jasp-stats / jaspttests Goto Github PK

As a result of the discussion on slack, we would like to change back to reporting mean differences and the resulting effect sizes. The PR we want to revert is (change from mean differences to location estimates) is

7f6cc1c

Parent is

2119573

When reverting back to mean differences it's good to consider

jasp-stats/jasp-issues#1036

The idea is that the t-test should collapse to the z-test if the mu is exactly mean(x) and sigma is exactly sd(x). Code suggested by Johnny (requires the sd(x) in the code below to be changed by the postulated sigma from options).

So it seems that the BSDA package does not account for the test value in the CI for the z-test, which is why the CI in the bug report does not include the effect size. We can manually adjust this:
dat <- read.csv("~/Downloads/survey5ED.csv")
x <- na.omit(dat$tslfest)
testVal <- 30
t.test(x, mu = testVal)
wilcox.test(x, mu = testVal)
a <- BSDA::z.test(x, mu = testVal,
sigma.x = sd(x),
conf.level = 0.95)
a$statistic / sqrt(length(x)) # Effect Size
(a$conf.int - testVal) / sd(x) # CI that captures ES

It's worthwhile to derive the sampling distribution in terms of the parameter delta=(mu - mu0)/sigma to confirm that the proposed confidence interval is correct.

.BANOVAdependentName and .BANOVAsubjectName are not in jaspTTests, but they are in jaspAnova.
I originally moved .summarySE() and summarySEwithin() to this module to avoid circular dependencies between jaspAnova and jaspTTests.
Should a copy of these functions exist in both or do we reckon they're important enough to give a few hacky arguments?
@vandenman

"Warning in install.packages :
package ‘jaspTTests’ is not available for this version of R"

What version of R was this library developed under?

Hi Johnny,

Can you also do the following PR into this module?

https://github.com/jasp-stats/jasp-desktop/pull/4228

aka this repository

jasp-stats/jaspTTests 

@TimKDJ that's correct isn't it?

Cheers,
Alexander