Need help please, should I use ANOVA or chi square?

#1
Hi there! I am trying to work on a university paper and I am having a hard time with the stats part.


I have asked some participants (29) to complete several tasks, and I have the results for the people that successfully completed it, and those who didn't.
I have a breakdown by age (4 groups) and I am trying to prove (Ho) ***that age is independent for the successful completion of the task.***


I am trying to understand whether I have to use chi square or ANOVA. I used anova and I was able to reject the Ho but if I use Chi Square, I fail to reject.


Can someone guide me on this, please?


Thanks in advance, I appreciate it
Pablo
 

trinker

ggplot2orBust
#2
Lets back up a second...can you name your predictor variable(s) and your outcome variable? What type of data are they? Then determine the type of data each variable is; is it numeric/continuous or categorical?
 
#4
Thanks a lot for both your replies. I appreciate it.
Please see below. It was a task each participant had to complete, and I am measuring whether they were able to do it or not.

Success Unsuccess
20/29 52 32
30/39 28 29
40/39 2 4
49 + 32 32
 

Karabiner

TS Contributor
#6
So you have n=29 participants, each with 2 measurements:
a) sucess/no success (categorical variable), and b) age (ordinal
scaled variable). You can compare age between the "success group"
versus the "no success group" using Mann-Whitney U-test.

With kind regards

Karabiner
 
#7
Thanks a lot, Karabiner. I appreciate your help.
In this case, ANOVA or chi cannot apply?
I am trying to prove that age is independent of performance
Thanks
 

Karabiner

TS Contributor
#8
No. They do not apply.

First of all, ANOVA does not apply at all. Because ANOVA is for interval scaled dependent variables. There is no interval scaled variable anywhere in your study description.

Chi square could be used possibly, but Chi square is useful most if both variables are categorical. Here, one of the variables is categorical, but the other variable is ordinal! It is better to treat an ordinal variable as ordinal, and not to treat an ordinal variable as categorical. Because, treating an ordinal variable as categorical does waste information.

Therefore, a test should be used which considers that one variable is categorical (binary, which means: with 2 groups), and the other variable is ordinal. Such a test which for the relationship between an ordinal variable (such as age group) and a binary variable (success yes/no) is the Mann-Whitney U-test.

By the way, you will absolutely not be able to "prove" that age is independent of performance. Maybe your test will fail to show a statistically significant relationship between age and performance. But with such a small sample size (n=29), your study is not sensitive enough to reliably rule out any relationship between age and performance.

With kind regards

K.
 
#9
Karabiner, thanks a lot for your help. I really appreciate it.
Understood, I will be using the test you mentioned.

I have also done two more groups but they were 2 X 2 tables; male/female, success-unsuccess, and if the person had or no previous experience on a similar task and if success or unsuccess.

I have used chi square test for those, do you recommend using a different kind of test or in this case per the type of data and group numbers, it makes sense to use chi?

Thank you very much for your guidance
Regards
Pablo
 

Karabiner

TS Contributor
#10
If you have 2 categorical variables (male/female and successful/unsuccessful, or, respectively, experience/no experience and successful/unsuccessful), then 2x2 tables with Chi square tests look absolutely reasonable.

With kind regards

K.
 
#11
Excellent, thanks a lot! May I ask the last question if you don't mind?
I am trying now to compare the results I got as metrics. The 29 participants would do specific tasks on 3 different websites. I have the average time that it took to complete each task on each website and the target for each task. Can I do ANOVA to check if there is a significant difference between each target and the time it took on each website to complete the task? of there is another tool i should use?
Thanks so much for your help
Pablo
 

Karabiner

TS Contributor
#12
If all participants had to perform all tasks at all websites, then this could be analysed using repeated-measures ANOVA with 2 within-subject factors, "website" (with 3 levels), and "task type".

With kind regards

K.