Introduction to

Statistical Inference
Statistical Inference: An Introduction

Aim of a statistical inference is to form a conclusion about population

parameter from a sample (statistic) taken from the population.

Population: Complete set of observations about which an investigator

wishes to draw conclusions.

Sample: A subset of population

Statistic: Descriptive index of a sample

**Population is defined in terms of observation rather than people.

 Steps used in statistical
inference
Step 1: Randomly draw a sample from the
population of interest.
Step 2: Use descriptive techniques to
calculate the sample statistic (for example,
X̅ ).
Step 3: Infer the value of μₓ using the
obtained value of our sample statistic (X̅)
Dr. Brown, the research director of a large school district, has read in a test
manual that the national norm for sixth-grade students on a mathematics
achievement test is 85. She wants to know whether students in her district are
performing at the same level. She selects at random 100 students from all the
sixth-grade students in the district and gives the same test to them. Dr. Brown
then calculates their mean score and compares it to the national norm.

If the mean of her group differs from the national norm, what should she
conclude?
Hypothesis Testing

In inference, we use a procedure called hypothesis testing.

Hypothesis: a statement about a population parameter to be
subjected to test and, on the outcome of the test, to be retained or
rejected.
Hypothesis in Dr Brown Example?
Mean of the mathematics achievement test in her school district
(her population) is equal to the mean of the national group.
In inference, the fundamental factor that must be taken into account is that: value of the characteristic we
are studying will vary from sample to sample.

ACTIVITY! Report your CA of any one core paper!

We draw a sample and compute the mean.
If we draw a second sample of the same size the mean would have a different value.
A third is likely to give still another value.
A distribution that will be formed by the means when the sampling is repeated again and again is known
as a sampling distribution.
Sampling Distribution

A sampling distribution is a theoretical relative frequency

distribution of the values of a statistic that would be obtained
by chance from an infinite number of samples of a particular
size drawn from a given population.
To learn what sample values will occur and with what
probability, there must be known rules that connect the
“behavior” of the samples to the population from which they
are drawn.
IT IS NOT THE SAMPLE OBTAINED THAT MATTERS, BUT
THE SYSTEMATIC AND CONSISTENT METHOD WHICH IS
USED TO OBTAIN THE SAMPLE.
Probability Sampling: Sampling method where probability of
inclusion of each of the element of the population within the
sample is known.
Random Sampling: Each element of the population has equal
probability of being selected as the sample.
Random Sampling

A random sample of a given population is a sample so drawn that each possible sample of
that size has an equal probability of being selected from the population.

2 properties of random sample:

1) If multiple random samples are drawn from the same population, the samples
will almost always differ, and therefore their characteristics will vary from
sample to sample.
2) Larger the random samples from a given population, less variation in
characteristics from sample to sample.
Sampling plans

2 sampling plans yield a random sample:

1) SAMPLING WITH REPLACEMENT (An element may

appear more than once in a sample)
2) SAMPLING WITHOUT REPLACEMENT (An element can
appear only once in a sample)
Hindrances in Obtaining a
Completely Random Sample
To obtain a random sample, every element in the population
must have an equal chance of being included in the sample.

In the real world of data, it is often difficult or costly (and

sometimes impossible!) to draw a simple random sample from
the population that we would truly like to study.

SHORTCUT: Casual Sampling (Convenience/ purposive)

Methods of Randomization

Using a Table of random numbers

Fishbowl Method (lottery picking)

The idea is to allot each element of the population a number,

and then randomly select the sample randomly, such that each
element has equal probability of being selected.
Random Sampling Distribution of
Means
A random sampling distribution of the mean is the relative frequency
distribution of X̅ obtained from all possible random samples of a given
size that could be drawn from a given population.
A sampling distribution of the mean shows all the possible values that
the sample mean can take on (for samples of a given size) and the
probability of each.
If the population is of finite size, the number of possible samples will
also be finite. This makes it possible to generate an entire sampling
distribution and to explore its properties
Let’s TRY!
Consider a population of four scores—2, 4, 6, and 8—from
which we are to select samples of size 2.

