review of factorial anova, correlations and reliability tests comm 420.8 fall, 2007 nan yu
TRANSCRIPT
Review of Factorial ANOVA, correlations and reliability tests
COMM 420.8
Fall, 2007
Nan Yu
Factorial ANOVA(ANOVAdata.sav)More than one IV, both are nominal,DV is interval or ratio-level
Behavioral intention will vary as a function of the type of ads that are featured on the web site and the gender of the participants.
IVs? DV?
Put the interval orratio-level variablehere. (DV)
Put the variablesrepresenting thegroups here. (IVs)
Then Click "Options"
Put the group variable in the right window. Then click "continue" and "Ok."
Select Compare main effects, Descriptive Statistics, Estimates of effect size, Homogeneity tests.
Tests of Between-Subjects Effects
Dependent Variable: binent-Behavioral Intention
26.478a 7 3.783 5.333 .000 .341
689.535 1 689.535 972.217 .000 .931
1.573 3 .524 .739 .532 .030
22.377 1 22.377 31.551 .000 .305
2.528 3 .843 1.188 .320 .047
51.065 72 .709
767.079 80
77.544 79
SourceCorrected Model
Intercept
condit
gender
condit * gender
Error
Total
Corrected Total
Type III Sumof Squares df Mean Square F Sig.
Partial EtaSquared
R Squared = .341 (Adjusted R Squared = .277)a.
Main Effects
Interaction
Test statisticsDegrees of freedom P-values
Significant main effect for Gender, F (1, 72) = 31.55, < . 001, partial η2 = .31
No significant Main Effect for Condition, F (3, 72) = .74, p =.53, partial η2 = .03
No significant Gender X Condition interaction, F (3, 72) = 1.19, p =.32, partial η2 = .05
Effect size
Creating Graphs
Highlight these two numbers with mouse, right click
Create Graph Bar
Significant main effect for Gender, F (1, 72) = 31.55, < . 001, partial η2 = .31
Male Female
gender-Gender of Respondent
0.000
1.000
2.000
3.000V
alu
es
2.407
3.465
Estimates
Dependent Variable : binent-Behavioral IntentionStatistics : Mean
Text Only Static Ad Animated Ad Pop-up Ad
condit-Condition
0.000
1.000
2.000
3.000
Val
ues
2.822
3.160
2.8132.949
Estimates
Dependent Variable : binent-Behavioral IntentionStatistics : Mean
No significant main effect for Condition, F (3, 72) = .74, p =.53, partial η2 = .03
How to Create Line Graphs for Interaction Effects?
3. condit-Condition * gender-Gender of Respondent
Dependent Variable: binent-Behavioral Intention
2.007 .266 1.476 2.538
3.636 .266 3.105 4.167
2.818 .266 2.287 3.349
3.502 .266 2.971 4.033
2.282 .266 1.751 2.813
3.344 .266 2.813 3.875
2.520 .266 1.989 3.051
3.377 .266 2.846 3.908
gender-Genderof RespondentMale
Female
Male
Female
Male
Female
Male
Female
condit-ConditionText Only
Static Ad
Animated Ad
Pop-up Ad
Mean Std. Error Lower Bound Upper Bound
95% Confidence Interval
Double click this table, then right
click
Select Pivoting Trays
Move the small square from
bottom to the left
Then, the means table will
look at this.
Highlight these numbers with mouse, then right click
Create Graph Line
Graphs for the Interaction Effects
Text Only
Static Ad
Animated Ad
Pop-up Ad
condit-Condition
Male Female
gender-Gender of Respondent
2.000
2.500
3.000
3.500
Val
ues
3. condit-Condition * gender-Gender of Respondent
Dependent Variable : binent-Behavioral IntentionStatistics : Mean
No significant Gender X Condition interaction, F (3, 72) = 1.19, p =.32, partial η2 = .05
These lines are not parallel, we can suspect that there might be interaction effects. But they are not statistically significant.
Correlations
Correlations indicate the strength and direction of the relationship between variables.
Correlation Coefficients
Dichotomous (dummy coded)Male=0Female=1
Low=0High=1
Hypothesis for correlations
People’s liking toward sad movies will be positively related to the liking toward horror movies.
Two variables, both interval.
SPSS and correlations
Go to Analyze Correlate Bivariate
SPSS and correlations
Put the two variablesin this box, click OK
Correlations
1 .276**
.002
124 124
.276** 1
.002
124 124
Pearson Correlation
Sig. (2-tailed)
N
Pearson Correlation
Sig. (2-tailed)
N
Liking of Movie Sad Films
Liking of Movie Horror
Liking ofMovie Sad
FilmsLiking of
Movie Horror
Correlation is significant at the 0.01 level (2-tailed).**.
r=.28, p<.01
Consistent with the Hypothesis, respondents’ rating of their liking of sad files was significantly positively associated with their ratings of horror movies, r=.28, p<.01.
Reliability Test
If multiple interval of ratio measures were used to measure one construct, we need to know whether these measures are “hanging together” or not.
Measurement of Fear:Scared 1 2 3 4 5Frightened 1 2 3 4 5Nervous 1 2 3 4 5
SPSS and Correlation
Please open Correlationdata.sav
Three items were used to measure “happiness”
happy 1 2 3 4 5 6 7
content 1 2 3 4 5 6 7
joyful 1 2 3 4 5 6 7
SPSS and Reliability Test
Go to Analyze Scale Reliability Analysis
Cronbach’s Alpha:
Reliability Statistics
.901 3
Cronbach'sAlpha N of Items
Item-Total Statistics
8.3468 11.269 .843 .823
7.7984 13.463 .753 .899
7.7903 12.135 .819 .844
Mood: happiness
Mood: content
Mood: joyful
Scale Mean ifItem Deleted
ScaleVariance if
Item Deleted
CorrectedItem-TotalCorrelation
Cronbach'sAlpha if Item
Deleted
If this number is greater than .70, meaning that three items
hang together well
Here you can see the changes of the reliability if one item is deleted,
1. Are both variables nominal?
2. Are both variables interval or ratio?
No (continue)
Chi-Square Test
No (skip to 3)
2a. Are you examining associations between variables? Correlation
2b. Are you comparing means of two variables?Paired T-Test
3. Assuming that some variables are nominal and some are interval / ratio...
3a. Are you comparing means between 2 groups? (IV is nominal with 2 levels, DV is interval or ratio)
ANOVA
3c. Are you comparing means as a function of more than 2 IV s (a factorial analysis)?
Factorial ANOVA
Decision Tree for Data Analysis
3b. Are you comparing means between more than 2 groups? (IV is nominal with more than 2 levels, DV is interval or ratio)
Independent sample T-test
In-Class Practice
Use the dataset of correlationdata.sav to
Test the following hypothesis:
Liking toward TV sports is positively related to total TV viewing per day. (tsports, totaltv)
Can we reject the null? (Report the correlations and significance levels).
Answers to In-Class Practice
Yes, we can reject the null.
r=.19, p<.05
Consistent with the hypothesis, liking toward TV sports is positively related to total TV viewing per day, r=.19, p<.05