PSY 315 Advanced Data Analysis for Psychology Study Guide (Stellenbosch): Exam Notes for Statistical Inference, Regression, ANOVA, and Interpretation

Psychology 315 at Stellenbosch University typically pushes beyond descriptive statistics into the logic of inference, model comparison, and the disciplined interpretation of psychological data. This study guide brings together the methods, assumptions, formulae, and exam-facing applications most likely to matter in an advanced undergraduate assessment. It is written for students who need a practical, structured revision document that links theory to analysis, reporting, and common pitfalls in psychology research.

1. What PSY 315 Covers and Why It Matters in Psychology

Advanced data analysis in psychology is not simply about calculating numbers correctly. It is about answering psychological questions with enough precision that the conclusions remain believable, explainable, and useful. In a course such as PSY 315 Advanced Data Analysis for Psychology at Stellenbosch University, the focus usually moves from basic summaries and introductory hypothesis testing toward deeper reasoning about sampling, uncertainty, effect size, model fit, and the assumptions that support statistical conclusions. The exam is therefore unlikely to reward memorisation alone. It rewards the ability to choose a suitable analysis, justify it, interpret output in plain language, and recognise when results are misleading.

The role of statistics in psychological science

Psychological data are often noisy, incomplete, and shaped by human behaviour rather than mechanical processes. This means the analyst must work with variability rather than pretend it does not exist. A test score, reaction time, attitude rating, or therapy outcome can shift because of many factors at once: measurement error, mood, fatigue, social desirability, age, culture, experimental manipulation, and chance. Advanced data analysis helps separate the signal from the noise.

The central statistical questions in psychology include:

Is there a difference?
Example: Do students who receive a mindfulness intervention report lower test anxiety than those who do not?
Is there a relationship?
Example: Is higher perceived stress associated with poorer sleep quality?
Can one variable predict another?
Example: Can early semester attendance predict final marks?
Do several variables jointly explain an outcome?
Example: Do stress, sleep, and coping style together predict depressive symptoms?
Are group differences still present when other variables are controlled?
Example: Does gender still predict anxiety after accounting for age and social support?

These are the kinds of questions that advanced statistical tools answer more efficiently than simple descriptive statistics. The choice of test, however, depends on design, measurement level, and the specific research question.

Typical course emphasis in an advanced psychology data analysis module

Although exact content can vary from year to year, a module like PSY 315 generally expects competence in the following areas:

Sampling and inference
Hypothesis testing
Effect sizes and confidence intervals
Correlation and simple linear regression
Multiple regression
Analysis of variance and related comparisons
Assumption checking and diagnostic reasoning
Interpretation of statistical software output
Basic research design logic

A student who understands these areas can move between theory and practice. For example, a regression output is not just a table of coefficients. It is evidence about how much one predictor contributes, whether the model improves prediction, and whether assumptions such as linearity and independence are plausibly satisfied.

Why advanced analysis is especially important in psychology

Psychology often deals with small-to-moderate effect sizes. Many important phenomena are not dramatic. A therapy may reduce symptoms by a modest but meaningful amount. A correlation between study habits and performance may be noticeable without being near perfect. Statistical significance can therefore be misleading if it is treated as the only criterion. A very small effect can be statistically significant in a large sample but practically trivial. A meaningful effect can fail to reach significance in a small sample simply because the study lacked power.

This is why exam answers in psychology should usually refer to more than one standard of evidence:

Statistical significance
Magnitude of effect
Direction of effect
Precision of estimate
Theoretical plausibility
Research design quality

A strong answer often sounds like: “The relationship is statistically significant, but the effect size is small, so the practical importance should be interpreted cautiously.” That sentence reflects the kind of balanced judgement that advanced psychology courses expect.

Key vocabulary to know cold

Several terms recur across most of the course:

Population: the full group of interest
Sample: the subset actually studied
Parameter: a population value, usually unknown
Statistic: a sample value used to estimate a parameter
Null hypothesis (H₀): usually no effect or no difference
Alternative hypothesis (H₁): an effect or difference exists
Alpha (α): the probability of Type I error, often 0.05
Type I error: rejecting a true null hypothesis
Type II error: failing to reject a false null hypothesis
Power: the probability of detecting a true effect
Effect size: the magnitude of a difference or relationship
Confidence interval: a range of plausible values for a parameter

These terms are foundational. If they are confused, later topics such as ANOVA or regression become much harder to understand.

