PSYCH 240:
Statistics for Psychologists
Interval
Estimation:
Understanding
the t
Distribution
A Probability Distribution
.30
.25
Rela tive Freq uency
.20
.15
.10
.05
0.00
0 1 2 3 4 5 6 7 8 9 10
Number of Tails
Probability Distribution with
Means
We want to assess the likelihood of getting a
particular result (rare events are interesting)
Based on population parameters and sample
statistics, we can use sampling distributions to
get this likelihood
But, just as with scores, sampling distributions
of the mean based on raw data are not helpful
Thus, we standardize our sample means to get
a more interpretable result
Standardizing Means
The standard normal distribution of z
scores before told us the likelihood
of getting individuals scores
Now, we can get the likelihood of
getting particular sample means
using what we know about sampling
distributions and the central limit
theorem
How rare is too rare?
Inferential Statistics
Definition: Numbers that will help us to
make conclusions about populations on
the basis of our samples
General Techniques
Hypothesis Testing: Determine statistical
significance (uniqueness) of a parameter
Confidence Interval Estimation: Estimate
the likely interval for a parameter
Logic of Hypothesis Tests
Hypothesize a parameter value upon which to
center the sampling distribution
Use Central Limit Theorem to identify shape of the
sampling distribution
Identify where data lies on the sampling distribution
Decide whether the data seems likely/consistent
with the hypothesized parameter
Conventional Decision Rules
But what exactly is the probability that an
outcome is considered too low to be purely by
chance?
Alpha (α ) reflects the cutoff probability where
we say the event has occurred because of
some characteristic of the study
Convention suggests that an appropriate value
for alpha is either .05 or .01
We almost always use .05
Characterizing Decision
(Rejection) Regions
The location of the cutoffs on the
distribution can be identified two
distinct but equivalent ways
Using the probability (alpha)
Using the z-score (now called a critical
value) associated with α
Determining Critical Values
α /2 α /2
Significant Not Significant Significant
Lower CV Upper CV
The Use of the Critical
Values
Statistical Significance refers to the
rarity of an event
If a sample mean is particularly rare,
we call it statistically significant
Over samples of size N, the
probability for an event is
determined by:
µ − (CV Z )(σ M ) ≤ µ ≤ µ + (CV Z )(σ M )
The p value revisited
Statistical significance is telling us that
there is something different or unique
about our sample
But compared to what?
Since we need a comparison, we use
what is called the Null Hypothesis
H0:M = µ
H1: M ≠ µ
The p value defined
The p value tells you the probability of
getting a particular result if the null
hypothesis were true
Thus, given an alpha of .05, any p
value less than .05 (p < .05) is called
“statistically significant”
Steps of Hypothesis Testing
1. Calculate the appropriate statistic
(such as z)
2. Estimate the p value (such as p < .
05 or p > .05) by using critical
values
3. Calculate the p value (if possible)
4. Make the appropriate conclusion
about statistical significance
Sampling as an Inexact
Science
In most cases, we do not know the
population parameters
The very reason we are sampling is
that we wish to infer the parameters
Both our samples’ means AND
standard deviations will have
sampling error
When Variances are
Unknown
When the population variance is unknown,
we need to estimate it on the basis of our
sample
Thus, our standard error of the mean is
directly influenced:
Standard Error based σ
on Population Scores
σ M =
N
Standard Error SD
estimated from Sample SEM =
Scores N
Distinguishing Sampling
Distributions
M −µM M −µM
z= t=
σM SE M
What is the t distribution???
When we don’t know the population
characteristics, we cannot use z
scores
We know this because our samples
are imperfect and would probably
cause use to have biased results
Thus, we make a correction with a
family of distributions that differ in
shape depending on sample size
Shape and Estimation
Is the overall shape affected in any
systematic way by estimation? Yes . . .
Degrees of freedom are what determine
the shape of the sampling distribution
Every statistic in a specific context has a
certain number of degrees of freedom
The degrees of freedom (df) for any
statistics is the number of components in its
calculation that are free to vary
Practically speaking, df = N - 1
Shape of the t Distribution
df = infinity
df = 25
df = 9
df = 1
qer F e vit al e R
-4 -3 -2 -1 0 1 2 3 4
Using the t Distribution
for Probabilities
Therefore, we can NOT use the normal
distribution for cases in which we estimate
the population variance
Instead, we use the t distribution
Directly calculating probabilities is too difficult
because we no longer have 1 distribution in
total, but instead 1 distribution for every degree
of freedom
Often, we are only interested in certain
locations (i.e., critical values) in the distribution
Determining Critical Values
α /2 α /2
Significant Not Significant Significant
Lower CV Upper CV
Calculating Probability of a
Statistic
In order to calculated the probability
of a statistic, we need to know
The mean of the possible means
The standard deviation of the possible
means
The shape of the distribution of means
Comparison of z scores for
individuals and samples
Using the z Formula for Scores:
Y −M
z= = ?, p = ?
