STAT301: Cheat Sheet

• Algebra
a+z×b−a
(i) b =z (ii) a(b + c) = a × b + a × c.
1
(iii) √
ab
= √a × √1b .
1

√
(iv) √a = a.
a
(v) a < b means a is less than b. a > b means a is bigger than b. a ≤ b means that a is less than or
the same as b.
(a+b)
– The point that cuts the interval [a, b] in half is 2 .

• Probability The chance of a certain event happening.

Example: Out of 44 calves, 12 weighed less than 90 pounds. The probability of randomly picking a
calf from this group which weighs less than 90 pounds is 12/44.

• Types of studies

– Observational Record data on individuals without attempting an intervention.

– Experimental Deliberatly impose a treatment on individuals. Usually this is done in a random-
ized fashion, where some are given a treatment and others a placebo.

• Confounding When a observed factor and unobserved factor are mixed-up, making it impossible to
decide what is influencing the response.

• Types of variable Numerical discrete, numerical continuous and categorical.

Figure 1: Shapes of distribution: Symmetric, Uniform (thick/heavy tailed), Left Skewed, Right Skewed
and corresponding QQplots

Data Analysis
• Checking for normality

– Use QQplot, see Figure 1.

1
 – A cruder method is to use the 68-95-99.7% rule (check to see whethe data is with one, two and
three sample sd of the sample mean).
• Measures of center given data X1 , . . . , Xn
(i) (Sample) mean X̄ = n1 ni=1 Xi .
P

Example: Average of 1, 1, 1, 3, 4, 5, 6 is x̄ = 3
(ii) Median is the point which cuts the data in half.
Example: Median of 1, 1, 1, 3, 4, 5, 6 is 3. Median of 1, 1, 3, 4, 5, 6 is 3.5.
• Measure of spread of given data X1 , . . . , Xn .
q
1 Pn 2
(i) (Sample) standard deviation s = n−1 i=1 (Xi − X̄) .
Example: The standard deviation of 1, 1, 1, 3, 4, 5, 6 is s = 2.08.
(iii) Quartiles and Interquartile range The first quartile cuts the first half of the data in half and the
second quartile cuts the top part of the data in half. IQR = 3rd quartile - 1st quartile.
Example: The first and third quartile of 1, 1, 1, 3, 4, 5, 6 is 1 and 5 respectively.
• Linear transformation Suppose the data X1 , . . . , Xn has sample mean X̄ and sample standard devia-
tion sX . We make a linear transformation of the data set using the transformation Yi = a + bXi . The
sample mean and standard deviation of the new data set is Ȳ = a + bX̄ and sY = |b|sX respectively.
Example: 0.5, 1.5, 2, 3.2, 3.8 has mean 2.2 and standard deviation 1.3. We transform it using Y =
12X. The new data set is 6.0, 18.0, 24.0, 38.4, 45.6 which has mean 12 × 2.2 and standard deviation
1.3 × 12.
• Z-score calculations Suppose X is a random variable with mean µ and standard deviation σ, the
z-transform is Z = X−µ
σ . The mean and standard deviation of the z-transform is zero and one.
The z-transform tells us how many standard deviations an observation, X, is from the mean µ.

Normal distribution
• Normal calculations. Note the normal distribution is (i) symmetric about the mean (ii) total area is
one (iii) the y-axis is positive.
– Question: Suppose the random variable X is known to come from a normal distribution with
mean 5 and standard deviation 2 N (5, 2). What is the chance X will be less than 6?
– Answer: Make z-transform z = 6−5 2 = 0.5 then look up 0.5 (from outside into the z-tables) to
give P (X ≤ 6) = P (Z ≤ 0.5) = 0.69:

=
• Question: Suppose that X is known to come from a normal distribution with mean 5 and standard
deviation 2 N (5, 2). If an observation X is in the 85th percentile what is X?
• Answer: Look up 0.85 (from inside to outside) the table, which corresponds to 1.04, so X = 5 +
1.04 × 2 = 7.08.
• Rule of Thumb: If data is normally distributed, the roughly speaking 68% of the data lies within
one standard deviation of the mean, 95% of the data lies within two standard deviations of the mean
and 99.8% of the data lies within 3 standard deviations of the mean.