What are the possible samples? (Sampling with replacement)

Calculate:

Population Mean

SD of Population

Mean of sampling distribution of means

SD of sampling distribution of Mean

CHEAT-SHEET

Standard Deviation Formulas:

OR
Random Sampling Distribution of
Means in Dr. Brown’s Case
Population in Dr. Brown’s Case: test scores of all 2,500 currently
enrolled sixth-grade students in her district.
Sample of 100, is just one possibility out of the many samples of
100 possible from the population.
Because of sampling variation, the X̅ would be different each time.
μX̅ = Average of X̅₁, X̅₂, X̅₃, …… X̅ո

(Random sampling Distribution of Means)

Characteristics of the Random
Sampling Distribution of the Mean (1)
The mean of any random sampling distribution of X, called the
expected value of the sample mean, is the same as the mean
of the population of scores from which the sample was drawn.

(regardless of n, 𝜎 , and the shape of the population)

Characteristics of the Random
Sampling Distribution of the Mean (2)
The standard deviation of the random sampling distribution of
the mean, called the standard error of the mean.

{depends on the standard deviation of the population, (𝜎X),

and the sample size (n)}
Characteristics (2 cont)

(a) sample means vary less than scores do (when the sample
size is at least 2),

(b) sample means vary less when scores in the population vary
less,

(c) sample means vary less when sample size is greater.

Characteristics (3)

If the population of scores is normally distributed, then sampling

distribution of the mean will also be normally distributed,
regardless of sample size.
If the population of scores is normally distributed, it does not
matter whether we take an infinite number of samples of size
2 or an infinite number of samples of size 200. In both cases,
the Sampling Distribution of Mean will also be normally
distributed.
Note: How the sampling distribution changes with a change in sample
size.
This figure reminds us that there is not just one random sampling
distribution of X corresponding to a given population, but a family of such
distributions, one for each possible sample size.
Population Normally distributed: ALL
IS WELL!
But what if Population is NOT
normally distributed??
CENTRAL LIMIT THEOREM to the rescue!

Central Limit Theorem says that the random sampling distribution of the
mean tends toward a normal distribution irrespective of the shape of the
population of observations sampled; the approximation to the normal
distribution improves as sample size increases.
Features of Normal Distribution

Symmetric

Asymptotic

Continuous

Unimodal
Random Sampling Without
Replacement
Both “Random Sampling With Replacement” & “Random
Sampling Without Replacement” satisfy the condition of
random sampling, but certain sample outcomes that are
possible when sampling with replacement are not possible
when sampling without replacement.
The standard error of the mean is smaller when sampling
without replacement.
However, the difference is trivial if the sample is small relative
to the population.
 Parametric
&
Non-Parametric
Test
Reference Text:
Minimum & King
S.K. Mangal
Nonparametric Test (Assumption
Freer Test)
Assumption Freer tests are tests that make fewer assumptions than
the traditional parametric tests such as t and F (Ury, 1967).
These procedures are often referred to as:
nonparametric statistics : because the statistical hypotheses tested
are not about parameters such as μ,
OR
distribution-free methods: if the procedure does not make any
assumptions about the shape of the population.
Note: These tests are not completely free of assumptions, but the
assumptions are less restrictive than t, F or 𝛘2
When the assumption of normality is met, the parametric
tests (e.g., t and F) are more powerful than the
assumption-freer tests.

When assumptions are not met (which is often the case), the
assumption-freer methods are more powerful. i.e., they are
more likely than t or F to give a significant result when the null
hypothesis is false.
When to Use a Parametric Test

1) Populations from whom the sample has been drawn should

be normally distributed. (Assumption of normality)
2) The variables involved must have been measured in interval
or ratio scale.
3) The observations must be independent. The inclusion or
exclusion of any case in the sample should not unduly affect
the results of the study.
4) These populations must have the same variance, or in
special cases must have a known ratio of variance.
(homoscedasticity)
When to use a Non-Parametric Test

1) When sample size is small.

