Inferential Statistics

Probability & hypothesis testing

Inferential statistics
Inferential statistics are used to make judgements of the probability that an observed difference
between groups is statistically significant or that the difference between groups happened
simply by chance

Sample vs. Population

sample population

statistic parameter

mean X μ

variance s2 σ2

standard
s σ
deviation
Sampling distribution of means
Consider 3 different types of distributions:

1 population distribution

20,000 students
Grip strength of university students

2 sample distribution
(100 subjects)

From the population (UW students), we

measure grip strength of 100
undergraduate students (sample).
Grip strength of university students
Sampling distribution of means

3 sampling distribution of means

Measure an infinite number of

population samples with 100 undergraduate
students in each sample.

sample sample sample sample

#1 #2 #199 ∞

The sampling distribution of means

is a frequency distribution of all of
the infinitely possible sample means N = numbers of scores used to
from a population. determine sample mean (standard error of the mean)
Three distributions:

1. Population distribution
- data from which sample is chosen

2. Sample distribution (one sample)

- assume σ and s to be equal

3. Sampling distribution of means (many samples)

- theoretical
Sampling distribution of means
Probability in everyday life:

1. Texas Hold’em

2. Insurance Rates - How much you pay in premiums is

based on probability.

3. Weather

4. Flipping a coin

5. Rolling a die
Inferential statistics and probability
Probability forms the basis for inferential statistics and statistical conclusions.

Most scores happen by chance = 95%

z score of -1.96 to +1.96

area/proportion of the
normal curve = 95 %
Rare = 5%
few scores occur here or here
2.5% of scores
z score
2.5% of scores -3 -2 -1 0 +1 +2 +3

Proportion of the total area under the curve for particular

scores equals the probability of measuring those scores.
Hypothesis testing

Sample data is used to make inferences about the population. For

interval/ratio data, the mean is the best representation of this data.

Hypothesis Testing:
• Allows you to determine if the sample is representative of the
population
• Allows you to determine if groups are from the same population or
from different populations

Uses sampling distribution of means to represent the population

Sampling distribution of means
•

N = 200

N = 100

standard error (SE) of the mean

Sampling distribution of means = normal distribution

Therefore, it possesses characteristics and properties of a normal curve.
remember z scores? – Table C.1 (probabilities)
Using sampling distribution of means

•
Using sampling distribution of means

What is the probability of

having a sample mean
between 39 – 41 kg?

z score of -1 to +1:
68.26% (0.6826)
35 36 37 38 39 40 41 42 43 44 45

-3 -2 -1 0 +1 +2 +3

Table C.1 -3 -2 -1 0 +1 +2 +3
Using sampling distribution of means
What is the probability of having a sample mean between 38 – 42 kg?

z score of -1.96 to +1.96: 95% Table C.1

Therefore, the probability of having a sample mean less than

38 kg is <2.5% and the probability of having a sample mean
greater than 42 kg is <2.5%

35 36 37 38 39 40 41 42 43 44 45

-3 -2 -1 0 +1 +2 +3

-3 -2 -1 0 +1 +2 +3
Using sampling distribution of means

Using a criterion of 5%, is it possible that a sample with a

mean of 36 (edge of blue)is from the population of interest?

35 36 37 38 39 40 41 42 43 44 45

• If a sample with a mean of 36 is not from the population of -3 -2 -1 0 +1 +2 +3

interest, then they must be from a different population.

-3 -2 -1 0 +1 +2 +3
• We call this a “rare event” the fact that the sample mean
was 36 did not simply happen by chance. There’s a reason
why the sample mean is 36 (could be from another
population OR in the case of experimental research, the
independent variable/intervention caused an effect).
Hypothesis Testing
The scientific method

• Make an observation Generate Identify a

New Ideas Problem

• Formulate an question
• Formulate a hypothesis Communicate Develop a
the Results Research Plan
• Design an experiment
• Execute the experiment
Analyze & Conduct the
Evaluate Data Study
• Analyze the results Hypothesis testing

• Draw conclusion
• Formulate a new hypothesis
STUDY DESIGN DICTATES WHAT STATISTICAL APPROACH IS USED TO ANALYZE YOUR DATA
Each bullet point influences the subsequent bullet point
Hypothesis Testing

Significance Level:
• Probability is used to define the sample means as being too unlikely to represent the
underlying raw score population (rare event)
i.e. This sample is NOT from this population.

