Categorical variables in SPSS - Chi-square and then what?

#1
Good morning from Portugal! I´m a biologist and I use SPSS for my statistical analysis.

I have some categorical data from questionnaires and made a chi-square analysis (most of the data had assumptions violated, so I used Fisher´s Exact test instead) followed by Phi or Cramer´s V to analyse the strength of association. My question is ... and then what? I know two variables are associated, I know if it is a strong or weak association, but I would like to know which category from the variable is really making the difference (sorry if I can´t explain correctly).

I´ll give an example:

I tested if respondents age were related to what they know about an invasive species - it is, and is a strong association. How do I know which of the age classes knows better the species?

Most of my tables are nxn. I send a crosstab in the attachment. age vs plant name crosstab.PNG

Thanks in advance!
 

Miner

TS Contributor
#2
I am a Minitab user, and it provides a "contribution to Chi-square" statistic for each cell. You then look for the largest statistic by cell. The formula for this is:

1625227305530.png
 

gianmarco

TS Contributor
#6
Hello.
First, a general comment on the cross-tab itself. If your goal is to test if there is any association between age and knowledge, I would drop the variable regarding nationality. Such complex table is difficult to read.

If you want to spot which cell significantly contribute to the rejection of the Null Hypothesis, you may want to inspect the table of standardized residuals. These are the residuals (the ones pointed out by Miner) but expressed as z-score. Any residual larger than 1.96 (in terms of absolute value) indicates a significant departure from independence (at alpha 0.05).

SPSS should have this option, as far as I recall.

Hope this helps
Gm
 
#7
Thanks for your reply!

Hello.
First, a general comment on the cross-tab itself. If your goal is to test if there is any association between age and knowledge, I would drop the variable regarding nationality. Such complex table is difficult to read.

If you want to spot which cell significantly contribute to the rejection of the Null Hypothesis, you may want to inspect the table of standardized residuals. These are the residuals (the ones pointed out by Miner) but expressed as z-score. Any residual larger than 1.96 (in terms of absolute value) indicates a significant departure from independence (at alpha 0.05).

SPSS should have this option, as far as I recall.

Hope this helps
Gm