NEWS

---

ratesci 1.0.0.9000

New features

• New functions rdci(), rrci() and orci() for giving a selection of confidence intervals for each contrast for 2 independent binomial or Poisson rates.

In rateci():

• New default point estimate is the "common sense" crude estimate x / n, instead of an estimate that is consistent with a 0% CI. std_est allows the user to choose. (#32, thanks to Chelsea Dickens for raising the issue.)
• New output object containing several confidence interval methods in an array format, including the inferior Wald, Wilson and Agresti-Coull methods (for reference, not recommended), plus an alternative formulation of the Clopper-Pearson or mid-p interval based on Beta or Gamma distributions (for comparison with the version based on binomial/Poisson tail probabilities). These match Clopper-Pearson and Garwood precisely when cc is TRUE, and approximately match the corresponding mid-p intervals when cc is FALSE.
• New output object for the Blaker exact method (for both binomial and Poisson rates)
• precis for setting the precision of the exact and mid-p method.

In pairbinci():

• New method = "Wald" option to give approximate normal method - not recommended but provided for reference and evaluation.

Bug fixes

In rateci() and moverci():

• Permit n=0 to produce output interval as [0, 1] for binomial, or [0, Inf] for Poisson with point estimate displayed as NaN (prevents error in pairbinci() with method = "BP").

In moverci():

• Present NaN MOVER-R Wilson upper limits for RR and OR as Inf.

ratesci 1.0.0 (2025-06-20)

New features

• New function clusterpci() for CI and test for a single binomial proportion from clustered data.
• Example datasets are now included.
• Improved documentation with pkgdown website & vignettes.

In pairbinci():

• skew for skewness correction.
• bcf for variance bias correction.
• Default paired RD and RR method changed to new SCAS method (i.e. including skewness correction - manuscript under review).
• method_RD, method_RR and method_OR are replaced with method.
• Bonett-Price methods for RD and RR (including proposed Jeffreys variant option for RR).
• TDAS method is deprecated.
• Default for MOVER method changed to Jeffreys.
• MOVER calculations now use x/N as point estimate instead of median from the Beta distribution.
• cc uses a new form of correction for RR giving equivariant intervals. Also allows consistency with the continuity-corrected McNemar test (or an intermediate correction of the user's choosing). cctype is deprecated.
• Default conditional odds ratio method changed to SCASp (with closed-form calculation).
• Output object now includes estimates of p1, p2, phi (correlation) and psi (odds ratio used by Fagerland et al).
• Output object now includes function call.

In scaspci():

• bcf option now implemented for contrast = "p" (default = FALSE).
• bign allows a different sample size to be used in the bias correction (used within transformed SCASp method for paired OR in pairbinci, for consistency with 'N-1' test).

In scoreci():

• bcf option now implemented for contrast = "p" (default = FALSE).
• (Note adjusted sample size for bias correction can be achieved by including a non-zero value for n2.)
• ORbias, RRtang, and MNtol arguments renamed as or_bias, rr_tang and mn_tol.
• Implementation of precis argument is improved for RR and OR contrasts.
• For contrast = "RD", weighting = "Tang" provides optimal test if RR is constant across strata.

Bug fixes

In exactci():

• Corrected duplicate point estimate reported for vector inputs.
• Derive point estimates to match LCL and UCL with level = 0.
• Corrected LCL for Poisson mid-p method.

Other

• Output object column names are updated (lower, est, upper) for consistency and style conformity.
• Tests added to confirm consistency of score methods vs McNemar test.
• Dependence on polynom package removed.
• Edition 3 of testthat implemented.

ratesci 0.5.0 (2025-01-10)

New features

In pairbinci():

• cc continuity correction is now available for all methods for all contrasts.
• cctype controls the type of correction to apply for contrast = "RR".
• New default method_RD = "Score_closed" for non-iterative calculation of the Tango score interval for contrast = "RD". Thanks to Tony Yang for permission to use the code in his 2013 paper.
• New default method_RR = "Score_closed" for non-iterative calculation of the Tang score interval for contrast = "RR". Thanks to Guogen Shan for contributing code via email.
• Added paired MOVER methods with method_RD = "MOVER" and method_RR = "MOVER". Also "MOVER_newc" incorporates Newcombe's correlation correction.
• Added moverbase, for specifying different versions of the MOVER methods (Wilson, Jeffreys, midp or SCAS).
• Added "jeff" and "wilson" method_OR options for transformed binomial methods for OR.
• Confirmed and documented that the 2-sided significance test is equivalent to the McNemar test (with or without continuity correction).

In scoreci():

• Confirmed that continuity corrections for all stratified (fixed-effects) binomial contrasts are consistent with the Mantel-Haenszel correction.
• Updated heterogeneity test to consistently omit non-informative (but non-empty) strata, and output the degrees of freedom.