2) When assumptions like normality of the distribution of scores in
the population are doubtful. In other words, where the
distribution is free i,e., the variates under question need not be
distributed in a specific way on the population.
3) When the measurement of data is available only in the form of
ordinal (rank) or nominal (categories) scale.
NOTE: Although, NPT are simpler and easier to be carried out, their use
should be restricted to those situations where the required conditions
for using parametric tests are not met.
BCZ. NPT are less powerful than PT in the same situation.
 Parametric Test Non-Parametric Test

t-test (independent groups) Mann-Whitney U-Test

Or
Wilcoxon Rank sum test (sample size
should be greater than 10)

t-test (Dependent groups) Sign Test

Or
Wilcoxon Signed-Ranks test

ANOVA (independent) Kruskal Wallis Test

Introduction to Statistical Inference:
Testing Hypotheses about a Single
Mean (z)
Statistical Inference
Statistical inference draws a conclusion about a population parameter
from a sample statistic.

Dr. Brown, the research director of a large school district, has read in
a test manual that the national norm for sixth-grade students on a
mathematics achievement test is 85. She wants to know whether
students in her district are performing at the same level. She selects at
random 100 students from all the sixth-grade students in the district
and gives the same test to them.

Dr. Brown then calculates their mean score and compares it to the
national norm. If the mean of her group differs from the national
norm, what should she conclude?

*This is what drawing a statistical inference entails.

Testing a Hypothesis about a Single
Mean
Dr. Brown first translates her question into a statistical
hypothesis: 𝜇ₓ= 85 (the mean of the population of sixth-grade
mathematics achievement scores in her district is 85).

To examine the validity of her hypothesis, Dr. Brown asks

what sample means would occur by chance if many samples of
the same size were drawn at random from her population if
the hypothesis that the population mean is 85 is true.
Testing a Hypothesis about a
Single Mean
 The Null and Alternative Hypotheses
Null hypothesis: Alternate hypothesis:

Denoted by H₀ Denoted by HA

a hypothesis about a population a hypothesis about a population

parameter (e.g., μₓ) that a researcher parameter that contradicts the null
tests hypothesis; in research, the hypothesis
the researcher wishes to prove is true

For every null hypothesis, there is also an alternative hypothesis.

Note: Both H0 and HA are always statements about the population parameter.

They are never statements about the sample statistic.

Although H0 is referred to as the null hypothesis, no special meaning should be attached to the term null. It does not
mean “zero” or “no difference.” The null hypothesis is simply whatever hypothesis we choose to test.
Alternate Hypothesis :
Directional & Non-Directional
If Dr. Brown is interested in discovering whether a difference exists irrespective of
direction, her alternative hypothesis is expressed as follows:

HA: μₓ≠85

But what if the researcher is interested in knowing only whether the scores in her
district are worse than the national norm, then,

HA: μₓ < 85

researcher is interested in knowing only whether the scores in her district are BETTER
than the national norm, then,

HA: μₓ > 85
When Do We Retain and When Do
We Reject the Null Hypothesis?
On drawing a random sample from a population, our obtained value of X̅ will almost
never exactly equal μₓ (because of chance when randomly drawing observations for our
sample).

Decision to reject or to retain the null hypothesis depends on the selected criterion for
distinguishing between those X̅ ’s that would be common and those that would be rare
if H0 is true.

REJECT Ho: When sample mean is so different from what is expected, that its
appearance would be unlikely to have occurred by chance

BUT, HOW MUCH RARE IS RARE ENOUGH?

Reject H0 if the sample mean is so deviant that its probability of occurrence by chance in
random sampling is 0.05 (less stringent)/ 0.01 (more stringent) or less (Measured against
the criterion chosen wrt level of significance)
Level of Significance (𝞪)

The probability value that is used as a criterion to decide that

an obtained sample statistic (X̅) has a low probability of
occurring by chance if the null hypothesis is true (resulting in
rejection of the alternate hypothesis)

Such that at𝞪 = 0.05 there is only 0.05 probability (5 in 100

trials) of the said result occurring by chance.
 The p Value
If the p value is less than or equal to the level of significance, the sample result is considered
to be sufficiently RARE to “Reject” Ho.
P <0.05
Summary for Hypothesis Testing

Step 1: null hypothesis (H0), and alternative hypothesis (HA) are formulated about a
parameter of the population (e.g., μₓ) .

