Uji Normalitas 1505659669

Using the Normal Distribution
Mark Muldoon
School of Mathematics, University of Manchester
Mark Muldoon, January 30, 2007 Using the normal - p. 1/19

Overview
● Overview
● PDFs on different
temperature scales
Today’s lecture is about uses of the normal distribution,
● Shifting and scaling the
normal
including ways to use standard tables.
● Check: shifting and scaling
● Tables for the normal
● Tables of the upper tail
■ Shifting & scaling: revisited briefly
Properties: changing µ and σ for the normal
● Tables of critical values
● Application: testing extremity
■
● z -Scores
● Approximating the binomial ■ Integrals: without pain
distribution
● Application: designing a study ■ Two things to do with the normal:
(a) testing for extremity;
(b) measuring frequencies and planning studies;

PDFs on different temperature scales
Probability distribs. for hourly

● Overview temperature data
temperature scales
0.3
Prob. Density
normal PDF (ºC)
● Tables of critical values 0.2
● Application: testing extremity PDF (ºF)
● z -Scores
● Approximating the binomial
distribution
● Application: designing a study 0.1
-16 -8 0 8 16 24 30 36 42 48
Temperature (both scales)
Fahrenheit histogram is:

■ shifted to the right (by 32 degrees);
■ both wider and flatter (by factor of (9/5)).

Shifting and scaling the normal
● Overview
The normal distribution has the property that shifted and
temperature scales
scaled versions of it are still normal distributions (though with
normal
different means and variances):
● Tables of critical values 0.4
Two normal distribs. with different means
0.4 Two normal distribs. (µ = 0, σ = 1)
● z -Scores
● Approximating the binomial (µ = 2, σ = 1) with different 0.3
distribution (µ = 0, σ = 1) variances
0.3
● Application: designing a study
0.2
0.2
0.1
(µ = 0, σ = 2)
0.1
-2 2 4 -6 -4 -2 2 4 6

Check: shifting and scaling
● Overview
If x is a normally-distributed random variable with mean µx and
temperature scales
standard deviation σx , then a scaled, translated random
normal
variable y = 2x + 5 has:
(a) a normal distribution centered on µy = 2µx ;
● z -Scores
● Approximating the binomial (b) a normal distribution centered on µy = 2µx + 5 ;
distribution
● Application: designing a study (c) a binomial distribution;
(d) variance σy2 = 4σx2 ;
(e) variance σy2 = 2σx2 + 5.
Answer:

Check: shifting and scaling
● Overview
If x is a normally-distributed random variable with mean µx and
temperature scales
standard deviation σx , then a scaled, translated random
normal
variable y = 2x + 5 has:
(a) a normal distribution centered on µy = 2µx ;
● z -Scores
● Approximating the binomial (b) a normal distribution centered on µy = 2µx + 5 ;
distribution
● Application: designing a study (c) a binomial distribution;
(d) variance σy2 = 4σx2 ;
(e) variance σy2 = 2σx2 + 5.
Answer: Only items (b) and (d) are true.

Tables for the normal
● Overview
The main point about the shifting and scaling that we have
temperature scales
been discussing is that it means that we never have to do any
normal
integrals to use the normal distribution. We will learn to use
two sorts of tables:
● Tables of critical values ■ those giving 1 − Φ(z) where
● z -Scores Z z
1 2
distribution
Φ(z) = √ e−z /2 dz
2π −∞
is the probability of drawing a value ≤ z from a standard
normal distribution;
■ those giving critical values of z for various interesting
probability levels.

Tables of the upper tail
● Overview
A table of 1 − Φ(z) is attached to these notes.
temperature scales
normal
● Check: shifting and scaling 0.4
Relative Frequency

● z -Scores
0.3
distribution
0.2
Φ(z)
0.1 1 − Φ(z)
-3 -2 -1 0 1 z 2 3

Tables of critical values
● Overview
These tables are a sort of inverse of the tables of the upper tail:
temperature scales
they tell you how what z needs to be so that 1 − Φ(z) has a
normal
certain value:
Target
● z -Scores value (%) z
distribution
10 1.28
5 1.64
2.5 1.96
1 2.33
0.5 2.58

Application: testing extremity
● Overview
Acme Salts aims to produce contact lens solutions with tightly
temperature scales
controlled salinity: the 500 ml bottles contain a mean of 5 gm
normal
of NaCl with a standard deviation of 0.05 gm. Bottles are
rejected if their salt content lies outside the band 4.92 to 5.08.
● Tables of critical values Assuming the salinity of bottles to be normally distributed:
● z -Scores
distribution
(a) Out of 100 bottles, how many would you expect to be
● Application: designing a study rejected?
(b) How many rejections would you expect if the acceptable
range narrowed to 4.98–5.03?