2
 Figure 2: The distribution of averages

The sample mean

• The sample mean Suppose a random sample X1 , . . . , Xn is drawn from a population, where the
mean is µ and the standardPdeviation is σ. The average, usually called the sample mean, X̄ =
1 1 n
n (X1 + X2 + . . . + Xn ) = n i=1 Xi is an estimator of the sample mean.

• Mean and standard error of the sample mean

– The mean of the sample mean (the average of the average) is µ.
√
– The standard error (variability) of the sample mean is σ/ n. The standard error informs us
how variable the estimator is. The smaller the standard error, the less variable it will be.
– Example: A population has mean µ = 67 and standard deviation σ = 3.8. A sample of 5
is drawn and the average is taken. The average will change from sample to sample, but it is
estimating the population mean. The mean of the average, X̄, (the average of the average) is
again µ = 67 (it is estimating this value, so it unbiased) and the standard error of the average,
√ 3.8
X̄, is s.e = σ/ n = √ 5
.

• The distribution of the sample mean

– Normal data If the distribution of the population is normal (examples include heights of one
gender) then the distribution of the sample mean (no matter how big or small the sample size)
will be normal.
Example 1: Female heights are normally distributed with mean 64 inches and standard deviation
2.5
2.5 inches (N (64, 2.5)). A sample of size three is taken the average is N (64, √ 3
).
Example 2: Female heights are normally distributed with mean 64 inches and standard deviation
2.5 inches (N (64, 2.5)). A sample of size 50 is taken the average is N (64, √2.5
50
).
– Non-normal data If the distribution of the population is not normal (examples include the
number of M&Ms in a bag) then the distribution of the sample mean will be close to normal
if the sample size is sufficiently large. How large is large depends on how close to normal the
original distribution.
Example 1: The mean number of M&Ms in a bag is µ = 13.54 with standard deviation σ√ = 4.64.
The average in 5 bags of M&Ms will have mean µ = 13.54 and standard error se=4.26/ 5, but
it will NOT be normally distributed because the original data is not normal.
Example 2: The mean number of M&Ms in a bag is µ = 13.54 with standard deviation
√ σ = 4.64.
The average in 5 bags of M&Ms will be close to normal with N (13.54, 4.26/ 40).
• If the sample mean is close to normal we can use all the usual normal calculations (using the mean
and standard error) to calculate probabilities.

