Skip to contents

Changelog

Source: NEWS.md

inferr 0.3.3

Bug Fixes

  • ifr_cochran_qtest() now maps each factor by its own declared level order, accepting factors with different labels and using explicit numeric evidence to resolve the shared coding of numeric columns.
  • ifr_cochran_qtest() now resolves the shared numeric 0/1 or 1/2 coding before incomplete rows are dropped, so a homogeneous numeric column of all 1s follows numeric evidence or consistent factor-label hints, and ambiguous mixtures error instead of being silently misread.
  • ifr_os_prop_test() now requires the null probability to be strictly inside (0, 1), because the normal approximation is undefined at prob = 0 or prob = 1.
  • ifr_runs_test() now computes its p-value from the upper normal tail, so small representable p-values are no longer returned as zero.
  • ifr_os_t_test() now detects essentially constant data relative to the mean, so a change of units no longer changes whether the t statistic can be estimated, and numerically degenerate data are rejected.
  • ifr_cochran_qtest() now keeps the declared levels of a binary factor, so a homogeneous factor column is accepted when the pairs are discordant elsewhere.
  • ifr_runs_test() and ifr_cochran_qtest() now report a clear error when the number of runs has zero variance or the data contain no within-subject discordance, instead of returning NaN.
  • ifr_mcnemar_test() now computes both chi-square p-values from the upper tail, so small representable p-values are no longer returned as zero.
  • ifr_ts_prop_group() now defines one success level before tabulating the groups, so character, logical and 0/1 outcomes work when a group contains only successes or only failures.
  • ifr_binom_calc(), ifr_binom_test(), ifr_os_prop_test(), ifr_os_t_test(), ifr_runs_test() and ifr_cochran_qtest() now reject invalid counts, probabilities, sample sizes and degenerate samples instead of returning plausible-looking values. The runs and Cochran tests remove incomplete observations with a documented, consistent policy.
  • ifr_cochran_qtest() now computes its p-value from the upper tail, so small representable p-values are no longer returned as zero.
  • ifr_ts_ind_ttest() now drops unused grouping levels after removing incomplete cases, so a factor that declares a third level does not reject data with two observed groups.
  • ifr_os_prop_test() and ifr_chisq_assoc_test() now compute p-values from unrounded z and chi-square statistics, so the reported decimals match the statistical reference.
  • ifr_os_var_test() now reports the correct confidence interval for the standard deviation.
  • ifr_ts_ind_ttest() now uses t quantiles, rather than normal quantiles, for the confidence interval of the mean difference and reports both the pooled and Satterthwaite intervals.
  • ifr_cochran_qtest() now recodes logical, 0/1, 1/2 and two-level factor columns to 0/1, and rejects other inputs.
  • ifr_oneway_anova() now drops incomplete cases and computes the test statistics from unrounded sums of squares; the printed table shows the p-value instead of 0 and keeps it within the Sig. column.
  • ifr_chisq_assoc_test() no longer returns NaN for the likelihood ratio statistic when a contingency table contains an empty cell.
  • ifr_chisq_gof_test() applies Yates correction only when correct = TRUE.
  • ifr_os_t_test() and ifr_ts_ind_ttest() now compute p-values and confidence intervals from unrounded statistics.
  • ifr_levene_test() now works with a character grouping variable and reports the correct number of observations and denominator degrees of freedom when the data contain missing values.
  • ifr_ts_var_test() now works with a character grouping variable and requires exactly two groups.
  • ifr_ts_prop_test(), ifr_ts_prop_group() and ifr_os_prop_test() now reject variables that are not binary.
  • ifr_ts_paired_ttest() now drops incomplete pairs instead of failing.
  • ifr_os_t_test(), ifr_ts_ind_ttest() and ifr_os_var_test() now drop missing values instead of failing.
  • ifr_mcnemar_test() now requires a 2 x 2 table when variables are supplied.
  • ifr_runs_test() now reports the number of cases remaining after drop = TRUE.
  • ifr_chisq_assoc_test() now returns NA for the phi coefficient when the table is not 2 x 2.
  • The two-sample proportion test output now labels the two-sided p-value as Pr(|Z| > |z|).
  • Missing suggested packages now produce an informative error message.
  • The documentation for ifr_ts_var_test() now lists the accepted alternative values, and the README example runs again.
  • ifr_binom_test() and ifr_binom_calc() now return and print the two-sided exact p-value, computed with the same rule as stats::binom.test().
  • ifr_binom_test() now returns an object of class ifr_binom_test (which also inherits from ifr_binom_calc), as documented.
  • The binomial test documentation now states that the success is the second factor level.
  • ifr_ts_var_test() now returns the documented two-sided p-value.
  • ifr_ts_prop_group() and ifr_ts_prop_calc() now compute all p-values from the unrounded z statistic.
  • ifr_ts_paired_ttest() now computes p-values and the confidence interval of the mean difference from the unrounded t statistic and standard deviation.
  • ifr_binom_test(), ifr_os_prop_test(), ifr_ts_prop_test(), ifr_ts_prop_group() and ifr_chisq_gof_test() no longer count missing categorical values as failures or scale expected counts by missing observations.
  • ifr_mcnemar_test() caps the exact p-value at 1, bounds the continuity correction at zero, and defines the no-discordant-pairs case.
  • ifr_chisq_assoc_test() and ifr_chisq_gof_test() bound the continuity correction at zero; the goodness-of-fit deviation and standardized residual use the signed bounded correction.
  • ifr_ts_paired_ttest() computes the correlation p-value from the unrounded correlation and the upper tail, so it no longer exceeds 1 for negative correlations.
  • ifr_levene_test() no longer interchanges Levene’s mean-centred statistic with Brown and Forsythe’s median-centred statistic.
  • ifr_ts_ind_ttest() reports the pooled standard deviation using n1 + n2 - 2 in the denominator.
  • ifr_oneway_anova() handles singleton groups, rejects designs without residual degrees of freedom, and uses the upper tail for the p-value so representable p-values are not returned as zero.
  • ifr_oneway_anova() and ifr_ts_ind_ttest() work when the data contain columns named x or y.
  • ifr_ts_var_test() requires exactly two columns, or exactly one outcome column with group_var, and validates the groups retained after missing values are removed.

