χ2 –test (Chi square test)

Chi –square denoted by Greek letter χ2 and has developed by Karl Pearson in 1990. This theory
describes the magnitude of observed and theoretical frequency (expected frequency) distribution.
Since Chi square test does not make any assumption about population parameter. It is also called
a distribution of free test.
Properties of chi- square test
1. Chi square distribution which lies between 0 to ∞ is continuous probability distribution.
2. Since chi square is sum of square, its value cannot be negative.
3. The value of chi- square will be zero if difference of each pair of chi-square is zero.
Application of chi-square
1. Chi- square test for goodness of fit.
2. Chi -square test for independence of attributes
3. Chi- square test for population variance.
Chi -Square test for goodness of fit.
The goodness of fit describes the differences between the observed and expected frequency
distribution. If the observed values are close to expected value under a hypothesis, the fit is said
to be good. The test goodness of fit is also used to test the significant difference between
observed and expected frequency distribution of binomial, Poisson and normal etc.
Steps
1. Setting of hypothesis
Null hypothesis (H0): There is no significance difference between observed (experimental) and
the expected (theoretical) frequency.
Alternative hypothesis (H1): there is significant difference between observed (experimental) and
the expected (theoretical) frequency.
2. Compute the test statistic
(𝑂−𝐸)2
χ2 =∑ 𝐸
ΣO
Where, E = 𝑛

3. Level of significance
4. Find the tabulated value (critical) values of ∑𝜒 2 for (n-1) degree of freedom.
5. Decision: if calculated value is less than its tabulated value, Ho is accepted and if calculated
value is greater than its tabulated value Ho is rejected and alternative hypothesis is accepted.
Conditions (assumptions) for the validity of chi-square test
1. N, total number of observed frequency greater than 50.
2. The sample observation should be independent i.e. no individual item should be included
twice or more in the sample.
3. Total observed and expected frequency should be equal. i.e. ΣO = ΣE =N
4. Each expected frequency should be large than 10. But in practice, no expected frequency is
less than5. If any frequency less than 5, we should be pooled the expected frequencies and
observed frequencies.
Then, d.f. = n-1 - k1 - k2
Example:
The following tables give the number of accidents on the road during a week. Test whether the
accidents are uniformly occurred over the week.
Day Sun Mon Tue Wed Thus Fri Sat
No. of accidents 16 12 14 18 9 11 8
Solution:
Step 1: Null hypothesis HO: The accidents are uniformly occurred over the week.
Alternative hypothesis H1: The accidents are not uniformly occurred over the week.
Step 2: Test statistic:
(𝑂−𝐸)2
χ2 =∑ 𝐸
ΣO
Where, E = Expected frequencies = Where, E = 𝑛

Calculation of χ2
O E O-E (𝑂 − 𝐸)2 (𝑂 − 𝐸)2
𝐸
16 13 3 9 0.6923
12 13 -1 1 0.0769
14 13 1 1 0.0769
17 13 4 16 1.2308
9 13 -4 16 1.2308
15 13 2 4 0.3077
8 13 -5 25 1.9231
ΣO=91 ΣE=91 ∑
(𝑂−𝐸)2
= 5.5385
𝐸
(𝑂−𝐸)2
χ2 = ∑ = 5.5385
𝐸

Step 3: Degree of freedom= n-1=7-1= 6

Step 4: Level of significance (α) = 5%
Step 5: Tabulated value of χ2at 5% level of significance for 6 degree of freedom is 12.592.
Step 6: Since the calculated value of χ2 is less than its tabulated value. So HO is accepted.
Example:
In a set of random numbers, the digits 0, 1, 2,……..9 were found to have the following
frequencies:
Digits 0 1 2 3 4 5 6 7 8 9
Frequency 43 32 38 27 38 52 36 31 39 24
Test whether they are significantly different from those expected on the hypothesis of uniform
distribution. (TU 2050 MBA)
Solution:
Step 1:
Null hypothesis HO: The digits are not significantly different from those expected on the
hypothesis uniform distribution.
Alternative hypothesis H1: The digits are significantly different from those expected on the
hypothesis uniform distribution.
Step 2: Test statistic:
(𝑂−𝐸)2
χ2 =∑ 𝐸

