Data, R and multiple imputation

Supported imports, the LinkEDA–R bridge and current multiple-imputation support.

Data workflow

Data, R and multiple imputation

Import data

Choose File > Import Data…. The native chooser exposes a Data format menu so the file list can be filtered to one of the formats implemented below.

Data format	Extensions	R component used
Delimited text	`.csv`, `.tsv`, `.txt`	Base R import functions
Excel	`.xls`, `.xlsx`	`readxl`
SPSS	`.sav`, `.zsav`	`haven`
Stata	`.dta`	`haven`
SAS	`.sas7bdat`, `.xpt`	`haven`
R objects	`.rds`, `.rda`, `.RData`	Base R serialization

The program reads the file through R, registers a data set and opens its data sheet. An .RData or .rda file must contain at least one data-frame-like object. Labelled values from foreign formats are preserved where the import path supports them.

The programmatic equivalent is ls_import_data(path), with specialised helpers such as ls_import_spss(), ls_import_excel() and ls_import_rds().

The native application and R

Calling LinkEDA() starts a separate native application process. The R package and the native process communicate through a local session protocol (named pipes and local task notifications on macOS). The native layer owns windows, events and drawing; the shared core owns plot state, case identities and hit testing; R retains the registered data and performs statistical calculations.

This division has practical consequences:

LinkEDA does not replace the active R session.
Data can be supplied at launch with LinkEDA(data) or chosen from the connected R session.
Menu commands can open data from R or return edited data/selected rows to R.
Model fitting, tests, smoothing and pooling are requested from R and displayed in native result windows.
Quit LinkEDA closes the workbench; it is not intended to quit R.

Exchange data with R

The File menu contains the connection commands used by the current implementation. Open Data from R… lists suitable data-frame-like objects. Return commands send the active data or selected rows back to the connected R session. Cancelling a return leaves the R environment unchanged.

When starting from code, these are the two principal forms:

library(LinkEDA)

LinkEDA()          # start first; choose or import data later
LinkEDA(mtcars)    # register and open a data frame immediately

Selections and analysis scopes

Selections are row identities, not a new copy of the data. The current selection can be named and retained as an analysis scope. Saved selections appear alongside Selected, Panel, Plot and All scopes, and may be combined or used to make a subset. An analysis output should therefore identify the named selection it used rather than merely saying “selected observations”.

How linked selection works across plots

Multiple imputation

Multiple imputation is an experimental workflow built around mice. LinkEDA can configure imputation methods, construct a predictor matrix and retain multiple completed versions of a data set. Available methods depend on variable type; current choices include predictive mean matching, normal regression, logistic or polytomous regression, proportional odds, CART and random forest where supported by mice.

Some analyses can pool estimates over the completed data sets, including selected descriptive, mean-comparison, linear-model and generalized-model workflows. Coverage is not uniform: non-parametric mean tests and contingency tables are currently excluded from the multiple-imputation path, and individual analysis windows may impose additional restrictions. Treat pooled output as experimental and check the method reported in the result window.

Note

The presence of a command in the interface does not imply that every variant can pool imputed data. LinkEDA should report an explicit limitation instead of silently analysing a single completed data set.