Negative Binomial & Geometric Distributions

Vijay Kumar

Department of Mathematics and Statistics

DDU Gorakhpur University, Gorakhpur
vkgkp@rediffmail.com

February 2023
 Negative Binomial Geometric PMF MGF CGF S & K Limiting form Recurrence

Negative Binomial distribution:

A random variable X is said to follow a negative binomial distribution if its probability mass
function is given by
 
 x+r−1 r x
p q : x = 0, 1, 2......
p(x) = P (X = x) = r−1
0, otherwise


The parameter p satisfies 0 ⩽ p ⩽ 1 and (1 − p) = q

Notation: X ∼ N B(r, p)
Suppose we have a succession of n Bernoulli trails. We assume that
the trials are independent,
the probability of success p in a trial remains constant from trial to trial.

V. Kumar, DDU Gorakhpur University, Gorakhpur B.Sc.-I : Negative Binomial Distribution Feb. 2023 2 / 16
 Negative Binomial Geometric PMF MGF CGF S & K Limiting form Recurrence

Negative Binomial distribution contd...

Let f (x; r, p) denote the probability that there are x failures preceding the rth success in x + r
trials. Now, the last trial must be a success, whose probability is p. In the remaining (x + r − 1)
trials we must have (r − 1) successes whose probability is given by
 
x + r − 1 r−1 x
p q
r−1
Therefore by compound probability theorem, f (x; r, p) is given by the product of these two
probabilities, i.e.    
x + r − 1 r−1 x x+r−1 r x
p q .p= p q
r−1 r−1
which is the (x + 1)th term in the expansion of pr (1 − q)−r , a binomial expansion with a
negative index. Hence the distribution is known as negative binomial distribution.

V. Kumar, DDU Gorakhpur University, Gorakhpur B.Sc.-I : Negative Binomial Distribution Feb. 2023 3 / 16
 Negative Binomial Geometric PMF MGF CGF S & K Limiting form Recurrence

Negative Binomial distribution contd...

Also ∞ ∞  
−r
(−q)x = pr (1 − q)−r = 1
X X
r
p(x) = p
x
x=0 x=0

Therefore p(x) represents the probability function and the discrete variable which follows this
probability function is called the negative binomial variate.
Geometric Distribution: If we take r = 1 we have

p (x) = q x p ; x = 0, 1, 2, . . . ,

which is the probability function of geometric distribution. Hence negative binomial distribution
may be regarded as the generalization of geometric distribution.
Consider a sequence of independent Bernoulli trials with success probability p. The distribution
of the random variable that represents the number of failures until the first success is called
geometric distribution.

V. Kumar, DDU Gorakhpur University, Gorakhpur B.Sc.-I : Negative Binomial Distribution Feb. 2023 4 / 16
 Negative Binomial Geometric PMF MGF CGF S & K Limiting form Recurrence

Relation with Binomial

In the binomial distribution, the number of successes out of a fixed number of trials is a
random variable, whereas in the negative binomial, the number of trials required to have a
given number of successes is a random variable.
The following relation between the negative binomial and binomial distributions is worth
noting.
P (X = k) = P (observing k or less failures before the rth success) = P ((k + r) or less
trials are required to 
have exactly r successes) = P (observing r or more successes in
k+r

P k+r i
(k + r) trials) = p (1 − p)k+r−i = P (Y ⩾ r) ,
i=r i
where Y is a binomial (k + r, p) random variable.

V. Kumar, DDU Gorakhpur University, Gorakhpur B.Sc.-I : Negative Binomial Distribution Feb. 2023 5 / 16
 Negative Binomial Geometric PMF MGF CGF S & K Limiting form Recurrence

PMF: Alternative form

     
x+r−1 r x n n
p(x) = p q since =
x r n−r

(x + r − 1)(x + r − 2)......(r + 1)r r x

= p q
x!
x
(−1) (−r)(−r − 1).....(−r − x + 2)(−r − x + 1) r x
= p q
   x!
−r r x −r r
= (−1)x p q = p (−q)x
x x
 
 −r r
p (−q)x , 0, 1, 2....
p(x) = x
0, otherwise


V. Kumar, DDU Gorakhpur University, Gorakhpur B.Sc.-I : Negative Binomial Distribution Feb. 2023 6 / 16
 Negative Binomial Geometric PMF MGF CGF S & K Limiting form Recurrence

PMF: Alternative form contd...