Calculation of χ2
O E O-E (𝑂 − 𝐸)2 (𝑂 − 𝐸)2
𝐸
43 36 7 49 1.36
32 36 -4 16 0.44
38 36 2 4 0.11
27 36 -9 81 2.25
38 36 2 4 0.11
52 36 16 256 7.11
36 36 0 0 0
31 36 -5 25 0.69
39 36 3 9 0.25
24 36 -12 144 4
ΣO =91 ΣE =360 ∑
(𝑂−𝐸)2
= 16.32
𝐸

(𝑂−𝐸)2
χ2 =∑ = 16.32
𝐸

Step 3: Degree of freedom= n-1=10-1= 9

Step 4: Level of significance (α) =5%
Step 5: Tabulated value of χ2 at 5% level of significance for 9 degree of freedom is 16.92.
Step 6: Since the calculated value of χ2 is less than its tabulated value. So HO is accepted.
Example:
Among 64 of springs of a certain cross between guinea pigs, 34 were red, 10 were black and 20
were white. According to the genetic model, these numbers should be in the ratio 9:3:4. Are the
data consistent with the model at 5percent level? (TU 2045 MBA)
Solution:
Null hypothesis H0: The data are consistent with the model.
Alternative hypothesis H1: The data are not consistent with the model.
Under H0, the test statistic is
(𝑂−𝐸)2
χ2 =∑
𝐸

Calculation of χ2
O E O-E (𝑂 − 𝐸)2 (𝑂 − 𝐸)2
𝐸
34 36 -2 4 0.1111
10 12 -2 4 0.3333
20 16 4 16 1
ΣO =64 ΣE =64 ∑
(𝑂−𝐸)2
= 1.44
𝐸
 (𝑂−𝐸)2
χ2 =∑ =1.44
𝐸

Degree of freedom = n-1= 3-1=2

Level of significance (α)=5%
Tabulated value of χ2 at 5% level of significance for 2 degree of freedom is 5.991
Since the calculated value of χ2 is less than its tabulated value. So HO is accepted.
Working note:
Calculation of expected value (E)
Total ratio = 9 + 3 + 4 =16
9
For 34, 16 x 64 = 36
3
For 10, x 64 = 12
16
4
For 20, 16 x 64 = 16

Example
Fit a binomial distribution to the following data, assuming that the coin is unbiased.
No. of heads 0 1 2 3 4 5
Frequency 90 560 1000 900 600 50
Test the goodness of fit for above data. (TU 2053 MBA)
Solution:
Let r be the number of heads occurred. Then, probability of getting r heads out of n coins is
given by
P(X= r) = n𝑐𝑟 𝑝𝑟 𝑞 𝑛−𝑟 where, r =0, 1, 2,…..n
Here, n = 5, p = 0.50, q =1-0.50 = 0.50
∴ P(X=r) = 5𝑐𝑟 (0.5)𝑟 (0.5)5−𝑟
= 5𝑐𝑟 (0.5)5
Where r = 0, 1, 2,3,4,5
The expected frequencies of getting r heads is
E=f (r) = N.P(X=r)
=N x 0.0313 x 5Cr
= 3200 x 0.0313 x 5Cr
Where, N=3200
Calculation of expected frequency ( E ) = f (r)= 100 x5Cr
f(0)=100 x 5C0 =100.16 =100
f(1)= 100 x 5𝑐1 =500
f(2)=100 x 5𝑐2 =1000
f(3)=100 x 5𝑐3 = 1000
f(4)=100 x 5𝑐4 =500
f(5)=100 x 5𝑐5 =100
Hence, the expected frequencies of heads are
No. of heads 0 1 2 3 4 5
Frequency 100 500 1000 1000 500 100

In order to test the goodness of fit for above data, the steps for χ2-test are as follows:
Null hypothesis H0: the binomial distribution with p = q= 0.50 is a good fit to the given data.
Alternative hypothesis H1: the binomial distribution with p = q= 0.50 is not good fit to the given
data.
Test statistic: Under H0, the test statistic is
Calculation of χ2
(𝑂−𝐸)2
χ2 =∑ 𝐸

