeneity of effect sizes is rejected (p < .0001). This indicates that the samples in the present study are not drawn from the same population. This heterogeneity can be corrected by deleting "outliers" until homogeneity is obtained (i.e., until a non-significant result is reached). In this case, however, eleven outliers had to be deleted out of twenty-one effect sizes before a non-significant result was reached at p < .05. This heterogeneity might be expected from the fact that complementarity is such a multi-faceted phenomenon.

Document Source

The effect of the moderating variable of document source (published versus unpublished) was tested. There tends to be a publication bias in favor of statistically significant findings. One way of obtaining statistically significant findings is to have a large effect size. However, in the present study a reverse publication bias seems to have been operational: The seven unpublished studies had far larger average effect sizes than the fourteen published studies (d = .7964, p < .0001). Homogeneity was rejected within the categories published (Qw = 138.3109, p < .0001) and unpublished (Qw = 61.3474, p < .0001), however, so this moderating variable cannot account for the overall heterogeneity in effect sizes.

Global or Summative Ratings Versus Sequence of Behaviors

The effect of the moderating variables of global or summative ratings versus sequence of behaviors was tested. There is a far larger average effect size for studies of behavioral sequences (d = 1.2267) than for studies of global ratings (d = .6889; difference of d = .7964, p < .0001). The six studies of behavioral sequences were homogeneous (Qw = 12.3183, p > .05), while the fifteen studies of global ratings were heterogeneous (Qw = 183.8516, p < .0001).

Other Study Characteristics

Eleven study characteristics were coded for each study. In addition to document source and global versus summative ratings, the following study characteristics were coded: year of publication, coder, status differences, setting, time in relationship, quadrant/octant/other description of behavioral tendencies, instrument used, statistic from which effect size is calculated, and sample size. Some of these variables could be very important to an understanding of complementarity and in remediating the heterogeneity among studies in this sample. However, only the first two were dichotomous, and thereby lent themselves readily to analysis in DSTAT, so only the first two were used in the present study. In a more elaborate meta-analysis, all of these study characteristics would be tested as potential moderating variables.


The present meta-analysis included only a subset of the total population of studies on complementarity. Given the modest assumption that alphabetical order of author last name is equivalent to a random method of sampling, this procedure should add only a slight amount of error to the results that would be obtained using all studies of complementarity, allowing us to draw relatively firm conclusions.

This meta-analysis found that the effect size of complementarity across studies is large. However, this result must be tempered by the recognition that half the effect sizes had to be deleted as "outliers" before homogeneity across studies was achieved. One study characteristic that seems to mediate effect size is global ratings versus sequence of behaviors; another is document source (published versus unpublished). A more elaborate meta-analysis is needed to determine the mediating effects of other study characteristics of great theoretical importance.


References marked with an asterisk indicate studies included in the meta-analysis.

*Beery, J. W. (1970). Therapists' responses as a function of level of therapist experience and attitude of the patient. Journal of Consulting and Clinical Psychology, 34, 239-243.

*Bluhm, C., Widiger, T. A., & Miele, G. M. (1990). Interpersonal complementarity and individual differences. Journal of Personality and Social Psychology, 58, 464-471.

*Brokaw, D. W., & McLemore, C. W. (1983). Toward a more rigorous definition of social reinforcement: Some interpersonal clarifications. Journal of Personality and Social Psychology, 44, 1014-1020.

*Buss, D. M. (1984). Marital assortment for personality dispositions: Assessment with three different data sources. Behavior Genetics, 14, 111-123.

*Campbell, S. R. (1991). The relationship of interpersonal complementarity to marital satisfaction and security (Doctoral dissertation, University of Maryland, College Park, 1991). Dissertation Abstracts International, 52, 1051.

Carson, R. C. (1969). Interaction concepts of personality. Chicago: Aldine.

*Coates, T., & Mazur, S. (1969). Personality characteristics and interpersonal attraction. Psychology, 6, 2-9.

*Cole, D. A., Vandercook, T., & Rynders, J. (1988). Comparison of two peer interaction programs: Children with and without severe disabilities. American Educational Research Journal, 25, 415-439.

*Edquist, M. H. (1973). Interpersonal choice and social attraction among four interpersonal types (Doctoral dissertation, Duke University, 1973). Dissertation Abstracts International, 34, 1722B. (University Microfilms No. 72-22, 977)

*Ferleger, N. A. (1993). The relationship of dependency and self-criticism to alliance, complementarity and outcome in short-term dynamic psychotherapy (Doctoral dissertation, Fordham University, 1993). Dissertation Abstracts International, 54, 2748.

Hedges, L. V., & Olkin, I. (1985). Statistical methods for meta-analysis. San Diego, CA: Academic Press.

Kiesler, D. J. (1983). The 1982 interpersonal circle: A taxonomy for complementarity in human transactions. Psychological Review, 90, 185-214.

Leary, T. (1957). Interpersonal diagnosis of personality. New York: Ronald Press.

Orford, J. (1986). The rules of interpersonal complementarity: Does hostility beget hostility and dominance, submission? Psychological Review, 93, 365-377.

Sullivan, H. S. (1953). The interpersonal theory of psychiatry. New York: Norton.

Appendix A

Last modified July 1998
Visited times since July 2001

Home to Great Ideas in Personality