Other changes

  • The README and pkgdown site were regenerated, so their printed results and version match the released package.
  • The binomial and runs test documentation now use the correct replacement function name, describe obs_p as observed, and describe the runs threshold correctly.
  • library(inferr) no longer contacts CRAN or offers to update packages when the package is attached.
  • The maintainer email is now aravindhebbali@rsquaredacademy.com.
  • The documented source of the exam data now points to a stable page.

inferr 0.3.2

CRAN release: 2024-11-11

This is a patch release to fix minor issues.

inferr 0.3.1

CRAN release: 2021-05-28

This is a patch release to fix error on R-devel on Debian and Fedora.

inferr 0.3.0

CRAN release: 2018-02-13

This is a minor release for bug fixes and API changes. We have completely revamped the API. All the functions now take a data.frame or tibble as the first argument followed by the variable names. The variable names need not be surrounded by single/double quotes anymore. Please view the guide for more details.

Bug Fixes

  • using if/while statement is used with a condition of length greater than 1 will result in a runtime error and not warning (#9)

  • error in binomial test (#11)

The shiny app has been updated to reflect the changes in the API.

inferr 0.2.0

CRAN release: 2017-12-11

New Features

  • shiny app for interactive analysis

inferr 0.1.1

CRAN release: 2017-05-02

Bug Fixes

  • binom_test() accepts non-binary variables (#1).
  • ind_ttest() should throw an error when the grouping variable has more than 2 levels (#2).
  • ts_prop_test() should accept only binary variables (#3).
  • var_test() should accept only binary variables for the group_var input. The number of continuous variables must also not exceed two. (#4).

inferr 0.1.0

CRAN release: 2017-02-14

  • First release