Module 4: Inferential Statistics
The Applied Research Center
Module 4 Overview
} Inferential Statistics
} Brief Introduction to Probabilities
} Hypothesis Testing
Parameter vs. Statistic
} A population is the entire set of individuals that we are
interested in studying
} A sample is simply a subset of individuals selected from
the population
} In most studies, we wish to quantify some characteristic
of the population à parameter
} Parameters are generally unknown, and must be
estimated from a sample
} The sample estimate is called a statistic
Inferential Statistics
} Techniques that allow us to make inferences about a
population based on data that we gather from a sample
} Study results will vary from sample to sample strictly due
to random chance (i.e., sampling error)
} Inferential statistics allow us to determine how likely it is
to obtain a set of results from a single sample
} This is also known as testing for statistical significance
Statistical Significance
} Consider a small weight loss study of 40 patients.
} After such a study is over, we want to make
generalizations about a larger group (e.g. all similar
patients in the nation), but, since it is a small study, the
results will be inexact.
} Statistical significance helps us by giving us a "ballpark
range" (i.e., confidence interval) around the number (for
example the amount of weight lost), encompassing the
true number.
Statistical Significance (cont d)
} If the range is small enough (p < .05), we say we are
confident that the true amount of weight lost is "more
than zero" and "statistically significant.
} Naturally, it says nothing about the practical significance,
since the patients might have lost just a gram of weight!
Statistical Significance Testing
Hypothesis Testing
A Brief Introduction to Probability
} A basic understanding of probability is needed
} The probability of an outcome (A), can be thought of as a
fraction, proportion, or percentage
num ber of A outcom es
probability of A =
total num ber of outcom es
Probability Examples
} What is the probability of rolling a single die and coming
up with a six?
} There is only 1 outcome A (a six)
} There are 6 possible outcomes (1 to 6)
} The probability is 1/6 = .1667 = 16.7%
} What is the probability of obtaining a red number in the
game of roulette?
} There are 18 red numbers (A)
} There are 38 numbers total
} The probability is 18/38 = .4737 = 47.4%
The Null Hypothesis
} The null hypothesis always states that nothing is going
on
} There is no difference, no relationship, no treatment effect, etc.
} H0: X = Y
} The alternate hypothesis states that there is a difference
} Ha: X ≠ Y (non directional)
} Ha: X > Y or Ha: X < Y (directional)
The (Somewhat Twisted) Logic of
Significance Testing
} Compute a probability value that tells how likely our data
(or results) would occur just by chance
} If the probability is low (e.g., p = .02), this means our
data is inconsistent with the null
} There is evidence that there is a difference
} If the probability value is high (e.g., p = .30), this means
our data is consistent with the null
} There does not seem to be evidence that there is a difference
More on the Logic
} The confusing thing is that we are not directly testing
whether or not there is a treatment effect, or relationship
} We are testing how consistent the data is with the
hypothesis that there is no treatment effect, relationship,
etc.
} Thus, a treatment effect is demonstrated indirectly if the
data is inconsistent with the null hypothesis
Rule of Thumb (p < .05)
} How inconsistent with the null does the data have to be
to demonstrate an effect?
} Conventional rules use a p < .05 cutoff
} If the data yields a probability value less than .05 (p < .
05), that means the data is inconsistent with the null,
which states no treatment effect or relationship exists
(H0: X = Y)
} Therefore, we reject the null
Two Outcomes
} If p < .05, our data is inconsistent with the null
} We reject the null and declare our results statistically
significant
} If p > .05, our data is consistent with the null
} We fail to reject the null and declare our results statistically
non-significant
Example 1
} Suppose we were comparing how males and females
differed with respect to their satisfaction with an online
course
} The null hypothesis states that men (X) and women (Y)
do not differ in their levels of satisfaction
} H0: X = Y
Example 1 (cont d)
} On a 25-point satisfaction scale, men and women differed
by about 5 points (means were 18.75 and 23.5,
respectively)
} They were not identical, but how likely is a 5 point
difference to occur just by chance?
Example 1 (cont d)
} An analysis was conducted, and the p-value for the gender
comparison was p = .11
} Thus, there was about a 11% chance that this data (the 5
point difference) would occur by chance
} The p-value is greater than .05, so we would fail to reject
the null (results are not significant)
} Thus, there is no evidence that males and females differ in
their satisfaction
Example 2
} Suppose we were comparing how males and females
differed with respect to how likely they would be to
recommend an online course (measured on a 5 point
scale)
} The null hypothesis states that there is no difference
between men and women in their recommendation of an
online course.
} (H0: X = Y)
Example 2 (cont d)
} On a 5-point satisfaction scale, men and women differed
by about 1 point (means were 4.3 and 3.1, respectively)
} They were not identical, but how likely is a 1 point
difference to occur by chance?
Example 2 (cont d)
} An analysis was conducted, and the
p-value for the gender comparison was
p = .03
} Thus, there was only a 3% probability that this data would
occur by chance
} The p-value is less than .05, so we would reject the null
(results are significant)
} Thus, there is evidence that males and females differ in
their recommendations
The Meaning of Statistical Significance
} p-values tell how likely it was that our sample was drawn
from a hypothetical population where nothing was going
on
} Thus, the term statistical significance simply means that
the obtained results are unlikely to represent a situation
where there was no relationship between variables
} The difference is big enough to be unlikely to have
happened simply due to chance
Cautionary Note
} Just because results are statistically significant, does not
mean that the results are of practical importance
} It ends up that large samples are more likely to yield
significant results , even if the differences are rather
trivial
} Don t equate statistical significance with a large or
important effect
Module 4 Summary
} Inferential Statistics
} Brief Introduction to Probabilities
} Hypothesis Testing
Review Activity and Quiz
} Please complete the Module 4 Review Activity:
Hypothesis Testing located in Module 4.
} Upon completion of the Review Activity, please complete
the Module 4 Quiz.
} Please note that all modules in this course build on one
another; as a result, completion of the Module 4 Review
Activity and Module 4 Quiz are required before moving
on to Module 4.
} You can complete the review activities and quizzes as
many times as you like.
Upcoming Modules
} Module 1: Introduction to Statistics
} Module 2: Introduction to SPSS
} Module 3: Descriptive Statistics
} Module 4: Inferential Statistics
} Module 5: Correlation
} Module 6: t-Tests
} Module 7: ANOVAs
} Module 8: Linear Regression
} Module 9: Nonparametric Procedures