Package 'irw'

Description

A dataset of item difficulty estimates and associated standard errors compiled from a curated subset of IRW datasets. Used internally by irw_simu_diff() to generate empirically grounded item parameters for simulation.

Usage

diff_long

Format

A data frame with N rows and 4 variables:

dataset

Dataset identifier (character). Indicates the source dataset in IRW.

item

Item identifier (character). A label for each item.

difficulty

Estimated item difficulty (numeric).

SE

Standard error of the difficulty estimate (numeric).

Details

This pool includes binary-response items from IRW datasets that passed basic quality checks— specifically, datasets with less than 50% missing data and without repeated or duplicate responses. It is accessible to users and is used as the default source in irw_simu_diff().

Source

Curated and compiled by the IRW package authors from selected datasets in the IRW repository.

Description

irw_fetch() and irw_itemtext() keep each table they download in this folder, so later sessions do not export it again – every export counts against the Redivis 30-day export cap.

Usage

irw_cache_dir()

Details

The folder is the IRW_CACHE_DIR environment variable if set, otherwise tools::R_user_dir("irw", "cache") (e.g. ⁠~/.cache/R/irw⁠ on Linux). The Python package reads and writes the same files, but its default folder is ⁠~/.cache/irw⁠; to share one cache between the two languages, set IRW_CACHE_DIR to the same folder for both.

Switch the cache off with options(irw.cache = FALSE) for a session, or IRW_CACHE=0 for all of them.

Value

Path to the cache folder (it may not exist yet).

See Also

irw_cache_info(), irw_clear_cache()

Description

irw_fetch() and irw_itemtext() keep each table they download so that later sessions do not export it again. A copy is used for as long as the table is unchanged on Redivis, and replaced when it changes. See irw_cache_dir() for where the files are and how to switch this off.

Usage

irw_cache_info()

Value

A tibble with one row per cached file: kind ("tables" or "itemtext"), table, hash (Redivis' content hash, which decides whether the copy is current), version (the dataset release it was fetched from), reference, size_mb, fetched_at and path.

Examples

info <- irw_cache_info()
sum(info$size_mb)

Description

Summarize basic response diagnostics in IRW data, including items with a single observed response category and sparse categories in polytomous items.

Usage

irw_check_resp(x, min_count = 5L, min_prop = 0.01, resp_col = "resp")

Arguments

Details

When x is long-format (with columns id, item, resp), diagnostics are recomputed using min_count and min_prop. When x is the wide-format result of irw_long2resp(), any attached diagnostics from check_resp = TRUE are returned directly.

A response category is treated as "sparse" if it has fewer than min_count observed responses OR a within-item proportion smaller than min_prop.

Value

A list with two elements: single_category_items (character vector of item IDs) and sparse_category_items (named list of data frames with columns resp, count, prop).

Description

Delete cached tables

Usage

irw_clear_cache(table = NULL)

Arguments

Value

Invisibly, the number of bytes freed.

See Also

irw_cache_info(), irw_cache_dir()

Description

Returns the names of every IRW table in one collection, so they can be passed straight to irw_fetch(), irw_save_bibtex() or any other vectorised accessor.

Usage

irw_collection(name, quiet = FALSE)

Arguments

Details

Prints the collection's definition and, when the collection does not cover the whole warehouse, says so — see the Coverage section of irw_collections(). Suppress with quiet = TRUE.

Value

A sorted character vector of table names.

See Also

irw_collections(), irw_filter()

Examples

tabs <- irw_collection("depression")
irw_save_bibtex(head(tabs, 3), output_file = file.path(tempdir(), "depression.bib"))

# Compose with the numeric filters
irw_filter(collection = "depression", n_participants = c(500, Inf))

# Cross-collection: math assessments administered in booklets
intersect(irw_collection("math"), irw_collection("multistage"))

Description

The long membership table: one row per (table, collection) pair, with the basis on which that membership was decided — a rule expression, or a curator. Use this to ask the inverse question: what collections is a given table in?

Usage

irw_collection_members(tables = NULL, collection = NULL)

Arguments

Value

A tibble with columns table, collection, basis.

Basis in the exact-instrument collections

In the exact-instrument collections (phq9, gad7, rses, cesd, dass21, panas, tipi, pcl5, gse, erq, mspss, sd3), basis starts with how the table was identified as that instrument:

wording

The study's own item text matches the instrument. Identity is observed.

supplied

IRW identified the instrument and supplied its canonical wording, so a wording match only confirms that identification. Identity is asserted.

name

There is no usable English item text; only the table's metadata names the instrument.

A note may follow after "; ": "supplied; subscale: anxiety (7/21)" for a table holding one whole subscale, "name; partial: 18/20 items" for one holding most of the scale. For the strictly verified set, keep rows whose basis starts with "wording", not rows equal to it.

See Also

irw_collections(), irw_collection()

Examples

irw_collection_members(tables = "frac20")
irw_collection_members(collection = "q_matrix")

# PHQ-9 tables whose own item text was checked against the instrument
m <- irw_collection_members(collection = "phq9")
m[startsWith(m$basis, "wording"), ]

Description

Returns the collection registry: one row per collection, with its kind, definition, the rule that produced it, and how complete its coverage is.

Usage

irw_collections(kind = NULL)

Arguments

Details

Collections are labelled groupings of IRW tables — study designs (rct, clustered, q_matrix), instrument families (big_five, promis), exact instruments (phq9, gad7, rses, ...) and constructs (depression, math). A table can belong to any number of them.

Value

A tibble, one row per collection.

