1 Probability: 1.1 Normal Distribution

1
Probability
1.1
Normal distribution
The hemoglobin values of an undoped athlete1 are known to be normally

distributed with = 148 and variance 2 = 85. Let X denote a measured
value of hemoglobin, i.e. X N (148, 85).
Calculate the probability that the hemoglobin value of an individual
exceeds 166. Derive the general formula:
Let X N (, 2 ), find the distribution of the random variable
Y = aX + b, where a > 0?
Hint: Start by expressing the distribution function, then derive
the density. Can this density be written as a density of a normal
covariate?
FY (y) = P (Y y) = P (aX + b y) = P (aX y b)
yb
yb
) = FX (
)
= P (X
a
a
yb
1
fX (
)
fY (y) =
a
a
Since X is normal,
(
)
2
[ yb
]
1
fY (y) =
exp a 2
2
2
a 2

1
[y (b + a)]2
exp
= p
2(a )2
2(a )2
We get Y N (a+b, (a)2 ). Linear transformation of a normal
variable remains normal.
1
In the hope for better performance, the athletes withdraw blood and then reinfuse it
before an important competition. In this way, the number of red blood cells is artificially
increased, improving the endurance of the athlete. The method is known as blood doping
and is highly dangerous and prohibited. Since no foreign substances can be found in the
body, there is no way to directly detect the method. Statistics is used to determine unusual
values of hemoglobin.
What do we have to take as a in b, if we want Y to be a standardized normal variable?

a should be equal to 1 , b should be equal to . The formula
Y = X
transforms X to the standardized normal scale. We can
then use the tables (or the appropriate numerical method) to get
the probabilities.
= 166148
= 1.95.
In our case, X = 166 and therefore Y = X148
85
85
Tables for the standardized normal variable (computer can be used
instead) give P (X 166) = P (Y 1.95) = 0.974, the probability
therefore equals P (X > 166) = 0.026.
Calculate (symmetrical) limits, that are exceeded with probability less
than 0.01 by an undoped athlete.
Let Y be a standardized normal variable. We wish to calculate the
limits that encompass 99% of the values. Since we want to have symmetrical limits, the tails are left with probability 0.005 each. From the
tables, we learn that P (Y 2.58) = 0.995, i.e. the limit value of the
standardized normal variable equals 2.58.
, therefore
Y = X148
85
X 148
X 148
2.58) + P (
> 2.58)
0.01 = P (
85
85
= P (X 148 2.58 85) + P (X > 148 + 2.58 85)

= P (X 124.2) + P (X > 171.8)
Consider the calculated limits and let each athlete be tested 10 times
in a year. What is the probability that he exceeds the limits at least
once (we assume the the athlete is tested within intervals of time that
are long enough to ensure no correlation between test results)?
Let U be a Bernoulli variable U Ber(0.01), where {U = 1} =
{the value is out of limits}. We have 10 independent realizations of
this random variable, Ui , i = 1, . . . , 10, with P (Ui = 1) = 0.01 for each
one of them. Since they are independent, P (U1 = 0, U2 = 0, . . . , U10 =
0) = {P (U1 = 0)}10 . The probability that the limits are not exceeded
in 10 measurements equals 0.9910 , the probability that they are ex2
ceeded at least once, equals P = 1 0.9910 = 0.096.

Let Y N (0, 1). What is the distribution of Y 2 ?
Hint: The density of the gamma distribution is given by fT (t) =
t1 et
.
()
FZ (z) = P (Z z) = P (Y 2 z) = P ( z Y z)
= FY ( z) FY ( z)
1
1
fZ (z) = fY ( z) + fY ( z)
2 z
2 z
1
= [fY ( z) + fY ( z)]
2 z
1
1
[ez/2 + ez/2 ] =
ez/2
=
2 z 2
z 2
1 t
The density of the gamma distribution is given by fT (t) = t ()e .
If = 12 and = 12 ( ( 21 ) = ), we get exactly the formula above.

Therefore, Y 2 ( 21 , 12 ) (this is also equal to the 21 distribution).
Studies have shown, that the hemoglobin of a biathlon athlete is distributed as N (150, 80) in an out-of-competition phase and as N (146, 80)
in the competition phase. Both phases in this sport last for approximately half a year. We wish to know the distribution of hemoglobin if
we do not know when the athlete was tested. Is this distribution still
normal?
S
T
P
Hint: P (A) = ni=1 P (A|Bi )P (Bi ), if P ( ni=1 Bi ) = 0 and P ( ni=1 Bi ) =
1.
Define a Bernoulli variable Y , that denotes the phase (0=out-of.competition,
1=competition, probability of each outcome equals 0.5). We know the
conditional distributions:
Z|Y = 0 N (150, 80), Z|Y = 1 N (146, 80). The distribution of Z
equals (we use the hint with B1 = {Y = 0} in B2 = {Y = 1}, and
plug-in probabilities instead of densities)

