Question about Chi-Square test for independence

#1
Hi everyone!
I'm carrying out some research in linguistics. I am trying to show if some discourse markers (e.g. causal, additive, etc.) are significantly associated to particular genres (news, conversation, etc.).

My problem/doubt is that everywhere I look for info. about chi-squared test for independence, the examples show 2 by 2 or 3 by 3 tables. Mine would be of 12 rows (different types of connectors) and 5 columns (different genres). Is there any problem with that? Should I use another technique? (I'm using SPSS, just in case)

Thank you in advance.
 

gianmarco

TS Contributor
#2
Hello,
I think there is no problem in analysing a 12x5 crosstabulation using chi-sq test. The only limitation is that, according to some rules of thumb (which someone consider too conservative), you should be sure that no expected frequency is smaller than 5, plus some other warnings. You may find information about these just googling a little.

More often than not, once you have assessed if there is a significant association between rows and columns, the next step could be to locate which cell significantly contributes to the rejection of the Null Hypothesis of independence. You may want to inspect the table of standardized residuals. As far as I recall, SPSS provides that facility. Further, you may want to assess the strenght of the association using some categorical correlation coefficient, such as Cramer's V for instance.

Moreover, sometimes (as tables become larger) representing a crosstabulation graphically in a low-dimensional space may help the interpretation of the association between row and column categories. In this case, you may want to use Correspondence Analysis. I have build a website which provides an introduction to CA and describes a R package aimed at helping the interpretation of the CA results: http://cainarchaeology.weebly.com/

Hope this helps,
Gm