NEWS.md
In roundRobinR: Manipulate and Analyze Round Robin Dyadic Data

roundRobinR 2.0.0

New: Co-Partner Social Relations Model (`cpsrm()`, `cpsrm_run()`)

Added cpsrm(), a friendly entry point for the Co-Partner SRM (CP-SRM) that takes raw long-format data plus actor_id/group_id and builds the required actor/partner dummy matrices internally via create_cp_dummies(), then fits the model with cpsrm_run().
Added cpsrm_run(), the full-control fitting function underneath cpsrm(), together with create_cp_dummies(), print.cpsrm(), and summary.cpsrm(). cpsrm_run() fits the CP-SRM -- in which each observation involves one actor and all other members of a group acting simultaneously as partners (e.g. three-person teams) -- via REML, using either a Woodbury-matrix-identity ("block") or full-covariance-loop ("loop") formulation, and supports variable group sizes.
print.cpsrm()/summary.cpsrm() report variance-of-total percentages under BOTH the RAW convention (each component's own share of total variance; the correct basis for comparing Partner's effect size to Actor's) and the COMBINED convention (Partner's contribution scaled by the number of partners summed into each row; the correct basis for a full total-variance decomposition), whenever weight_partners = FALSE and the number of partners per row is constant.
?cpsrm_run documents (a) the RAW vs. COMBINED distinction and why conflating them can make partner effects look several times more or less important than actor effects than they really are, and (b) the boundary-corrected likelihood-ratio test procedure for testing whether a variance component is zero (one-sided halved chi-square for a single component; the three-part 1/4-1/2-1/4 mixture for a joint two-component test), since variance components are boundary-constrained and the naive two-sided/unhalved chi-square test is not valid here.

Renamed several existing functions to snake_case for consistency with the new cpsrm()/cpsrm_run()/create_cp_dummies() naming: createDummies() -> create_dummies(), srmRun() -> srm_run(), srmVarPct() -> srm_var_pct(), srmPseudoRSq() -> srm_pseudo_rsq().
Every renamed function's old name still works exactly as before and remains exported; calling it emits a one-time .Deprecated() warning pointing to the new name and new (also renamed) arguments. No user-facing breaking changes.

roundRobinR 1.0.0

First submission to CRAN.
Functions for manipulating round-robin dyadic data: createDummies().
Functions for fitting the Social Relations Model via multilevel modeling: srmRun(), srmVarPct(), srmPseudoRSq().
Custom nlme covariance class pdSRM implementing the SRM variance-covariance constraints (equal actor variances, equal partner variances, single actor-partner covariance).
Sample dataset sampleDyadData with simulated round-robin data from two time points.
Vignette: "Introduction to roundRobinR".

Fixed reference to groupId column in srmRun() that used the variable name string rather than the column contents when creating the pdSRM_group_id grouping variable.
Replaced class(x) == "try-error" with inherits(x, "error") in pdMatrix.pdSRM() per CRAN policy.
Added stats:: namespace prefixes to calls to formula(), na.omit(), and coef() throughout.
Wrapped long-running examples in \donttest{} to comply with CRAN example time limits.