To fully account for the Complexities of real-life behaviors and interpersonal perceptions, it is necessary to represent the two-sided and dynamic nature of interpersonal perceptions and behaviors in one’s research design.

A research design thus need to be polydiadic (People need to be members of multiple dyads: Each perceiver must be observed with multiple targets and each actor must be observed with multiple partners) and bidirectional (Within each dyad people are both actors/perceivers and partners/targets: They perceive each other and interact with each other).

The two most commonly used designs for this aim are round robin design and the full block design. In the case of the round robin design each person rates or interacts with each other person (see left part of Figure 1). The full block design is preferable when one only wants to analyze asymmetric dyads, i.e. dyads that are not interchangeable because the two persons of a dyad can be distinguished by a certain feature that is crucial for one’s research question. When one is for example interested in smiling as a heterosexual flirting behavior, sex is a crucial feature of the two partners of a dyad and they are not interchangeable. In this example, one is only interested in the cross-sex smiles and not in smiling behavior of men towards men and women towards women. Thus, in a block design a group of people is broken into two subgroups (e.g., men and women). Each person then rates or behaves towards everyone else in the other subgroup (see right part of Figure 1).


Figure 1: The Round Robin Design and the Full Block Design.

Both designs, the round robin design and the full block design, are perfectly suited to disentangle components and model the bidirectionality of interpersonal perceptions and behaviors, respectively (see Concepts).

However, a round robin design or a full block design is not always possible or feasible. Sometimes it might be hard to obtain bidirectional perceptions or behaviors, for example when young children are judged by family members but do not judge them back. Moreover, sometimes it is not possible to realize a design with multiple dyadic partners. For example, in Western societies, most husbands and wives are a member of only one marital relationship.

All "simpler" designs that can be used in such cases have some shortcomings that should be carefully beared in mind when interpreting the findings. Unidirectional designs (ego-centered network, half-block design, see Table) cannot account for the reciprocal nature of interpersonal perceptions and behaviors and monodyadic designs (distinguishable or indistinguishable, see Table) cannot seperate social relations components (even though they of course still exist and ignoring them potentially obscures an accurate interpretation of social phenomena).

The following table lists 3 major groups of designs that are often employed by researchers. For further information on statistical analyses, key findings and further readings please go to stats Stats and Key Findings.

Design Description Illustration/Example Possible analyses Shortcomings
Unidirectional/

Ego-centered

Ego-centered Network
  • Participants rate all of their social network partners
  • Network partners are not part of the study (don’t reciprocate the ratings)
  • Persons rate the perceived closeness to and conflict with all network partners
  • Multivariate pseudo-dyadic analyses using multilevel modeling
  • Aggregation of network characteristics across raters
  • No SRM variance decomposition
Half-Block-Design
  • All members of Group A (Perceivers/Actors) interact with all members of Group B (Targets/Partners)
  • Ratings/Behaviors are unidirectional
  • Female participants rate the same photographs of four male targets (stimuli) on an attractiveness scale
  • Limited SRM variance decomposition: Actor/Perceiver effects for members of Group A, Partner/Target effect for members of Group B, relationship effects for each member of Group A towards each member of Group B (but not vice versa)
  • Correlations between individual difference scales and corresponding SRM effects
  • SRM effects generalize only to interactions with members of other group
  • Actor/Perceiver and Partner/Target effects are not simultaneously available for the same participant (and hence cannot be correlated)
  • No dyadic analyses possible
  • No reciprocities (as symmetric relationship effects are lacking)

Single partner dyadic

Distinguishable

Dyads

  • Dyad members interact with each other
  • Dyad members are distinguishable by a specific characteristic (e.g. gender, role, age)
  • Interaction between mother and child, both providing self-ratings about their communication style and interaction quality
  • Estimation of APIM effects
  • In APIM, Actor Effects refer to the influence of one's score on X on one's own score on Y; Partner Effects refer to the influence of one’s own score on X on the partner's score on Y
  • No SRM variance decomposition
  • APIM effects are biased as a result
  • APIM effects generalize only to interactions with persons of the other group
Indistinguishable Dyads
  • Dyad members interact with each other
  • Dyad members are not distinguishable by a specific characteristic
  • Interaction between same sex friends, both providing self-ratings about their communication style and their interaction quality
  • Separation of APIM effects possible
  • APIM effects generalize to both (indistinguishable) interaction partners
  • No SRM variance decomposition
  • APIM effects are biased as a result

Multiple partner

Full Block
  • All members of one group interact with all members of another group and vice versa
  • Members of both groups serve as Perceiver/Actor and Target/Partner
  • Symmetric block design: the two groups are indistinguishable
  • Asymmetric block design: the two groups are distinguishable (e.g., male and female)
  • A group of males talks with all members of a group of females and vice versa (e.g., speed dating)
  • SRM variance decomposition for all individuals/dyads
  • Bi- and multivariate SRM analyses
  • Correlations of individual difference scales with SRM effects
  • SRM effects generalize only to interactions with persons of the other group
  • No estimation of social network parameters
Round-Robin
  • One big or several smaller groups
  • Every participant interacts with every other participant of his group and is thus Perceiver/Actor and Target/Partner at the same time.
  • In a random group of students each member rates the attractiveness of each of the other members (reciprocal measurement)
  • SRM variance decomposition for all individuals/dyads
  • Bi- and multivariate SRM analyses
  • Correlations of individual difference scales with SRM effects
  • SRM effects generalize to the entire sample
  • No estimation of social network parameters
Social network analysis
  • One big or several smaller groups
  • Focus on dichotomous dyadic ties (nominations) between indistinguishable group members
  • Cross-sectional and longitudinal data can be analyzed
  • In several work groups of between 20-25 people, group members nominate themselves as either friends or non-friends; in addition, information about personality and demographic characteristics of all members is collected

Network effects

  • Average rate of change in number of dyadic ties
  • Number of outgoing and incoming dyadic ties
  • Proportion of dyadic ties in relation to the total number of possible ties
  • Reciprocity of dyadic ties
  • Tendency to form new dyadic ties with others who belong to the network of an existing dyadic tie (transitivity)

Multivariate effects

  • Effect of individual differences on selecting others as a tie
  • Effect of individual differences on being selected as a tie
  • Effect of dyadic similarity in individual differences on forming a tie
  • No SRM variance decomposition of ties
  • No estimation of SRM effects on ties
  • Dependent variable needs to be (transformed into) dichotomous format