fZ (z) = fZ|Y =0 (z)P (Y = 0) + fZ|Y =1 (z)P (Y = 1)
1
1
= fZ|Y =0 (z) + fZ|Y =1 (z)
2
2
(z146)2
168(z146)
1
=
e 280 [1 + e 280 ]
2 280
This variable is not normally distributed.
Understanding the ideas with R:
Generate 10000 realization of a normal variable X N (148, 85) (rnorm).
Plot a histogram (hist), calculate the proportion of values above 166
(sum(x>166)/10000).
Use the function qnorm to find the limits, that an athlete exceeds with
probability 0.01. Compare with the proportion in your example.
Transform the values of X to get a standardized normal variable
(y=(x-148)/sqrt(85)). Check graphically with a histogram.
Generate 10 values for 10000 individuals. Calculate the proportion of
individuals, that have at least one value outside the interval [124.2, 171.8].
Plot a histogram of values X 2 , compare with the histogram of values
generated with functions rgamma or rchisq.
Plot the distribution of a random variable described in the last part of
the exercise (let the means in the two phases differ more to make sure
that the overall distribution is not normal).
1.2
Generating random variables using the uniform

distribution
A generator of (pseudo-)random variables results in a number of values xi ,

distributed as X U [0, 1].
How could you use this generator to get 10 Bernoulli distributed variables Y , with P (Y = 1) = 0.1?
We generate 10 values. For example:
> set.seed(4)
> runif(10)
[1] 0.585800305 0.008945796 0.293739612 0.277374958
[5] 0.813574215 0.260427771 0.724405893 0.906092151
[9] 0.949040221 0.073144469
The values below 0.1 get the value 1, the rest get the value 0. Therefore:
> set.seed(4)
> (runif(10)<0.1)*1
[1] 0 1 0 0 0 0 0 0 0 1
Say that we again have 10 units, but we wish to assign them different
probabilities to be drawn: we want to draw the first five units with
the probability 0.3, and the second five with the probablity 0.1 (as an
example, say we want to make a random draw, but assign a higher
probability to the women, that constitute half of our sample). How
can we use the same generator for this purpose?
> set.seed(4)
> (runif(10)<c(0.1,0.1,0.1,0.1,0.1,0.3,0.3,0.3,0.3,0.3))*1
[1] 0 1 0 0 0 1 0 0 0 1
Let Z = F (X), where F is the distribution function of the random

variable X.
Sketch the graph of the cumulative distribution function (variable
X on the x-axis, and variable Z on the y-axis.)
What values can be taken by the variable Z?
Between 0 and 1
5
Let X N (0, 1). Which value of X results in Z = 0.5? What is

the probability that Z 0.5?
The probability equals to 0.5
Let X N (0, 1). Which value of X results in Z = 0.975? What
is the probability that Z 0.975?
The probability equals 0.975. The values of Z are the quantiles of
the X distribution.
Express the function FZ (z) for a given F (assume that the inverse
F 1 is defined for all possible values of X).
FZ (z) = P (Z z) = P (FX (X) z) = P (X FX1 (z))
= FX (FX1 (z)) = z
The variable Z is uniformly distributed.
Let U U [0, 1] in X = F 1 (U ). Show that F is the cumulative
distribution function of the variable X.
If U is uniform, then FU (u) = u:
P (X x) = P (F 1 (U ) x) = P (U F (x)) = FU (F (x)) = F (x)
F is therefore the cumulative distribution function of the variable X.
We wish to simulate values from the exponential distribution (f (x) =
ex , za x > 0). How can we simulate them using a generator of
uniform distribution?
We first need to theoretically express the function F :
Zz
FZ (z) =
ex dx
x z
e
0
z
= 1[e
1] = 1 ez
and its inverse:

u = 1 ex
1 u = ex
log(1 u) = x
log(1 u)
x =
If the values u are uniformly distributed, the values x shall represent

the realizations of an exponentially distributed variable X.
How would we simulate values for individuals with with different values
of ?
Same as above, but the values of can differ.
Understanding the ideas with R:
Generate data for 10000 drivers, let 500 of them be drunk and 9500
sober. Let the probability of being involved in a car accident equal 0.3
and 0.003, for a drunk and a sober driver respectively. Calculate the
proportion of the accidents on the simulated data and compare it to
the actual probability of the accident.
Generate data for 100 individuals, so that their age is uniformly distributed between 50 and 80. Check graphically with a histogram.
Say that these individuals have been diagnosed with a lethal disease.
Generate the survival times using an exponential distribution, so that
the older patients have a higher probability of dying earlier.
Hint: for example, = age/100.

1 Probability: 1.1 Normal Distribution

Uploaded by

Copyright:

Available Formats

1 Probability: 1.1 Normal Distribution

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

1 Probability: 1.1 Normal Distribution

Uploaded by

Copyright:

Available Formats

1

The hemoglobin values of an undoped athlete1 are known to be normally

What do we have to take as a in b, if we want Y to be a standardized normal variable?

= P (X 148 2.58 85) + P (X > 148 + 2.58 85)

ceeded at least once, equals P = 1 0.9910 = 0.096.

The density of the gamma distribution is given by fT (t) = t ()e .

If = 12 and = 12 ( ( 21 ) = ), we get exactly the formula above.

plug-in probabilities instead of densities)

Generating random variables using the uniform

A generator of (pseudo-)random variables results in a number of values xi ,

Let Z = F (X), where F is the distribution function of the random

Let X N (0, 1). Which value of X results in Z = 0.5? What is

and its inverse:

If the values u are uniformly distributed, the values x shall represent

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.