 
 −r r
p (−q)x , 0, 1, 2....
p(x) = x
0, otherwise


If
1 P
p= and q = so that Q − P = 1, ( p + q = 1)
Q Q
Then  
 −r
 x
Q−r − QP
, x = 0, 1, 2.....
p(x) = x
0, otherwise


This is the general term in the negative binomial expansion (Q − P )−r .

V. Kumar, DDU Gorakhpur University, Gorakhpur B.Sc.-I : Negative Binomial Distribution Feb. 2023 7 / 16
 Negative Binomial Geometric PMF MGF CGF S & K Limiting form Recurrence

MGF:

∞
X
MX (t) = E(etX ) = etx p(x)
x=0
∞    tx x
X −r Pe
= Q−r = (Q − P et )−r
x Q
x=0
 
′ d h
−r−1
i
µ1 = M (t) = −r(P et )(Q − P et ) = rP
dt t=0 r=0
 
′ d
µ2 = M (t)
dt2 t=0
= (rP et (Q − P et )−r−1 − (−r − 1)rP et (Q.P et )−r−2 (−P et ))t=0
= rP + r(r + 1)P 2
∴ µ2 = µ′2 − (µ′1 )2 = rP Q

V. Kumar, DDU Gorakhpur University, Gorakhpur B.Sc.-I : Negative Binomial Distribution Feb. 2023 8 / 16
 Negative Binomial Geometric PMF MGF CGF S & K Limiting form Recurrence

CGF:

KX (t) = logMX (t) = −r log(Q − P et )

t2 t3
  
= −r log Q − P 1 + t + + + . . .
2! 3!
2 t3
  
t
= −r log 1 − P t + + + . . .
2! 3!
2 t3 t4 P2 t2 t3 t4
    
t
= −r −P t + + + + ... − t + + + + ..
2! 3! 4! 2 2! 3! 4!
3 2 3 4 4 2 3 t4
   
P t t t P t t
− t + + + + .. − t + + + + ... + . . .
3 2! 3! 4! 4 2! 3! 4!
M ean = k1 = coeff. of t in KX (t) = rP
t2
µ2 = k2 = coeff. of in KX (t) = rP (1 + P ) = rP Q
2!

V. Kumar, DDU Gorakhpur University, Gorakhpur B.Sc.-I : Negative Binomial Distribution Feb. 2023 9 / 16
 Negative Binomial Geometric PMF MGF CGF S & K Limiting form Recurrence

β1 and β2 :
t3
µ3 = k3 = coeff. of in KX (t) = nP (1 + P )(1 + 6P + 6P 2 )
3!
= rP Q (Q + P )
µ4 = k4 + 3k2 2 = rP Q[1 + 3P Q(r + 2)]
1 q
Since Q = p and P = qQ = p we have

rq rq
Mean = ; variance = 2
p p
rq(1 + q) rq[p2 + 3q(r + 2)]
µ3 = 3
; µ4 =
p p4
(1 + q)2 p2 + 3q(r + 2)


µ3 2 µ4
β1 = 3 = ; β2 = 2 =
µ2 rq µ2 rq
p (1 + q) 2
p + 6q
γ1 = β1 = √ ; γ2 = β2 − 3 =
rq rq

V. Kumar, DDU Gorakhpur University, Gorakhpur B.Sc.-I : Negative Binomial Distribution Feb. 2023 10 / 16
 Negative Binomial Geometric PMF MGF CGF S & K Limiting form Recurrence

NB tends to Poisson :
Negative binomial distribution tends to Poisson distribution as p → 0, such that rP = λ
(finite).Proceeding to the limits ,we get
     x
x+r−1 r x x+r−1 −r P
lim p (x) = lim p q = lim Q
r→∞ r→∞ r−1 r→∞ r−1 Q
 x
(x + r − 1)(x + r − 2)....(r + 1)r −r P
= lim (1 + P )
r→∞ x! 1+P
      x 
1 x−1 1 x −r P
= lim 1+ ... 1 + .1.r (1 + P )
r→∞ x! r r 1+P
  x 
1 rP
= lim (1 + P )−r
x! r→∞ 1+P
−r
x λ −x λx −λ e−λ λx
  
λ λ
= lim 1 + lim 1 + = e .1 =
x! r→∞ r r→∞ r x! x!
which is the PMF of Poisson distribution with parameter λ.
V. Kumar, DDU Gorakhpur University, Gorakhpur B.Sc.-I : Negative Binomial Distribution Feb. 2023 11 / 16
 Negative Binomial Geometric PMF MGF CGF S & K Limiting form Recurrence