O E O-E (𝑂 − 𝐸)2 (𝑂 − 𝐸)2

𝐸
90 100 -10 100 1
560 500 60 3600 7.2
1000 1000 0 0 0
900 1000 -100 10000 10
600 500 100 10000 20
50 100 -50 2500 25
ΣO =3200 ΣE =3200 ∑
(𝑂−𝐸)2
= 63.2
𝐸

(𝑂−𝐸)2
χ2 =∑ = 63.2
𝐸

Level of significance (α) =5%

Degree of freedom (d.f.) = n-1= 6-1 =5
Tabulated value of χ2 at 5% level of significance for 6 degree of freedom is 11.070
Since the calculated value of χ2 is greater than its tabulated value. So HO is rejected and H1 is
accepted.
Example:
Fit a Poisson distribution to the following data and test the goodness of fit.
X 0 1 2 3 4
F 120 60 15 4 1
Test the goodness of fit for above data. (TU 2053 MBA)
Solution:
Null hypothesis H0: Poisson distribution is good fit to the given data.
Alternative hypothesis H1: Poisson distribution is not a good fit to the given data.
𝜮𝒇𝑿
Now, λ= mean of the Poisson distribution which can be obtained by using, 𝑋̅= 𝑵
𝜮𝒇𝑿
̅=
∴λ=m=𝑿 𝑵
𝟎 𝒙 𝟏𝟐𝟎 +𝟏 𝒙 𝟔𝟎 +𝟐 𝒙 𝟏𝟓 +𝟑 𝒙 𝟒+𝟏 𝒙 𝟒
= = 0.53
𝟐𝟎𝟎

Solution:
O= Observed frequency
E= Expected frequency
=N x P(X= r)
𝑒 −𝑚 . 𝑚 𝑟
=N x P(X= r) =N. 𝑟!

Where, r = 0, 1, 2, 3, 4
𝑒 −0.53 . 0.53𝑟 2.7183−0.53 . 0.53𝑟
E= f(r) =200 x P(X= r) =200 x =200 x
𝑟! 𝑟!

=Where, r = 0, 1, 2, 3, 4
Calculation of expected frequency ( E)
(0.53)0
f(0)= 117.72 x = 117.72
0!
(0.53)1
f(1)= 117.72 117.72 x = 62.39
1!
(0.53)2
f(2)= 117.72 x =16.53
2!
(0.53)3
f(3)= 117.72 x = 2.921
3!
 (0.53)4
f(4)= 117.72 x =0.381
4!

Calculation of χ2
O E O-E (𝑂 − 𝐸)2 (𝑂 − 𝐸)2
𝐸
120 117.72 2.28 5.1984 0.0442
60 62.39 -2.39 5.7121 0.0916
15 16.53
4 20 2.921 19.83 0.17 0.0289 0.0015
1 0.381
ΣO =200 ΣE =200 ∑
(𝑂−𝐸)2
= 0..1373
𝐸

(𝑂−𝐸)2
χ2 =∑ = 0.1373
𝐸

K1= 1 d.f. lost due to computed of parameter λ

K2= 2 d.f. lost due to pooling last three expected frequencies to make 5 or more
∴ d.f. = n-1 –k1-k2 = 5-1-1-2 =1
Level of significance (α) = 5%
Tabulated value at 1 d.f. and 5% level of significance = 3.841
Since the calculated value of χ2 is less than its tabulated value. So HO is accepted.
Chi- square test for independence of attributes
Chi-square test for independence of attributes is applied to test whether two or more attributes
are associated or not. i.e. whether the attributes are related or independent. In other words, if
contingency table (row or column) are given, chi-square independence of attributes is used.
(𝑂−𝐸)2
χ2 =∑ 𝐸
𝑹𝑻 𝑿 𝑪𝑻
Where, E= 𝑵

𝑵(𝐚𝐝−𝐛𝐜)𝟐
Or, χ2 =
(𝐚+𝐛)(𝐜+𝐝)(𝐚+𝐜)(𝐛+𝐝)

Degree of freedom (D.F.) = (r-1) (c-1)

Where r = no. of rows
And c = no. of columns
Example:
Do the following data provide evidence of the effectiveness of inoculation in preventing
tuberculosis? (TU 2046 MBA)
Attacked Not attacked Total
Inoculated 20 300 320
Not inoculated 80 600 680
Total 100 900 1000

