Lecture: Independent Samples T-Tests
Experimental Design Foundations
Experimental research relies on structured methodologies to assess causal relationships by manipulating independent variables and observing their effects on dependent variables.
Between-Subjects vs. Within-Subjects Designs
Experimental designs are categorised primarily based on how participants are allocated across experimental conditions:
Between-Subjects Design (Independent Groups):
Different participants are assigned to each experimental condition.
Each participant contributes exactly one independent data point to the dataset; there is no overlap and no repeated measurements across conditions.
Example: Comparing the research methods examination scores of Year 1 students directly against the scores of Year 2 students.
Strengths: Eliminates order effects, practice effects, and fatigue effects because participants experience only a single condition.
Weaknesses: Statistically less powerful due to unsystematic individual differences (participant variance) between the distinct groups.

Within-Subjects Design (Repeated Measures):
The same participants take part in all experimental conditions.
Each participant serves as their own control, providing repeated measurements across levels of the independent variable.
Example: Measuring the same cohort of students on their Research Methods performance versus their Social Psychology performance.
Strengths: Statistically more powerful because individual baseline differences are held constant, reducing error variance.
Weaknesses: Highly susceptible to order effects, carryover effects, practice, and fatigue.

Continuous and Scale Data Characteristics
Inferential mean comparisons require the dependent variable to be continuous. In SPSS, continuous measurements are classified under the Scale measure:
Interval Data: Measured along a continuum where intervals between adjacent values are equal, but there is no absolute, non-arbitrary zero point (e.g., temperature in Celsius).
Ratio Data: Continuous measurement possessing an absolute true zero, allowing meaningful ratio comparisons (e.g., speed, time, word recall counts).
Parametric Representation: Continuous data are summarized through measures of central tendency and dispersion, specifically the mean ( or ) and standard deviation ( or ).
Skewed Data Representation: When distributions deviate from normality, data are summarized using the median and interquartile range ().
Theoretical Logic of T-Tests
Evaluating Differences in the Context of Variance
A common statistical question is whether the observed difference between two sample means is statistically significant or merely the product of random sampling variation. Evaluating whether two means differ requires evaluating the magnitude of the difference relative to the spread (variance) of the underlying distributions.

When two distributions display substantial spread and overlap heavily, identifying whether an individual score belongs to Condition 1 or Condition 2 is uncertain. Conversely, when the variance within each condition is narrow, the distributions show minimal overlap, making the identical numerical mean difference substantially more likely to reflect a genuine, statistically significant effect.


The Statistic as a Variance Ratio
The statistic represents a signal-to-noise ratio comparing between-condition variance to within-condition variance:
Numerator (Between-Condition Variance): The difference between group means, attributed to the manipulation of the independent variable.
Denominator (Within-Condition Variance): Uncontrolled variability resulting from individual differences, measurement error, and environmental noise.
Null Hypothesis Significance: The -test evaluates how likely the observed mean difference would occur if the null hypothesis () were true (i.e., if the true population distributions overlapped completely). If is sufficiently large, the observed effect is unlikely to be an artifact of chance, yielding a low probability value ().
Statistical Assumptions and Diagnostic Testing
Parametric vs. Non-Parametric Frameworks
Every inferential statistical procedure relies on specific mathematical assumptions regarding the distribution and structure of the data:
Parametric Tests: Statistical techniques that estimate population parameters and require data to meet stringent criteria (e.g., continuous measurement, normal distribution, homogeneity of variance). They have higher statistical power to detect an effect when their assumptions are fulfilled.
Non-Parametric Tests: Distribution-free tests that make fewer distributional assumptions. They transform continuous raw scores into ranks or assess medians, and serve as direct alternatives when parametric assumptions are violated.
The Assumption of Normality
Parametric tests require the continuous dependent variable to be normally distributed within each condition. A normal distribution exhibits a symmetrical bell-shaped curve where the highest frequency of observations clusters around the central mean, tapering off symmetrically toward the positive and negative tails.

Deviations from normality invalidate parametric probability calculations. Common violations include:
Negative Skew: Scores cluster at the upper end of the scale, creating an elongated tail toward the negative (lower) values (e.g., a test that is excessively easy, creating a ceiling effect).
Positive Skew: Scores cluster at the lower end of the scale, creating an elongated tail toward the positive (higher) values (e.g., a test that is excessively difficult, creating a floor effect).
Bimodal Distributions: The distribution features two distinct peaks, indicating that the sample comprises two distinct sub-populations.