Step 2: Random sample is drawn from the population and sample statistic (e.g., X̅)
is obtained.

Step 3: The random sampling distribution of the statistic under consideration is

examined to learn what sample outcomes would occur by chance over an infinite
number of repetitions (and with what relative frequencies) if H0 is true.

Step 4: H0 is retained if observed sample outcome is in line with the outcomes

expected if the hypothesis is true; otherwise, it is rejected and HA is accepted.
Region of Acceptance and Rejection

Region of rejection The portion of the

sampling distribution of X̅ (consisting
of values of X̅ that are unlikely to have
occurred by chance if H0 is true) that
leads to rejection of H0

Region of retention the portion of the

sampling distribution of X̅ that leads
to retention of H0. (the 95% and/or
99% of the sampling distribution)
Continuing on Dr. Brown’s Problem

Ho: μₓ = 85 HA: μₓ ≠ 85
𝛂 = 0.05
Reject Ho only if obtained sample mean is
so deviant that it falls in the upper 2.5% or
lower 2.5% of all the possible sample means
that would occur by chance when Ho is true
(0.025 + 0.025 = .05).
Arriving at Critical Values

If a random sampling distribution of X̅ is assumed to approximate a normal

curve, (Central Limit Theorem)

Then the normal curve table is used to calculate the z values that separate the
upper 2.5% and lower 2.5% of sample means, from the rest of the area of
retention; Revealing Critical Value (the value(s) that separates the region of
rejection from the region of retention).

Area under the curve (Table A Appendix D)

Calculate Zcrit for non directional hypothesis with 0.05?

Calculate Zcrit for directional hypothesis (are students performing worse) with
0.05?
 Non- Directional Hypothesis (The area beyond the z would be divided in 2, since the region of rejection
falls on both tails in a non directional hypothesis)
α = 0.05 (Zcrit = ±1.96) Zcrit will be calculated from area beyond 0.025
α = 0.01 (Zcrit = ± 2.57) Zcrit will be calculated from area beyond 0.005

Region of Rejection of
Ho Region of Rejection of Ho

Region of Region of Retention Region of

Rejection of of Ho Rejection of
Ho Ho

X̅2 =65 X̅1 =75

-3 -2 -1 0 1 2 3
+1.96 +2.57
-2.57 -1.96 μX̅ =85
 Directional Hypothesis (Region of rejection falls in one tail only, based on < or > sign in
Alternate Hypothesis)
α = 0.05 (Zcrit = ±1.64)
α = 0.01 (Zcrit = ± 2.33)

Region of Rejection of
Ho

Region of Retention
Region of of Ho
Rejection of Ho

-3 -2 -1 0 1 2 3
-2.33 -1.64
Calculating z score for testing
Hypothesis of single mean

z value here shows by how many standard errors of the mean the
obtained sample mean deviates from the hypothesized
population mean stated in Ho.
Calculate the z for obtained sample
mean of 90, in Dr. Brown’s case?
Ho: μₓ = 85

HA: μₓ ≠ 85

n=100

SD = 20

z= ?

𝞪 = 0.01

Will you retain or reject the null hypothesis?

What does each of these scenarios indicate?

Statistical Decision v/s
Research Conclusion
Statistical Decision: Since the Mean (x̅) corresponding to z
score of __ fall in the region of rejection/retention of Ho at
alpha =0.01 level, we reject/retain the null hypothesis.

Research Conclusion: The maths scores of the district

students in the maths test are different from (Higher/lower)/
same as the scores of the students as per the national norm.
What does retention & rejection of
Ho really mean?
The decision to “retain” Ho does not mean that it is likely that Ho is true.

It MEANS THAT…

We do not have sufficient evidence to reject the null hypothesis.

