sigurdjanson / u.t.-lity Goto Github PK

Some of the ... arguments are hard-coded. If a user would try to override, that would fail (which it should not).

See also "DiscountingDiscountUsabilityEvaluation"

• Simulate one or many sample
• Provide a "test sketching" function (similar to a Kurvendiskussion) that binds everything together and brings the complete results to you.
• [ ]

The function has dead code at the end. Remove that code and the the plot argument (because this seems to be the only ci-function with a plot-argument).

See Sauro & Lewis, p 150
See also http://rob.schapire.net/papers/good-turing.pdf

• Estimate defects found given "sample size" and "visibility"
• Estimate sample size given "found defects" and "visibility"
• Estimate visibility given "sample size" and "found defects"

Including the methods to ...

• get average p.occ
• get vector with the number of defects for each participant
  • same as proportion
• get vector with the number of participants showing a defect
  • same as proportion
• total count

Go through all plots and check the strings

It is not clear, how it works. There is also no reference to the t.test underlying the confidence intervals.

It does not show how many problems you find. It describes the probability to find each defect at least once.

Function is a stub

• gg computes percentages per cell; gr does it per column.
• gg has no percentages per col at the bottom printed out (see #14).
• gr grid does not have quadratic tiles.
• gr does not have a legend for tile colours.

• nDefectsPlot() for binomial statistics using the defect discovery rate.
• nDefectsPlot() for binomial statistics using the defects found.
• Estimate visibility given "sample size" and "found defects"
• Estimate Defect Discovery Rate found given "sample size" and "visibility"
• Estimate sample size given the "Defect Discovery Rate" and "visibility" (p.occ).
• Estimate sample size given "found defects" (given a specified confidence) and "visibility"

There are inconsistencies. FInd them and fix them.

See test-binomci.R

# This test could be improved - it is not very precise
test_that("one-sided intervals are correct", { ...

Problem: MKinfer seems to give slightly different results than binom::binom.confint. But the difference are too large for a good test.