3
Inference for the sample mean
• Confidence Intervals A confidence interval is an interval where we believe with C% confidence the
population mean lies. Typically C = 95%, 99%, 90%. To construct a confidence interval using the
sample mean X̄ (which is evaluated from the data) we need to be sure that the sample mean is
normally distributed (either by normality of the data or the sample size being large enough for the
CLT to kick in). We consider the two cases, which depends on whether the population standard
deviation in known or not.
(i) Known population standard deviation σ If for some reason the population standard
deviation
h is known
i but the population mean is unknown the the 95% CI for the mean is
σ
X̄ ± 1.96 × √n (we look up 2.5% in the z-tables to get 1.96).
(ii) unknown population standard deviation σ If the population standard deviation is unknown
then we need to estimate it from the data. If the sample size is n, we replace the normal
distribution
h with the t-distribution
i with (n − 1) degrees of freedom. The 95% CI for the mean
s
is X̄ ± tn−1 (2.5%) × √n (remember we need to look up 2.5% each side). As the sample size
grows the difference between the normal and the t-distribution becomes less.
Example The sample size is 30, the sample mean is 0.5 and sample standard deviation s = 4,
√
the 95% CI is [0.5 ± 2.04 × 4/ 30].
• Margin of Error This is half the length of the confidence interval.
Example The margin of error of 95% CI [3, 8] is MoE = (8 − 3)/2.
• Formula for Margin of Error If the population standard deviation is known, then the MoE for a 95%
CI is 1.96 × √σn . We can use this to find the minimum sample size to obtain a given margin of error:
n = (1.96 × σ/M oE)2 . Notes:
– The larger the standard deviation σ the larger the sample size we will need.
– If the standard deviation is unknown then bounds are given say, its somewhere between σ1 to
σ2 . Use the largest standard deviation to get the smallest margin of error.
– To decrease the margin of error from m to m/P you need to increase the sample size by a factor
P 2.
• Testing the mean Depending on what the alternative of interest is, there are three different possible
test set-ups. To reduce algebra we will assume the mean under investigation is 5.
– H0 : µ = 5 against HA : µ 6= 5.
– H0 : µ ≤ 5 against HA : µ > 5.
– H0 : µ ≥ 5 against HA : µ < 5.
Which hypothesis you use depends on the alternative that you want to ‘prove’.
Example: The mean height of females 30 years ago was known to be 63 inches. It is believed that
female heights have increased over the past 30 years, what is the hypothesis of interest? Answer
H0 µ ≤ 63 against HA : µ > 63.
• Let us suppose that X1 , . . . , Xn (these are numbers) is a random sample of size n, drawn from
a population with mean µ (this is what we are investigating) and standard deviation σ. We will
assume that the sample size is large enough such that the sample mean is normally distributed with
√
mean µ and standard error σ/ n.
If the population standard deviation is unknown and is instead estimated from the data, then in all
the calculation use a t-distribution with n − 1-degrees of freedom rather than the standard normal
distribution.

4
• Calculating the p-value The p-value is always calculated under the null. This means determining the
chance of the observations if the null were true (how viable is the null)?
Example 1: We test the hypothesis H0 : µ = 5 against HA : µ 6= 5. We collect a random sample of
size 30, the sample mean based on this sample is X̄ = 6 and the sample standard deviation is s = 3.
X̄−µ 6−5
The t-transform is t = √
s/ n
= √
3/ 30
= 1.825. To calculate the p-value:

1 Calculate the smallest area under the plot, in this case it is the area to the RIGHT of 1.825.
Using t-tables with 29df, we see that it is between 2.5-5%.
2. The p-value for the two-sided test, is two times this area, which is between 5-10%.

Example 2: We test the hypothesis H0 : µ ≤ 5 against HA : µ > 5. We collect a random sample of

size 30, the sample mean based on this sample is X̄ = 6 and the sample standard deviation is s = 3.
X̄−µ 6−5
The t-transform is t = √
s/ n
= √
3/ 30
= 1.825. To calculate the p-value:

1 Check to see the direction of the alternative. Since HA : µ > 5, the alternative is pointing
RIGHT.
2. The p-value for this one-sided test, is the area to the RIGHT of t = 1.825. From tables this
area, is between 5-10%.
3. For the one-sided test the p-value is this area, which is between 5-10%.

Example 3: We test the hypothesis H0 : µ ≥ 5 against HA : µ < 5. We collect a random sample of

size 30, the sample mean based on this sample is X̄ = 6 and the sample standard deviation is s = 3.
X̄−µ 6−5
The t-transform is t = √
s/ n
= √
3/ 30
= 1.825. To calculate the p-value:

1 Check to see the direction of the alternative. Since HA : µ < 5, the alternative is pointing
LEFT.
2. The p-value for this one-sided test, is the area to the LEFT of t = 1.825. From tables this area,
is between 90-95%.
3. For the one-sided test the p-value is this area, which is between 90-95%.