SD
Using the z Formula for Means:
M −µ
z= = ?, p = ?
σM
Comparison of z and t
Using the z Formula M −µ
for Means: z= = ?, p = ?
σM
Using the t Formula
for Means:
M − µM
t (df ) = = ?, p = ?
SEM
Using the t Distribution
Example: College students exercise 3
times per week on average. I select
a sample of 16 students who
exercise 3.5 time per week (SD =
1.5). What is was the probability of
getting a sample this distinct?
M −µM
t ( df ) = =?, p =?
SE M
Using the t Distribution
Example: College students exercise 3
times per week on average. I select
a sample of 16 students who
exercise 3.5 time per week (SD =
1.5). What is was the probability of
getting a sample this distinct?
3.5 −3
t (15 ) = =1.333 , p ≅.20
1.5
16
Some Problems
Determine the two-tailed probabilities for
each of the following samples (μ = 3.0):
N = 100, M = 3.2, SD = 1.0
N = 25, M = 3.6, SD = 1.3
N = 9, M = 2.75, SD = 1.9
What conclusions would you make about
the likelihood of each of these samples?
Problems
For N = 100, M = 3.2, SD = 1.0
3.2 − 3.0 0.2 0.2
t (99) = = = = 2.0
1.0 1 0 .1
100 10
t (99) = 2.00, p < .05
Problems
For N = 25, M = 3.6, SD = 1.3
3.6 − 3.0 0.6 0.6
t (99) = = = = 2.308
1.3 1.3 0.26
25 5
t (24) = 2.31, p < .05 Or you could report:
t (24) = 2.31, p ≅ .03
Problems
For N = 9, M = 2.75, SD = 1.9
2.75 − 3.0 − 0.25 − 0.25
t (8) = = = = 0.395
1.9 1.9 0.633
9 3
t (8) = 0.40, p > .05
How Can We Interpret These
Outcomes?
What implications might these have
for how we think about our samples?
Possibilities:
Our sample happened by total chance
Our sample is not representative of the
population
Reporting non-significance
There are several ways of
reporting non-significant
inferential stats
t (8) = 0.39, p ≅ .70
This is the most
specific
This is least specific
t (8) = 0.39, p > .05
t (8) = 0.39, n.s.
This is most common
Reporting non-significance
t <1
This is the most succinct for t values < 1
No t value less than 1 for any level of df can be significant, so why bother reporting everything?
Point vs. Interval Estimation
Point Estimation: The process of deriving a
single best estimate of a parameter (such
as a mean)
Interval Estimation: The process of
deriving a range of values that may
include a parameter (such as a mean)
Example: Polling and the prediction of
election results
Confidence Intervals
This creates a range of values that
is expected to include the
population parameter
Width is therefore determined by:
Standard Error (Standard deviation
and Sample size)
Probability level
The Concept of Sampling as an
Inexact Process
The “center”
is the
parameter –
the value of
interest
The “arrows”
are the
samples
taken
One-Sample Problem #1
The Scenario: An instructor is interested in
the exam performance of his students.
His newest class (N = 29) had a M =
88.14 and a SD = 10.53.
With 95% confidence, what is the likely
mean for the population exam score?
The Logic Behind Calculating a
Confidence Interval
In research, “skill
level” is equivalent
to the standard error
of the mean
The “radius” is
therefore
determined by our
confidence level
AND the standard
error
Steps in Creating the
Confidence Interval
1. Establish the center of the interval.
2. Obtain the t score appropriate for the
level of confidence.
3. Estimate the standard error.
4. Calculate the confidence limits in raw
score values by using the following
equation:
CI M = M ± (CVt )( SEM )
One-Sample Problem #1
The Scenario: An instructor is interested in
the exam performance of his students.
His newest class (N = 29) had a M =
88.14 and a SD = 10.53.
With 95% confidence, what is the likely
mean for the population exam score?
Calculating Our Example
CI M = M ± (CVt )( SE M )
Since CVt = 2.048 and the SE =
1.96, the 95% confidence interval
ranges from 92.14 to 84.14.
One-Sample Problem #2
The Psychology Department is interested
in learning about its statistics teachers. A
sample of 15 psychologists have a mean
teacher evaluation of 1.2 and a SD = .3.
With 99% confidence, what is the likely
mean for the population evaluation?
Interpreting the Meaning of
Confidence Intervals
Each interval
is an
estimate of
the
parameter
Not every
interval is
guaranteed
to include
the
parameter