4x2 Chi-Square HELP!

#1
Hi,

I'm doing my dissertation for my Masters and I am in desperate need of help!

I need to run 4x2 chi-squares in SPSS. I understand the assumptions (all expected counts must be above 1 and there should be no more than 20% of expected counts below 5), however as I am running multiple tests I am at risk of type 1 error.

1. Do I need to apply the Monte Carlo method? And if so, what for?

2. My supervisor mentioned the Bonferroni test - how do I apply this to chi-squares and when would it be necessary?

3. If a significant result is found in the 4x2 chi-square, what follow-up tests do I then need to run? Do I then just carry out chi-squares for each of the groups to identify where the significant relationship is?

4. Any other suggestions of what may need to be done?

I would realllllllyyyyy appreciate any help you can give me. I've tried the internet and multiple stats text books, but I can only find 2x2 chi-square tests in detail!! It is due next week (and it is my birthday! :shakehead) - it is not that I have left it til the last minute. I had just ran loadssssssss of 2x2 chi-squares and after reading my first draft my supervisor has said it would look a lot less complicated by running fewer 4x2's.

Anyone who responds, thank you so much, you're a lifesaver!!!!
 

gianmarco

TS Contributor
#3
Hi,
Just an handful of hints.
Take into account the table of standardized residuals to identify what cell has an obs value significantly different from the exp one unger the null-hypothesis.

As for ch-sq assumptions, I have read that it is quite robust in relation to the presence of few small values. In any instance, permutation ch-sq test can be used when assumptions are not met.

I hope this can help, and I look forward for further hints by other Forum's member

Best wishes
Gm
 

gianmarco

TS Contributor
#4
By the way, re-reading your post, I realized that you have not provided details about your research, your data, your goals, etc

May be that other members can give you more appropriate suggestions once they get a clearer picture of your data and goals.


Regards,
Gm
 

trinker

ggplot2orBust
#5
becky52 said:
If a significant result is found in the 4x2 chi-square, what follow-up tests do I then need to run?
From my experience I will add about this question...

The O-E values will show you where a significanct difference resides. Also some have suggested using a phi coefficient from a 2 x2 table as an effect size. This requires combining cells to have a group of interest against all others (almost like a planned contrast when we combine groups).
 

trinker

ggplot2orBust
#6
becky52 said:
however as I am running multiple tests I am at risk of type 1 error
May I ask why you're running multiple tests if you have a 4x2 design? I agree with GM that more info would help us give you better answers.