Mann Whitney Test

Sample Size: Test Statistics Critical Value Decision

Small Size: 𝑼𝒂, (𝒏𝟏, 𝒏𝟐) If Uo >
n1 ≤ and n2 ≤ 10 Uo = minimum if {U1 and U2} 𝑼𝒂, (𝒏𝟏, 𝒏𝟐)
n1 + n2 ≤ 20 where U1 and U2 = n1.n2 Do not reject Ho
U1 = n1.n2 +
𝑛1(𝑛1+1)
− 𝑅1 else reject Ho
2
𝑛2(𝑛1+1)
U2 = n1.n2 + − 𝑅2 From man If p-value > a,
2
Whitney table then we do not
find p value for reject Ho,
n1, n2 and Uo otherwise reject
For two tailed, Ho
p-value = 2po,
for one tailed p-
value = Po
Large Size: For large samples size, the distribution of Za from z-table If Za table > Z calc,
n1 > , n2 > 10 Uo is approximated to normal distn with We accept Ho,
n1 + n2> 20 𝑛1.𝑛2
mean = and else reject Ho
2
𝑛1.𝑛2(𝑛1+𝑛2+1)
variance =
12
Test statistics:
𝑛 ⋅𝑛
𝑈𝑜− 1 2
2
𝑧=
𝑛1𝑛2(𝑛1+𝑛2+1)
√
12

In case of tied observation, corrected sd:

Where n = n1 + n2 and ti = number for

times ith rank is repeated.
Median Test

Sample Size: Test Statistics: Critical Value Decision

Small Size: For two-
n1 ≤ and n2 ‘a’ Po = P(A≥a); tailed:
≤ 10
n1 + n2 ≤ 20 Test statistics are obtained as If 2Po > a,
follows: accept Ho,
else reject
Combine both samples and Ho
arrange them in ascending
order of magnitude such that For one-
n = n1 + n2 tailed:

calculate the median of the If Po > a,

combined sampled and count accept Ho,
the number of observations else reject
less than or equal to the Ho
median in the first sample

This is ‘a’

𝑛1+ 𝑛2
K=
2

Large Size: No No Total Extract X2 value from X2 table for 1 If X2cal >
of of
n1 > 10 and obs obs degree of freedom and a level of X2tab, we
n2 > 10 <= > significance reject Ho,
n1 + n2 > 20 Md Md else
Sample I a b a+b
accept.
Sample II
Total
c
a+c
d
b+d
c+d
a+b+c+d X2 α,1
Kolmogorov Smirnov Test

Test Statistics Critical Value Decision

One Sample:
Do = Max | Fe(x) – Fo(x) | Dtabulated = Dn, α If DO > Dn, α:
To check if
there is a CFe Then we reject
significant Where, Fe(x) =
n H0 otherwise
difference Cfe = expected cumulative we do not
between frequency reject H0.
observed and
expected Fe = expected frequency
frequency. 𝛴𝑓
= npi = 𝑛𝑜 𝑜𝑓 𝑐𝑎𝑡𝑒𝑔𝑜𝑟𝑖𝑒𝑠
CFo
Fo(x) =
n

Cfo = observed cumulative

frequency

Two samples: For two-tailed If Dn1, n2,a > Do,

Dtabulated = D(n1, n2), α Then we do not
Do = Max | F(x) – F(y) |
Small sample CFx
reject H0
n1 ≤ = n2 ≤ 40 Where, F(x) = , otherwise we
n1
n1 != n2 ≤ 20 CFy reject H0.
F(y) =
n2
Dtabulated = D (n1, n2), α If Dn1, n2, a > Do,
Then we do not
For two-tailed reject H0
Do = Max | F(x) – F(y)| otherwise we
reject H0.

Large sample
n1, n2 > 40 If X2cal > X2tab,
for n1 = n2
X2 calculated = X2(α, 2) degree
𝑛1𝑛2 we reject Ho,
and X2 = 4Do2( ) of freedom else accept Ho
𝑛1+𝑛2
n1, n2 > 20
for n1! =n2
Wilcoxon Matched Pair Sign Rank Test (used for a small sample, n<=20)

One Sample: Test Statistics Critical Value Decision

Use for small
sample size T = min {s(+), s(-), If Ta, ne >= T,
Ttabulated = Ta, ne
we reject
where,
S(+) = sum of ranks of Ho, else
difference with ‘+’ sign ne = effective sample size accept Ho
S(-) = sum of ranks of
ne = n – t
difference with ‘-‘ sign t = number of differences with zero

Kruskal Walis H test

One Test Statistics Critical Value Decision

Sample:
Small Obtain p-values If p- value > a
sample: we accept Ho,
n <= 5 and else reject
k=3

where, n = n1 + n2 + … + nk
if there is a tie in observation, then corrected H,

𝐻
Hcorr = 𝐶.𝐹

Large Same as above If H < X2tab,

sample:
X2tabulated =
Accept ho
X2a, (k-1) (degree else reject Ho
ni > 5 and
k>3 of freedom)
Freidman F-Test

K= number of samples, n = size of each sample

Sample: Test Statistics Critical Value Decision

Small
sample: Obtain p-values If p- value > a
2 <= n <= 5 from table we accept Ho,
and else reject Ho
k=4

if there is a tie in observation, then corrected Fr is

used,

𝐹𝑟
Fr corrected =
𝐶.𝐹

ti = no time ith rank is repeated

Large Same as above

Sample
X2tabulated = If Fr < X2tab,
Accept ho
n>5&k>3 X2a, (k-1) (degree else reject Ho
of freedom)