• The decision process The decision process is made at the α (typically 5%) significance level. We
reject the null and say there is evidence to suggest the alternative is true (or equivalently there is
evidence to reject the null), if the p-value is less than α%.
α is often called the type I error (or significance level). The larger α the more likely we are to falsely
reject the null when the null is true.
Example In a tomato packing plant, the mean weight of tomato boxes is tested at the α% significance
level, every hour. If the machine is working correctly for every 100 tests, on average we will falsely
reject (determine the machine faulty) α times.

• Confidence intervals and p-values The (100 − α)% (eg. 95%) confidence interval and a test done at
the α% (α%) significance level are connected in the sense that bounds for p-values can be deduced
from the confidence interval (this is because the length of the confidence interval and the non-rejection
region are the same).
Example The 95% CI for the mean is [0.5, 4]. The sample mean is X̄ = (4 + 0.5)/2 = 2.25. Using
this we can deduce the following:

1. Two-sided tests

5
 (a) We test the hypothesis H0 : µ = 0 against HA : µ 6= 0. Since 0 is not in the interval [0.5, 4]
the p-value for this two sided test is less than 5%. This means the smallest area is less than
2.5% and is the area to the RIGHT of X̄ = 2.25 (centered about zero).
(b) We test the hypothesis H0 : µ = 1 against HA : µ 6= 1. 1 is inside the interval [0.5, 4], thus
it is plausible that the mean is 1. The p-value for this two-sided test is greater than 5%.
This means that p-value for the smallest area is greater than 2.5% and is the area to the
RIGHT of X̄ = 2.25 (centered about one).
2. One-sided test pointing RIGHT
We test the hypothesis H0 : µ ≤ 0 against HA : µ > 0. The p-value is the area to the RIGHT
of X̄ = 2.25 (centered about zero), which we have shown in (a) is LESS than 2.5%.
We test the hypothesis H0 : µ ≤ 1 against HA : µ > 1. The p-value is the area to the RIGHT
of X̄ = 2.25 (centered about one), which we have shown in (b) is GREATER than 2.5%.
3. One-sided test pointing LEFT
We test the hypothesis H0 : µ ≥ 0 against HA : µ < 0. The p-value is the area to the LEFT of
X̄ = 2.25 (centered about zero), which we have shown in (a) is GREATER than 97.5%.
We test the hypothesis H0 : µ ≥ 1 against HA : µ < 1. The p-value is the area to the LEFT
of X̄ = 2.25 (centered about one), which we have shown in (b) is less than 97.5%. But since
X̄ = 2.25 is on the right of 1 it has to be greater than 50%. Thus the p-value is between 50 to
97.5%.
• Type I, Type II errors and Power Whenever we do a test we always base the result on a pre-set sig-
nificance level (typically 5% usually denoted as α). These are features about the statistical procedure
and not the data itself.
Example, suppose we test for gestational diabetes and use the hypothesis H0 : µ ≤ 140 against
HA : µ > 140.
– Type I error is the same as the significance level =α% and is pre-set by us. What it means:
If we set α = 5%, then the proportion of healthy women with µ = 140, that we would falsely
diagnose as having gestational diabetes is 5%.
– Type II error Given an alternative of interest the type II error is the chance of rejecting the
null when actually that alternative is true. What it means: If someone is said to have severe
diabetes when their mean is 148 then the type II error is the probability our test does not detect
her. It can be calculated using: We see that when the sample size is 4 and the test is done at

the 5% level the chance of this happening is 1 − 0.9907 = 0.93%.

– Power The power is the chance of rejecting null when the alternative is true. Using the above
plot we see the power for the alternative µ = 148 is 99.07%.
• Increasing the type I error increases power. Increasing sample size increases power.

