Title: Hypothesis Tests, Statistical Significance
1Hypothesis Tests, Statistical Significance
Correlation Coefficients
2What is a hypothesis?
An assumption subject to verification or proof
An educated guess about what is true in the
world
3Two Types of Hypotheses
- The Null Hypothesis Ho
- There is no difference between groups.
- There is no relationship between variables.
- Ho mmales mfemales
- vs.
- The Alternative Hypothesis Ha
- There is a difference between groups.
- There is a relationship between the variables.
- Ha mmales ? mfemales (for two-tailed test)
- mmales gt or lt mfemales (for
one-tailed test) -
4Statistical Significance
- Are results due to random sampling error or
chance? Or, is it unlikely that the results
observed are due to chance? - If the test results are statistically
significant, reject the null hypothesis and
conclude there are real differences - The researchers decision is subject to error
-
5Errors in Hypothesis Testing
- Type I Errors (called alpha, or a error)
- The error of rejecting the null hypothesis when
it is in fact true, or concluding there are
relationships or differences when none really
exist. - To avoid a Type I error a conservative alpha
level like .01 might be used. - Type II Error (beta errors)
- The error of retaining the null hypothesis when
it is in fact false, or - concluding there are no relationships or
differences when in fact they - do exist.
- To avoid a Type II error a liberal alpha level
such as .10 might be used -
6Standard Normal Distribution
7Graphic Demonstrating Hypothesis Testing for a
Two-Tailed Test
8Steps in Statistical Significance
(Hypothesis)Testing
- STEPS
- Step 1 Set alpha level ( a ) reflects level
of Type I error the researcher is willing to risk - Results must have probability of error equal
to or less than alpha before the researcher will
reject the null hypothesis and conclude the
results are statistically significant
9Steps in Statistical Significance Testing
- Step 2 Conduct the appropriate data analysis
procedures for the test - Step 3 You will get a TEST statistic
(coefficient, chi square, t value, F value, etc.)
that measures how close the sample has come to Ho - Step 4 Look at the probability (p-value) of
error associated with getting your test statistic - Step 5 Compare the p-value to your alpha level
-
10The Research Decision
- Retain the null if the p-value is GREATER than
your alpha level - For example, if a .05 and p .10, retain the
null and conclude results are NOT statistically
significant - Reject the null if the p-value is equal to or
LESS than your alpha level, conclude Ha - For example, if a .05 and p .001, reject the
null, results are statistically significant
11Practical vs. Statistical Significance
- Results can sometimes be statistically
significant, but the difference or strength isnt
enough to be of any practical consequence - Statistical significance is easier to obtain when
sample is large (because SE is lower)
12Bivariate Measures of AssociationThe Pearson PM
Correlation Coefficient (r)
- Relationship, not causation, between 2 variables
- Both variables must be INTERVAL level
- Measures the direction degree of a relationship
Direction is positive or negative - POSITIVE The variables move in the same
direction (i.e., when one is high (low) the other
is high (low) - NEGATIVE The variables move in OPPOSITE
directions (i.e., when one is high the other is
low)
13The Pearson PM Correlation Coefficient ( r )
- Can range from 1.00 to 1.00
- Degree Generally r gt .60 considered strong,
between .40 and .60 moderate, r lt .40 weak (lt .10
to 0.00 no relationship) - A correlation of 0.00 indicates no linear
relationship between the variables - Coefficient of determination (r2) shows
proportion of change in one variable explained by
change in another (explained variance)
14Correlation Example
- Example A interpretation
- There is a strong, positive relationship between
yrs in school and income level (r .90, p
.000). More years in school is associated with
higher income level. - Example B interpretation
- There is a strong, negative relationship between
income level and of children (r -.72, p
.000). Lower income households tend to have more
children in the home. - Example C interpretation
- There is no relationship between income level and
weight.
151-way 2-way Chi Square (Crosstabs)
16When to use 1-Way Chi Square (c2)Goodness of
fit
- Analysis of frequency distributions to test
whether distribution of responses fits
presupposed proportions in population - Univariate, the variable is categorical (nominal
or ordinal), i.e, check only one type questions
- What did you like best about your dining
experience today?
17The Statistical Question
- There will likely be differences in responses to
the different categories. - For example What did you like best about us
today? - 43 said quality of food
- 34 said fast, friendly service
- 23 said restaurant atmosphere
- But are these differences statistically
significant? In other words, do these
differences reflect attitudes of the population,
or just this particular sample?
18Example 1-way Chi Square
- Research question Are Dominos, Pizza Hut, and
Papa Johns pizza equally preferred among UNM
students? - Questionnaire question Which of the following
do you prefer, Dominos, Pizza Hut, or Papa
Johns? (check one) - You sample 105 UNM students
19Hypothesis Testing
- Ho pd pph ppj
- (i.e., if preferences equal, expected value
should be 35/105 or 33.3 for each) - Ha pd ? pph ? ppj
- Assume an alpha of .05
20SPSS for 1-way Chi Square
- Analyzegtnonparametric testsgtchi square
- Move your variable to the test variable list and
hit OK - There is a difference in how many people actually
said they preferred each and how many were
expected (35) - The test statistic, c2 2.46, df 2
- Is this a big enough difference to make an
inference, or could differences be due to random
sampling error?
21Conclusion
- For c2 2.46, df 2, and a .05, the pvalue
would need to be equal to or less than .05 to be
significant. - p .293, Our conclusion here?
- Retain the null. There is a slight difference in
preferences but it is not a statistically
significant difference and could be due to random
sampling error.
22When to use2-way Chi Square (c2)
- For compare group research questions
- Involving 2 variables, the grouping variable and
whatever variable you wish to compare the
subgroups on (bivariate) - When both variables are categorical, i.e.,
nominal or ordinal
23Example 2-way Chi Square
- Research question Are Dominos, Pizza Hut, and
Papa Johns pizza equally preferred by male and
female UNM students? - Questionnaire question Which of the following
do you prefer, Dominos Pizza Hut, or Papa
Johns? (check one) - AND Gender Male Female
- You sample 105 UNM students
24Significant Test for 2-way Chi Sq
- Ho Males and females have equal preferences for
Dominos, Pizza Hut, and Papa Johns pizza. - Ha Males and females do not have not equal
preferences for Dominos, Pizza Hut, and Papa
Johns pizza. - Set alpha for testing here, a .01
25AnalyzegtDescriptive StatsgtCrosstabs
- Put one variable in Row and the other in Column
- Request observed expected counts, row, column
percentages (dont need total) so you will have
all the information you need to interpret each
cell
26Interpretation
- Look at the overall test statistic (c2 14.01, p
.001) - Conclusion
- Do you reject the null or accept the null at a
.01? - Reject null. There is a significant difference
between males females in terms of their pizza
preferences. - Now, interpret further by looking at the cells.
Decide which way you are reading (down or across
doesnt matter) and use the frequency
percentages to describe what is different between
the groups (NOT within a group!). Look at cells
with big difference between observed and expected
values.
27Interpreting Percentages for 2-way Chi square
- Always start with of the.. and insert either
your row or column title. If you are reading
rows, read across within the row variable. If
you are reading columns, read down within the
column variable - Conclusion It appears males prefer Dominos and
Papa Johns and females prefer Pizza Hut.
Reading down columns Of those preferring
Dominos, 75 are male and 25 are female. Of
those preferring Pizza Hut, 34.1 are male and
65.9 are female. Or read across rows Of the
males, 45.8 prefer Dominos as opposed to only
19.6 of females. Of the females, 58.7 prefer
Pizza Hut as opposed to only 23.7 of males.