search

crtahlin / medplot

Functions for drawing graphs in R visualizing medical information.
4 stars 2 forks source link

Summary : time #109

Closed crtahlin closed 10 years ago

crtahlin commented 10 years ago

Moving form #103 Crt: Regarding the sentence: "The descriptive tables report also the 95\% confidence intervals for medians (based on percentile bootstrap with 2000 iterations) or proportions (based on binomial distribution), which are also graphically displayed." At the moment, the confidence intervals for proportions throughout the app are calculated via prop.test() for which the documentation states: The confidence interval is computed by inverting the score test. As I understand the text in the draft, an exact binomial test is described there, e.g. like from binom.test()? What should we do: a) nothing, as I misintepreted the draft b) use binom.test() (where, only on this tab?) c) change text in draft

Lara: I will think about it - might change it in the paper describing the use of prop.test. (in principle I would prefer binom.test for this tab since it would always be ok also for extremely small or large proportions, this is the only tab where it would be used, but the priority for this change is super-low)

llaarraa commented 10 years ago

make consistent outputs for tables and figures - same order of the symptoms (add 95% to the title of the figure, for the CI)

llaarraa commented 10 years ago

I would reorder the multipicity adjustments so that the order is: Holm-Bonferroni, permutations, Q values (BH), q - values (BY)

llaarraa commented 10 years ago

What should we do: a) nothing, as I misintepreted the draft b) use binom.test() (where, only on this tab?) c) change text in draft

Lara: if it is not too much work to adapt the script, can you change to binom.test()?

crtahlin commented 10 years ago

It should be minimal work to change tto binom.test() (well, hopefully:), so I will do it once I get to changing that part.

crtahlin commented 10 years ago

Did everything but the binom.test(). I haven't noticed before, that although the function description is almost the same same as for prop.test(), there is a slight difference - it does not take a matrix as input and it does not compare the difference of proportions for two samples (groups), but only compares the sample to a hypothesis. @llaarraa : Is what we are looking for perhaps the exact Fisher test (fisher.test())?

llaarraa commented 10 years ago

binom.test() is fine for one proportion. fisher.test() would be used instead of chisq.test() or prop.test() for two proportions. the part for two proportions is fine, should not be changed. just what appears in the summary (not in summary by groping variable) should be calculated with binom.test instead of prop.test()

crtahlin commented 10 years ago

My mistake, sorry. Yes, a trivial "upgrade". Done, closing.

llaarraa commented 10 years ago

updated in the manuscript