Coverage

coverage says how much of the warehouse the collection's rule actually searched, and it matters when you need "all" of something:

metadata-complete

The rule read irw_metadata(), so it saw every documented table. Note this is not the same as the whole warehouse — irw_metadata() itself has gaps.

tagged-subset-only

The rule read the ⁠construct name⁠ tag, which is filled for only about half of documented tables, and far fewer in the newer warehouses. The definition states the exact count searched. These collections are biased toward older tables and should not be described as exhaustive.

curated-only

Every member was chosen by hand. The exact-instrument collections are of this kind; see irw_collection_members() for how each of their members was identified.

See Also

irw_collection(), irw_collection_members(), irw_filter()

Examples

irw_collections()
irw_collections(kind = "instrument")

Description

Conjoint tables (source = "conj") have one row per respondent, task and profile: id, task, profile, the outcomes (choice, rating, and any choice_<name> or rating_<name>), the profile's attribute levels in attr_* columns, and respondent covariates in cov_*. There is no item or resp, because every profile is a new random bundle of attributes.

Usage

irw_conj_long(df, outcomes = NULL)

Arguments

Details

This function stacks the outcomes into the core layout so that tools built for id/item/resp can be used: item is the name of the outcome question ("choice", "rating", ...), resp its value, and task, profile and the attributes become trial_ columns. Rows with a missing outcome are dropped.

Value

A data frame with columns id, item, resp, trial_task, trial_profile, trial_attr_*, and the remaining columns of df.

Examples

## Not run: 
d <- irw_fetch("kreps_2020_covid_vaccine", source = "conj")
long <- irw_conj_long(d)
table(long$item)

## End(Not run)

Description

Many covariates ship as bare codes – cov_gender in {1, 2} – whose meaning was recorded in the source file (an SPSS or Stata value label) but is not in the IRW table, and the meaning differs from table to table: cucchi_2018_rfq codes 1 = female, rowe_2016_cfs_strain 1 = male. This returns the source's own labels, one row per code, from irw_meta's covariate_labels table. It does not touch the response data, and irw_fetch() output is unchanged.

Usage

irw_covariate_labels(table)

Arguments

Value

A tibble with character columns table, covariate, code, label, one row per code, sorted by table, covariate and code. code is the shipped value written as text (a shipped 1.0 is "1"). label is the source's wording, verbatim and not harmonised across tables, or the literal "[institution name withheld]" where the codes name institutions. Only codes that occur in the shipped column are listed, and a covariate may be partly covered. Zero rows, with a message, when IRW has no labels for the table.

If the irw_meta version in use has no covariate_labels table (an older or pinned version), this is an error of class irw_covariate_labels_unavailable, not an empty result, so it cannot be mistaken for "no labelled covariates".

See Also

irw_covariates, whose labels = TRUE applies these labels.

Examples

irw_covariate_labels("cucchi_2018_rfq")

Description

Reduces IRW long-format data to one row per id, keeping the columns that describe the person rather than the response. Optionally aligns those rows to the row order of a wide response matrix, which is what most psychometric packages need when a grouping variable is passed alongside a response matrix.

Usage

irw_covariates(df, cols = NULL, align = NULL, labels = FALSE, table = NULL)

Arguments

Details

irw_long2resp() keeps only id, item, and the response column, so person-level covariates are dropped and the row order of its result is not sorted. Aligning a covariate to that result by hand requires a match() on id, and getting it wrong silently misassigns people to groups. Passing the wide result as align does that match for you.

When cols is NULL, the person-level columns are detected automatically: every column other than id and the response-level columns (item, resp, rater, rt, text, date, source_table) that takes exactly one value within every id. A column with a value for some of a person's rows and NA for others counts as varying and is not selected. Note that wave is treated as person-level only when each person appears in a single wave.

Value

A data frame with one row per id and columns id followed by the person-level covariates. When align is supplied, rows follow that order and ids not present in df yield NA rows.

See Also

irw_long2resp, irw_covariate_labels

Examples

df <- data.frame(
  id = c(1, 1, 2, 2),
  item = c("i1", "i2", "i1", "i2"),
  resp = c(1, 0, 1, 1),
  cov_group = c("A", "A", "B", "B"),
  stringsAsFactors = FALSE
)
irw_covariates(df)

# Aligned to the row order of a wide response matrix
wide <- irw_long2resp(df, id_density_threshold = NULL)
irw_covariates(df, align = wide)

# With the source's value labels (cov_gender 1/2 -> female/male)

if (irw_has_credentials()) {
  df <- irw_fetch("cucchi_2018_rfq")
  irw_covariates(df, labels = TRUE, table = "cucchi_2018_rfq")
}

Description

Inverse of irw_recode(): maps codes such as I0001 back to the original identifier values using the recode key.

Usage

irw_decode(x, key = NULL, cols = NULL)

Arguments

Details

Works on long-format data (columns named in the key are decoded in place) and on the wide output of irw_long2resp(), where item codes appear as column names. irw_long2resp() prefixes item columns with item_; that prefix is handled and preserved. Values or column names that are not found in the key are left untouched.

Value

x with codes replaced by the original values and the "irw_recode_key" attribute removed.

See Also

irw_recode, irw_recode_key

Description

Downloads a specified dataset table from the IRW datasource.

Usage

irw_download(
  table_name,
  path,
  overwrite = FALSE,
  source = "core",
  sim = FALSE,
  comp = FALSE,
  nom = FALSE
)

Arguments

Value

A message confirming the file download location.

Description

Retrieves one or more tables from IRW and returns them as tibbles. This downloads every row of each table and counts against the Redivis export quota; if you only need the item or response value sets, or per-item counts and ranges, use irw_table_sets(), which answers with a server-side query. If the table includes a character resp column, the function attempts to coerce it into numeric. Strings like "NA", "", and NA are treated as missing values. A warning is issued only if other non-numeric values are encountered.