Rejecting the null hypothesis means that it does not seem reasonable to
believe that it is true, but retaining the null hypothesis means only that
we believe that the hypothesis could be true. It does not mean that it
must be true, or even that it is probably true.
 What would be the effect of
Ho: μₓ = 85
changing sample means
HA: μₓ ≠ 85
Actual population mean really does differ from the
x̅ = 90 hypothesized mean
n=25 If Dr. Brown had drawn a random sample of 25 students
instead of 100, she would not have rejected the null
SD = 20 hypothesis, because z = (90 - 85)/(20/√25) = 1.25, and 1.25
falls in the region of retention when ∝ =.05.
z= ?
If sampling variation is large, it may mask a difference
𝞪 = 0.01 between μhyp and μtrue.
Choice of HA: One-Tailed and
Two-Tailed Tests
When the alternative hypothesis is nondirectional, a
two-tailed test results, and it is possible to detect a
difference between the true value and the hypothesized
value of the parameter regardless of the direction of the
difference.
In a directional (one-tailed) test, our interest is in discovering
whether or not there is a difference in a particular direction.
One tailed Test: Drawback
In a one-tailed test, the alternative hypothesis limits what the researcher
has a chance to discover.

If HA says that the population mean is less than 𝜇hyp, then the only thing
the researcher can discover in this study is that the mean is less than 𝜇hyp.

For example, look at Figure 12.7.

Here, H0: 𝜇X =100 ; HA: 𝜇X < 100.

The region of rejection is entirely in the left tail. A value of X̅ that falls
above 100 cannot lead to rejection of the null hypothesis, no matter how
far above 100 it may be.

Does not allow for any chance of discovering that reality is just the
opposite of what the alternative hypothesis states.
When to take decision of using a
one-tailed or two-tailed test?
● Decision should always flow from the logic of the research question.
● Nature of the alternative hypothesis: SHOULD be decided at beginning of study,
before data collection.
● NOT Proper to observe the sample outcome and then set the region of rejection in
the tail of the sampling distribution toward which the sample outcome tends.
● if one adopts the 5% level of significance and follow this erroneous procedure
systematically, we are really, in the long run, conducting two-tailed tests at the
10% significance level.
● Not satisfactory to set our one-tailed test in the direction in which we think the
outcome might go, only to switch to a two-tailed test if the sample mean appears
to fall in the opposite direction. If tests are conducted in this manner, using 𝛼 = .05,
they are equivalent to two-tailed tests at 𝛼 = .075, with an area of .05 in one tail
and an area of .025 in the other
Review of Assumptions in Testing
Hypotheses about a Single Mean
Whenever we perform a z test, we use the normal curve table. For the normal
curve model of statistical inference about single means to be precisely correct,
several conditions must hold:

1. A random sample has been drawn from the population. (Violation of this
assumption may affect the mean and standard deviation of the sampling distribution
in unpredictable ways)

2. The sample has been drawn by the with-replacement sampling plan. (it is
common practice to sample without replacement. The consequent error in inference is
quite small as long as the sample size is a small fraction of the population size)

3. The sampling distribution of X̅ follows the normal curve. (reasonably

approximated when the scores in the population are reasonably close to a normal
distribution. When the scores in the population are not normally distributed, the
central limit theorem comes to the rescue when the sample size is 25 or larger)

4. The standard deviation of the population of scores is known.

PRACTICE QUESTION 1
The chancellor at City University is worried that
the grades have fallen in the past year. The
grade point average for graduating students in
the previous 5 years was 2.80 with a standard
deviation of 0.25. The chancellor randomly
samples 10 seniors from this year’s graduating
class and obtains the following GPAs: 2.40, 2.80,
2.65, 2.15, 2.65, 2.72, 2.42, 2.70, 2.52, 2.49. What
can the chancellor conclude? Use a one-tailed
test at the .05 significance level.
PRACTICE QUESTION 2

A) For the question posed by Dr. Brown, if the

population standard deviation on the
mathematics achievement test for
sixth-grade students is 10, what will Dr.
Brown conclude if she uses a two-tailed
test at the .05 significance level and
obtains a mean of 87.1 for her sample of
100 students?
B) Assume the same data as are given in
Problem 2A. Suppose that the alternative
hypothesis reads HA: 𝜇X < 85. Test the null
hypothesis at the .05 significance level and
state your conclusions.
PRACTICE QUESTION 3

