LoginSign Up
← Back to all sparks
G

ggeffects

ANALYTICS
Velocity0.0

Compute and plot marginal effects and adjusted predictions from R models

github.com/strengejacke/ggeffects ↗

ggeffects hands its contrast engine to modelbased and keeps the interface

marginal-effectsr-statsstatisticsbreaking-changesmixed-modelsdependencies
◆Current state
ggeffects computes and plots marginal effects for a long tail of R model classes. Its recent line has two threads: steadily broadening model support and argument surface, and repeatedly absorbing breaking changes from the packages it computes on top of. In 2.2.0 it stopped absorbing them and delegated test_predictions() and johnson_neyman() to modelbased instead.
◆Where it's heading
The package is settling into a front-end role — a consistent predict_response() interface over other people's estimation engines — rather than owning the computation itself. The 2.x releases also show a pattern of removing deprecated arguments and clarifying mixed-model semantics, so the interface is being tightened as the backend is outsourced.
◆Prediction
Expect the features lost in the modelbased handover to return as that package's contrast and slope estimation matures, rather than being reimplemented locally.

◆Recent moves

  1. 1y ago

    ggeffects delegates contrasts and slopes to modelbased

    ⚡ SPARK

    Ends the cycle of absorbing marginaleffects breakage by handing the computation to modelbased. It is the clearest statement yet that ggeffects intends to be an interface, not an estimation engine.

    View source ↗
  2. 1y ago

    Five focal terms and formula-based contrast tests

    Raises the focal-term limit to five with patchwork-arranged facet grids, and lets the test argument take a formula for consecutive or reference-level contrasts. Interface breadth in the period just before the modelbased handover.

    View source ↗
  3. 1y ago

    Mixed-model predictions split type from interval

    Separates unit-level from population-level predictions and moves interval choice to its own argument. A breaking change, but a clarifying one — it makes the mixed-model semantics teachable rather than adding capability.

    View source ↗
  4. 1y ago

    Bias correction for back-transformed mixed-model predictions

    Adds bias_correction for non-Gaussian mixed models and drops several deprecated arguments. Part of the steady tightening that runs through the 1.7.x line.

    View source ↗
  5. 1y ago

    Support for WeightIt model classes

    Broadens model coverage to the WeightIt families and fixes multivariate-response and inverse-link interval issues. Typical of how the package grows: one more model class at a time.

    View source ↗
  6. 2y ago

    glmgee support and vcov controls for ggemmeans()

    Adds glmtoolbox model support and brings ggemmeans() to parity with ggpredict() on variance-covariance arguments. Consistency work across the estimation front ends.

    View source ↗