Usage

irw_fetch(
  name,
  source = "core",
  dedup = FALSE,
  sim = FALSE,
  comp = FALSE,
  nom = FALSE,
  resp = FALSE
)

Arguments

Details

Each table downloaded is also kept on disk (see irw_cache_dir()) and reused by later fetches, in this session or another, for as long as the table is unchanged on Redivis; the Python package shares the same copies. Switch this off with options(irw.cache = FALSE) or IRW_CACHE=0.

A table the IRW found through another collection, such as openESM, prints that collection's note and citation once per session per collection. Silence it with options(irw.source_note = FALSE) or IRW_SOURCE_NOTE=0.

Value

If a single name is provided, returns a tibble. If multiple, returns a named list of tibbles (or error messages, if retrieval failed).

See Also

irw_table_sets() for value sets and summaries without a download; irw_cache_info() for the tables kept on disk.

Examples

# Main IRW data
irw_fetch("environment_ltm")

# Deduplicated
irw_fetch("pks_probability", dedup = TRUE)

# Simulation data
irw_fetch("gilbert_meta_3", source = "sim")

# Competition data
irw_fetch("collegefb_2021and2022", source = "comp")

Description

Returns the names of datasets in the Item Response Warehouse (IRW) that match user-specified metadata, tag values, variable presence, and license criteria.

Usage

irw_filter(
  n_responses = NULL,
  n_categories = NULL,
  n_participants = NULL,
  n_items = NULL,
  responses_per_participant = NULL,
  responses_per_item = NULL,
  density = c(0.5, 1),
  var = NULL,
  age_range = NULL,
  child_age__for_child_focused_studies_ = NULL,
  construct_type = NULL,
  construct_name = NULL,
  sample = NULL,
  measurement_tool = NULL,
  item_format = NULL,
  primary_language_s_ = NULL,
  longitudinal = NULL,
  n_actors = NULL,
  n_respondents = NULL,
  n_attributes = NULL,
  outcome = NULL,
  country = NULL,
  domain = NULL,
  choice_object = NULL,
  source = "core",
  comp = FALSE,
  sim = FALSE,
  nom = FALSE,
  license = NULL,
  collection = NULL
)

Arguments

Details

Filtering is based on:

• Numeric metadata: number of responses, participants, items, etc.

• Tag metadata: e.g., construct type, sample, measurement tool

• Variable presence: e.g., rt, wave, cov_

• License type: e.g., "CC BY 4.0", "CC0 1.0"

Metadata and Tag-Based Filtering

To explore available metadata:

• summary(irw_metadata()) — numeric summaries (e.g., n_responses, density)

• irw_tags() — full tag metadata table (1 row per dataset)

• irw_tag_options("column_name") — valid values (with counts) for any tag column

• irw_license_options() — available license values with frequencies

• irw_collections() — the 22 collections, with kind, definition and coverage

• irw_collection("name") — the table names in one collection

Tag-based metadata (e.g., construct_type, sample, item_format) can be passed directly as named arguments. See the parameter list below for supported tag columns. Tag filters are available for source = "core" and source = "nom"; "comp" and "sim" have no tags by design and reject them.

Value

A sorted character vector of dataset names that match all specified filters, or character(0) if no match is found.

Examples

# Numeric filters
irw_filter(n_responses = c(1000, Inf), n_items = c(10, 50))
irw_filter(n_participants = c(500, Inf), density = c(0.3, 0.9))

# Variable presence
irw_filter(var = "rt")
irw_filter(var = c("wave", "cov_"), density = NULL)

# Tag metadata filtering
irw_filter(construct_type = "Affective/mental health", sample = "Educational")
irw_tag_options("construct_type")  # view tag values

# License filtering
irw_license_options()
irw_filter(license = "CC BY 4.0")

# Collection filtering
irw_collections()                                    # what exists
irw_filter(collection = "rct")
irw_filter(collection = "depression", n_participants = c(500, Inf))
irw_filter(collection = c("rct", "response_time"))   # union, not intersection

# Filter by response category complexity
irw_filter(n_categories = 2, density = NULL)           # binary
irw_filter(n_categories = c(3, 5), density = NULL)     # small multi-category
irw_filter(n_categories = c(10, Inf), density = NULL)  # large category sets
irw_filter(source = "sim", n_items = c(10, Inf))
irw_filter(source = "comp", n_actors = c(2, 10))
irw_filter(source = "conj", outcome = "rating", country = c("US", "GB"))
irw_filter(source = "conj", domain = "immigration_asylum")          # the immigration conjoints
irw_filter(source = "conj", domain = c("candidate_choice", "parties_platforms"),
           choice_object = "person")

Description

Returns the names of competition tables that match user-specified criteria. Competition filtering uses:

• Numeric metadata: n_responses, n_actors

• License: via comp_biblio (Derived_License)

Usage

irw_filter_comp(n_responses = NULL, n_actors = NULL, license = NULL)

Arguments

Details

Notes:

• Competition datasets do not have tags or item text metadata.

• Competition metadata table names are stored in lowercase.

Value

A sorted character vector of competition table names (lowercase), or character(0) if no match is found.

Description

Shows which Redivis version each IRW dataset is being read at, and whether that version is pinned for the session or is simply the current release. Record these tags to make an analysis reproducible; replay them with irw_set_version.

Usage

irw_get_version(dataset = NULL)

Arguments

Value

A data frame with columns dataset, version, and pinned.

