getFinalConfidenceInterval {rpact}R Documentation

Get Final Confidence Interval

Description

Returns the final confidence interval for the parameter of interest. It is based on the prototype case, i.e., the test for testing a mean for normally distributed variables.

Usage

getFinalConfidenceInterval(
  design,
  dataInput,
  ...,
  directionUpper = TRUE,
  thetaH0 = NA_real_,
  tolerance = 1e-06,
  stage = NA_integer_
)

Arguments

design

The trial design.

dataInput

The summary data used for calculating the test results. This is either an element of DatasetMeans, of DatasetRates, or of DatasetSurvival and should be created with the function getDataset. For more information see getDataset.

...

Further (optional) arguments to be passed:

normalApproximation

The type of computation of the p-values. Default is FALSE for testing means (i.e., the t test is used) and TRUE for testing rates and the hazard ratio. For testing rates, if normalApproximation = FALSE is specified, the binomial test (one sample) or the exact test of Fisher (two samples) is used for calculating the p-values. In the survival setting, normalApproximation = FALSE has no effect.

equalVariances

The type of t test. For testing means in two treatment groups, either the t test assuming that the variances are equal or the t test without assuming this, i.e., the test of Welch-Satterthwaite is calculated, default is TRUE.

directionUpper

Specifies the direction of the alternative, only applicable for one-sided testing; default is TRUE which means that larger values of the test statistics yield smaller p-values.

thetaH0

The null hypothesis value, default is 0 for the normal and the binary case (testing means and rates, respectively), it is 1 for the survival case (testing the hazard ratio).

For non-inferiority designs, thetaH0 is the non-inferiority bound. That is, in case of (one-sided) testing of

  • means: a value != 0 (or a value != 1 for testing the mean ratio) can be specified.

  • rates: a value != 0 (or a value != 1 for testing the risk ratio pi1 / pi2) can be specified.

  • survival data: a bound for testing H0: hazard ratio = thetaH0 != 1 can be specified.

For testing a rate in one sample, a value thetaH0 in (0, 1) has to be specified for defining the null hypothesis H0: pi = thetaH0.

tolerance

The numerical tolerance, default is 1e-06.

stage

The stage number (optional). Default: total number of existing stages in the data input.

Details

Depending on design and dataInput the final confidence interval and median unbiased estimate that is based on the stage-wise ordering of the sample space will be calculated and returned. Additionally, a non-standardized ("general") version is provided, the estimated standard deviation must be used to obtain the confidence interval for the parameter of interest.

For the inverse normal combination test design with more than two stages, a warning informs that the validity of the confidence interval is theoretically shown only if no sample size change was performed.

Value

Returns a list containing

See Also

Other analysis functions: getAnalysisResults(), getClosedCombinationTestResults(), getClosedConditionalDunnettTestResults(), getConditionalPower(), getConditionalRejectionProbabilities(), getFinalPValue(), getRepeatedConfidenceIntervals(), getRepeatedPValues(), getStageResults(), getTestActions()

Examples


design <- getDesignInverseNormal(kMax = 2)
data <- getDataset(
    n = c(20, 30),
    means = c(50, 51),
    stDevs = c(130, 140)
)
getFinalConfidenceInterval(design, dataInput = data)



[Package rpact version 3.3.0 Index]