STATISTICAL INFERENCE

Dr. Samuel Adjorlolo

 OUTLINE
• Fundamentals of Statistical Inference

• Hypothesis Testing
 INTRODUCTION

Inferential statistics allow researchers to draw

conclusions about population parameters, based on
statistics from a sample.

Inferences are based on likelihood, hence there is a

risk of error
 Fundamentals of Probability
• There are two mutually exclusive possibilities in testing the
effectiveness of transcutaneous nerve treatment for
alleviating pain;

The experimental treatment is not effective in reducing pain

• This is called the null hypothesis (H0)

The experimental treatment is effective in reducing pain

• This is called the alternative hypothesis (H1)
 Probability of (consecutive) Event
• P(event) = number of ways the specified event can
occur/total number of possible events

• Probability of obtaining heads when a coin is tossed is 1/2 =

0.5

• The probability of obtaining 6 when a die is played is 1/6 =

0.17
 Probability of Consecutive Events
• The multiplicative law of probability:
p(A and then B) = p(A)*p(B);
• A is the first independent event
• B is the second independent event

• Probability of obtaining two consecutive heads when a coin is

tossed = 0.5*0.5 = .25
• What is the probability of obtaining ten consecutive heads?
Toss No. Possible Outcomes and Probabilities
1 H = .500
2 HH = .250 This table can be used to test the
3 HHH = .125
hypothesis that a coin tossing is
4 HHHH = .063
biased
5 HHHHH = .031
H0: The coin is fair
6 HHHHHH = .016
H1: The coin is biased
7 HHHHHHH = .008
8 HHHHHHHH = .004
What is the probability of
9 HHHHHHHHH = .002
obtaining 10 consecutive heads?
10 HHHHHHHHHH = .001
• We can conclude that the coin tossing is bias.

• This analogy extend to the research arena where

researchers often based their decisions by consulting
some probability tables or values.
 Sampling Distribution and Error
• A sample statistic does not necessarily equate the
corresponding population parameter because of sampling
error

• Sampling error reflects the tendency of statistics to fluctuate

from one sample to another

• It is the difference between the obtained sample value and

population parameter
 Central Limit Theorem
• Stipulates that the mean of the sampling distribution is
identical to the population mean of raw scores.

• That is, as sample size increases, its mean approximate the

population mean
Standard Error of the Mean (SEM)
• It is referred to as the standard deviation of a sampling
distribution of the mean.

• The error implies that various sample means of a

population contain some error as estimates of the
population mean.

• Standard signifies that SEM is an index of the average

amount of error for all possible sample means
 Standard Error of the Mean (SEM)
• The smaller the SEM, the more accurate are the
sample means as estimates of the population value.
Example:
• Sx̅ = SD/√N
Let SD = 5 and N= 25
• Where
– Sx̅ = estimated SEM SEM= 1
– SD = standard deviation of the sample
– N = number of cases in the sample How can we decrease
the SEM?
 ESTIMATION OF PARAMETERS
• Inferential statistics address two broad goals:

• Estimate the value of population parameters

– can be either point estimates and/or interval
estimates

• Test hypothesis (most widely used)

 Point Estimation
• Calculating a single value to estimate the parameter.
– E.g., a mean of 10 could be taken as representing a
population mean

• Major challenges
– It offers no context for interpreting their accuracy

– It gives no information regarding the probability that it is

correct or close to the population mean.
 Interval Estimation
• Calculating a range of value that has a high probability of
containing the population value.

• This involves constructing a confidence interval (CI)

with values at the boundaries serving as confidence
limits

• E.g., CI: 59-63 (59 and 63 are lower and upper limits
respectively).
 Calculating Confidence Interval
• When the population SEM is known
• In calculating CI, confidence level, sample mean and
SEM are used.
When 99% is used
instead to reduce error
• E.g., 95% CI = x̅ ± (1.96* σx̅ )
by substitution
• Let x̅ = 61 and population SEM = 1
99% CI = 61 ± (2.58*1)
• 95% CI = 61 ± (1.96*1)
• 95% CI = 59.04≤µ≤62.96 99% CI = 58.42≤µ≤63.58
 Calculating Confidence Interval
• When the population SEM is not known
• t distribution is used to calculate the CI

• t distribution is similar to normal distribution but the

former is affected by sample size.

• There is t distribution for different sample sizes and

confidence level.
 Calculating Confidence Interval
• When the population SEM is not known
• E.g., 95% CI = x̅ ± (t* sample SEM)

• Let N= 25, SD = 5, x̅ = 61 and sample SEM = 1, t = 2.06

• 95% CI = 61 ± (2.06*1)

• 95% CI = 58.94≤µ≤63.06
 HYPOTHESIS TESTING
• Begins with the assumption that H0 is true, and it
involves making a decision to accept or reject H0
• H0 always state absence of, no difference, no effect, or
innocence

• Only H1 is stated in research work by researchers

• Researchers are interested in H1 but only via testing Ho

 ERRORS IN RESEARCH

True Ho False Ho

Accept Ho Correct decision Type II error

Probability = 1- α Probability is β
Correct decision
Type 1 error Probability = 1 - β
Reject Ho
Probability = α
 Controlling the Risk of Type I Error
• Type I errors can be controlled through level of
significance (α)

• α =.05 implies the risk of rejecting a true H0 five times in

100 samples.

• α=.01 implies the risk of rejecting a true H0 once in 100

samples.

• How about α = .001?

Controlling the Risk of Type II Error
• This is affected by several factors;
Sample size
Research design
Type of statistical test
Strength of relationship between variables
 DECREASING TYPE I & II ERRORS
• There is a trade-off between decreasing type I or type
II errors

• Establishing a strict α to decrease

type I increases the risk of type II
error
 STEPS IN HYPOTHESIS TESTING
Determine the appropriate test statistic
Establish the level of significance
Determine to use one-tailed or two-tailed test
Calculate the test statistic
Determine the degrees of freedom
Compare the computed test value against a tabled
value
 Determine Appropriate Test Statistic
• The selection of statistical test depends on several
factors;
The nature of comparison
The number of groups being compared
Level of measurement of IV and DV
Statistical assumptions
 Establish the Level of Significance

• Determine the criterion for the decision to reject H0

before the analyses are undertaken.
• e.g., α = .05, .01 .001
Determine Whether One-tailed or Two-Tailed Test

• Two tailed test uses both tails of a sampling distribution to

determine the critical region for rejecting H0
– E.g., There would be a significant difference between males and
females in medical nursing exam (non-directional
hypothesis)

• A one-tailed test uses one tail of a sampling distribution to

determine the critical region for rejecting H0
– E.g., Males will perform significantly better than females in
medical nursing exam (directional hypothesis)
 Calculate the Test Statistic
• Compute the value of the test statistic using the right
formula.

• The value of the test statistic is called “statistic

obtained”
– E.g., t-obtained, F-obtained
 Determine the Degrees of Freedom
• Degree of freedom refers to the number of
components that are free to vary about a parameter.
 Compare Computed test value against
Critical Value
• If the absolute value of computed statistic is greater
than the tabled value, the result is statistically
significant at the specified probability level; if the
computed statistic is smaller, then the results are
nonsignificant.
 Hypothesis Testing With Statistical
Software
• Rather than comparing the computed statistic with
critical values to reject or retain Ho, statistical software
provides the exact probability levels for each statistical
analysis.

• The decision to reject or retain Ho is made by comparing

the probability level obtained to a predetermined
statistical significance level