See Also

irw_set_version, irw_reset_version

Examples

irw_get_version()
irw_get_version("item_response_warehouse")

Description

Fetching and listing IRW tables needs no Redivis login: the warehouse is public, and redivis 0.12.14 and later send those requests anonymously. A login is still needed for the server-side queries behind irw_table_sets, and for every warehouse call on an older redivis, which authenticates unconditionally. In an interactive session irw opens the browser login when one is needed; in a script or other non-interactive session it stops with instructions instead. This function reports whether the redivis client and a login are already in place, so code can skip the calls that would otherwise stop.

Usage

irw_has_credentials()

Value

TRUE if redivis is installed and credentials are available (a REDIVIS_API_TOKEN environment variable or a cached login), otherwise FALSE.

Examples

irw_has_credentials()

Description

Compare two sets of predicted probabilities for the same binary outcomes. The IMV expresses how much better the second model predicts than the first, on the scale of a weighted coin: each model's geometric mean likelihood is converted to the weight of a coin with the same entropy, and the IMV is the proportional gain in that weight.

Usage

irw_imv(data, p1, p2, resp = "resp", eps = 1e-06)

Arguments

Details

An IMV of 0 means the two models predict equally well; 0.05 means model 2 is equivalent to a coin 5\ is not symmetric: irw_imv(data, "p1", "p2") is the gain from moving from p1 to p2, and swapping the arguments does not simply flip the sign.

Predicted probabilities are clamped to [eps, 1 - eps] before the likelihoods are computed, so predictions of exactly 0 or 1 do not produce an infinite log-likelihood.

The computation is imv.binary from the imv package; irw_imv() adds the data-frame interface and input checks.

Out-of-sample predictions are the intended use: comparing in-sample predictions will favour the more flexible model by construction. The "Comparing Models Out of Sample with the IMV" article works through a cross-validated comparison of a Rasch model against a 2PL.

Value

A single numeric value: the proportional gain in coin weight from p1 to p2.

References

Domingue, B. W., Rahal, C., Faul, J., Freese, J., Kanopka, K., Rigos, A., Stenhaug, B., & Tripathi, A. (2025). The InterModel Vigorish (IMV) as a flexible and portable approach for quantifying predictive accuracy with binary outcomes. PLOS ONE, 20(3), e0316491. doi:10.1371/journal.pone.0316491

See Also

irw_predict for producing p1 and p2 from fitted mirt models.

Examples

set.seed(1)
truth <- rbinom(500, 1, 0.7)
# A model that knows the base rate beats one that guesses at random.
irw_imv(truth, p1 = rep(0.5, 500), p2 = rep(0.7, 500))

Description

Prints information about the IRW datasets or a specific table.

Usage

irw_info(
  table_name = NULL,
  details = FALSE,
  source = "core",
  comp = FALSE,
  sim = FALSE,
  nom = FALSE
)

Arguments

Details

If table_name is NULL, shows combined totals across all IRW datasets. Use details = TRUE to also print a per-dataset breakdown (main IRW only). If a table name is provided, shows details for that table.

Value

Invisibly returns NULL.

Examples

irw_info()                 # Combined totals
irw_info(details = TRUE)   # Combined + per-dataset breakdown
irw_info("frac20")      # Specific IRW table
irw_info("t20_hyper", source = "comp")  # Competition table
irw_info("gilbert_meta_3", source = "sim")

Description

Returns item-level text metadata for a given IRW table, if available. If no item text table is available for the specified table, the function returns NULL and prints a message.

Usage

irw_itemtext(table_name)

Arguments

Value

A tibble containing item text metadata, or NULL if unavailable.

Disclaimer

IRW item text is reconstructed from published articles, codebooks, and supplementary materials using a largely automated pipeline with partial human review, and is provided for research purposes only. We make no guarantee that the text is accurate, complete, or correctly aligned with the item identifiers in the corresponding IRW response data; the original source document is always authoritative. Users are responsible for verifying item text against that source before drawing substantive conclusions. Documented discrepancies are listed at https://itemresponsewarehouse.org/itemtext_issues.html.

Note also that the licenses recorded for IRW response data do not extend to the instruments themselves. Inclusion of item text here implies no license to reuse an instrument; copyright and related rights remain with the original rights holders, and users are responsible for obtaining any permissions their intended use requires. The exception is a table whose wording its source released under an open licence: that licence is printed after the fetch and is returned by irw_itemtext_licence.

Examples

irw_itemtext("gilbert_meta_49")

Description

Most IRW item text carries no licence of its own: the response-data licence does not extend to the instrument, and inclusion implies no licence to reuse it. A minority of tables are different: their wording was released by its source under an open licence (CC BY-SA 4.0, CC BY-NC 4.0, ...), and IRW republishes it under that licence, with attribution and a note of its changes. This function returns that record for one table, or NULL when the wording carries no licence of its own.

Usage

irw_itemtext_licence(table_name)

Arguments

Details

The record is the IRW repository's itemtext/itemtext_licences.csv, fetched once per session. irw_itemtext() prints the same line after fetching a licensed table.

Value

A one-row data frame with columns table, licence, licence_url, attribution, changes and basis, or NULL if the wording carries no licence of its own (or the record could not be fetched, in which case a warning says so).

Examples

irw_itemtext_licence("iowa_gambling_task")

Description

Returns a data frame showing the number of datasets associated with each license.

Usage

irw_license_options(source = "core", comp = FALSE, sim = FALSE, nom = FALSE)

Arguments

Value

A data.frame with 'license' and 'count' columns.

Description

Returns a character vector of IRW table names for which item text is available.

Usage

irw_list_itemtext_tables()

Value

A sorted character vector of IRW table names with item text metadata.

Disclaimer

IRW item text is reconstructed from published articles, codebooks, and supplementary materials using a largely automated pipeline with partial human review, and is provided for research purposes only. We make no guarantee that the text is accurate, complete, or correctly aligned with the item identifiers in the corresponding IRW response data; the original source document is always authoritative. Users are responsible for verifying item text against that source before drawing substantive conclusions. Documented discrepancies are listed at https://itemresponsewarehouse.org/itemtext_issues.html.