Two years ago, a complete survey of all students in a

large university indicated that the mean number of
cigarettes smoked per day by male students was 8.3
with a standard deviation of 3.7. The director of student
health services wishes to determine whether or not
smoking habits among male students at the university
have changed over the 2-year period. She obtains the
following results for the men from a recently selected
random sample of students (in cigarettes smoked per
day): X bar = 7.7, n = 108. (a) Set up H0 and HA.
(b) Perform the statistical test (𝛼 = .05). (c) Draw final
conclusions.
What to do when 𝛔ₓ UNKNOWN!

If we know the value of 𝛔ₓ, we probably already know the

value of μₓ .

If we know the value of μₓ, why would we want to estimate it

from X̅ ?

Thus, we go for estimation of 𝛔ₓ from the sample.

 Quick Review
Conceptual Revision
VARIANCE = (STANDARD DEVIATION)²
STANDARD DEVIATION = √VARIANCE
Sum of Sqauares = Sum of squared deviations from the mean = SSₓ = Σ(X-X̅)²
Population Variance = 𝛔ₓ² Population SD = 𝛔ₓ

Sample Variance (Biased est.) = Sₓ² = SSₓ/n Sample SD (Biased est of population’s SD) =Sₓ = √SSₓ/n

Sample Variance (Un-Biased est.) = sₓ²=SSₓ/ (n-1) Sample SD (un-Biased est. of pop’s SD) =sₓ =√SSₓ/(n-1)
Estimating the Standard Error of the
Mean When 𝛔ₓ Is Unknown
To test hypotheses about means using z, we must calculate the standard error of the
mean 𝛔ₓ̅ (Read: Sigma x bar)

SEM requires 𝛔ₓ, which in actual practice is not known, since most times the population
parameter is unknown.

Substituting 𝛔 ₓ with Sₓ; Creates TROUBLE!

Sₓ², is a biased estimator of the population variance, 𝛔ₓ²

WHY? Bcz, just dividing by (n), we are underestimating the true population value.

This underestimation is corrected through by dividing by (n-1) rather than n, which is

called as the unbiased estimate of population variance sₓ².

Let’s Witness.
Decoding the PROOF

Here the Population

VARIANCE is calculated from
the population data, and it is
being used to compare how Sₓ² = Σ(X-X̅)²/n
biased / unbiased the 0
estimator (Estimate of 1
population variance) is. 4
9
1
OBSERVATION 0
Sₓ² is a biased estimator as it 1
would give a value less than 5 4
4
(𝝈ₓ²). {Sₓ² is not calculated here, you 1
may calculate it and check}
0
1
sₓ² is an un-biased estimator 9
as it gives a value equal to 5 4
1
(𝝈ₓ²). 0

Mean value of Sₓ² =

2.67
Unbiased Estimator : The mean of the estimates made from
all possible samples equals the value of the parameter
estimated (X̅ is an unbiased estimator of μₓ; Sₓ is not an
unbiased estimator of μₓ)
Now that Estimated SEM is achieved,
Substituting..
When we substitute sₓ for 𝛔ₓ, the result is called the estimated
standard error of the mean, symbolized sₓ‾ (Read: s xbar)

When we substitute sₓ for 𝛔ₓ in the formula for z, it result in a

new statistic, called t (approximation of z).

Note: t is not normally distributed.

WAS

NOW..
If we were to draw an infinite number of samples of the same
size n from a population, calculate X̅, sₓ, and sₓ‾ for each
sample, and then find t for each sample, we would have the
sampling distribution of t.

British mathematician William S. Gosset, presented the proper

distribution for t, called the Student’s distribution or Student’s
distribution of t.
Characteristics of Student’s
Distribution of t
1) Student’s distribution of t is not a single distribution, but rather a
family of distributions, which differs in the degree of approximation
to the normal curve, based on sample size and df .
2) t is not normally distributed. WHY?

