Same datas, two ShapiroWilk results

[loic.bahu.etu]

Hello,

notice i am very poor in using statistics (maybe worth than in english talking).
I have to make a few stats for a university work that is not about stats, but it takes me a lot of time.

I read a lot of things this 2 days, even a consequent part of the french documentation on jamovi (https://hal.archives-ouvertes.fr/hal-02 ... 2/document)

And now i see spomething i don't understand.

I have understand that i have to check the normality of my "scale data" and that Shapiro-Wilk's "p" value seems to be a good idea for that (please, dont try to make me a detailled course about stats, i dont need to know HOW IT WORK i just need to know what value is good or bad for cotinue working).

So i have first checked my scales Data with the Exploration Tool,
And i say :
- those who S-W "p" value is < .05 are NOT normal
- those who S-W "p" value is > .05 are normal

And with that i can say
- Not normal will go to Spearman or U Mann Whitney
- Normal will go to Pearson or Student/Welch

It seems simple, like that (well, hem, no, not so simple, 2 days of work to have this conclusion, i will die soooooon).

But something went wrong.

i was trying to make some samples T-Test, then i see a "normality test" square button.
Ohhh !
I tried it. It shows ShapiroWilk Values.
But the values are not the same !

Let have a look to this.

The 3 datas (SUS, Age and FlowD1) are the same at every time.
But T-Tests > Independant> assumption shows me some values
and Exploration > Desfritives show me others

See this screen capture :

ShapiroWilkX2.png (16.68 KiB) Viewed 3594 times

Age's SW's p value can ben 0.018 or 0.001
What the **** is this ?

what value of ShapiroWilk can i believe !?

[MAgojam]

Hey Loïc Bahu,
I'll be short and concise, because I don't want to increase your suffering ...

If you go back to the text you quoted on p. 350 find this fragment:

Normalité. On suppose que les résidus sont normalement répartis. Comme nous
l’avons vu à la section 11.8, nous pouvons l’évaluer en examinant les graphiques QQ
(ou en effectuant un test Shapiro-Wilk). J’en parlerai davantage dans le contexte de
l’analyse de variance à la section 13.6.4.

Thus, for an Independent Samples T-Test, the assumption of normality will be verified with a Shapiro-Wilk test on the residuals and not on the N values of the variables, as is the case in Exploration> Descriptives.

Good life.
Maurizio

[loic.bahu.etu]

Hi Maurizio.

First, i have to say how i am amazed than you can find the solution in french, in that so long document, as it is not your own language.

And then, i thank you a lot to not increasing my suffering, you are nice ! Grazie mille !

So,if i understand what you say.
The Shapiro-Wilk test that is done inside the T-Test menu, is a specific Shapiro-Wilk to test normality on something important to make the T-test (the famous "residual") ?
if i say this on another way : if i have to make a T test, it will probably be better to use the "inside T test menu" SW test than to use the "generic in exploration-descriptive menu" SW test ?
Is this seems a reasonnable summary ?

And one last thing.
When i write
- when S-W "p" value is < .05 the scale data is NOT normal
- when S-W "p" value is > .05 the scale data is normal
is this something that is reasonnably correct (in a approximatively general way, i mean)

[MAgojam]

...
Is this seems a reasonnable summary ?
....
is this something that is reasonnably correct (in a approximatively general way, i mean)

Bien sûr, cher ami, vos résumés sont "raisonnablement" corrects.
Maurizio

[loic.bahu.etu]

Perfect. I feel reassured.
Have a nice day.