Note also that the licenses recorded for IRW response data do not extend to the instruments themselves. Inclusion of item text here implies no license to reuse an instrument; copyright and related rights remain with the original rights holders, and users are responsible for obtaining any permissions their intended use requires.

Examples

irw_list_itemtext_tables()

Description

Retrieves a summary of available tables in the IRW database, including their name, number of rows, and variable count, sorted alphabetically by table name.

Usage

irw_list_tables(source = "core", sim = FALSE, comp = FALSE, nom = FALSE)

Arguments

Value

A data frame with the following columns:

Examples

irw_list_tables()               # Main IRW database
irw_list_tables(source = "sim")   # Simulation dataset
irw_list_tables(source = "comp")  # Competition dataset

Description

Converts an IRW-compliant dataset from long format to wide format while handling optional metadata elements and filtering by wave.

Usage

irw_long2resp(
  df,
  wave = NULL,
  id_density_threshold = 0.1,
  agg_method = NULL,
  check_resp = FALSE,
  resp_col = "resp"
)

Arguments

Details

This function applies a default sparsity filter with a threshold of 0.1 for id density to avoid excessive sparsity. Users can disable filtering by setting id_density_threshold = NULL.

If check_resp = TRUE, the function will run a basic diagnostic using default thresholds (min_count = 5, min_prop = 0.01), attach the results as an attribute "resp_checks" on returned object, and add a short NOTE with counts of flagged items. A response category is considered sparse if it has fewer than min_count responses OR a within-item proportion smaller than min_prop. For more control, use irw_check_resp() directly.

Value

A data frame in wide format where rows represent id values and columns represent item_* responses. The column used for response values is recorded in the "resp_col" attribute. If check_resp = TRUE, an attribute "resp_checks" is attached:

• single_category_items: character vector of item IDs with only one observed category.

• sparse_category_items: named list; each element is a data.frame with sparse categories for that item (columns: resp, count, prop).

Description

Identifies and merges tables that share the same DOI or, if DOI is missing, the same BibTex entry. If tables do not have the same structure, only those with identical structures will be merged. This function performs several consistency checks before merging, and prompts the user for confirmation if inconsistencies are found.

Usage

irw_merge(table_name, add_source_column = TRUE)

Arguments

Details

irw_merge() performs the following checks before proceeding with the merge:

1. Number of Respondents Check:

   • The function checks if the number of respondents is consistent across all tables.

   • If the number of respondents is consistent, the merge proceeds.

   • If the number of respondents is inconsistent across tables, a WARNING message is displayed:

     === WARNING ===
     The number of respondents (N) is not consistent across tables.
     Merging tables with inconsistent N respondents may require human judgment.
     

2. ID Consistency Check:

   • Checks if all tables share the same set of IDs.

   • Checks if the IDs form a strict sequence (e.g., 1, 2, 3, ...).

   • If IDs are not consistent across tables, a NOTE message is displayed:

     === NOTE: ===
     - IDs do not match across tables.
     - IDs are sequential (1...n). You may need to manually verify the IDs, 
       as there could be multiple studies with different subjects, where IDs 
       are the same in both studies.
     

3. Item Column Overlap Check:

   • Verifies that there is no overlap of item columns across tables.

   • If overlapping items are found, a NOTE message is displayed:

     === NOTE: ===
     - There are items that overlap across tables.
     

Tables whose bibliography entry is missing (recorded as "NA" in the biblio table) are never treated as merge candidates for one another.

In a non-interactive session the confirmation prompts cannot be answered, so the merge proceeds by default; the checks above are still reported as messages.

Value

A merged data frame containing all tables with the same DOI or BibTex, or NULL if no merge candidates are found.

Description

Fetches the metadata table from Redivis and returns it as a tibble. Automatically checks for updates and refreshes only when needed.

Usage

irw_metadata(source = "core", sim = FALSE, comp = FALSE, nom = FALSE)

Arguments

Details

Each source has its own columns. For source = "conj" a row is one conjoint experiment: n_respondents, n_rows, n_tasks, n_profiles, n_attributes, outcomes, n_optout_tasks, and the design facts country, display_language, label_language, restrictions (none, yes, observed or unknown), restrictions_note, task_source and profile_source (recorded or inferred from row order), and the topic tags domain (what the study is about; one or more values joined with ";", primary first) and choice_object (what one profile represents). Both are filters in irw_filter, which documents their vocabularies.

For source = "rank" a row is one rankings table: the counts n_respondents, n_rows, n_tasks, n_items, max_resp (highest position marked) and n_duplicate_marks, and the design record (method, seats, n_positions, n_candidates, n_ballots_cast, n_ballots_with_marks, jurisdiction, date, office, ...), whose codes are documented in the irw repo's data/rankings/README.md.

Value

A tibble containing metadata information.

Description

Attach per-response predicted probabilities from a fitted unidimensional, dichotomous mirt model to long-format IRW data. This is the piece needed to compare models out of sample with irw_imv.

Usage

irw_predict(model, resp, newdata = NULL, ability = "EAP")

Arguments

Details

Person abilities are taken from the fit itself with mirt::fscores(), so row i of the scored matrix is person resp$id[i]. Passing the wide frame that produced the fit is what lets the function recover that mapping; supplying ids in a different order than the rows given to mirt::mirt() would silently mispair people with abilities.

Probabilities come from mirt::probtrace(), so any dichotomous mirt item type (Rasch, 2PL, 3PL, 4PL, ...) is handled without assuming a particular parameterisation.

newdata may contain any id/item pairs, including ones held out of the fit, as long as each person appears in resp and each item was estimated. That is the cross-validation case: fit on part of the data, predict the rest.

