Chi Squared for Ordinal level Data ??

[Pippy]

Hi all

Please could someone explain to me in what circumstances you can use Chi squared with ORDINAL level data?

I have collected data which is of ordinal level and want to check for differences between variables. I initially used Mann Whitney U tests however my tutor has advised me that although I can do this, with the results I have obtained (which are mainly graded into low, moderate or high categories) he suggested Chi squared is better. However I thought Chi Squared was mainly used for nominal level data. Any advice and explanation would be greatly appreciated.

Thank you!

[noetsi]

You can use chi square with ordinal or interval data. It wont violate the assumptions. But why would you? When you have ordinal level data there are better methods that give you more information.

One possibility, if the data is normally distributed, is to do ordered logistic regression.

[Pippy]

Thank you Noetsi for your reply

Im doing a Masters degree so im new to statistics. We have been taught a variety of different tests however the only two tests we have been taught that would suit the data i have collecetd are the Mann WHItney U Tests or apparently the Chi Square test. We have not done what you have mentioned about the ordered statistic regression im afraid. Do you think Mann WHitney U test is better than Chi square for what im trying to find out?

Thanks again

[noetsi]

I don't work much with non-parametric methods such as Mann Whitney so I am not a good person to answer that question. Generally speaking chi square is the simplest statistic you can use requiring the fewest assumptions. It is used when you have nominal data or are not familar generally with statistics. If your data is ordinal then there are certainly better tests although which one to use would have to come from someone else.

This might be a starting point

http://www.wadsworth.com/psychology_...e_stat_24.html

[Pippy]

Thank you Noetsi for the link. It was helpful

I will continue to look into it!

[noetsi]

Somewhere on this board, maybe in the FAQ I think we have a diagram discussing this issue (what test to use)

[Eugenie]

I woke up this morning and realised why one might use Chi-squared to test ordinal level data instead of a U test.
It depends of course on what null hypothesis you are interested in testing. A U test will examine the null hypothesis
that there is no tendency of one group to rank things higher than the other group. A Chi-squared test will examine
the null hypothesis that the two groups tend to rank things in the same way. Consider a situation in which
men and women are asked to rank something on a 3 point scale, and suppose that the men always choose the rank 2,
whereas the women are equally split between 1 and 3. then there is no tendency of one group to rank things
higher than the other group, that is, the U test will be insignificant. However, certainly the two groups rank things
differently, so the chi squared test will be significant.

So the question is, are you interested in if one group ranks things higher than the other group, or are you interested
in telling if, eg, the responses of one group tend to be more polarised than in the other group?