3 categorical data significant analysis

#1
Dear all, how to conduct significant test for 3 group categorical data and ordinal varibales? i have student grade 1, 2 and 3, label as 1-2-3, and asking theri perception on study with likertscale 0-4 (completely disagree to completely agree). what way is best to conduct significant test compare the 3 groups of student? Chisquare? Thansk very much in advance!
 

Karabiner

TS Contributor
#2
This is not a Likert scale, but just a single Likert-item?
If you consider "grade" as categorical, then you could analyse the effect of the variable "grade" on "perception" using a Kruskal-Wallis H test.
If you consider "grade" as ordinal, then you could use a Spearman rank correlation to analyse its association with "perception".

With kind regards

Karabiner
 
#3
Dear Karabiner,
Thansk very for your help, BUT I am still not clear here! I survey student grade 1-3, and their perception about study. eg. how do you confident in reading? 0= no confidence, 1=slight confidence, 2=Moderate, and 3=High confidence, My question is there any differnt btween on the confident bewteen student grade 1, grade 2 and grade 3? and want to test the level of significant. so what do you suggest the best way to analyst this? again thansk very much! I am look forward to hear from you.
 

Karabiner

TS Contributor
#4
You can use Kruskal-Wallis H test for testing whether the 3 groups differ with regard to confidence. If the H-test is statistically significant, then in the next step you can use Mann-Whitney U-tests for pairwise comparisons (grade1 versus grade2; grade1 versus grade3; grade2 versus grade3).

If, alternatively, you just want to know whether confidence monotonically increases (or monotonically decreases) with grade , then you could do a Spearman-correlation.

With kind regards

Karabiner
 
#5
You can use Kruskal-Wallis H test for testing whether the 3 groups differ with regard to confidence. If the H-test is statistically significant, then in the next step you can use Mann-Whitney U-tests for pairwise comparisons (grade1 versus grade2; grade1 versus grade3; grade2 versus grade3).

If, alternatively, you just want to know whether confidence monotonically increases (or monotonically decreases) with grade , then you could do a Spearman-correlation.

With kind regards

Karabiner
Thansk very MUCH that help alot Sir
 
#6
Thansk very MUCH that help alot Sir
Dear Karabiner, two more help please:
First: is that possible there is no significant with H-test, but when i run Mann-W-U test compare 1 & 2, 1 &3 and 2&3 itis signifiant!

Second: i been asked a lot of quesiton to student about their motivation in study (many factor that might influence their motivation) include, School [, enviroment, study materials, teachers, ..etc], at their home.....community...The quesiton is I want to construct the model to say which student are motivate and which are not, and those motivated which aspects or factor that strongly influen (mostly the quesiton are binery, categorical, ordinal ..so which method that I can construct or test this? regression? ...please help me out. Again thank very much! !
 

Karabiner

TS Contributor
#7
Dear Karabiner, two more help please:is that possible there is no significant with H-test, but when i run Mann-W-U test compare 1 & 2, 1 &3 and 2&3 itis signifiant!
If you perform 3 separate pairwise comparisons, while ignoring the non-signficant result of the global test, then you should contemplate whether to adjust your significance level, to avoid false-positive results due to multiple testing. E.g. you might want to use Bonferroni-correction - if your singificance level is 5%, then the adjusted level would be 5%/3 = 1.667.

..so which method that I can construct or test this? regression?
If you want to analyse whether multiple variables jointly predict "motivation", then some multiple regression approach would come to mind, yes.

With kind regards

Karabiner
 
Last edited:
#8
If you perform 3 separate pairwise comparisons, while ignoring the nion-signficant result of the global test, then you should contemplate whether to adjust your significance level, to avoid false-positive results due to multiple testing. E.g. you might want to use Bonferroni-correction - if your singificance level is 5%, then the adjusted level would be 5%/3 = 1.667.


If you want to analyse whether multiple variables jointly predict "motivation", then some multiple regression approach would come to mind, yes.

With kind regards

Karabiner
Thansk very much Karabiner, you are the best!