Requires the mirt package.

Value

newdata with an added numeric column p, the predicted probability of a response of 1.

See Also

irw_imv, irw_long2resp

Examples

df <- irw_simdata(n_id = 200, n_item = 10, model = "2PL", seed = 1)
wide <- irw_long2resp(df)
fit <- mirt::mirt(wide[setdiff(names(wide), "id")], 1, "Rasch")
preds <- irw_predict(fit, wide)

Description

Rankings tables (source = "rank") keep the core column names, but resp is a position: one row per respondent (id), ranking (task) and marked position, resp = 1 for the first choice, as recorded. Two things in the stored table need care before a person-by-item pivot or a Plackett-Luce fit: a skipped position stays a gap (a ballot marked at positions 1 and 3 keeps resp 1 and 3), and a candidate marked twice keeps both rows. Counting rules for both differ by jurisdiction, so the table does not apply one; this function applies the common one.

Usage

irw_rank_compress(df, drop_repeats = TRUE)

Arguments

Details

Within each id x task, rows are taken in resp order. With drop_repeats = TRUE an item that has already been marked at an earlier position is dropped at every later one (most counting rules treat the later mark as a skip). The reserved items (write-in) and (overvote) are never dropped: two write-in rows are usually two different people, and two overvote rows are two separate events. Then resp is renumbered 1..k in rank order, so the gaps close. Every other column is kept as it was. The result is sorted by id, task, resp.

The Python package's irw.rank_compress() does exactly the same; the two are tested on one shared fixture.

Value

A data frame with the same columns as df, resp an integer running 1..k within each id x task.

See Also

irw_fetch, irw_metadata with source = "rank" for each table's design record.

Examples

d <- data.frame(id = c(1, 1, 1, 2), task = 1, item = c("c01", "c03", "c01", "c02"),
                resp = c(1, 3, 4, 2))
irw_rank_compress(d)
irw_rank_compress(d, drop_repeats = FALSE)
## Not run: 
d <- irw_fetch("eurovision_2016_final", source = "rank")
first_choices <- subset(irw_rank_compress(d), resp == 1)

## End(Not run)

Description

Replaces the values of identifier columns (by default id and item) with short ASCII codes such as P0001 and I0001. This is useful when identifiers are in a script or use characters that are awkward to type on a US keyboard, which makes subsetting, writing model formulas, and reading wide-format column names difficult.

Usage

irw_recode(df, cols = c("id", "item"), prefix = NULL)

Arguments

Details

Codes are assigned in sorted order of the unique non-missing values of each column, so they are deterministic for a given set of values. They are, however, only meaningful relative to their key: two different subsets of the same table will generally not produce the same codes.

The mapping is attached to the result as an attribute and can be retrieved with irw_recode_key(). Save it if you need it later, because most base R operations on a data frame drop attributes. The original values can be restored with irw_decode().

Note that recoding item breaks manual joins against irw_itemtext() output, which is keyed on the original item identifiers. Use irw_recode_key() to map back before joining.

Value

df with the requested columns replaced by character codes and an "irw_recode_key" attribute holding a data frame with columns column, original, and code.

See Also

irw_decode, irw_recode_key

Examples

df <- data.frame(
  id = c("María-01", "María-01", "René-02"),
  item = c("Q_alpha", "Q_beta", "Q_alpha"),
  resp = c(1, 0, 1),
  stringsAsFactors = FALSE
)
d <- irw_recode(df)
head(d)
irw_recode_key(d)
identical(irw_decode(d)$item, df$item)

Description

Retrieve the Recode Key from an Object Returned by irw_recode()

Usage

irw_recode_key(x)

Arguments

Value

A data frame with columns column, original, and code.

See Also

irw_recode, irw_decode

Description

Returns one dataset, or the whole session, to reading the current release.

Usage

irw_reset_version(dataset = NULL)

Arguments

Value

Invisibly, the named character vector of pins still in effect.

See Also

irw_set_version, irw_get_version

Examples

irw_reset_version("item_response_warehouse")
irw_reset_version()   # unpin everything

Description

Inverse of irw_long2resp(): converts a wide response data frame/matrix into long format with columns id, item, resp.

Usage

irw_resp2long(x, id = TRUE)

Arguments

Value

A data.frame with columns id, item, resp.

Description

Saves BibTeX entries for one or more IRW tables to a specified output file. Updates the BibTeX key to match the table name.

Usage

irw_save_bibtex(
  table_names,
  output_file,
  source = "core",
  comp = FALSE,
  sim = FALSE,
  nom = FALSE
)

Arguments

Details

A table the IRW found through another collection, such as openESM, also gets that collection's own entry (e.g. siepe2026openesm), added once however many of its tables are requested.

Value

Invisibly returns BibTeX entries as a character vector.

Description

Fixes the version of one IRW dataset for the remainder of the R session, so that fetches, listings, and metadata are reproducible. Pin at the top of an analysis script and the script will return the same data when it is re-run, even after IRW has been updated.

Usage

irw_set_version(dataset, version)

Arguments

Details

The pin covers every lookup that reads the dataset, not just fetches: irw_fetch(), irw_filter(), irw_list_tables(), and metadata all read the pinned version.

Each IRW dataset is versioned independently on Redivis, so pinning is per-dataset. irw_get_version() reports the versions currently in use, which is how you find the tags to write into a script.

A table that does not exist in the pinned version is an error rather than a silent fetch from the current release: quietly mixing versions would break the reproducibility the pin exists to provide.

Value

Invisibly, the named character vector of all pins in effect.