Solution:
Null hypothesis H0: Inoculation and attack by T.B. are independent. In other words, Inoculation
is not effective in preventing tuberculosis.
Alternative hypothesis H1: Inoculation and attack by T.B. are dependent. In other words,
Inoculation is effective in preventing tuberculosis.
Test statistic: Under H0, the test statistic is
(𝑂−𝐸)2
χ2 =∑ 𝐸
𝑅𝑇 𝑋 𝐶𝑇
Where, E= Expected frequency in a cell = Where, E = 𝑁

Attacked Not attacked Total

Inoculated 20 300 320 RT
Not inoculated 80 600 680 RT
Total 100 CT 900 CT 1000 Total

Calculation of expected frequencies and χ2

O E O-E (𝑂 − 𝐸)2 (𝑂 − 𝐸)2
𝐸
20 32 -12 144 4.5
300 288 12 144 0.5
80 68 12 144 2.12
600 612 -12 144 0.24
ΣO =1000 ΣE =1000 ∑
(𝑂−𝐸)2
= 7.36
𝐸
(𝑂−𝐸)2
χ2 =∑ = 7.36
𝐸

Alternative method:
The χ2 for 2x2 Contingency table can also be computed by using following formula directly.
a=20 b=300 a + b =320
c=80 d =600 c + d=680
a + c =100 b + d=900 N=1000
𝑵(𝐚𝐝−𝐛𝐜)𝟐 𝟏𝟎𝟎𝟎(𝟐𝟎 𝐱 𝟔𝟎𝟎−𝟖𝟎 𝐱 𝟑𝟎𝟎)𝟐
χ2 = (𝐚+𝐛)(𝐜+𝐝)(𝐚+𝐜)(𝐛+𝐝) = = 7.35
𝟑𝟐𝟎 𝐱 𝟔𝟖𝟎 𝐱 𝟏𝟎𝟎 𝐱 𝟗𝟎𝟎

Degree of freedom (d.f.) = (r-1) (c-1) = (2-1) (2-1) = 1

Level of significance (α) = 5%
Tabulated value at 5% level of significance for 1 d.f. =3.841
Since calculated value is greater than its tabulated value. So null hypothesis is rejected and
alternative hypothesis is accepted.
Yates correction for continuity for 2 x2 tables
In 2x2 contingency table, the number of degree of freedom is (r-1) (c-1) = (2-1) (2-1) =1. After
pooling degree of freedom=1-1=0 which is meaningless since chi square must be at least 1
degree of freedom. In this case we apply a correction given by F-Yates which is usually known
as “Yates” correction for continuity. To remove this deficiency, Yates suggested that 0.50 be
added to the small frequency for which the expected frequency is less than 5 and other frequency
be adjusted by adding or subtracting 0.50 in such a way that the marginal totals remain the same.
𝑵
𝑵(‫𝐝𝐚׀‬−𝐛𝐜‫׀‬− )𝟐
χ2 = 𝟐
(𝐚+𝐛)(𝐜+𝐝)(𝐚+𝐜)(𝐛+𝐝)

Degree of freedom (D. F.) = (r-1) (c-1)

Chi –square test for population variance
Following conditions used/need for testing the hypothesis about a single population variance of
chi-square distributions.
i) The population from which the sample is drawn is normally distributed.
ii) The data is from a random sample.
iii) The observation must be independent of each other.
Example:
In an experiment on the immunization of goats from Anthrax, the following results were
obtained. Derive your inference on the efficiency on the vaccine.
Died of anthrax Survived Total
Inoculated with vaccine 4 10 14
Not inoculated 7 7 14
Total 11 17 28
Solution:
Since first cell frequency is 4 which is less than 5, we should apply Yates correction for
calculating χ2. For this, add0.5 to cell frequency which is less than 5 and adjust the remaining
frequencies by fixing row and column total. Thus, adjust 2 x 2 contingency table is presented in
the following table.

