# Social Relation Model (SRM) Analyses for Single or Multiple Groups

Social Relation Model (SRM) analyses for single or multiple round-robin groups are performed. These analyses are either based on one manifest variable, one latent construct measured by two manifest variables, two manifest variables and their bivariate relations, or two latent constructs each measured by two manifest variables. Within-group t-tests for variance components and covariances are provided for single groups. For multiple groups two types of significance tests are provided: between-groups t-tests (as in SOREMO) and enhanced standard errors based on Lashley and Bond (1997) . Handling for missing values is provided.

# ------------------------------------------------------------

• Fixed a bug where the p-value of negative bivariate covariances computed with Lashley-Bond standard errors was incorrect.

# ------------------------------------------------------------

• Added warning when SOREMO standard errors are chosen with <= 8 groups
• Added warning when SOREMO standard errors are chosen with a single group

# ------------------------------------------------------------

Added more information in the help files about the new standard errors; new reference:

Lashley, B. R. & Kenny, D. A. (1998). Power estimation in social relations analyses. \emph{Psychological Methods, 3}, 328-338. doi:10.1037/1082-989X.3.3.328

# ------------------------------------------------------------

• New computation of significance tests in multiple groups.
• Standard errors and p-values of multiple groups are now based on the single-group SE derived by Lashley & Bond (1997)
• You can get the old behavior by setting RR(..., se = "SOREMO")
• PLEASE NOTE: We changed the default behavior. From version 1.5 on, you get the new Lashley & Bond (1997) standard errors by default.
• Fixed a minor bug in bivariate computations of single-group Lashley-Bond standard errors
• Improved output
• Refactored a lot for a cleaner structure of the package

# ------------------------------------------------------------

• Compability update: Had to react to changes in package plyr/reshape2. Now requires plyr >= 1.8.2

# ------------------------------------------------------------

• getEffects function can now export effects with group means added (set gm=TRUE; see ?getEffects)
• Some changes for CRAN compability, internal restructuring

# ------------------------------------------------------------

• changed order of authorship
• much less time for checking examples

# ------------------------------------------------------------

• Now with correct citation information, see citation("TripleR")
• New sanity check: If double entries are present, now a warning is given. (Before version 1.2, only the uninformative "Aggregation requires fun.aggregate: length used as default" was printed)
• General R 3.0.2 compability

# ------------------------------------------------------------

• Minor changes to be compatible with ggplot 0.9.2
• Finally got rid of the R CMD check warnings about global bindings

# ------------------------------------------------------------

NEW FEATURES

• An error is printed if actor or partner IDs are not unique (they have to be unique across groups). Non-unique IDs sometimes caused strange errors.

BUG FIXES

• removed an unnecessary print-out of the effects at each calculation
• established compatibility to the forthcoming 0.9.0 release of ggplot2

IMPROVEMENTS

• reduced size of vignette

# ------------------------------------------------------------

BUG FIXES

• corrected bug in calculation of p-values for the single group significance test as well as in the parCor function. t-values now are correctly translated into p-values (p-values have been too conservative before; t-values were always correct)

# ------------------------------------------------------------

NEW FEATURES

• added argument minVar to RR, getEffects, and RR.style. Actor and partners effects are only calculated if the respective relative variance component is greater than minVar. Set minVar to NA if this cleaning should not be performed. For small groups, Kenny (1994) suggests a minimum relative variance of 10% for the interpretation of SRM effects. In any case, actor/ partner effects and correlations with these variables should not be interpreted if these components have negative variance estimates. minVar defaults to zero; with RR.style this default can be changed for all subsequent analyses. If minVar is set to NA, all effects are calculated, regardless of the size of the variance component.
• Bivariate correlations are not printed for variance components < minVar.

# ------------------------------------------------------------

NEW FEATURES

• added new user-visible function 'parCor' which computes partial correlations between variables, controlled for group membership. This can be used to compute partial correlations between SRA effects and external variables like age, personality scales, etc. See help file for an example.

• If self ratings are present, partial correlations of self-ratings with actor and partner effects are printed in the standard output (controlled for group membership if multiple groups are present)