See Also

irw_get_version, irw_reset_version

Examples

irw_set_version("item_response_warehouse", "v32.0")
irw_set_version("item_response_warehouse_2", "v1.8")

df <- irw_fetch("kfcovid_fear_Li2020")   # served from v32.0

irw_reset_version()                      # back to the current release

Description

Simulates item response data from 1PL, 2PL, or 3PL logistic IRT models. Returns a long-format data frame with id, item, and resp columns.

Usage

irw_simdata(
  n_id = 1000,
  n_item = 20,
  model = "1PL",
  a = NULL,
  b = NULL,
  g = NULL,
  theta = NULL,
  theta_mean = 0,
  theta_sd = 1,
  seed = NULL,
  return_params = FALSE
)

Arguments

Value

A data frame with columns id, item, resp, or a list with additional parameters if return_params = TRUE.

Examples

# 1PL (default)
dat <- irw_simdata(n_item = 5)

# 3PL with latent traits drawn from N(-0.5, 1)
sim <- irw_simdata(n_item = 5, model = "3PL", theta_mean = -0.5, return_params = TRUE)
head(sim$data)

Description

Generates pairwise outcomes among n_agent agents using the Davidson model with ties. (See https://link.springer.com/article/10.3758/s13428-021-01714-2). Returns one row per comparison with agent_a, agent_b, and winner.

Usage

irw_simdata_comp(
  n_agent = 100,
  n_pairs = 10000,
  nu = 0,
  theta = NULL,
  theta_mean = 0,
  theta_sd = 1,
  seed = NULL,
  return_params = FALSE
)

Arguments

Value

A data.frame with columns:

• agent_a (int): index of first agent

• agent_b (int): index of second agent

• winner (chr): one of "agent_a","agent_b", or "draw".

If return_params = TRUE, returns a list with:

• data: the simulated data frame.

• theta: vector of simulated thetas used

• nu: tie parameter

Examples

# --- Basic usage ---
d <- irw_simdata_comp(n_agent = 100, n_pairs = 10000, nu = 0)
head(d)

# Simulate pairwise comparison data and recover abilities with a Davidson model

if (requireNamespace("gnm", quietly = TRUE) &&
    requireNamespace("BradleyTerry2", quietly = TRUE)) {
  th <- rnorm(n = 50, mean = 0.5, sd = 3)
  pairs <- irw_simdata_comp(
    n_agent = 50, n_pairs = 3000, nu = 0.1,
    theta = th, seed = 1, return_params = TRUE
  )

  x <- pairs$data
  x$winner <- ifelse(x$winner == "agent_b", -1, 1)
  x$winner <- ifelse(x$winner == "draw", 0, x$winner)
  x$agent_a <- as.factor(x$agent_a)
  x$agent_b <- as.factor(x$agent_b)
  x$pair <- seq_len(nrow(x))  # add pair id for expandCategorical

  library(gnm)
  library(BradleyTerry2)

  pairs.tri <- expandCategorical(x, "winner", idvar = "pair")

  dav <- gnm(
    count ~ GenDavidson(
      winner == 1, winner == 0, winner == -1,
      player1 = agent_a,
      player2 = agent_b
    ) - 1,
    eliminate = pair,
    family = poisson,
    data = pairs.tri
  )

  plot(
    pairs$theta, coef(dav)[-1],
    xlab = "True theta",
    ylab = "Estimated ability (shifted)",
    main = "Davidson recovery using gnm"
  )
  abline(lm(coef(dav)[-1] ~ pairs$theta), lty = 2)
}

Description

This function generates simulated item difficulties by drawing from normal distributions centered around existing difficulty estimates and their associated standard errors. The result is a mixture distribution, from which new difficulties are sampled using inverse CDF sampling.

Usage

irw_simu_diff(
  num_items = 10,
  num_replications = 1,
  irw_names = NULL,
  difficulty_pool = NULL
)

Arguments

Details

By default, the function uses diff_long, a built-in dataset included in the irw package. This dataset contains item difficulty estimates and standard errors from a curated subset of IRW datasets. You can:

• Use the full IRW difficulty pool (diff_long)

• Filter to specific IRW datasets via irw_names

• Provide your own difficulty pool via difficulty_pool

This method is based on Zhang et al. (2025), which constructs realistic empirical distributions by accounting for uncertainty around item difficulty estimates.

Value

A numeric vector of difficulties (if num_replications = 1), or a data frame with replication and difficulty columns (if num_replications > 1).

References

Zhang, L., Liu, Y., Molenaar, D., & Domingue, B. (2025). Realistic Simulation of Item Difficulties. https://doi.org/10.31234/osf.io/jbhxy_v1

Examples

# Use all IRW data (default)
irw_simu_diff(num_items = 5)

# Filter to specific IRW datasets
irw_simu_diff(num_items = 5, irw_names = c("psychtools_epi", "psychtools_blot"))

# Use a custom difficulty pool
my_pool <- data.frame(dataset = "x",
                      difficulty = c(-0.2, 0.1),
                      SE = c(0.1, 0.2))
irw_simu_diff(num_items = 5, difficulty_pool = my_pool)

# Explore built-in IRW difficulty pool
head(diff_long)
unique(diff_long$dataset)

Description

Answers set and summary questions about a table – which item codes it contains, which response values occur, how many rows each item has – using server-side aggregate queries rather than exporting the table. This matters for large tables: computing the item set of a 68-million-row table with irw_fetch() downloads all 68 million rows, while irw_table_sets() returns the same answer in seconds and does not consume the Redivis export quota.