Using the standard normal: z-scores
● Overview
As the mean salinity is 5 gm. per bottle and the standard
temperature scales
deviation is 0.05, the scaled, translated random variable z
normal
given by
s − (mean salinity)
z =
● z -Scores
(standard deviation of salinity)
distribution (s − 5)
● Application: designing a study =
0.05
has a standard normal distribution (µ = 0, σ = 1). This is an
example of a z-score.

Salinity: both distributions
Salinity (gm. NaCl per bottle) Salinity z-Score (dimensionless)

8.0 0.4
Prob. Density
Prob. Density
6.0 0.3
4.0 0.2
µ = 5.0 µ = 0.0
2.0 σ = 0.05 0.1 σ = 1.0
4.95 4.9 4.95 5.0 5.05 5.1 5.15 s -3.0 -2.0 -1.0 0.0 1.0 2.0 3.0 z
(µ − 3σ) (µ − σ) µ (µ + σ) (µ + 3σ) (µ − 3σ) (µ − σ) µ (µ + σ) (µ + 3σ)
s < 4.92 s > 5.08 z < -1.6 z > 1.6
Notice that the vertical scale changes in concert with the horizontal scale: this
is so that the area beneath the bell-curve remains equal to 1.0.

Salinity: part (a)
● Overview
■ Allowed range is 4.92 to 5.08
temperature scales
● Shifting and scaling the ■ z-values corresponding to limits are -1.6 (for 4.92) and 1.6
normal
● Check: shifting and scaling (for 5.08).
● Tables of the upper tail ■ z has standard normal distribution:
● z -Scores
P ((s < 4.92) or (s > 5.08) ) =
distribution
● Application: designing a study = P ((z < −1.6) or (z > 1.6))
= P (z < −1.6) + P (z > 1.6)
≈ 0.0548 + 0.0548 = 0.1096
■ Expected number of rejections is
100 × P ( unacceptable s) = 100 × 0.1096 ≈ 11 bottles.

Salinity: part (b)
● Overview
Calculations are similar:
temperature scales
normal P ((s < 4.98) or (s > 5.03)) = P (z < −0.4) or (z > 0.6))
● Tables for the normal = P (z < −0.4) + P (z > 0.6)
≈ 0.3446 + 0.2743 = 0.6189
● z -Scores
distribution
Expected number of rejections is

100 × P (unacceptable s) = 100 × 0.6189 ≈ 62 bottles.

Approximating the binomial distribution
For large numbers of trials the binomial distribution

p is well-approximated by a
normal distribution with µ = pN and σ = p(1 − p)N :
0.4
4 Coins 8 Coins
Relative freq.
Relative freq. 0.25

0.3
0.2 0.15
0.1 0.05
0 1 2 3 4 0 2 4 6 8
Number of Heads Number of Heads
0.2 0.14 Approximation is valid

16 Coins 32 Coins
when both pN and
Relative freq.
Relative freq.
0.15 0.1
0.1
0.06
(1 − p)N exceed 5.
0.05
0.02
0 2 4 6 8 10 12 14 16 0 8 16 24 32
Number of Heads Number of Heads

Application: designing a study
● Overview
Suppose you are interested in studying a trait known from
temperature scales
previous genetic surveys to be carried by 20 people per 100 in
normal
the UK.
(a) What is the probability of finding 220 or more carriers in a
● Application: testing extremity sample of 1000 people?
● z -Scores
distribution
(b) When planning the study, how large a sample would one
need to screen in order to be 99% certain of finding 100
carriers?

