Lecture 12 – Chi-square – summary page
- ie male/female white/block/asian
-chi-square tests tell you whether differences across categorical variables are significant or not.
-compares the frequencies you actually find by the frequencies you would expect to get by chance.
• Only time we don’t need to worry about normality and homogeneity of variance
o Not based on normal curve which is why
-independence of cases
-expected cell counts are sufficient (no more than 20% of cells should have expected counts less than 5;
none should have expected counts less than 1). If any are less than 5, that’s a problem.
- makes the estimates unreliable if 20% is less than 5. If you don’t have enough cases with results you
can trust its prone to random fluxuation and we run into issues.
Example: Sex by contact with courts.
Male Female Total
Finish HS N expected N expected A
Did not Finish HS N expected N Expected
Total C B
To get sell count:
• Expected cell count = (row total/n) x (column total/n) x number of cases.
• = (A/B) + (C/B) x B
• To check expected cell counts:
-enter dependent variable into “Row” and independent variable into “Column”
-under “Cells”, check “Expected”
To get Chi-Square value need to calculate degrees of freedom as well
• Degrees of freedom = (# of rows-1)x(# of columns -1)
• Spss does this for