Two-Way Factorial ANOVA in SPSS: Model Setup, Diagnostic Tests, Post Hoc Analysis, and Interaction Plots
Introductory Classroom Anecdotes and Banter
- Humor and wordplay shared prior to data analysis:
- A joke regarding the Seine: A man jumped off a bridge in Paris because he was "insane" (in Seine).
- A joke regarding the classic riddle: Why did the chicken cross the road? The realization that "getting to the other side" signifies dying.
- A joke regarding mycology: Why did the fun guy leave the party? Because there was "not mushroom" (not much room).
- An incomplete reference regarding pop culture: Why the Goblin King was afraid of David Bowie.
- Graveyard shortcut anecdote:
- Walking to class from the Killian Building, taking the customary shortcut through the graveyard.
- A skeleton suddenly emerges from a grave and begins giving chase, moving faster than the runner.
- Upon reaching the graveyard gate, the skeleton catches up, delivering a hug and a kiss because it was an "exoskeleton."
Dataset Structure for Isaac's Factorial ANOVA Study
- Overview and context:
- The dataset represents the exact experimental data collected and analyzed by Isaac to execute his memory study.
- The study design involves two independent variables, requiring two dedicated grouping variables in the spreadsheet layout, alongside a single outcome measure.
- Column configuration:
- Column 1 (
Age grouping variable): Coded using numerical identifiers 1 and 2.- Value 1 denotes older adult participants.
- Value 2 denotes younger adult participants.
- Column 2 (
Condition grouping variable): Coded using numerical values 1 through 5, representing the five discrete experimental memorization conditions. - Column 3 (
Score dependent variable): Represents the total number of words memorized by each participant.
- Data preparation requirements:
- When setting up grouping variables in statistical software, all numerical values must have descriptive labels explicitly assigned to preserve interpretive clarity.
SPSS GLM Univariate Setup: Fixed vs. Random Factors
- Navigation pathway in SPSS:
- Select
Analyze → General Linear Model → Univariate. - Move the outcome measure ("words memorized") into the field labeled
Dependent Variable. - Move both grouping variables (
Age and Condition) into the box labeled Fixed Factor(s).
- Distinction between Fixed Factors and Random Factors:
- Factor classification alters mathematical calculations and substantively changes output interpretation.
- Fixed Factor:
- Employed when an investigation examines the entire universe of interest for the independent variable.
- Analyzes every single level of the independent variable present in the research framework (e.g., investigating all five memorization conditions established in the design).
- Random Factor:
- Employed when an investigation examines only an arbitrary or sampled subset of all possible levels of the independent variable.
- For example, choosing to sample and evaluate only levels 1, 3, and 5 out of the total five memorization conditions.
- For this specific study, and across standard factorial designs, both independent variables represent the complete universe of interest and are specified as fixed factors.
Post Hoc Testing Configuration and Pairwise Restrictions
- Accessing post hoc tests:
- Click the button labeled
Post Hoc on the univariate dialog interface.
- Methodological rule regarding factor levels:
- Post hoc tests cannot be run on independent variables containing only two levels (such as
Age, which only consists of young versus old). - With only two levels, only a single pairwise comparison exists; thus, pairwise post hoc tests are mathematically redundant and SPSS will trigger an error message if attempted.
- Factors with more than two levels require post hoc testing to identify specific group differences.
- For the
Condition factor containing k=5 levels, the total number of pairwise comparisons is:
(25)=25×4=10
- Software execution steps:
- Transfer only the
Condition variable from the factor list on the left to the right-hand testing box. - Do not transfer the
Age variable. - Select
Tukey (Tukey's Honestly Significant Difference / HSD) from the list of available post hoc tests. - Click
Continue on the bottom left to return to the primary univariate dialog.
Generating Interaction Profile Plots
- Setting up profile plots:
- Click the button labeled
Plots at the top right of the primary dialog interface to create interaction plots. - Generating two complementary plots allows researchers to evaluate which visualization illustrates the interaction effect more clearly.
- Plot 1 configuration:
- Move
Condition into the box marked Horizontal Axis. - Move
Age into the box marked Separate Lines. - Click
Add beneath the lines box, producing an interaction plot specification labeled condition * age.
- Plot 2 configuration:
- Move
Age into the box marked Horizontal Axis. - Move
Condition into the box marked Separate Lines. - Click
Add, producing a reverse specification labeled age * condition.
- Completing plot setup:
- Ensure both
condition * age and age * condition appear in the plots list. - Click
Continue in the bottom left.
Display Options and Output Diagnostics: Sample Sizes, Descriptives, and Levene's Test
- Display options setup:
- Click the
Options button at the top right of the primary dialog. - Within the group box labeled
Display, select the four primary diagnostic check boxes (including descriptive statistics, estimates of effect size, observed power, and homogeneity tests), ignoring all unneeded settings below. - Click
Continue, then click OK to generate output.
- Diagnostic Output Tables:
- Between-Subjects Factors:
- Summarizes sample sizes across all factor levels.
- For
Age, confirms N=50 for older adults and N=50 for younger adults (Ntotal=100). - For
Condition, confirms N=20 across each of the five memorization conditions (20×5=100). - Confirms balanced cell sizes without accidental data entry errors.