The logic of evidence in a psychology exam

A good exam answer usually moves through four layers:

Identify the design or question
Select the correct analysis
Interpret the output correctly
State the psychological meaning

For instance, if two therapy groups are compared on anxiety after treatment, the likely choice is an independent-samples t-test or one-way ANOVA if more than two groups are included. The answer should then explain what the p-value means, whether assumptions are acceptable, what the effect size says, and what the result implies for therapy effectiveness.

Students often lose marks because they describe the numbers without explaining the psychological conclusion. A result should always be linked back to the question that generated it. Statistical language and psychological language must both appear.

2. Core Statistical Foundations: Variables, Distributions, Sampling, and Assumptions

Before analysis can be interpreted, the basic structure of the data must be understood. The best psychologists do not merely apply a test because software can produce an output. They ask whether the variables are measured appropriately, whether the sample is adequate, and whether the assumptions of the analysis are plausible. These foundations are essential in PSY 315 because many later topics are simply extensions of them.

Types of variables and measurement levels

A variable is anything that can vary across individuals, trials, or groups. In psychology, variables may be measured at different levels:

Level of measurement	Description	Examples in psychology	Suitable summaries
Nominal	Categories with no intrinsic order	Gender identity, diagnosis group, treatment condition	Counts, proportions, mode
Ordinal	Ordered categories, unequal spacing possible	Likert responses, rank order, severity levels	Median, percentiles, frequencies
Interval	Equal intervals, no true zero	Standardised test scores	Mean, standard deviation
Ratio	Equal intervals and a true zero	Reaction time, number of sessions attended, age	Mean, SD, ratios

In practice, psychology often treats Likert-type scales as approximately interval when several items are combined into a scale score. This is a pragmatic decision, not a claim that the data are truly interval in the philosophical sense. A good exam answer recognises this nuance.

Independent, dependent, and covariate variables

Independent variable (IV): the predictor, grouping factor, or explanatory variable
Dependent variable (DV): the outcome being explained or predicted
Covariate: a control variable included to adjust for other influences

For example, if the effect of a stress-reduction workshop is studied, the workshop condition is the IV and anxiety score is the DV. If baseline anxiety is included to improve precision, it becomes a covariate.

A recurring exam skill is identifying whether a variable is used as:

a grouping factor,
a continuous predictor, or
a control variable.

The same variable can play different roles in different analyses.

Populations, samples, and sampling error

A sample never matches a population perfectly. Even when selected carefully, it will differ by chance. That mismatch is sampling error. Statistical inference exists because researchers must estimate population patterns from partial information.

If a class of 60 students takes a memory test and the sample mean is 72, that number is only an estimate of the broader population mean. Another sample could yield 69 or 75 simply because of random variation. This is why confidence intervals and significance tests are needed. They quantify how uncertain the estimate is.

Important consequences of sampling error:

Small samples tend to have more variability in estimates.
Larger samples tend to produce more stable estimates.
A statistically significant result in a tiny sample may still be unstable.
Non-random samples may produce biased estimates, not just noisy ones.

Distributions and what they tell you

A distribution describes how scores are spread across values. Exam questions often require recognising or describing the shape of a distribution.

Common distribution features include:

Central tendency: mean, median, mode
Dispersion: range, variance, standard deviation
Shape: symmetric, positively skewed, negatively skewed, bimodal
Outliers: extreme values that may distort analysis

Normal distribution

The normal distribution is bell-shaped and symmetric. Many inferential tests assume approximate normality of errors or residuals. In psychology, variables such as cognitive scores, composite scale scores, and many biologically influenced measures may be roughly normal, while variables like reaction time, symptom counts, and income are often skewed.

Skewness in psychological data

Positive skew: long tail to the right; common for reaction times and symptom counts
Negative skew: long tail to the left; possible in high-performing samples where many scores cluster near the top
Bimodality: two peaks; may indicate two subgroups, such as clinical and non-clinical participants

Understanding skew matters because mean-based analyses can be distorted by extreme values in skewed distributions. That does not always invalidate the analysis, but it signals the need for caution or transformation.