Total
a = 4.5 b = 9.5 14
c = 6.5 d =7.5 14
Total 11 17 28

Null hypothesis H0: The vaccine is not effective in preventing from died of Anthrax.
Alternative hypothesis H1: The vaccine is effective in preventing from die of Anthrax.
Test statistic: Under H0, the test statistic is
O E O-E (𝑂 − 𝐸)2 (𝑂 − 𝐸)2
𝐸
4.5 5.5 -1 1 0.1818
9.5 8.5 1 1 0.1176
6.5 5.5 1 1 0.1818
7.5 8.5 -1 1 0.1176
ΣO =28 ΣE =28 ∑
(𝑂−𝐸)2
= 0.5988
𝐸
(𝑂−𝐸)2
χ2 =∑ = 7.36
𝐸
𝑅𝑇 𝑋 𝐶𝑇
Where, E= Expected frequency in a cell = Where, E= 𝑁

Degree of freedom (d.f.) = (r-1) (c-1) = (2-1) (2-1) = 1

Level of significance (α) = 5%
Tabulated value at 5% level of significance for 1 d.f. =3.841
Since calculated value is less than its tabulated value. So the null hypothesis is accepted.
Chi-square test for population variance is used to test whether the population variance 𝜹𝟐
has some specified value.
(𝑛−1)𝑆 2
χ2 = 𝛿2

𝛿 2 =Population variance
𝑆 2 = an unbiased estimated of population variance δ2
1
𝑆 2 = 𝑛−1 Σ[𝑋 − ̅̅̅
𝑋)2
1 (𝑋)2
𝑆 2 = 𝑛−1 Σ[𝑋 2 - ]
𝑛

Where, d=X - A
If 𝑆 2 denote the biased estimate of population variance 𝛿 2 , then
1
̅̅̅2
𝑆 2 = 𝑛−1 Σ[𝑋 − 𝑋)
If sample variance is given
𝑛𝑆 2
χ2 =
𝛿2

Note:
𝑛𝑆 2
i) For small sample case (i.e. n < 30), χ2 = 𝛿2

ii) For large sample case (i.e. n ≥ 30)

Z=√2x 2 - √2n − 1
d.f. = n-1
Example:
A random sample of size 25 from a normal population gives the sample standard deviation of 9.
Test the hypothesis that the population variance is 121.
Solution:
Given
Standard deviation(s) = 9 (biased estimate)
Population variance (𝛿 2 ) =121
Sample size (n) = 25
Null hypothesis H0: 𝛿 2 =121 i.e. Population variance is 121.
Alternative hypothesis H1: 𝛿 2 ≠ 121 i.e. population variance is not equal to 121. (Two tailed
test)
Test statistic, under H0, the test statistic is
𝑛𝑆 2 25𝑥 92
χ2 = = =16.7355
𝛿2 121

Degree of freedom (d.f.) = 25-1=24

Level of significance (α) =5%
Tabulated value at 5% level of significance and 24 d.f. is 36.415
Since calculated value of χ2 is less than its tabulated value. So the null hypothesis is accepted.
Example:
A random sample of 10 bulbs has the following life in months: 24, 26,32,28,20,23,34,30 and 43.
Can we say that variance of distribution of life of all bulbs from which the above sample of 10
bulbs was drawn is equal to 20 months?
Solution:
Sample size (n) = 10
Population variance (𝛿 2 ) =20
Null hypothesis H0: 𝛿 2 =20 i.e. Variance of distribution of life of all bulbs is 20 months.
Alternative hypothesis H1: 𝛿 2 ≠ 20 i.e. Variance of distribution of life of all bulbs is not equal to
20 months.
Test statistic, under H0, the test statistic is
(𝑛−1)𝑆 2
χ2 = 𝛿2

Calculation of sample variance (𝑺𝟐 )

Life (months) d= X-20 𝒅𝟐
24 4 16
26 6 36
32 12 144
28 8 64
20 0 0
23 3 9
34 14 196
302 10 100
43 23 529
Σd =80 Σ𝒅𝟐 =1094
 1 (∑𝑑)2
Now, 𝑆 2 = 𝑛−1 [Σ𝑑 2 - ]
𝑛
1 (∑80)2
= [1094 - ] = 50.44
10 −1 10
(𝑛−1)𝑆 2
χ2 = 𝑆2
(10−1)𝑥 50.44
= =22.6980
20

Degree of freedom (d.f.) = n-1= 10-1= 9

Tabulated value at 9 d.f. and 5% level of significance is =16.919
Since calculated value of χ2 is greater than its tabulated value. So null hypothesis is rejected and
alternative hypothesis is accepted.