In moverci():

• Added continuity correction for type = "wilson".
• Added options for type = "SCAS" and "midp" intervals.
• Standardised output to include lower CL, midpoint, upper CL, in that order.

Bug fixes

In scoreci():

• Improved handling of special cases for MN weighting (#25, thanks to Vincent Jaquet for reporting the issue and proposed solution. Also #27 for RR, thanks to @lovestat.) As a result, double-zero strata need not be excluded when weighting = "MN".

In moverci():

• Corrected calculation of score intervals for single Poisson rate, using Rao score interval.
• Same correction affects MOVER method for comparison of Poisson rates [i.e. moverci() with distrib = "poi" and type = "wilson"]

Other

• Improved documentation of hypothesis tests and continuity corrections, clarifying links to Chi-squared tests and CMH test with selected weights.
• Correction to documentation of default weights for OR.
• Added tests confirming equivalence of iterative and closed-form methods in pairbinci.

ratesci 0.4-0 (2021-12-05)

New features

In scoreci():

• MN weighting now iterates to convergence (@jonjvallejo, #20).
• Added optional prediction interval for random effects method (also in tdasci()).
• Added xlim and ylim arguments to control plot output.
• Added sda & fda arguments for optional sparse/full data adjustment when x1 + x2 = 0 or x1 + x2 = n1 + n2 in a stratum.
• Added INV option for weights that omit the variance bias correction.
• Added RRtang argument to apply Tang's alternative score for RR (recommended for stratified analysis with INV/IVS weights. Experimental for Poisson RR). Stheta = (p1hat - p2hat * theta) / p2d (see Tang 2020)
• Added simplified skewness correction option (causes p-values to be omitted, see Tang 2021 & Laud 2021).
• Introduced warning and plot features for very rare occasions when quadratic skewness correction cannot be calculated due to a negative discriminant.
• p-value suppressed where affected by negative discriminants.
• Changed ORbias default to TRUE (see Laud 2018).
• Changed weighting default to MH for RD & RR, INV for OR (for consistency with CMH test).
• Added hetplot argument to separate heterogeneity plots from score function plot.
• Uninformative strata are now retained in the analysis except if:
  • contrast = OR with MH weighting;
  • contrast = RR with IVS/INV weighting if RRtang = FALSE;
  • random = TRUE (needs further evaluation);
  • excluded using new option dropzeros = TRUE.

In tdasci():

• Default uses skew = TRUE for stratum CIs.

Bug fixes

• MN weighting in scoreci() corrected for distrib="poi".
• Fixed bug in scoreci() for calculation of stratum CIs with random=TRUE.
• Fixed bug in scoreci() for distrib = "poi" and contrast = "p" (#7).
• Fixed finite precision bug in scaspci().
• Fixed bug in rateci() for closed-form calculation of continuity-corrected SCAS.
• Fixed bug in scoreci() for stratified zero scores calculated as NA, resulting in UL = 0. (Thanks to Lidia Mukina for reporting the bug.)
• Fixed variable plot ranges for vectorised inputs.

Other

• Renamed tdas argument to 'random'.
• Removed redundant t2 variable.

ratesci 0.3-0 (2018-02-15)

New features

• Added bias correction in scoreci() for OR SCAS method (derived from Gart 1985).
• Added score methods (Tango & Tang) as default for paired binomial RD and RR in pairbinci().
• Added transformed mid-p method for paired OR for comparison with transformed SCAS.
• Added scaspci() for non-iterative SCAS methods for single binomial or Poisson rate.
• Added rateci() for selected methods for single binomial or Poisson rate.

Bug fixes

• Fixed bug in pairbinci() for contrast="OR".
• Fixed bug in moverci() for contrast="p" and type="wilson".
• Corrected error in cc for stratified SCAS method for OR.
• Clarified documentation regarding continuity corrections.
• Set Stheta to 0 if |Stheta|<cc in scoreci()
• Fixed stratified calulations for contrast = "p" in scoreci().

ratesci 0.2-0 (2017-04-21)

New features

• Added pairbinci() for all comparisons of paired binomial rates.
• Added option to suppress warnings in scoreci.
• Added Galbraith plot (for assessing stratum heterogeneity) to scoreci().
• Added qualitative interaction test to scoreci().
• Added stratum estimates & CIs to scoreci() output when stratified = TRUE.

Bug fixes

• Fixed bug for contrast = "p" in moverci().
• Fixed bug in tdasci() wrapper function.
• Fixed bug for stratified OR.
• Altered adjustment options for boundary cases in moverci().
• Changed point estimate used in moverci() to posterior median for type = "jeff", to ensure consistent calculations with informative priors.