Because of the presence of the variable in the denominator.

3) When the sample are large, the value of sₓ̅ will be so close to 𝝈ₓ̅
that distribution of t will be nearly normal. Such that, t=z, (true for
infinitely large sample).
Similarities btw t and z

1. Have a mean of zero.

2. Are symmetrical.

3. Are unimodal.
Differences btw t and z distributions

1. t is platykurtic compared to the normal distribution (i.e., it is

narrower at the peak and has a greater concentration in the tails
than does a normal curve).
2. t has a larger standard deviation (remember that 𝛔z =1).
3. t distribution depends on the number of degrees of freedom.
Because the tails of the t distribution descend less rapidly to the
horizontal axis, it is necessary to go to a more extreme location to
find the point beyond which .025 of the area (in one tail) falls.
As df increases, the t distribution, gets closer to z
distribution (at df = ∞, t distribution identical to z)
How the difference btw t and z impact
STATISTICAL INFERENCE drawn
If null hypothesis is tested at α = 0.05, z critical = + 1.96

Since t is platykurtic, tails descend less rapidly, thus, it

becomes necessary to go to a more extreme location to find
the point beyond which 0.025 of the area lies.

Since the shape of the curve changes for each value of df, the
critical values of t also change.
 Degrees of Freedom
The degrees of freedom corresponds to the number of observations that are completely free to vary.

Suppose a sample consists of three scores: X₁, X₂, and X₃.

sₓ = √Σ(X-X̅)²/(n-1)

The value of sₓ will be determined by the values of the three deviation scores: (X₁ -X̅), (X₂ - X̅), and (X₃ - X̅).

Now only two (any two) of the deviation scores are free to vary.

If we were to, arbitrarily assign to (X₁ -X̅) = +3, and to (X₂ - X̅) = -7. As soon as two of the three have been given
numerical values, the third is no longer free to vary.

NOTE: sum of deviations taken about the mean is always zero

So as soon as the first two deviations are assigned, the third must be a value such that the three deviations sum to
zero. In our numerical example, the value of (X₃ - X̅) must be +4.

The number of degrees of freedom in a problem involving the calculation of s is (n - 1).

When testing hypothesis
about mean, t critical
arrived at from Table D
in Appendix D, at the
intersection to the row
corresponding to
appropriate degree of
freedom (left) and the
column corresponding
to level of significance
for each tail (top).

Calculate t critical for

df= 15, at α = 0.05, one
tailed test.

Calculate t critical for

df= 15, at α = 0.05, two -
tailed test.
 Let’s Tackle a Problem
Children average about 4 hours a day watching television, with one-fourth of them spending
more than 6 hours a day. In the late 1980s, the average child watched about 100 violent acts
on television every week. Professor Telley believes that the media glorifies violence and as a
result the amount of violent content that children see on prime-time television shows has
increased even further since the late 1980s.
To test her belief, Professor Telley draws a random sample of 25 eighth-grade school children
from the local school district and calculates the number of violent acts that each of them had
seen on television in the previous week. Her research question leads to a statistical
hypothesis to be tested, she decides, at the .05 level of significance. Although Professor
Telley believes that children are seeing more violent acts on television than in the 1980s, she
also wants to know whether they are seeing fewer violent acts. The mean number of violent
acts seen on television in 1 week by her sample of 25 children turns out to be 120.
Let us assume that sₓ proves to be 35.5.
Calculating t

Ho: μₓ = 100

HA: μₓ ≠ 100

X̅ = 120

sₓ= 35.5

n=25

Find t?
What does a t value of 2.82 indicate?

It means that Professor Telley’s sample mean was 2.82

estimated standard error of the mean units greater than the
value stated in H0.

degrees of freedom is n - 1 = 24

Calculate t critical values? at α = 0.05

Conclusion (Research)

Professor Telley can conclude that eighth-grade children in

her local school district see more violent acts on television
during prime time than children did 20 years earlier because
television and the media glorify violence.
Case 1

If Professor Telley had been concerned only with the possibility that
eighth-grade students see more violent acts on television than in the
past, then how would the alternative hypothesis change and the
following analysis change?
HA ?
tcalculated =?
tcritical = ?