• function RR now has a new parameter "varname". In latent analysis where two variables are combined to one latent construct, this latent variable now is called "varname" in RR$effects, e.g.: RR3 <- RR(liking_a / liking_b ~ actor.idpartner.id, data=likingLong) #latent (construct-level) univariate SRM analysis, define new variable name for the latent construct RR3b <- RR(liking_a / liking_b ~ actor.idpartner.id, data=likingLong, varname="liking") # compare: head(RR3$effects) head(RR3b$effects) ADDITIONALLY • renamed variables in demo data sets to perceiver.id and target.id (as they are about interpersonal perception) BUG FIXES • fixed a minor bug in getEffects / long2matrix were corrupt groups led to a crash • fixed wrong column names in multiple group # ------------------------------------------------------------ # ---- Changes in TripleR version 1.0.7 ---------------------- # ------------------------------------------------------------ • added a group id column to the relationship effects • prepared a helper function to convert relationship effects from long format (which is the default output of the RR function) to wide format. Long format is best suited for subsequent MLM analyses, wide format is needed for SEM analyses. # ------------------------------------------------------------ # ---- Changes in TripleR version 1.0.6 ---------------------- # ------------------------------------------------------------ • moved ggplot2 to "Depends" • minor updates in vignette # ------------------------------------------------------------ # ---- Changes in TripleR version 1.0.5 ---------------------- # ------------------------------------------------------------ BUG FIXES • the labeling of missing values in plot_missings now is correct (swapped 'yes' and 'no') • minData now works correctly in the bivariate latent case # ------------------------------------------------------------ # ---- Changes in TripleR version 1.0.3 ---------------------- # ------------------------------------------------------------ IMPROVEMENTS • in bivariate analyses now both univariate and the bivariate results are based on the same set of participants (in previous versions, data sets with many NAs could cause the exclusion of participants in one of both variables). # ------------------------------------------------------------ # ---- Changes in TripleR version 1.0.2 ---------------------- # ------------------------------------------------------------ NEW FEATURES • added a function to display group statistics (group sizes, excluded groups, etc.): ?RR.summary • added the possibility to calculate self enhancement index (Kwan-Index) in the RR and getEffects function (set index="enhance"). These indices are reported along with the actor and partner effects, see RR_object$effects
• effects-with-group-mean-added are now also provided in latent analyses

BUG FIXES

• the warning message about excluded participants now is correct

IMPROVEMENTS

• warning messages are more concise
• many under-the-hood improvements

# ------------------------------------------------------------

• Adjusted the recommendations for missing values based on simulations (they now are a bit more liberal)

# ------------------------------------------------------------

As everything works now beautifully, we celebrate version 1.0!

# ------------------------------------------------------------

• renamed the labels of the dyads in the effectsRel variable: now each dyad has a unique id, which is important for subsequent APIM analyses

# ------------------------------------------------------------

• Reliabilities are always printed with 3 decimals
• Warnings about excluded groups and/or participants are more meaningful and cleaned up (double warnings should not occur anymore)
• Increased reliability with many NAs (especially in latent analyses)

# ------------------------------------------------------------

BUG FIXES

• self ratings are now correctly centered

MINOR IMPROVEMENTS/ CHANGES

• updated vignette with sections on missing values and installation of R
• Handling of multiple groups now is more robust in cases where some groups contained self ratings and others do not

# ------------------------------------------------------------

• checked 2.12.0 compatibility ( ... it works)
• introduced new parameter to RR: minData. minData (default = 1) controls the minimum amount of data points which should be in each row or column of the round robin matrix. For example, minData=2 means that for each participant, at least two data points have to be present both as actor and as target.
• Updated the procedure for missing values: For the calculation of variance components and effects, missing values are imputed as the average of the respective row and col means. Relationship effects which were missing in the original data set then are set to NA again.

# ------------------------------------------------------------

NEW FEATURES

• added significance test for covariances in the univariate latent case

MINOR IMPROVEMENTS/ CHANGES

• improved vignette
• added a link to TripleR.pdf in the main help page
• fixed a small bug in printing multiple groups: The variable name now is displayed
• fixed a namespace conflict with imported packages

# ------------------------------------------------------------

MINOR IMPROVEMENTS/ CHANGES

• updated vignette with an instruction how to calculate partial correlations between SRM effects and external variables
• speed improvements

# ------------------------------------------------------------

MINOR IMPROVEMENTS/ CHANGES

• Improved help files, added help file for plot.RR

BUG FIXES

• Fixed minor bug in styles: Labels of reliabilities now always are displayed correctly

# ------------------------------------------------------------

BUG FIXES

• Standardized variances in bivariate manifest analysis are corrected

# ------------------------------------------------------------

NEW FEATURES

• Added a NEWS file (you're reading it) to document changes
• Relationship effects now have a new column called 'dyad'. This facilitates subsequent dyadic analyses.
• Relationship effects now are sorted that each pair of dyadic effects is in consecutive lines
• Added new labeling system: all labels can now centrally controlled by one function: RR.style. You set your styles once at the beginning of the project (e.g., RR.style("perception")), and all subsequent calls of RR are styled accordingly (i.e., in RR.print appropriate labels are taken, and the suffixes of the effects are appropriately set)
• Reliability of relationship effects now is printed (provided only in latent analyses)

MINOR IMPROVEMENTS/ CHANGES

• Improved help files
• Removed warning, that variance components < 10% should not be interpreted (the relevant indicator is the significance of these components)
• The labels of the reliabilities now are correctly set to actor/partner or perceiver/target, depending on the style set.
• Fixed line breaks in the vignette

BUG FIXES

• Labeling of perception/metaperception now is correct

# Reference manual

install.packages("TripleR")

1.5.3 by Felix Schönbrodt, 3 years ago

Browse source code at https://github.com/cran/TripleR

Authors: F. D. Schönbrodt , M. D. Back , S. C. Schmukle

Documentation:   PDF Manual

Task views: Psychometric Models and Methods

GPL (>= 2) license

Imports reshape2, plyr

Depends on ggplot2

Suggests lme4

Imported by pointRes.

See at CRAN