Usage

irw_table_sets(name, source = "core", per_item = FALSE)

Arguments

Value

For a single name, a list with elements:

table

Fully qualified Redivis reference for the table.

n_rows

Total number of rows.

items

Sorted character vector of distinct item values.

resp

Sorted vector of distinct resp values, numeric when all values are numeric and character otherwise. "NA" and empty strings are treated as missing, matching irw_fetch().

per_item

Data frame of per-item summaries, or NULL when per_item = FALSE or the table has no item or resp column.

For several names, a named list of those lists, one per table, in the order given. Each element is exactly what the single-name call returns.

Examples

sets <- irw_table_sets("rosenberg_selfesteem")
sets$items
sets$resp

irw_table_sets("condon_2024_sapa_personality", per_item = TRUE)$per_item

# Several tables at once: no export, one set of queries per table
sweep <- irw_table_sets(c("rosenberg_selfesteem", "environment_ltm"))
lengths(lapply(sweep, `[[`, "items"))

Description

Returns a data frame of unique tag values from a given tag metadata column, along with the number of datasets each tag appears in. Multi-tag fields hold a comma-joined list; no tag value contains a comma (see .irw_split_tags).

Usage

irw_tag_options(column, source = "core")

Arguments

Value

A data.frame with columns: tag and count, sorted by descending frequency.

Description

Fetches the tags table from Redivis and returns it as a tibble. Link to Redivis table: https://redivis.com/datasets/bdxt-4fqe5tyf4/tables/7nkh-6m7wq01yv Automatically checks for updates and refreshes only when needed.

Usage

irw_tags(tables = NULL, source = "core")

Arguments

Value

A tibble containing tags information.

Examples

irw_tags()                  # core
irw_tags(source = "nom")    # nominal

Description

Pins every IRW dataset at once to what a single IRW version held, so that irw_fetch() and everything around it return the same data whenever the script is re-run – even after IRW has been corrected or extended. This is the reproducibility switch: put one call at the top of an analysis and the rest of the script is frozen.

Usage

irw_use_version(version = NULL, date = NULL, quiet = FALSE)

Arguments

Details

irw_use_version(332)
df <- irw_fetch("gilbert_meta_12")   # v332's copy, today and next year

Called with no arguments it pins the newest version and reports its number: run it at the start of a project and record the number it prints.

The pin covers the metadata and item text datasets too, not just response data, and lasts for the session; irw_reset_version() lifts it.

Value

Invisibly, a data frame of dataset and version as pinned, with the IRW version number in the attribute irw_version.

Each dataset is pinned to its own version

The response data is spread over several Redivis warehouses ("shards"), each released on its own schedule, so one IRW version means a different Redivis version in each. That is handled: every dataset is pinned to the tag the manifest records for it, and a single irw_fetch() call may read several shards at different versions.

irw_use_version(200)   # shard 1 at v36.0, shard 2 at v3.3
L <- irw_fetch(c("idcr_martinez_2023_numSeries", "gcb5_2025"))

A dataset that did not exist yet

Not every IRW dataset has a release in every IRW version – the sixth warehouse shard is younger than the first. Rather than let those fall through to the current release and silently mix versions into a "reproduced" run, they are pinned to nothing and error if something reads them. Fetches are unaffected: a table that did not exist then is simply not found.

Prefer a version number to a date

A date has to be resolved to a version, and before 21 July 2026 that resolution is approximate for the older warehouse shards – see irw_version(). Pinning by date therefore warns; pinning by number never has to.

See Also

irw_version(), irw_get_version(), irw_reset_version()

Examples

irw_use_version(332)              # read the corpus as v332 held it
irw_use_version()                 # freeze today's version; record the number
irw_use_version(date = "2026-08-01")
irw_reset_version()               # back to the current release

Description

IRW is made of eleven Redivis datasets that are versioned independently, so no single Redivis version describes the corpus. The IRW version number does: it increments whenever any dataset is published and names one exact combination of the eleven. Cite it in a paper and a reader can reconstruct the data you used.

Usage

irw_version(date = NULL, version = NULL)

Arguments

Details

Called with no argument, reports the newest IRW version. Called with a version, reports exactly what that version held. Called with a date, reports the version that was live then – which is how you recover what an analysis run months ago was actually reading.

Use irw_use_version() to make the session read one of these versions.

Value

A data frame with one row per dataset and columns dataset, version, released_at, and approximate. The IRW version number is attached as the attribute irw_version and printed as a message.

Dates before 21 July 2026 are approximate

Redivis overwrote its own release timestamps for the older warehouse shards during a platform migration: 142 of the corpus' 332 released versions claim to have been released inside one 80-minute window on 2026-07-21. For those, the manifest records the earliest date the version could have been live (its creation date) and marks the row bracketed. A lookup that lands on one warns, because the tag may then be wrong too – a later version could already have been released inside the bracket. IRW version numbers themselves are always exact; only the mapping from a date to a version is affected. This is why a paper should cite the number, not the date.

See Also

irw_use_version() to read a whole IRW version, irw_set_version() to pin one dataset, irw_get_version() for the versions the current session is actually reading.

Examples

irw_version()                # the newest IRW version, and its eleven pins
irw_version(version = 332)   # exactly what v332 held
irw_version("2026-08-01")    # what was live on 1 August 2026

Description

A sample dataset sourced from the IRW database to illustrate package functionality and enable users to experiment without needing authentication.

Usage

swmd_mokken

Format

A data frame with 4557 rows and 4 variables:

id

Unique identifier for each subject.

item

Item identifier, specifying the question or measurement.

resp

Response variable, indicating the response provided.

rater

Rater identifier, specifying the rater who provided the response.

Details

This dataset is part of the IRW database and serves as an example for using package functions. It includes basic variables typically present in psychometric datasets such as subject identifiers, item identifiers, and response variables.

Source

IRW Data Warehouse