- Descriptive Statistics:
- Reports cell means and standard deviations across all 10 factor combinations (old and young across all five memory conditions).
- Levene's Test of Equality of Error Variances:
- Serves as an "apples to apples" test of distribution equivalence across all 10 cells in the experimental grid.
- Null hypothesis asserts that error variance is equal across all cells.
- Evaluating the top row ("Based on Mean") yields a significance value of p=0.02.
- Because p=0.02<0.05, the assumption of homogeneity of variance is violated, demonstrating that at least one of the ten cells has an unusual or non-equivalent distribution relative to the other nine.
Interpreting the Tests of Between-Subjects Effects
- Reading the ANOVA table:
- The initial two rows (Corrected Model and Intercept) can be set aside for advanced graduate coursework.
- Rows three, four, and five report the core factorial ANOVA results:
- Row 3: Main effect of
Age. - Row 4: Main effect of
Condition. - Row 5: Interaction effect of
Age \times Condition (Age * Condition).
- All three tests yield statistically significant results (p<0.05), perfectly aligning with hand calculations.
- Effect size benchmarks (Partial Eta Squared, ηp2):
- Standard interpretation guidelines:
- Small effect threshold: ηp2≈0.01
- Moderate effect threshold: ηp2≈0.25 (with intermediate marks around 0.10)
- Large effect threshold: ηp2≥0.40
- Model effect evaluations:
- An effect size around 0.25 represents a moderate effect.
- An effect size reaching or exceeding 0.40 represents a very large effect.
- An effect size falling below 0.25 but above small thresholds represents a small-to-moderate effect.
- Statistical power:
- Observed power across the tests approaches 1.0 (nearly perfect power), indicating negligible risk of committing a Type II error.
Post Hoc Pairwise Comparisons Analysis
- Interpreting homogeneous subsets (Tukey HSD table):
- Pairs of group means that share the same subset column are not statistically significantly different from one another.
- Pairs of group means located in different subset columns are statistically significantly different from one another.
- Pairwise comparison outcomes:
- Among the 5 memorization conditions, there are 10 total pairwise comparisons.
- Exactly 8 of the 10 pairwise comparisons reach statistical significance.
- Exactly 2 pairwise comparisons fail to reach significance:
- The difference between counting and writing.
- The difference involving intention.
Questions & Discussion: Evaluating Interaction Visualizations
- Audience Inquiry: What are the specific age cutoff criteria to fall into the older versus younger category?
- Response: The operational definitions and age cutoff criteria were established and covered during the preceding Wednesday class session.
- Audience Exchange: Comparing the two interaction plots:
- Plot Structure 1: Two lines representing
Age (young vs. old) across the five memorization conditions arranged on the horizontal x-axis. - Plot Structure 2: Five lines representing the five
Condition levels plotted across the two Age categories on the horizontal x-axis. - Student Feedback regarding Plot 2 (Five lines across Age):
- One student preferred having the older group clustered on one side and the younger group on the other, though noted it forced their brain to think more intensely.
- Another student noted preference for its appearance, feeling it made more sense visually.
- Another student stated it was easier to absorb the data at a single glance and separated the categories neatly.
- Student Feedback regarding Plot 1 (Two lines across Condition):
- One student favored Plot 1 because it presents data left-to-right rather than bottom-to-top, making direct condition differences clearer.
- Another student observed that differences between experimental conditions stand out more directly across the two lines.
- Student Sarah noted that Plot 1 is significantly "not as busy."
- Communication Synthesis:
- Plot 1 exemplifies the principle of "maximum information, minimum aid": it conveys the identical underlying data with only two moving lines rather than five.
- While Plot 2 is technically accurate, its increased visual density forces the audience to expend higher cognitive effort to decode the findings.
- Effective data communicators must balance personal graphical preferences against audience comprehension needs.
- Audience Inquiry: Is it necessary or beneficial to show both charts simultaneously?
- Response: While presenting both charts is technically possible, doing so introduces visual redundancy because both figures display identical mathematical coordinates and contrasts. Dual presentation should be reserved solely for instances where an audience member specifically struggles to comprehend the primary chart and requests an alternative viewpoint.
Principles of Data Communication and Concluding Remarks
- Core presentation standard:
- If a graphical relationship can be explained in two sentences rather than seven sentences, always choose the two-sentence visualization.
- A communicator must set aside personal preferences and prioritize the format that maximizes clarity for the audience.
- Concluding cat ghost story:
- In the middle of the night, inside a closed bar where the bartender is sweeping up, the ghost of a cat enters carrying its severed tail in its mouth.
- The ghost cat jumps onto the bar, drops its tail, and explains that exactly one year ago today, it was run over and killed outside the establishment.
- Explaining that it is cursed to wander the earth in torment until its tail is reattached, the cat asks the bartender for assistance.
- The bartender replies: "I'd really like to. But, unfortunately, I am not allowed to retail spirits after 2 AM."
- Course deadline reminder:
- All scheduled assignments and submissions must be finalized and submitted prior to the midnight deadline.