Movatterモバイル変換

propertee 1.0.3

• Fix bug incov_adj() when covariance adjustment modelformula includes transformations of variables

propertee 1.0.2

• Fix bug in variance estimation code when the mean response/offset ina subpopulation of the control condition is not identified

propertee 1.0.1

• Fix small test suite bugs

propertee 1.0.0

• First CRAN submission

propertee 0.7.0

New Features

• HC2 and CR2 standard errors are now available forteeMod objects. CR2 corrections use a new, fast computationthat obviates the need for obtaining the spectral decomposition oforthocomplements of cluster-specific projection matrices.
• Standard errors forteeMod objects now optionallyinclude methodological developments from a forthcoming paper fromWasserman and Hansen:
  1. When computing the meat of the sandwich standard errors, the biascorrection to the residuals from estimating intent-to-treat effects isallowed to differ from the bias correction to residuals from acovariance adjustment model fit. Thetype argument ofvcov_tee() determines the correction for the former, whilethecov_adj_rcorrect argument determines the latter. Thefindings from the simulation study in the paper inform the defaultarguments ofvcov_tee() (and thus,summary.teeMod()): an HC2/CR2 correction for the residualsof the intent-to-treat effect estimates, and an HC1/CR1 correction orthe residuals from the covariance adjustment model fit.
  2. When units used for estimating intent-to-treat effects alsocontribute to fitting the covariance adjustment model, the booleanloco_residuals argument ofvcov_tee()indicates the residuals from effect estimation associated withindividuals in the overlapping units should be replaced with residualsthat use a leave-one-out estimate of the covariance adjustment modelparameters.
• Degrees of freedom provided insummary.teeMod() nowreflect clustering: for CR0 and CR1 standard errors, the associateddegrees of freedom are one less than the number of clusters used forestimation; for CR2 standard errors, degrees of freedom are computedusing the approach of Imbens and Kolesár (2016) (see documentation of.compute_IK_dof() for citation) and leverage the fastcomputational routine mentioned above
• Units of assignment inStudySpecification are nowoptional, though recommended.

Bug Fixes

• Fix minor bugs with thedichotomy argument oflmitt()

propertee 0.6.1

Bug Fixes

• unit_of_assignment(),unitid(),cluster(),uoa(),block(), andforcing() no longer fail automatically when passed anon-numeric or non-character variable. Now, they will first attempt toconvert the variable to a character variable.
• An internal routine invcov_tee() no longer fails whenthe environment in whichvcov_tee() is called differs fromthe environment in which theStudySpecification associatedwith theteeMod is created.

propertee 0.6.0

New Features

• In theircoefficients element,teeModobjects now report estimates of mean quantities in the control condition(response and, if applicable, predictions of response). See thelmitt() man page for further details.
• Introducesetc() (effect of the treatment on controls)andato() (overlap-weighted effect) weighting functions.atc is an alias foretc(), whileolw,owt, andpwt are aliases forato().
• Bump minimum R version to 4.1.0 to allow internal usage of pipes andanonymous functions.

Bug Fixes

• Whenate()-type functions are called withdata arguments that do not include rows associated with allunits of assignment specified in theStudySpecificationobject, the resulting weights reflect assignment probabilities acrossall units of assignment in theStudySpecification, not onlythose represented indata.

propertee 0.5.2

New Features

• Users fitting multipleteeMod objects to test multipleoutcome variables or different levels of a factor treatment variable canpass those models to anmmm object from themultcomp package, then pass themmm object tomultcomp’sglht() function to obtain standarderrors estimated usingvcov_tee(). This requires passingvcov=vcov_tee toglht(). On a technical note,**propertee** now “suggests” installingmultcomp.

Bug Fixes

• Weights produced byate() andett() nolonger produce 0’s or NA’s forStudySpecification objectscreated with multiple columns specifying the blocking scheme

propertee 0.5.1

Bug Fixes

• Variance estimation routine fixes miscalculations in the case whenthere is covariance adjustment and rows are omitted from thelmitt() fit due toNA’s in the covariates ortreatment assignment

propertee 0.5.0

Major Changes

• All references to “design” have been changed to “specification”.
  • *_design is now*_spec(e.g. rct_design is nowrct_spec)
  • Design objects are nowStudySpecificationobjects
  • Thedesign= argument tolmitt() is nowspecification=.

propertee 0.4.1

New Features

• Passingabsorb = TRUE tolmitt withoutspecifying a block proceeds as if the entire sample is a singleblock.

Bug Fixes

• Fixed bug where use ofdichotomy and moderatorvariables inlmitt() could lead to errors due to too longof a formula.

propertee 0.4.0

New Features

• lmitt(), weights calculation functionsate() andett(), and assignment vectorgeneration functionassigned() now accept adichotomy argument that can be used for studies withtime-varying treatment assignment. TheDesign object,unlike before, will not carry information about this dichotomization.Instead, the information stored there reflecting when units wereassigned to treatment (if they were assigned to treatment) will beleveraged to create inverse probability of assignment weights andassignment indicators for datasets that have longitudinal data for thestudy units.