• Use probability of 0.05 or 0.01

• Use symbol a (alpha) to represent significance level

• As a researcher, you choose the α value (either 5% or 1%). As a KIN232 student, you will
always be told/given the α value to use. Working with an α value of 1% is what we call
being more conservative because it will be harder to have a mean that is NOT from the
population. In other words, it will be harder to find a significant difference between
means/groups.
Graphically: α = 0.05, one tailed α = 0.01, one tailed

5% 1%

One tail = directional, examines a higher mean or lower mean (examine one only)

α = 0.05, two tailed α = 0.01, two tailed

2.5% 2.5%
0.5% 0.5%

Two tailed = non–directional, examines a difference. We’re not sure if the mean should be higher or lower than
the population mean. Therefore, we examine both possibilities
Converting alpha and 1 or 2-tails into z score

α = 0.05, one α = 0.01, one tailed

tailed Least conservative
5% 1%
α = 0.05, one tail (z = 1.65)

α = 0.05, two tail (z = 1.96)

α = 0.05, two tailed
α = 0.01, two tailed α = 0.01, one tail (z = 2.33)

2.5%
α = 0.01, two tail (z = 2.58)
2.5%
0.5% 0.5%

Most conservative
Hypothesis testing - 7 steps

1) state null hypothesis in symbols and words

The scientific method

2) state alternative hypothesis in symbols and words • Make an observation

• Formulate an question
3) use α level and decide if one or two-tailed • Formulate a hypothesis
• Design an experiment
4) state rejection and retain rule • Execute the experiment
• Analyze the results
5) compute appropriate statistic • Draw conclusion
• Formulate a new hypothesis
6) make decision by applying rejection / retain rule

7) write conclusion in context of study

 1 2 3 4 5 6 7
State State Rejection
α level Calculate Decision Conclusion
null alternate rule

NOTE – THIS IS SIMPLY AN EXAMPLE. You will see specific aspects of each step with
the various types of analyses we will examine.
Example:
Is the starting salary of University of Waterloo graduates higher than other Ontario
University graduates who average $50,000?

Questions you must ask yourself when reading the descriptions of the study (this one is simple, they will get
more complex).
Is there a reference to a population or a comparison being made to a population?
Is randomization mentioned?
How many groups are there?
Is there a control group?
Is there an independent variable (or more than 1)? How many levels of the independent variable are there?
Is there directionality (e.g., lower, higher, smaller, larger) or not (altered, difference, change)?
Is there a pre-test and post-test?
Is there repeated measures?
 1 2 3 4 5 6 7
State State Rejection
α level Calculate Decision Conclusion
null alternate rule

Recall the question: Is the starting salary of University of Waterloo graduates higher than other Ontario
University graduates who average $50,000?
μ: University of Waterloo mean salary
1. Ho : μ: = μo
μo: Other Ontario Universities mean salary

University of Waterloo graduates have the same starting salary as other

Ontario University graduates.

We will use these 7 steps to perform several different statistical

analyses. With each step there are similarities and differences
between statistical tests. I will do my best to highlight these
similarities and differences to you.
 1 2 3 4 5 6 7
State State Rejection
α level Calculate Decision Conclusion
null alternate rule

Recall the question: Is the starting salary of University of Waterloo graduates higher than other Ontario
University graduates who average $50,000?

2. H1: μ > μo

University of Waterloo graduates have a higher starting salary than the $50,000
average of other Ontario University graduates.
 1 2 3 4 5 6 7
State State Rejection
α level Calculate Decision Conclusion
null alternate rule

Recall the question: Is the starting salary of University

3. α = 0.01,
of Waterloo graduates higher than other Ontario
University graduates who average $50,000?

You will be told whether to use 0.05 or 0.01. The only decision you need to
make is whether it is a one or two tailed test.
 1 2 3 4 5 6 7
State State Rejection
α level Calculate Decision Conclusion
null alternate rule

4. using Table C.1, α = 0.01 / one tail How you structure the rejection rule is dependent
on several things:
• What type of statistical test you are performing
zcritical = + 2.33 • Is it a one tail or two tail analysis
• What is your data? What I mean by that is it
blood pressure (reduction is good), is it
reject Ho : if the test statistic > or = zcritical hypertrophy (more muscle is better), is it memory
loss (more is bad)
retain Ho : if the test statistic < zcritical

NOTE: must also include retain rule

= 0.01
 1 2 3 4 5 6 7
State State Rejection
α level Calculate Decision Conclusion
null alternate rule

5. compute test statistic using appropriate test

This is the math side of things – we will have to do some (t-tests, confidence
intervals and Chi square). Other you will be given the output from the
calculations and you have to interpret the results.
 1 2 3 4 5 6 7
State State Rejection
α level Calculate Decision Conclusion
null alternate rule

6. Make a decision by applying rejection rule:

Possibility 1: If test statistic is ≥ zcritical (+ 2.33) then, reject null hypothesis

Possibility 2: If test statistic is < zcritical (+ 2.33) then, retain null hypothesis
 1 2 3 4 5 6 7
State State Rejection
α level Calculate Decision Conclusion
null alternate rule

7. Conclusion must reflect the decision (step 6)

Possibility 1: If our decision was to reject Ho
University of Waterloo graduates have a significantly higher starting salary (> 50 000) than other
Ontario University graduates. (p < 0.01)

Possibility 2: If our decision was to retain Ho

University of Waterloo graduates have a statistically, the same starting salary (50 000) as other Ontario
University graduates (p > 0.01)

The conclusion must:

• Include mention of the dependent variable;
• The comparison being made (i.e. control vs. exercise group);
• The statistical results (i.e. significantly different, no significant difference, statistically);
• The probability.