Introduction to roundRobinR

Overview

The roundRobinR package provides tools for analyzing directed dyadic data collected with a round-robin design, in which every person in a group rates or interacts with every other person. The core analytical framework is the Social Relations Model (SRM; Kenny, Kashy, & Cook, 2006), estimated via multilevel modeling following Snijders and Kenny (1999) and Knight and Humphrey (2019).

The SRM decomposes the total variance in a directed dyadic outcome into four components:

Component	Interpretation
Group	How much groups differ in their mean level
Actor	How much people differ in what they give or emit
Partner	How much people differ in what they receive or elicit
Dyad	Variance in the unique relationship between each specific pair

In addition, two reciprocity parameters are estimated:

Generalized reciprocity: Do people who give a lot also tend to receive a lot? (Actor-partner correlation)
Dyadic reciprocity: Does person A’s unique relationship with person B predict B’s unique relationship with A? (Within-dyad relationship correlation)

The Sample Dataset

The package includes sampleDyadData, a simulated dataset with 1,548 rows from a round-robin design collected at two time points.

head(sampleDyadData)
#>      groupId actId partId timeId contact liking actEx actAg actMale actAge
#> 4          1     1      2      1       1      7   4.4   4.4       1     19
#> 1300       1     1      2      2      NA      7   2.7   3.2       1     19
#> 7          1     1      3      1       1      7   4.4   4.4       1     19
#> 1303       1     1      3      2      NA      7   2.7   3.2       1     19
#> 11         1     1      4      1      NA      7   4.4   4.4       1     19
#> 1307       1     1      4      2      NA      7   2.7   3.2       1     19
#>      partEx partAg partMale partAge groupCohesion groupEfficacy
#> 4       4.7    1.5        0      20            NA            NA
#> 1300    3.8    1.5        0      20            NA            NA
#> 7       5.0    1.5        1      21            NA            NA
#> 1303    5.6    1.3        1      21            NA            NA
#> 11      3.5    2.0        1      20            NA            NA
#> 1307    4.7    1.2        1      20            NA            NA
nrow(sampleDyadData)
#> [1] 1584
length(unique(sampleDyadData$groupId))
#> [1] 66

Key variables:

groupId, actId, partId — group, actor, and partner identifiers
timeId — time point (1 or 2)
liking — directed dyadic rating of liking (the main outcome)
actEx, partEx — actor and partner extraversion scores
contact — dyad-level contact frequency
groupCohesion — group-level cohesion

Step 1: Create Dummy Variables

The SRM requires a set of dummy variables for each actor and partner position within a group. createDummies() generates these automatically.

d <- createDummies(
  group.id       = "groupId",
  act.id         = "actId",
  part.id        = "partId",
  d              = sampleDyadData[sampleDyadData$timeId == 1, ],
  merge.original = TRUE
)
head(d[, c("groupId", "actId", "partId", "pdSRM_dyad_id",
           "a1", "a2", "a3", "a4", "p1", "p2", "p3", "p4")])
#>   groupId actId partId pdSRM_dyad_id a1 a2 a3 a4 p1 p2 p3 p4
#> 1       1     1      2             1  1  0  0  0  0  1  0  0
#> 2       1     1      3             2  1  0  0  0  0  0  1  0
#> 3       1     1      4             4  1  0  0  0  0  0  0  1
#> 4       1     2      1             1  0  1  0  0  1  0  0  0
#> 5       1     2      3             3  0  1  0  0  0  0  1  0
#> 6       1     2      4             5  0  1  0  0  0  0  0  1

The key output columns are:

a1–a4: actor dummy variables (one equals 1 per row, indicating which group position is acting)
p1–p4: partner dummy variables (same, for the partner)
pdSRM_dyad_id: undirected dyad identifier (shared by i→j and j→i)

Step 2: Fit the Null SRM

The simplest model decomposes total variance into the four SRM components with no predictors. srmRun() handles dummy creation and model fitting in one step.

null_mod <- srmRun(
  dv      = "liking",
  groupId = "groupId",
  actId   = "actId",
  partId  = "partId",
  data    = sampleDyadData[sampleDyadData$timeId == 1, ]
)
null_mod$srm.output
#>                         variances.and.covariances percents.and.correlations
#> Group                                       0.122                     9.926
#> Actor                                       0.808                    65.596
#> Partner                                     0.027                     2.154
#> Dyad                                        0.275                    22.324
#> Generalized Reciprocity                     0.002                     0.011
#> Dyadic Reciprocity                          0.009                     0.033

Reading the output:

variances.and.covariances: raw variance estimates for each component
percents.and.correlations: for the four variance components, the percentage of total variance; for the two reciprocity rows, the correlation coefficients

Step 3: Fit a Model with Predictors

Adding fixed effects tests whether actor-level, partner-level, or dyad-level predictors explain variance in the outcome.

full_mod <- srmRun(
  dv      = "liking",
  groupId = "groupId",
  actId   = "actId",
  partId  = "partId",
  feVars  = c("actEx", "partEx", "contact"),
  data    = sampleDyadData[sampleDyadData$timeId == 1, ]
)
full_mod$srm.output
#>                         variances.and.covariances percents.and.correlations
#> Group                                       0.119                     9.790
#> Actor                                       0.800                    66.073
#> Partner                                     0.020                     1.667
#> Dyad                                        0.272                    22.470
#> Generalized Reciprocity                     0.000                    -0.002
#> Dyadic Reciprocity                          0.005                     0.018

The fixed-effect estimates can be examined with:

summary(full_mod$lme.output)$tTable
#>                   Value  Std.Error  DF    t-value      p-value
#> (Intercept)  5.20484636 0.26625736 683 19.5481783 6.346570e-68
#> actEx        0.07387859 0.06060827 683  1.2189522 2.232832e-01
#> partEx      -0.01342750 0.02468930 683 -0.5438589 5.867161e-01
#> contact      0.07888947 0.01591042 683  4.9583522 8.979027e-07

Step 4: Pseudo R-Squared

srmPseudoRSq() compares the null and full models to estimate the proportion of variance in each component explained by the predictors.

srmPseudoRSq(
  null.model    = null_mod$lme.output,
  predict.model = full_mod$lme.output
)
#>               null    predict    pseudoR2
#> Group   0.12228080 0.11860539 0.030057131
#> Actor   0.80811232 0.80048626 0.009436881
#> Partner 0.02654036 0.02019843 0.238954181
#> Dyad    0.27502661 0.27222552 0.010184796

A positive pseudoR2 for a given component means the fixed effects reduced that component’s variance, suggesting the predictors partly explain the group-, actor-, partner-, or relationship-level variation.

References

Kenny, D. A., Kashy, D. A., & Cook, W. L. (2006). Dyadic Data Analysis. Guilford Press.

Knight, A. P., & Humphrey, S. E. (2019). Dyadic data analysis. In S. E. Humphrey & J. M. LeBreton (Eds.), The Handbook for Multilevel Theory, Measurement, and Analysis (pp. 423–447). American Psychological Association. https://doi.org/10.1037/0000115-019

Snijders, T. A. B., & Kenny, D. A. (1999). The social relations model for family data: A multilevel approach. Personal Relationships, 6, 471–486. https://doi.org/10.1111/j.1475-6811.1999.tb00204.x