Bug Fixes

• Standard error calculations no longer error when abycolumn is used to uniquely identify rows in the covariance adjustment oreffect estimation sample that cannot be distinguished with informationin theDesign alone

propertee 0.3.10

Bug Fixes

• Linking unit of assignments to clusters for variance estimation nolonger errors whenDesign objects are created with atibble
• cov_adj() does not error with covariance adjustmentmodels fit withrobustbase::glmrob()

propertee 0.3.9

Bug Fixes

• Scaling constants have been updated inestfun.teeMod()to account for a previously missing factor of sqrt(n / n_C) applied tocontributions to the covariance adjustment model estimatingequations

propertee 0.3.8

Breaking Changes

• When model-based standard errors clustered at the level ofassignment are called for in a blocked design,vcov_tee()clusters units of assignment in small blocks, blocks with only onetreated or control unit, together.

propertee 0.3.7

Breaking Changes

• vcov_tee() scales estimating equations using differentconstants than it did before

propertee 0.3.6

Bug Fixes

• Previous procedure for aligning contributions to estimatingequations from first-stage and second-stage models failed when column(s)used for alignment had NA’s. Outputs ofvcov_tee() wereliable to change from call to call as a result. This has beenfixed.

propertee 0.3.5

Improvements

• Diagonal elements ofvcov_tee() matrices lackingsufficient degrees of freedom for estimation are returned as NA’s ratherthan numeric zeros. This is a deviation from thesandwichpackage that aims to provide clarity to results that may otherwiseappear as negative diagonal elements of the vcov matrix

Bug Fixes

• Whenlmitt() is called with a blocked design andabsorb=TRUE, the block-centered assignment and, ifapplicable, moderator and assignment:moderator interaction columns, areno longer centered on the grand mean of the column. This ensures blocksthat do not satisfy positivity of the assignment variable (or positivitywithin a factor level) do not contribute to effect estimation
• lmitt() now accepts references to formula objects

propertee 0.3.4

Improvements

• Computational performance forestfun.teeMod has beenimproved

Bug Fixes

• No more errors due to under-the-hood duplication of a moderatorvariable
• absorb=TRUE estimates have been corrected in the casewhen all observations in a stratum have 0 weights due to only treated orcontrol units of assignment existing in the stratum

propertee 0.3.3

Added Features

• vcov_tee() can accept user-created variance estimationfunctions that start with the prefix.vcov_; thetype argument should take the rest of the function name asan input
• Variance estimation for robust GLM’s (models fit usingrobustbase::glmrob) is now accommodated
• HC1 variance estimates are now accommodated

propertee 0.3.2

Added Features

• Effect estimation for continuous moderator variables is nowsupported

Non-Breaking Changes

• vcov_tee() will return NA’s for the entries of thecovariance matrix that lack sufficient degrees of freedom for anestimate. Informative warnings will accompany the matrix, furtherindicating which standard errors have been NA’d out.

Bug Fixes

• Functions for generating weights,ate() andett(), return weights of 0 rather than infinity for blocksthat contain treated units but no control units.
• Prior covariate adjustment fits were previously incorporated intovariance estimation differently depending on whether one created aSandwichLayer object before callinglmitt() orcalledcov_adj() in theoffset argument of thelmitt() call. This has been corrected, and both ways returnthe same variance estimates.
• Covariate adjustment models that admit rectangular bread matrices,such as those produced byrobustbase::lmrob, are nowaccommodated given the reformulated estimating equations in versionsv0.1.1 and later.
• A contrasts error raised bymodel.matrix() in certaincov_adj() calls has been resolved.

propertee 0.3.1

Breaking Changes

• We now orderteeMod objects’ matrix of estimatingequations based on user-specified ID columns or unit of assignmentID’s.
• Thestats::update function can no longer be called onteeMod objects.

Non-Breaking Changes

• teeMod objects now havelmitt_callslots.
• summary calls onteeMod objects acceptvcov.type arguments to specify the desired standard errorcalculation shown in the output. Acceptable types follow thedocumentation forvcov_tee.
• Shown or printedteeMod objects return morecomprehensible labels for ITT effect outputs.

R Version Compatibility

• Now compatible with R 4.3. Particularly, we advise users workingwith R 4.3 to avoidexpand.model.frame calls onteeMod objects and instead use the internal function.expand.model.frame_teeMod when necessary.

propertee 0.2.1

Breaking Changes

• Stratum fixed effects and subgroup moderating effects can now beaccounted for via theabsorb argument. Previous versionsdid not properly support this functionality. Valid standard errors underabsorption, however, have not been confirmed.

propertee 0.1.1

Breaking Changes

• We have reformulated the estimating equations used to derivestandard errors. In estimation settings we accommodate, testing has notrevealed any differences in standard error estimates between theprevious and current estimating equations, but we do not assure this isthe case for all possible situations.

propertee 0.0.1

• Compatible with R 4.2.3
• Introduces functionality for direct adjusted and design-informedstandard errors accommodating covariance adjustment in the model-basedsetting
• Cluster-robust standard errors can only be estimated using the HC0estimator