Practical problems
1. The following table gives the number of aircraft accidents that occurred during the various
days of the week. Test whether the accidents are uniform distributed over the week? (T.U. 2015
M, 2016 R, 2017 M)
Day Sun Mon Tue Wed Thu Fri Sat
No. of accidents 14 16 8 12 11 9 14
(Ans: χ2= 4.1665)
2. The numbers of automobile accidents per week in a certain community were as follows:
12 8 20 2 14 10 15 6 9 4. Are these frequencies in agreement with the belief that accidents
conditions were the same during the 10 week periods under considerations? (T.U.2041 MBA)
(Ans: χ2= 26.6)
3. A sample of 500 workers of a factory according to sex and nature of work is as follows.
Sex Total
Male Female
Technical 200 100 300
Non-technical 50 150 200
Total 250 250 500
Test at 5% level of significance whether there exists any relationship between sex and nature of
work. (T.U. 2057, 2014 R MBS) (Ans: χ2= 83.33)
4. Four hundred employees of a factory are classified according to their level of and decision. Do
you agree with the statement that decisions vary according to level of employee? (T.U. 2057
MBS)
Decisions Senior officer officer Junior officer Total
Quick 60 55 70 210
Slow 40 45 90 190
Total 100 100 160 400
(Ans: χ2=8.377)
5. Marketing manager of a company is concerned that the Brand’s share is unevenly distributed
throughout the country. In a survey of 100 consumers in each geographic religions of the
country, following result was obtained. (T.U. 2056 MBA)

Region Total
North east North west South east South east
Purchase the brand 40 55 45 50 190
Do not purchase the brand 60 45 55 50 210
Total 100 100 100 100 400

Is the marketing manager’s concern correct? (Ans: χ2 =5.012)

6. A manufacturer of TV sets was trying to find out what variable influenced the purchased of
TV set. A sample of 500 households was selected and the information obtained is classified as
shown below:
Have TV set Do not have TV set
Low income group 0 250
Middle income group 50 100
High income group 80 20

Is there evidence from the above data of a relation between ownership of TV sets and level of
income? (T.U. 2055 MBA) (Ans: χ2=243.59)
7. The department of Business Administration at a university would like to determine whether
there is a relationship between student’s interest in business administration and ability in
mathematics. A random sample of 200 students is selected and they are asked whether their
ability in mathematics and interest in business administration are low, average or high. The
results were as follows:
Interest in Business Administration Ability in mathematics Total
Low Average High
Low 60 15 15 90
Average 15 45 10 70
High 5 10 25 40
Total 80 70 50 200
Test whether there is any relationship between student interest in Business Administration and
ability in mathematics. (T.U. 2057 MBA, 2015 M MBS) (Ans: χ2= 84.75)
8. The following table shows the reaction of 1000 audience about the new movie after leaving
the theatre in different locations of the country.
Audience Reaction Location Total
Kathmandu Pokhara Bharatpur
Excellent 250 120 80 450
Good 200 50 50 300
Poor 150 50 50 250
Total 600 220 180 1000

Test at 5% level of significance, the hypothesis that the audience reaction is independent of the
location. (T.U. 2018 R, MBS) ( Ans: χ2= 9.488)
9. Following table provides data with regard to structure of the fathers and their sons at the age
of 20 years.
Structure Structure of fathers Total
of sons Tall Short
Tall 8 3 11
Short 6 7 13
Total 14 10 24

Test that the structure of sons is independent of the structure of the fathers. ( Ans: χ2= 0.8104)
10. Test the hypothesis that δ =18, given that s = 21 for random sample of size 29 from a normal
population. ( Ans: χ2= 39.47)
11. Height in inches of 8 students is given below:
72 74 75 76 62 64 68 62
Can we say that variance of distribution of heights of all students from which the above sample
of 8 students was drawn is equal to 25 inches? ( Ans: χ2= 0.3886)
4. Test the hypothesis that δ =12, given that s=18 for a random sample size of 40.
(Ans: χ2 =4.5846)