6
Comparing populations
One of the most common statistical methods is when samples are collected from two different populations
and compared (for example does a drug work verses a placebo). Hence there are two different data sets.
How we analyze the data depends on how the data was collected.
Matched paired t-procedure:

• This a method for analyzing two data-set where there appears to be a clear matching/pairing in the
data. If there is, a difference between the data is take and a one sample method (as described above)
is applied to these differences (note because there is matching the size of the two samples have to be
the same).
Example: Nutritionist want to investigate where red wine increases polyphenol levels in healthy
males. To do a random sample of of 15 healthy males were taken (before the start of the experiment
none drank red wine on a regular basis). Their polyphenol level was take before they started the
experiment, their polyphenol level was measure. For two weeks they drank one glass of red wine each
night. At the end of two weeks their level was taken again. The results of the test H0 : µA − µB ≤ 0
against HA : µA − µB > 0 is give below. We see from the output that there is evidence to suggest

4.3
that drinking red wine increases polyphenol levels. Observe the t-value = 0.79 = 5.4. The 95% CI
for the difference in the means [4.3 ± 2.14 × 0.79] (where 2.14 was found using the t-distribution with
14df with 2.5%).

The independent (two) sample t-test:

• If the two samples are take from two populations and there is no dependence between the two samples
(no relationship), then an independent (two) sample procedure must be done. Because there is no
matching between the two samples the sample sizes can be different.
Example: It is believed that male students at A&M are taller than female students at A&M. We do
this by comparing their population means. The hypothesis of interest is H0 : µM − µF ≤ 0 against
HA : µM − µF > 0. The results of the test is given below. Note t = 0.466−0
0.064 = 7.27. The same output
can be used to deduce the 95% confidence interval (if the 2.5% value is given).

7
Proportions
In this section we consider reponses which are ‘binary’, eg. Yes or No. Guess or not Guess, Vote for
Candidate B not vote for candidate B. We take a random sample of ‘individuals’ and use this to estimate
the proportion of the population who say ‘Yes’, Guessed or vote for candidate A.

• One sample proportions Suppose the proportion of successes (this can be anything) ‘population’ is
p and a random sample of size n is drawn from the sample. The number of success in a sample of
size n follows what is known as a Binomial distribution Bin(p, n). The Binomial is NOT the same
as a normal distribution, however as the sample size n grows the ‘closer’ it gets to normality (this is
the central limit theorem), the size of the sample depends on how close to p is to half (the further p
is from 1/2 the more skewed the distribution and the larger a sample size is required for the central
limit to set in). We can use these results to make tests on proportions and construct confidence
intervals (do not mix p and p̂ with p-value!).
Example Out of a random sample of 140 people 90 said they supported Gay Marriage. Based on this
data, is there any evidence that the majority of the population support Gay Marriage?
The hypothesis of interest in this case is H0 : p ≤ 0.5 against HA : p > 0.5. If the null were true
the distribution of numbers out of 140 is given in the binomial distribution below. We see from both

the Statcrunch crunch test and the binomial calculator that the p-value is extremely small (less than
0.4%), thus there is strong evidence against null, suggesting that majority do support Gay-marriage.
The slight discrepency is because the Statcunch output uses a normal approximation, which is cal-
culated by making the z-transform
0.6428 − 0.5
z= = 3.38
0.0422
and using the z-tables to calculate the area to the RIGHT of 3.38.

8
 The 95% confidence interval for the location of the mean is [0.6428 ± 1.96 × 0.0405] = [0.56, 0.72],
which tells us that the with 95% confidence the proportion that do support Gay marriage lies in this
interval.
q
• Margin of Errors and Sample sizes The standard error for proportions p(1−p) n . Thus the margin of
error (for a 95% confidence interval) and sample size for a given margin of error is
r
p(1 − p) 1.962 p(1 − p)
M oE = 1.96 and sample size n = .
n M oE 2
Of course, p is unknown. Thus we need to choose the p which maximises the variability. If p is
completely unknown then choose p = 1/2 (since p(1 − p) is maximised when p = 1/2). If some
bounds on p are given, say is it is known that p is in a given range, then choose the p in the range
which is closest to 1/2 Eg. if it is known that p lies in the interval [0, 0.3], then use p = 0.3 in the
sample size calculation. A plot of standard errors is given in Figure 3.

Figure 3: Standard error for a given sample size