Which statistical test do I use?

#1
I have a survey project and I'm not the best at stats.
I'm comparing the means of answers given based on the continent the responder is from.
So I have North America (7.20), South America (5.40), Europe (5.84), Africa (6.95), Asia (5.91) and Australia (5.90). The mean answers they gave for a question are in the brackets. I also have the total mean of all the responders as (6.20). I want to compare this data in Rstudio. Do I want to run a one-way ANOVA or Kruskal Wallis test? There are a total of 8 questions but they are all independent of each other so I'll have to do the same thing 8 times correct?
 

obh

Active Member
#2
Hi Ani,

I'm not sure what do you try to do

One way ANOVA will only tell you if based on the samples, you can reject the H0 that Mean north america = Mean South America = ...= Mean Australia.

The Kruskal Wallis is the non-parametric test of the One way ANOVA, doing the same with a different assumption.

Do all the 8 questions measure the same? or different aspects?
 
#4
Hello,
I do have the raw data.
What I want to do is compare the means of all the continents for each question they were asked in the survey. I'm just unsure of how I do this.

Thank you.
 

obh

Active Member
#5
I have a survey project and I'm not the best at stats.
There are a total of 8 questions but they are all independent of each other so I'll have to do the same thing 8 times correct?
Correct if each question describes different measurement, otherwise you may calculate the average of the questions that describe the same measurement.

So you can do the "One way ANOVA" if meet the assumptions or otherwise Kruskal Wallis.
What is your sample size?
Do you use the Likert scale, how many possible answers?
 

obh

Active Member
#7
Hi Ani,

The data doesn't distribute normally since it is a discrete range of limited range value. (can't be more than 10 or less than 0)

But since the number of option is 10
And the sample size is more than 30.
I assume it should be okay Central_limit_theorem)

You don't need to test the data for normality as the average should distribute toward the normal.
 

obh

Active Member
#9
I assume it should be okay for the normality assumption, as the sample size is very big (500).
Another Anova assumption you need to know is for equal variances between the groups. (more important if the groups' sizes aren't similar)
 
#12
Sorry for the brief absence. You have been so helpful Obh.
I decided to run the ANOVA for each individual question. I've been going over the P-values and they just don't seem right but this project is more complicated than I am used to in terms of number of questions and variables for each question.

Question 1: P=0
Question 2: P=0
Question 3: P= 0.0007
Question 4: P= 0.0823
Question 5: P=0
Question 6: P= 0.8138
Question 7: P= 0.0003
Question 8: P= 0.0029
 
#14
After reviewing the them more closely I think they do make sense.
There are no connetions between the questions.

So my null hypothesis would be that there is no statistical sig between what continent you are from and how you answer each question.
So questions 4 and 6 were the only ones with p-values >0.05 they are the only ones were I failed to reject the null hypothesis correct?
That would seem to be true as those are the only two where the means didn't vary greatly. They were are pretty well grouped around the average mean.
The other questions have large variation in the means from each country compared to the total average mean.

I wish I could show you my data but I can't seem to upload it. I hope my description helps. Again, I can't thank you enough got your incredibly valuable help!
 
#15
Hi Ani,

Even question 4 is potentially significance.
0.05 is commonly used, but actually the is no big difference between 0.05 and 0.08, it is only a rule of thumb.
It only says that if you decide the difference is significance, the possibility of error is relatively big : 0.08

But since your sample size is very big 500, and we need to have a cut of value I would say not significance. (even if significance the effect size is probably very small)

"So questions 4 and 6 were the only ones with p-values >0.05 they are the only ones were I failed to reject the null hypothesis correct?"
Yes

You can always attached a zip file