Recurrence Relation:

Prove that the recurrence formula for negative binomial distribution is:
x+r
f (x + 1; r, p) = q f (x; r, p)
x+1
Proof: We have  
x+r−1
f (x; r, p) = pr q x
r−1
 
x+r
f (x + 1; r, p) = pr q x+1
r−1
f (x + 1; r, p) (x + r)! (r − 1)! x! x+r
= q= q
f (x; r, p) (r − 1)! (x + 1)! (x + r − 1)! x+1
Hence,
x+r
f (x + 1; r, p) = q f (x; r, p)
x+1

V. Kumar, DDU Gorakhpur University, Gorakhpur B.Sc.-I : Negative Binomial Distribution Feb. 2023 12 / 16
 Negative Binomial Geometric PMF Moments MGF Memoryless Property

Geometric distribution:

Suppose we have a series of independent trials or repetitions and on each repetition or trial the
probability of success p remains the same. Then the probability that there are x failures preceding
the first success is given by q x p.
A rv X is said to have a geometric distribution if it assumes only non-negative values and its
probability mass function is given by
 x
q p ; x = 0, 1, 2, . . . ; 0 < p ⩽ 1.
p (x) = P (X = x) =
0; otherwise

Since the various probabilities for x = 0, 1, 2, . . . are the various terms of geometric
progression, hence the name geometric distribution.
Clearly, assignment of probabilities is permissible, since
∞ ∞
q x p =p 1 + q + q 2 + . . . = p(1 − q)−1 = 1
X X

P (X = x) =
x=0 x=0

V. Kumar, DDU Gorakhpur University, Gorakhpur B.Sc.-I : Negative Binomial Distribution Feb. 2023 13 / 16
 Negative Binomial Geometric PMF Moments MGF Memoryless Property

Moments:

∞ ∞ ∞
pq q
x q x−1 =pq(1 − q)−2 =
X X X
µ′1 = x p(x) = x q x p =pq =
p2 p
x=0 x=1 x=1

V (X) = E(X 2 ) − (E(X)) = E (X(X − 1) + X) − (E(X))2

2

= E (X(X − 1)) + E(X) − (E(X))2

∞
X x(x − 1) 2q 2
E (X(X − 1)) = 2pq 2
q x−2 = 2pq 2 (1 − q)−3 =
2 p2
x=2

q 2q 2 q 2 qp + q 2 q
V (X) = + 2 − 2 = 2
= 2
p p p p p

V. Kumar, DDU Gorakhpur University, Gorakhpur B.Sc.-I : Negative Binomial Distribution Feb. 2023 14 / 16
 Negative Binomial Geometric PMF Moments MGF Memoryless Property

MGF:

∞ ∞

X X
x
−1
MX (t) = E etX = etx p(x) =p qet =p 1 − qet
x=0 x=0
   
d d
−1
µ′1 = MX (t) = p 1 − qet
dt t=0 dt t=0
h
−2
i q
p qet 1 − qet = pq(1 − q)−2 =


t=0 p
 2
q 2q 2

′ d
µ2 = M X (t) = + 2
dt2 t=0 p p
2
µ2 = µ′2 − µ′1
q 2q 2 q 2 qp + q 2 q
= + 2 − 2 = 2
= 2
p p p p p

V. Kumar, DDU Gorakhpur University, Gorakhpur B.Sc.-I : Negative Binomial Distribution Feb. 2023 15 / 16
 Negative Binomial Geometric PMF Moments MGF Memoryless Property

Memoryless Property: :

For nonnegative integers k and m,

P (X > m + k and X > m)

P (X > m + k|X > m) =
P (X > m)
P (X > m + k)
= = P (X > k)
P (X > m)

Thus, the probability of observing an additional k failures, given the fact that m failures have
already been observed, is the same as the probability of observing k failures at the start of the
sequence.
That is, geometric distribution forgets(’lacks memory’) what has occurred earlier.

Remark: Let the two independent random variables Xl and X2 have the same geometric
distribution. Show that the conditional distribution of X1 |(X1 + X2 = n) is discrete uniform.

V. Kumar, DDU Gorakhpur University, Gorakhpur B.Sc.-I : Negative Binomial Distribution Feb. 2023 16 / 16