Which statistical test do I use?

obh

Active Member
#21
Hi Ani

oh I see. So does that mean that ANOVA tests are one-tailed?
There are two types of one-tailed: right-tailed and left-tailed. ANOVA test is right-tailed.

PS maybe try to look for a used laptop, some people want a new one and don't know what to do with the old one.
Cheers
 
#24
Thank you.

Now with Tukey results I again care about the P-value right?
From what I've researched it appears that it just compares the groups to each other individually so again I only care about the significant results or the results that are P= <0.05.
 

obh

Active Member
#25
Correct, it won't be correct to try it if the ANOVA test is insignificance. (PS you chose to use 0.05, but it may be other value like 0.01, depending on the risk of rejecting a correct null assumption)
 
#26
So with the Tukey. Since I care about significant results the 1st set of data showed that 1 vs 3 (p=0.00<0.05) and 1 vs 4 (p=0.0278<0.05) those are the results I report right? And the probability of there not being a disparity between 1 and 3 is 0.00% and 2.78% between 1 and 4, correct?
 

obh

Active Member
#27
Hi Ani,

After reporting the ANOVA results you should report the Tukey's HSD results.
I assume that you should report first the significant pairs with the p-values and then the insignificant pairs with the p-values
 
#28
Right. Okay so for my results if for the first picture I got significant results back for North America vs Australia, North America vs Asia and North America vs Europe that would mean that those three results are what are making the ANOVA significant?

Is a Tukey possible when the data sets are independent?
 
#29
You should test the assumptions of ANOVA first. If the distributions are roughly normal and the variances aren’t too heterogeneous then ANOVA should be fine. If the distributions are extremely skewed and/or heterogeneous you may opt for the non parametric option or transform the data. See here for more information on transforming the data for ANOVA or other parametric tests. https://www.anatomisebiostats.com/biostatistics-blog/transforming-skewed-data
...You may also want to think about adjusting the p values for multiple comparisons.
 
#30
Sarah Baker-
I'm a little confused by your response. I think it is a bit complicated for what I'm doing.
I already did all the ANOVA stuff.
I have responses to a survey from 6 continents. My number of subjects for each continent ranges from 5-290.
They responded to 8 pictures using a scale of 1-10.
I needed to compare the means from each continent for each picture. I already ran the ANOVA for each picture and go significant results for 6 of the 8 pictures.
I then did a Tukey's HSD and got significant results again for a few of the pairs.
None of the significant pairs included the continent with the lowest number responders (5). Is that because there weren't enough responders?
The fact that there is significant results does that mean that these are the continents that differ the most in opinion?
 

obh

Active Member
#31
Hi Ani,

As I wrote, to assume normality sample size of each continent should be more than 30. (say 5 is not so good ...)
you need to assume normality for each continent.
I also suggested you run the Brown–Forsythe test for equality of variances.

None of the significant pairs included the continent with the lowest number responders (5). Is that because there weren't enough responders?
correct. Standard deviation (average)= S/Sqrt(n)

"...You may also want to think about adjusting the p values for multiple comparisons."
This says the "actual significance level" is bigger than the "significance level" you used for each question. you can understand if you calculate the following simple question:
The probability of a mistake in one question is p=0.05. what is the probability for mistake asking 8 questions?

PS test for equality of variances. : http://www.statskingdom.com/230var_levenes.html

Cheers
 
Last edited: