Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Scribd
Upload
38 views

Sampling Distributions

The central limit theorem (CLT) states that as sample size increases, the sample mean will be approximately normally distributed around the population mean, regardless of the population distribution. For the CLT to apply, samples must be independent and identically distributed, and the sample size must be large enough (usually n≥30). The CLT allows us to use normal distributions to calculate probabilities related to sample means and proportions.

Uploaded by

Jawad Sandhu
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
38 views

Sampling Distributions

The central limit theorem (CLT) states that as sample size increases, the sample mean will be approximately normally distributed around the population mean, regardless of the population distribution. For the CLT to apply, samples must be independent and identically distributed, and the sample size must be large enough (usually n≥30). The CLT allows us to use normal distributions to calculate probabilities related to sample means and proportions.

Uploaded by

Jawad Sandhu
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 25

Central Limit Theorem

General Idea: Regardless of the population


distribution model, as the sample size increases,
the sample mean tends to be normally distributed
around the population mean, and its standard
deviation shrinks as n increases.

Certain conditions must be met to use the CLT.


● The samples must be independent
● The sample size must be “big enough”
CLT Conditions
Independent Samples Test
● “Randomization”: Each sample should
represent a random sample from the
population, or at least follow the population
distribution.
● “10% Rule”: The sample size must not be
bigger than 10% of the entire population.
Large Enough Sample Size
● Sample size n should be large enough so that
np≥10 and nq≥10
Example: Is CLT appropriate?
It is believed that nearsightedness affects about
8% of all children. 194 incoming children have
their eyesight tested. Can the CLT be used in
this situation?
●
Randomization: We have to assume there isn't some
factor in the region that makes it more likely these kids
have vision problems.
●
10% Rule: The population is “all children” - this is in the
millions. 194 is less than 10% of the population.
●
np=194*.08=15.52, nq=194*.92=176.48
We have to make one assumption when using the CLT in
this situation.
Central Limit Theorem
(Sample Mean)
● X , X , ..., X are n random variables that are
1 2 n
independent and identically distributed with
mean μ and standard deviation σ.
● X = (X +X +...+X )/n is the sample mean
1 2 n

● We can show E(X)=μ and SD(X)=σ/√n


● CLT states: X −μ
N 0,1
σ/ n
as n→∞
Implication of CLT
● We have: X −μ N 0,1
σ/ n
● Which means X N μ , σ2/n
● So the sample mean can be approximated with
a normal random variable with mean μ and
standard deviation σ√n.
Proportions of a Sample
Let's say we have a population with probability
p of a certain characteristic (and q=1-p). We
have a random sample of n from the population.
What is the mean and standard deviation of
the proportion of our sample that has the
characteristic?
● We can use the CLT if n is large enough
● If X is the number of times the characteristic is
found in our sample, p=X/n, our sample
proportion, has mean p and standard deviation
√(pq/n)
Central Limit Theorem
(Sample Sum)
● X , X , ..., X are n random variables that are
1 2 n
independent and identically distributed with
mean μ and standard deviation σ.
● S = X +X +...+X is the sample sum
n 1 2 n
● We can show E(S )=nμ and SD(S )=σ√n
n n

● CLT states: Sn−nμ


N 0,1
σ n
as n→∞
Applications of CLT
It is believed that nearsightedness affects about
8% of all children. 194 incoming children have
their eyesight tested. Can the CLT be used in
this situation?
● X should be approximately normally distributed
have a mean of .08*194=15.52 and SD of
√(.08*.92*194)=3.7787
CLT for Proportions
● How is the proportion of nearsighted children
distributed?
● Divide by n=194: mean is .08, SD is .0195
Applying the 68-95-99.7 Rule
● You can be 68% sure the sample mean is within
1 standard deviation (of the population mean)
● You are 95% sure the sample mean is within 2
standard deviations
● You are 99.7% sure the sample mean is within
3 standard deviations.
● To cover virtually all possibilities, we can go
3 standard deviations from the sample
mean.
Example: Nearsightedness (cont)
With 192 incoming children, what is a
reasonable range of nearsighted children the
school can expect?
● Because 3 standard deviations covers 99.7% of
the data, we use this for 'reasonable'.
● 3 standard deviations is 11.3361.
● 15.52-11.361=4.1839, 15.52+11.361=26.881
● The school should expect between 4 and 27
nearsighted children.
Example: College Retention Rates
Nationally 74% of college freshman continue as
sophomores. A particular school has 486 out of
all 600 freshmen stay. Is this unusual?
● We consider “unusual” anything above or below
3 standard deviations from the mean.
● We expect the retention to be .74*600=444,
with a standard deviation of
√npq=√(600*.74*.26)=10.744
● μ+3σ=444+3*10.744=476.232, so 486 is
“unusual”
Example: Pregnancy Length
Human pregnancies follow a normal distribution
with mean of 268 days and s.d. 11 days. We
study the mean pregnancy length of 70 women
(call this random variable X). What is this
statistic's Expected Value and S.D.?
● It is reasonable to use the CLT (conditions are
met)
● X is large enough to approximate with a normal
distribution, E(X)=268, SD(X)=11/√70=1.314
Example: Dice Game
Awards for a dice game are as follows:
● Roll and Odd number: $0
● Roll a 2 or a 4: $2
● Roll a 6: $26

