Changes in version 0.3.0 (2026-11-13)                  

After upgrading, run update_cvpa_data() to refresh your local cache: the
cached question catalog from earlier versions is out of date.

Breaking changes

  - update_cvpa_data(questions = "all") has been removed. Name the
    question codes you need (see question_catalog()); it now errors with
    a pointer to the full dataset.parquet in the cvpa_dataset repository
    for anyone who wants everything.

  - wtd_vote() and gap_analysis() now default to years = NULL, meaning
    every year in data, instead of 1945:2023 and 1945:2022. The old
    defaults silently dropped any year after 2023 (and, for
    gap_analysis(), 2023 itself). Results on the bundled vote_data are
    unchanged for wtd_vote(); gap_analysis() now includes 2023. Pass
    years explicitly to restore the old behaviour.

Data

  - The default cvpa.data_url is now
    https://policyattitudes.ca/cp3_parquet/, hosted with the project
    site, instead of the copy on quantoid.ca, which was out of date and
    is no longer read by the package. The new data includes the merge of
    ces_soim_009 into bri_soim_014 (same-sex marriage; bri_soim_014 now
    covers the CES 1993, 1997, 2000, 2004, 2006 and 2008 surveys and has
    the wording "Support or oppose allowing same-sex couples to marry"),
    and respondent ages recovered for five Environics surveys.

  - question_catalog() now lists only questions and surveys that have
    respondent-level data, so n_waves counts only surveys that actually
    have data, and questions with no data (such as env_intd_045) are no
    longer listed. Surveys with no region variable cannot be weighted
    and are not included.

Tests

  - cvpa now has a testthat suite (run offline against the bundled
    example data) covering the data download and caching, the question
    catalog, decoding, the weighted-share and gap-analysis functions,
    the unofficial-data warnings and the legacy vote functions.

Examples and the example data

  - Every help page now has examples, and all of them run (none are
    wrapped in \dontrun{} any more). The ones that need data use a small
    bundled example dataset -- four questions, with a random sample of
    respondents from each survey year -- through the new
    cvpa_example_url(), in a temporary cache, so they run offline and
    never touch your own cache. The example data is for illustration
    only.

  - update_cvpa_data() now warns whenever options(cvpa.data_url = ...)
    is anything other than the official live data
    (https://policyattitudes.ca/cp3_parquet/), and says how to switch
    back. The functions that read the cache (question_catalog(),
    question_data(), question_response_labels()) also warn, once per
    session, if the cache was downloaded from an unofficial source;
    update_cvpa_data() records the source in a small cvpa_source.txt
    file in the cache directory.

Bug fixes

  - gap_analysis() and response_gap_analysis() reported conf.low and
    conf.high the wrong way round: the confidence interval for diff
    (group 2 minus group 1) was negated but its limits were not swapped,
    so conf.low was always larger than conf.high. The interval itself
    was right and plots that draw between the two values looked fine,
    but anything that treated conf.low as the lower bound got the wrong
    number. conf.low is now the lower limit and diff always lies between
    the two. If you saved or published these two columns from an earlier
    version, they are simply swapped: the old conf.high is the correct
    lower limit and the old conf.low the correct upper limit.

  - update_cvpa_data() no longer leaves a corrupt file in the cache when
    a download fails. Downloads are written to a temporary file and
    checked to be valid Parquet before they replace anything, so a
    failed, truncated, or error-page download can't overwrite a good
    cached copy. A failure to download the catalog or code labels stops
    with an error and leaves the cache untouched.

  - update_cvpa_data() skips a question whose file can't be downloaded,
    with a warning naming it, instead of stopping at the first missing
    file. The return value gains a "skipped" attribute listing those
    questions.

  - A local file:// mirror set through options(cvpa.data_url = ...) now
    works; a cache-busting query string was being appended to every URL.

  - Two question wordings (ces_dpin_016, env_intd_012) no longer show
    their quotation marks repeated several times over.