Statistics & the Chi Square function

[rujedi]

Hi everyone,

I'm teaching statistics and started incorporating Excel alongside R. When we got to running a Chi-Square test, I noticed something very odd that I cannot figure out. Can someone tell me why I get different results with the same formula function.
Note, I'm using Excel 365 on Windows 10.

	A	B	C	D	E
1	Observed	past	present	future	row.total
2	up	12	45	78	135
3	down	23	56	89	168
4	col.total	35	101	167	303

<tbody>
</tbody>

	A	B	C	D	E
7	Expected	past	present	future	row.total
8	up	15.59	45.00	74.41	135
9	down	19.41	56.00	92.59	168
10	col.total	35	101	167	303

<tbody>
</tbody>

When I use =CHISQ.TEST(B2:D3,B8:D9) I get a p-value of 0.4051 which is correct.
BUT.....
If I rearrange the tables to look like below, I get a different p-value.

	H	I	J
1		observed	expected
2	up past	12	15.59
3	up present	45	45.00
4	up future	78	74.41
5	down past	23	19.41
6	down present	56	56.00
7	down future	89	92.59
8	total	303	303

<tbody>
</tbody>

When I use =CHISQ.TEST(I2:I8,J2:J8) I get a p-value of 0.8741 which is not correct.

What's going on that I am missing? Does it have something to do with how the degrees of freedom are being counted?
I didn't see an function option to specify a Goodness of Fit or Test for Independence, which would change how df is counted.

Is there a function to produce the Chi-Square statistic?
At first I thought it was =CHISQ.TEST(), because =CHISQ.DIST.RT also gives me the p-value if I provide it the Chi-Square statistic and the degrees of freedom.
When I start to type =CHI all of the options that pop-up have something to do with the probability and not providing the statistic.
I don't see Chi-Square listed in the Data Analysis Tookpak.

Thanks for the assist!
Shawn

 

[AhoyNC]

You are right it has to do with the df.
df=(rows-1)(columns-1)

In your first table you have 2 rows and 3 columns = (2-1)(3-1)=2 df
Your second table has 6 rows and 2 columns = (6-1)(2-1)=5 df

If you calculate chi-square in both cases you will get 1.803
I don't know of a function in Excel that will only calculate chi-square.

Excel Workbook
	A	B	C	D	E	F
1	Observed	past	present	future	row.total
2	up	12	45	78	135
3	down	23	56	89	168
4	col.total	35	101	167	303
5
6
Table 1

Excel Workbook
	A	B	C	D	E	F
1		observed	expected		chi-square
2	up past	12	15.59		0.82669
3	up present	45	45		0.00000
4	up future	78	74.41		0.17320
5	down past	23	19.41		0.66399
6	down present	56	56		0.00000
7	down future	89	92.59		0.13920
8	total	303	303		1.80308	= chi-square
9
10
11	chi-square	1.803
12	df	5
13	p value	0.8756765
Table 2

 

[AhoyNC]

I noticed that Excel Jeanie didn't post everthing above.

Excel Workbook
	A	B	C	D	E	F
1	Observed	past	present	future	row.total
2	up	12	45	78	135
3	down	23	56	89	168
4	col.total	35	101	167	303
5
6
7	Expected	past	present	future	row.total
8	up	15.59	45	74.41	135
9	down	19.41	56	92.59	168
10	col.total	35	101	167	303
11
12
13	chi-square	1.803082
14	df=	2
15	p value	0.405944
16
Table 1

Excel Workbook
	A	B	C	D	E	F
1		observed	expected		chi-square
2	up past	12	15.59		0.82669
3	up present	45	45		0.00000
4	up future	78	74.41		0.17320
5	down past	23	19.41		0.66399
6	down present	56	56		0.00000
7	down future	89	92.59		0.13920
8	total	303	303		1.80308	= chi-square
9
10
11	chi-square	1.803
12	df	5
13	p value	0.8756765
Table 2

 

[rujedi]

Thank you AhoyNC!
I had a feeling it was the degrees of freedom tripping up the function. Too bad I cannot specify the degrees of freedom in that function or specify if it is a Goodness of Fit or Test of Independence Chi-Square since df is different for each.

This also means that I will have to be very specific if allow my students to reshape the table to manually calculate Chi-Square (less space & easier to drag formulas) since it will change the df. They'll have to use the =CHI.DIST.RT(x, df) function and input the df since the =CHISQ.TEST(obs, exp) function doesn't have the ability to set the df.

Excel Jeanie HTML 4--Very neat trick. That will surely save me work trying to replicate my spreadsheet in a forum post.

I find it interesting, that Excel offers two ways to get the p-value:
=CHI.DIST.RT(x, df) when we already have the Chi-Square and df, and
=CHISQ.TEST(observed, expected) when we don't yet have the Chi-Square or df.

Curious though--one of my students started using =CHISQ.INV.RT(probability, df) to generate the Chi-Square statistic. Sometimes this works and seems exact, while other times it is in the ball park but not the same value.

Using the example of Excel tables, =CHISQ.INV.RT(B13, B14) results in a Chi-Square of 1.807. Pretty spot on this time.

 

[AhoyNC]

Your students could add a 3rd table to calculate the (obs-exp)^2/exp as in cell B14 below and the just copy the formula across and down. Then just use the SUM function to get the Chi-square.

Excel Workbook
	A	B	C	D	E
1	Observed	past	present	future	row.total
2	up	12	45	78	135
3	down	23	56	89	168
4	col.total	35	101	167	303
5
6
7	Expected	past	present	future	row.total
8	up	15.59	45	74.41	135
9	down	19.41	56	92.59	168
10	col.total	35	101	167	303
11
12
13	(Obs-Exp)^2/Exp	past	present	future
14		0.8267	0.0000	0.1732
15		0.6640	0.0000	0.1392
16
17	Chi-square	1.80308
Sheet

 

[rujedi]

I wholly agree AhoyNC. I actually make them do the Chi-Square step by step in Excel, to eventually sum the Chi-Square, look up the critical value on stats tables and determine the statistical significance. This was usually to help make sure they knew how to do it 'by hand' and double-check the results they get from computing the value in R (or SPSS or Stata depending which class I'm teaching).

Only recently did we start exploring Excel's ability calculate p-values and try to calculate the Chi-Square statistic, which is what led to my initial confusion.

Thanks again for the assist!