● E(X)=$5, SD(X)=√[E(X2)-52]=9.43
(We already know how to find these)
Example: Dice Game (cont.)
● If you play the dice game 30 times, what is the
expected value and standard deviation of your
winnings? What is the probability you win at
least $200?
● Let S30=X1+X2+...+X30
● E(S30)=30*$5=$150
● SD(S30)=9.43*√30=51.65
● P(S30≥200)=normalcdf(200, 1000, 150, 51.65)
=.1665
Example: AP Scores
● A Teacher has a class of 68 students taking the
AP Physics test. Assuming they are typical of
the population, the result of whose scores are
given, what is the probability the average score
will be at least 3?
Score 1 2 3 4 5
Probability 21.5% 18.8% 24.7% 21.7% 13.3%
Example: AP Scores (cont.)
Score 1 2 3 4 5
Probability 21.5% 18.8% 24.7% 21.7% 13.3%

● If X1 is the score of a student, E(X1)=2.865,


SD(X1)=1.334
● If X is the average of all 68 students, CLT states
E(X)=2.865 and SD(X)=1.334/√68=.162
● P(X≥3)=normalcdf(3, 5, 2.865,.162)=.202
Example: Restaurant Tips
A waitress earns varying tips with a mean of
$10.90 and sd of $5.60. Assume on the
weekend she gets 60 tips.
● What is the probability she earns $750 or
more?
● Oh the best 5% of weekends, she earns at least
how much $$? (i.e. what does she make less
than on 95% of weekends)?
Example: Restaurant Tips (cont)
● We can use CLT to answer the problems.
● Let S60 be the sum of 60 tips following this
model. E(S60)=60*10.90 = $654,
SD(S60)=5.60*√60=$43.3774
● P(S60≥750)= normalcdf(750, 1000, 654,
43.3774)=.0134
● Upper 5th percentile is invNorm(.95, 654,
43.3774)=$725.35
Sums of Independent Normal r.v.s
●
If you have two random variables, each following a
normal distribution:
●
X~N(μ ,σ2 ) and Y~N(μ ,σ2 )
X X Y Y

●
Then Let W=X+Y
2 2
●
W~N(μ +μ ,σ +σ )
X Y X Y

●
So the sum of two indpt. Normal random variables is
Normal, its mean is the sum of their means and its
variance is the sum of their variances.
●
Note: its sd is NOT the sum of their sds!!
Example: Comparing IQ Scores
IQ Scores are normally distributed. Students at
University A have IQ scores mean 130 and sd
6. At University B the mean is 120 and sd is 9. If
you compare a random student from each
university, what is the probability the IQ score
from the University A student will be higher by 5
points or more?
● IQ Scores from University A ~ μA=130 σA=6
● IQ Scores from University B ~ μB=120 σB=9
Example: IQ Scores (cont)
● Let A and B be the samples from these
populations. Let's define X=A-B, the difference
of their scores.
● E(X)=E(A-B)=130-120=10
● SD(X)=√Var(X)=√(Var(A)+Var(B))
=√(62+92)=10.8167
● So X ~ Normal with μX=10 σX=10.8167
● P(X≥5)=normalcdf(5, 100, 10, 10.8167)
=.678
Example: IQ Scores (cont)
Recall B~Normal with μ =120 σ =9. What is the
B B
probability the average of 3 University B students'
IQs is at least 125?
●
If B , B and B are the IQ scores of 3 univ. B
1 2 3
students, Let B=(B +B +B )/3
1 2 3

●
E(B)=120,
●
SD(B)=SD[ (B1+B2+B3)/3 ]
=SD(B1+B2+B3)/3=√3*9/3=5.1962
●
B~Normal μB=120 σB=5.1962 so
P(B≥125)=normalcdf(125,200,120,5.1962)=.1680
Example: IQ Scores (cont)
Say we want to compare the average of 3
random Univ. A students' IQ scores with the
average of 3 random students from Univ. B.
What is the probability the Univ. A students' avg
score is at least 5 points higher?
● First define A similar to B on the previous page,
and find A~Normal with μA=130 σA=3.4641.
● Define a new X=A-B. We know X will be
normal, so we just need its mean and standard
deviation.
Example: IQ Scores (cont)
● E(X)=E(A-B)=E(A)-E(B)=130-120=10
● SD(X)=SD(A-B)=√Var(A-B) =√[Var(A)+Var(B)]
=√(3.46412+5.19622) =6.245
●
So the probability the A students' avg IQ is at least
5 points higher than the B students is the same as
the probability X≥5.
●
Pr(X≥5)=normalcdf(5, 100, 10, 6.245)

You might also like