Carriers: part (a)
● Overview
Convert “220 carriers in a sample of 1000” into a z-score:
temperature scales
normal
■ As p = 0.2, the expected number of carriers is
µ = 1000p = 200.
■ Standard deviation is
● z -Scores
p √
● Approximating the binomial σ = 1000p(1 − p) = 160 ≈ 12.65.
distribution
■ So the z-score is:
220 − µ 220 − 200
z = = √
σ 160
≈ (20/12.65) ≈ 1.58

From binomial to normal
0.4
0.03
µ = 200 µ = 0.0
Prob. Density
0.3
σ ≈12.65 z ≈ (m - 200)/12.65 σ = 1.0
0.02
0.2
0.01 0.1
170 180 190 200 210 220 230 -3 -2 -1 1 2 3

Number of Carriers z-Score
m > 220 z > 1.58
220 cases or more (in a sample of 1000) is a fairly unlikely outcome:

P (z ≥ 1.58) ≈ 0.0571.

Carriers: part (b)
● Overview
Use a z-score in reverse: find sample size N such that
temperature scales
probability of finding 100 or fewer carriers is < 1%.
normal
■ z-score of -2.33 corresponds to a probability of ≈ 0.0099
■ Want to choose N so large that the z-score corresponding to
● z -Scores
m = 100 is -2.33.
p = (100 − µ)/σ
That is, want to solve −2.33 √ with
distribution ■
µ = pN = 0.2 N and σ = p(1 − p)N = 0.16 N .

Carriers: part (b)
● Overview
Leads to a quadratic in N
temperature scales
normal
■ Larger of the two solution is N ≈ 616: this is the one we
want.
■ Note that if we only wanted 100 carriers in expectation we’d
● z -Scores
need only 100/p ≈ 500 subjects, but then we’d run a 50%
distribution
risk of getting fewer carriers than we wanted.

Uji Normalitas 1505659669

Încărcat de

Informații document

Descriere originală:

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Uji Normalitas 1505659669

Încărcat de

Drepturi de autor:

Formate disponibile

Using the Normal Distribution

Mark Muldoon, January 30, 2007 Using the normal - p. 1/19

Mark Muldoon, January 30, 2007 Using the normal - p. 2/19

Probability distribs. for hourly

Fahrenheit histogram is:

■ both wider and flatter (by factor of (9/5)).

Mark Muldoon, January 30, 2007 Using the normal - p. 3/19

Mark Muldoon, January 30, 2007 Using the normal - p. 4/19

Mark Muldoon, January 30, 2007 Using the normal - p. 5/19

Mark Muldoon, January 30, 2007 Using the normal - p. 5/19

Mark Muldoon, January 30, 2007 Using the normal - p. 6/19

● Tables of the upper tail

Mark Muldoon, January 30, 2007 Using the normal - p. 7/19

Mark Muldoon, January 30, 2007 Using the normal - p. 8/19

Mark Muldoon, January 30, 2007 Using the normal - p. 9/19

Mark Muldoon, January 30, 2007 Using the normal - p. 10/19

Salinity (gm. NaCl per bottle) Salinity z-Score (dimensionless)

s < 4.92 s > 5.08 z < -1.6 z > 1.6

Mark Muldoon, January 30, 2007 Using the normal - p. 11/19

Mark Muldoon, January 30, 2007 Using the normal - p. 12/19

Expected number of rejections is

Mark Muldoon, January 30, 2007 Using the normal - p. 13/19

For large numbers of trials the binomial distribution

Relative freq. 0.25

0.2 0.14 Approximation is valid

Mark Muldoon, January 30, 2007 Using the normal - p. 14/19

Mark Muldoon, January 30, 2007 Using the normal - p. 15/19

Mark Muldoon, January 30, 2007 Using the normal - p. 16/19

170 180 190 200 210 220 230 -3 -2 -1 1 2 3

220 cases or more (in a sample of 1000) is a fairly unlikely outcome:

Mark Muldoon, January 30, 2007 Using the normal - p. 17/19

Mark Muldoon, January 30, 2007 Using the normal - p. 18/19

Mark Muldoon, January 30, 2007 Using the normal - p. 19/19

S-ar putea să vă placă și