The Shapiro-Wilk Test of Normality
While visual inspection of histograms provides an initial evaluation of distribution shape, formal inferential tests determine whether empirical data deviate significantly from a theoretical normal curve. The Shapiro-Wilk () test is the preferred metric, particularly for small-to-moderate sample sizes (compared to the alternative Kolmogorov-Smirnov test).
Statistical Question: What is the probability that the observed distribution occurred by chance from a normally distributed population?
Hypothesis Formulation:
: The sample distribution is not significantly different from a normal distribution ().
: The sample distribution is significantly different from a normal distribution ().
Decision Criterion:
(Non-Significant): Retain . There is no significant difference between the data and a normal distribution; the assumption of normality is met.
(Significant): Reject . The data deviate significantly from a normal distribution; the assumption of normality is violated.
Step-by-Step Execution of Normality Testing in SPSS
To conduct normality testing in SPSS:
Navigate the top menu:
Analyze>Descriptive Statistics>Explore....

In the
Exploredialog box, move the continuous dependent variable into the Dependent List: field.For between-subjects designs with multiple conditions, transfer the categorical independent variable into the Factor List: field.
Click the Plots… button on the right-hand side.

Under the
Descriptivepanel, uncheckStem-and-leafand check Histogram.Check the checkbox for Normality plots with tests.
Click
Continue, then clickOK.

Interpreting Normality Test Outputs and Histograms
The SPSS output window displays two primary tables for assessing distribution shape:
Descriptives Table: Displays the Mean, Confidence Interval for Mean (Lower and Upper Bound), Trimmed Mean, Median, Variance, Standard Deviation, Minimum, Maximum, Range, Interquartile Range, Skewness, and Kurtosis.

Tests of Normality Table: Features columns for both the Kolmogorov-Smirnov and Shapiro-Wilk tests, listing the test Statistic (), degrees of freedom (), and significance value (
Sig., which denotes the -value).

Exemplar Normality Comparisons
Normal Dataset (Assumption Met): Statistic = 0.987, df = 100, Sig. = 0.425 (p > 0.05 -> Normal)
Skewed Dataset (Assumption Failed): Statistic = 0.931, df = 41, Sig. = 0.015 (p < 0.05 -> Non-normal)

One-Sample T-Test
Purpose and Experimental Setup
A One-Sample T-Test evaluates whether the mean of a single continuous sample differs significantly from a known, predetermined threshold, benchmark, or population mean (e.g., assessing whether a cohort of students performs significantly better than a known national average of ).
Independent Variable: None (no experimental manipulation or control group).
Dependent Variable: One continuous variable measured once per participant.
Comparative Reference: A single, fixed numerical test value.
Formal Assumptions
Independence of Data: Each observation is independent; each participant contributes exactly one measurement.
Continuous Measurement: The dependent variable is measured at the interval or ratio level.
Normality: The sample distribution is normally distributed ( on the Shapiro-Wilk test).
Worked Example: Motorway Driving Speeds
An investigation measured the driving speeds of motorists on a UK motorway to test whether speeds significantly exceeded the legal motorway speed limit of .
Sample size:
Observed sample mean and standard deviation:
Hypothesized test value ():
Data Setup in SPSS
Data View: Structured with one row per participant.
Variable View: A single continuous column titled
Speed, set toScaleunder theMeasureproperty.

Verifying Normality
Running Analyze > Descriptive Statistics > Explore... on the Speed variable produces:
Shapiro-Wilk: Statistic , ,
Sig..Kolmogorov-Smirnov: Statistic , ,
Sig..Evaluation: Because , the speed data are not significantly different from a normal distribution; the assumption of normality is met.

SPSS Procedure for One-Sample T-Tests
Navigate to
Analyze>Compare Means>One-Sample T Test....

In the
One-Sample T Testdialog box, transfer the dependent variable (Speed) into the Test Variable(s): box.In the Test Value: box at the bottom, enter the comparative benchmark (e.g.,
70).Click
OK.