Decision?
Case 1 solution
Case 2

Had Professor Telley had been interested in knowing only

whether the eighth-grade children had seen fewer acts of
violence than in the past. Then?
HA ?
tcalculated =?
tcritical = ?

Decision?
Case 2 Solution
 Calculating t from RAW SCORE

SSᵪ = Σ(X- X̅)² = ΣX² - (ΣX)²

n
Substituting
Suppose that Dr. Jones, a colleague of [Link],
wants to replicate her study with a sample of
10th-grade students.

Ho: μₓ = 100

HA: μₓ ≠ 100

α = 0.05

Dr. Jones now randomly selects fifteen 10th-graders from the

same school district used by Dr. Telley and determines how
many violent acts each had seen on television in the previous
week. Data on next slide, find t
Step 1: Calculate sᵪ̅

Step 2: Substitute
values in formula
for t.
Practice

(a) the value of t that for 15 degrees of freedom is so high that only
1% of ts would be higher,
(b) the centrally located limits for 8 degrees of freedom that would
include 95% of t’s,
(c) for 25 degrees of freedom, the proportion of ts that would be
less than t = 1.316,
(d) for 20 degrees of freedom, the probability in random sampling
of obtaining t ≥ +2.528.
We adopt 𝛼 = .05 and test the hypothesis H0∶ 𝜇X = 50.
What conclusion should we draw if
(a) n = 10, tcalc = +2.10, and HA∶ 𝜇X ≠ 50?
(b) n = 20, tcalc = +2.10, and HA∶ 𝜇X ≠ 50?
(c) n = 10, tcalc = +2.10, and HA∶ 𝜇X > 50?
Show the critical value of t for each part.
105

98 The PTA at Central High School is concerned that students who

101
graduate from the school do not score as well on mathematics
achievement tests, on average, as do students from other schools in
110
the district. To address their concerns, the school principal randomly
96 selects 15 students from the graduating class and administers the
103 standardized mathematics achievement test used in that school
104
district. The mean score for graduating students in the school
district is 98. The scores of the 15 students from Central High are as
101
follows:
98
105, 98, 101, 110, 96, 103, 104, 101, 98, 105, 112, 95, 105, 100, 108
105

112
(a) State formally the hypotheses necessary to conduct a
nondirectional test.
95

105
(b) Complete the test at the .05 level of significance and state your
conclusion.
100

108
A new brand of automobile tire is advertised to outlast the leading
brand, which has a mean lifetime of 80,000 miles. A consumer agency
randomly selects six tires from the manufacturer’s first production
run and tests the tires on a machine that guarantees even tread wear.
The results are as follows:
78,000 87,000 85,000 79,000 80,000 86,000
(a) State formally the hypotheses necessary to conduct a
nondirectional test.
(b) Determine the critical value of t (𝛼 = .05).
(c) Calculate t.
(d) State your conclusion. (Do these results cast doubt on the
advertised claim?)
A training director for a large company has been told that, on
completion of the training course, the average score of her trainees
on the final evaluation should be 100. Her only concern is whether
she will have to begin remedial steps to ensure that the population of
trainees is not below standard.

She draws a random sample of 10 scores of recent trainees:

94, 98, 101, 90, 86, 102, 95, 100, 98, and 92.

(a) State the null and alternative hypotheses best suited to the nature
of her inquiry. Test the null hypothesis

(b) at the .05 level of significance and state your conclusions and

(c) at the .01 level of significance and state your conclusions.

The chancellor at City University is worried that the grades
have fallen in the past year. The grade point average for
graduating students in the previous 5 years was 2.80. The
chancellor randomly samples 10 seniors from this year’s
graduating class and obtains the following GPAs:

2.40, 2.80, 2.65, 2.15, 2.65, 2.72, 2.42, 2.70, 2.52, 2.49.

(a)What can the chancellor conclude?

Use a one-tailed test at the .05 significance level

Topics covered so far include: