Why the Row percent Chisq value is the same as Col percent Chisq value?

[ybz12003]

Hello,

I use the Proc Freq to get Chisq value from Row percent and Column percent.  I found the result came back the same, why?

proc freq data=have;
	table  Insurance * PT_class / chisq nopercent norow;
	where 	PT_class in (1,2) and insurance in (1,2,4);
run;

proc freq data=have;
	table  Insurance * PT_class / chisq nopercent nocol;
	where 	PT_class in (1,2) and insurance in (1,2,4);
run;

[PaigeMiller]

It's the same data, and so you get the same Chi-squared value. The fact that different output options are chosen does not affect the Chi-squared value.

[ybz12003]

Should the chisq value in the comparison of column percent and row percent be different?

[StatDave · Accepted solution]

The chi-square is testing the association of the row and column variables in the table as a whole. It is not comparing rows irrespective of the columns or vice versa. Equivalently, the chi-square can be thought of as comparing the row profiles or the column profiles, but it is a matter of convenience to state it as one or the other since they are equivalent comparisons. Again, it is testing the table overall, not just the rows or just the columns. Consequently, the NOROW and NOCOL options have no effect on the chi-square computation.

[ballardw]

You can add the CellChi option to see the CELL contribution to the chi-square statistic. The "row" and "column" do not get a contribution. The counts are used to calculate the cell contributions but there is no actual "row chi-square".

[PaigeMiller]

---

@ybz12003 wrote:
Should the chisq value in the comparison of column percent and row percent be different?

---

No. All you have changed is the information that is displayed in the output from PROC FREQ. You have not changed the calculation of Chi-squared.