Interpreting and Reporting One-Sample T-Test Results
The test generates two primary output tables:
One-Sample Statistics:
One-Sample Test:
(reported as )
Confidence Interval of the Difference: Lower , Upper

Results Reporting Structure
Formal reporting of statistical analyses must follow this sequence:
Data Handling: Clarify data screening, extraction of descriptive parameters, and treatment of outliers.
Descriptive Statistics: State the mean and standard deviation (), including measurement units.
Test Identification & Assumptions: State the test utilized and confirm the status of assumption checks.
Inferential Statistics: Report the test statistic, degrees of freedom, and exact probability: .
Standard Model Report: "The mean speed of the participants was . The data met the assumptions of a one-sample -test, which revealed that this was significantly faster than the speed limit of (".
Non-Parametric Alternative: One-Sample Wilcoxon Signed-Rank Test
Rationale and Operational Differences
When continuous single-sample data violate the assumption of normality, the parametric One-Sample T-Test is invalid. In this scenario, the non-parametric equivalent is the One-Sample Wilcoxon Signed-Rank Test.
Mechanism: Non-parametric tests evaluate the median rather than the mean. Because the mean is sensitive to extreme values and skewness, the median provides a more robust indicator of central tendency for non-normal distributions.
Null Hypothesis: The population median of differences between data points and the hypothesized test value equals zero (or the sample median equals the hypothesized median).
SPSS Step-by-Step Procedure
Navigate to
Analyze>Nonparametric Tests>One Sample....

In the tabbed interface, select the Fields tab.
Ensure the dependent variable (
Speed) is positioned in the Test Fields: window.Select the Settings tab.
Select the radio button for Customize tests.
Check the box for Compare median to hypothesized (Wilcoxon signed-rank test).
In the Hypothesized median: field, enter the comparative value (
70).Click Run.

Interpreting Output and Median Visualization
The output window provides an automated hypothesis test summary alongside test statistics and median plots:
Hypothesis Test Summary:
Null Hypothesis: The median of Speed equals .
Test: One-Sample Wilcoxon Signed Rank Test.
Significance (
Sig.): .Decision: Reject the null hypothesis.
Test Summary Metrics:
Graphical Representation: Contrasts the observed sample median () against the hypothetical benchmark median ().


Independent-Samples T-Test
Purpose and Experimental Design Requirements
An Independent-Samples T-Test (Between-Subjects T-Test) compares the means of two mutually exclusive, unrelated groups to evaluate whether their population means differ significantly.
Design Structure: Between-subjects design with independent groups.
Independent Variable: Exactly one categorical factor comprising exactly two levels (dichotomous groups, such as experimental vs. control, noise vs. silence).
Dependent Variable: Exactly one continuous variable measured once per participant.
Assumptions Checklist
Independence of Observations: Data points must be collected from separate participants; no participant belongs to more than one group.
Continuous Measurement: The dependent variable must be interval or ratio.
Adequate Sample Size: A minimum sample size of participants per condition ( across two groups).
Normality: The dependent variable must be normally distributed within each condition independently.
Homogeneity of Variance: The variance of the continuous dependent variable must be equal (homogeneous) across both experimental groups.
Homogeneity of Variance & Levene's Test
Homogeneity of variance evaluates whether the dispersion of data points around the group mean is comparable across both conditions.


Evaluating homogeneity of variance involves three diagnostic steps:
Descriptive Comparison: Examining descriptive dispersion metrics across conditions (, variance, standard error, confidence intervals). Extreme disparities suggest potential violations.
Graphical Inspection: Visualizing distributions using histograms or boxplots to ensure equal dispersion width.
Inferential Testing via Levene's Test: Formal hypothesis testing for the equality of variances:
: (Variances between Condition 1 and Condition 2 are equal).
: (Variances between Condition 1 and Condition 2 are significantly different).
Interpretation: A non-significant result () indicates homogeneity of variance is met; a significant result () indicates homogeneity of variance is violated.
Quantile-Quantile (Q-Q) Plots
A Normal Q-Q Plot graphs the quantiles of the observed sample distribution directly against the theoretical quantiles expected from a standard normal distribution:
Ordered empirical observations are plotted along the x-axis, while expected normal values are plotted along the y-axis.
If data points fall roughly along the straight diagonal reference line, the data conform to the specified distributional properties.
Systematic curved departures, S-shapes, or severe scattering indicate skewness, kurtosis, or variance discrepancies.

Worked Example: Environmental Noise and Word Recall
An experiment tested whether environmental background noise impairs verbal memory retention:
Sample: participants allocated into two independent groups ( per condition).
Condition 1 (Noise): Participants memorized a list of words in a noisy environment.
Condition 2 (Silence): Participants memorized the identical list of words in complete silence.
Dependent Variable: Total number of words correctly recalled (
Score).
Data Setup in SPSS
One row per participant ( rows total).
Column 1:
Condition(Independent Variable, coded as1 = Noise,2 = Silence).Column 2:
Score(Dependent Variable, continuous scale).

