r2mlm 0.3.2
Major Changes
- Output now returns as numeric rather than characters. (#55)
Minor Edits
- Removes broomExtra dependency. (#52, #57)
- Changes how variable types are checked from if() to is().
r2mlm 0.3.1
Major Changes
- Exported
r2mlm_long_manual
to be user-facing.
Minor Edits
- Updated
r2mlm_manual
and r2mlm_comp_manual
documentation to reflect changes to teachsat
dataset
implemented in version 0.3.0. (#53)
r2mlm 0.3.0
Major Changes
- Adds two manual functions: one for 3-level models (r2mlm3_manual)
and one for models with heteroscedasticity, autocorrelation,
nonlinearity, and non-centered-within-cluster models
(r2mlm_long_manual)
- Bar graph output is now optional. The default behaviour is to output
bar graphs, but if you don’t want graphical output, the argument is
bargraph = FALSE
. For example,
r2mlm(model, bargraph = FALSE)
. (Issue #46)
Bug Fixes
- To test whether clusters are mean-centered, the code computes
cluster means for all level-1 variables, sees if the means are roughly
zero (< .0000001), and if yes then it assigns
clustermeancentered = TRUE
. This update changes the code to
test whether the absolute value of the means are roughly zero,
to address the case in which a cluster has a negative non-zero mean
(that would otherwise mistakenly be assigned to
clustermeancentered = TRUE
because the negative number is
less than 0.0000001). (Issue #41)
- Fixes an issue where models with non-cwc interaction terms were
returning results as though they were centered-within-cluster. r2mlm
returns non-cwc results, r2mlm_comp breaks. (Issue #42)
- Fixed an error thrown if certain groups only have one unit: “Error
in if (variance_tracker == 0) { : missing value where TRUE/FALSE
needed.” Fixed this (#44).
Minor Edits
r2mlm 0.2.0
Major Changes
- Can now accept data with missing points, handles it with listwise
deletion via broomExtra::augment(model). (#23, #29)
- Related to accepting missing data (#23, #29), this update changes
r2mlm_comp()
to accept optional data argument. You can now
call r2mlm_comp(modelA, modelB)
or
r2mlm_comp(modelA, modelB, data)
. If data is provided, the
function will use that data. If data is not provided and models are
hierarchically nested, the function will extract data automatically. If
data is not provided and models are not hierarchically nested, the
function will throw an error asking users to input data.
Bug Fixes
- Bug fix: when groups of 1 exist, variance was returning as NA,
generating “Error in if (variance_tracker == 0) { : missing value where
TRUE/FALSE needed” (#26)
Minor Edits
- Fixed typo in
r2mlm_manual()
documentation (#33)
- Updates documentation of
r2mlm()
and
r2mlm_comp()
to note that models run in lme4
must be formatted with random effects at the end of the formula.
(#30)
- Refactored to increase modularity, adding files: utils.R,
prepare_data.R