Testing Assumptions: Normality per Group
Normality must be evaluated separately for each condition by placing Score in the Dependent List: and Condition in the Factor List:.
Noise Condition ():
Shapiro-Wilk: Statistic , ,
Kolmogorov-Smirnov: Statistic , ,
Silence Condition ():
Shapiro-Wilk: Statistic , ,
Kolmogorov-Smirnov: Statistic , ,
Conclusion: Both Shapiro-Wilk significance values exceed , satisfying the assumption of normality for each condition.

SPSS Step-by-Step Procedure
Navigate to
Analyze>Compare Means>Independent-Samples T Test....

Move the continuous dependent variable (
Score) into the Test Variable(s): list.Move the categorical factor (
Condition) into the Grouping Variable: box.

Click the Define Groups… button.
In the
Use specified valuesinput fields, type the numerical codes assigned to each condition (e.g.,Group 1: 1,Group 2: 2).

Click
Continue, then clickOK.
Evaluating Assumptions: Normality & Levene's Test Outputs
The Independent-Samples T-Test output includes Levene's Test for Equality of Variances alongside inferential test statistics:
Levene's Test Results:
Sig.()Interpretation: Because , the variances between the noise and silence conditions do not differ significantly. The assumption of homogeneity of variance is met.

Interpreting Results and Reporting Standards
Because Levene's test is non-significant, read statistics from the Equal variances assumed row.
Group Statistics:
Noise (): , ,
Silence (): , ,
Independent Samples Test Output (Equal variances assumed):
(
Confidence Interval of the Difference: Lower , Upper


Standard Model Report: "The mean number of words recalled was calculated for the noise () and silence conditions (). The data met the assumptions for an independent-samples -test, which revealed that significantly fewer words were recalled in the noise condition compared to the silence condition (".
Adjusting for Violations of Homogeneity of Variance
When Levene's test is significant (), the assumption of homogeneity of variance is violated. Rather than abandoning the parametric framework, SPSS provides an automated correction via Welch's -test.
Procedure: Read directly from the second row: Equal variances not assumed.
Effect on Degrees of Freedom: The degrees of freedom are adjusted downward to a non-integer decimal value to compensate for unequal variances ( in this dataset).
Comparative Row Metrics:
Equal variances assumed:
Equal variances not assumed:

Non-Parametric Alternative: Mann-Whitney U Test
Indications for Use
When data violate other parametric assumptions—such as non-normal distributions in one or both conditions, sample sizes smaller than per group, or ordinal-level measurement—the Mann-Whitney U Test is the required non-parametric alternative for between-subjects comparisons.
Mechanism: Rather than comparing group means, the Mann-Whitney U test pools all observations across groups, converts continuous scores into ranks ( to ), and evaluates differences in the mean ranks (and medians) between groups.
SPSS Step-by-Step Procedure (Legacy Dialogs)
Navigate to
Analyze>Nonparametric Tests>Legacy Dialogs>2 Independent Samples....

In the
Two-Independent-Samples Testsdialog box, move the continuous dependent variable (Score) into the Test Variable List:.Move the independent variable (
Condition) into the Grouping Variable: box.Click Define Groups… and enter group codes (
1and2). ClickContinue.Under the
Test Typeoptions, check the box for Mann-Whitney U.Click
OK.

Interpreting Mann-Whitney U Outputs and Reporting
The output displays rank distributions and inferential test statistics:
Ranks Table:
Noise (): ,
Silence (): ,
Test Statistics Table:
Decision: When exact significance is available, report the exact two-tailed value (). The significant result indicates that word recall was significantly lower in the noise condition compared to the silence condition.

Statistical Decision Summary & Core Pairings
Research Objective | Parametric Test | Required Assumptions | Diagnostic Test | Non-Parametric Equivalent |
|---|---|---|---|---|
Compare 1 Sample to Benchmark | One-Sample T-Test | Independent observations, Continuous scale, Normal distribution | Shapiro-Wilk () | One-Sample Wilcoxon Signed-Rank Test |
Compare 2 Independent Groups | Independent-Samples T-Test | Independent observations, Continuous scale, per group, Normal distribution per group, Homogeneity of variance | Shapiro-Wilk (), Levene's Test () | Mann-Whitney U Test (or Welch's -test if only homogeneity fails) |
Key Reporting Rules in APA Style
Report descriptive statistics as Mean and Standard Deviation () for parametric tests, and Median and Interquartile Range () or Mean Ranks for non-parametric tests.
Italicize statistical notation: , , , , , , , , .
Degrees of freedom are presented in parentheses immediately following the test statistic: .
Report probability values to three decimal places. When SPSS displays
Sig. = .000, report as (never as ).Always report two-tailed (two-sided) significance values unless a directional hypothesis has been pre-specified.