Key assumptions in psychological statistics

Most advanced analyses rely on assumptions. These assumptions are not always perfectly met, but they must be considered.

1. Independence

Observations should not influence one another. If one participant’s score depends on another’s response, the independence assumption is violated.

Examples of possible non-independence:

Students working together in groups
Repeated measures on the same person
Twins or family members in a dataset
Patients nested within therapy groups

Non-independence matters because it can make standard errors too small, which inflates false positives.

2. Normality

Some tests assume that residuals or score distributions are approximately normal. Many tests are fairly robust to moderate deviations, especially with larger samples, but heavy skew or outliers can still cause problems.

3. Homogeneity of variance

When comparing groups, the spread of scores should be similar across groups. If one group is much more variable than another, a standard t-test or ANOVA may be less trustworthy.

4. Linearity

For correlation and regression, relationships should be approximately linear. If the true relationship is curved, a linear model may miss it.

5. Homoscedasticity

The spread of residuals should be similar across levels of the predictor. If residual variance increases as predicted values increase, the model may be misspecified.

6. Absence of multicollinearity

In multiple regression, predictors should not be so highly correlated that they become redundant. If two variables measure nearly the same thing, their individual effects become difficult to separate.

How to think about assumptions in an exam

It is not enough to list assumptions mechanically. Good exam answers explain what would happen if an assumption fails.

For example:

If normality is violated mildly, the test may still be usable with caution.
If independence is violated, the p-value may be unreliable.
If variances are unequal and group sizes are very different, the risk of error increases.
If predictors are highly collinear, regression coefficients may become unstable.

This is the kind of reasoning that distinguishes advanced analysis from introductory formula recall.

3. Hypothesis Testing, Confidence Intervals, Effect Sizes, and Practical Significance

A major theme in advanced statistics is the difference between evidence, uncertainty, and practical importance. Students often focus too heavily on p-values. A stronger approach recognises that hypothesis testing is only one way of summarising what the data suggest. Confidence intervals and effect sizes often tell a richer story.

The logic of hypothesis testing

Hypothesis testing begins with two competing statements:

Null hypothesis (H₀): there is no effect, no difference, or no association
Alternative hypothesis (H₁): there is an effect, difference, or association

The test asks: if the null hypothesis were true, how surprising would the observed data be? If the observed result would be very unlikely under H₀, the null is rejected.

This logic is often misunderstood. A p-value does not tell you the probability that the null hypothesis is true. It tells you the probability of obtaining data at least as extreme as the observed data, assuming the null is true.

The meaning of alpha and p-values

The significance level, α, is the threshold set before analysis, commonly 0.05. If p < α, the result is statistically significant.

Important cautions:

Statistical significance does not prove a theory.
Non-significance does not prove no effect.
A p-value does not measure effect size.
A significant result can still be trivial in practical terms.

An exam answer should ideally state:

whether p is below or above α,
what that means for H₀,
and whether the result seems meaningful in context.

Type I and Type II errors

Error type	Definition	Example in psychology	Consequence
Type I error	Rejecting a true null hypothesis	Concluding a new intervention works when it does not	False positive, wasted resources, misleading claims
Type II error	Failing to reject a false null hypothesis	Missing a real treatment effect because the sample is too small	False negative, missed opportunity

Type I error is controlled by alpha. Type II error is linked to sample size, effect size, and variability. Power is the probability of correctly detecting a true effect. Higher power is generally better, and it increases when the sample is larger, the effect is larger, and the measurement is more reliable.

Confidence intervals: what they do better than p-values

A confidence interval gives a plausible range of values for the population parameter. For a mean difference, it shows the likely size and direction of the effect. If the interval excludes zero, the result is typically significant at the matching alpha level. But even when significant, the interval may show the effect is small.

Example:

Mean difference = 4.2 points
95% CI = [1.0, 7.4]

This tells us the effect is likely positive, but the true difference may be modest or moderate. If the interval were [0.2, 8.0], the effect is still significant but uncertain in size. If it were [-1.5, 9.2], the evidence is too imprecise to confidently claim a difference.

Effect size: why magnitude matters

Effect size quantifies the size of a result. In psychology, effect sizes are essential because they tell us whether the result matters beyond statistical significance.

Common effect sizes include:

Cohen’s d for mean differences
Pearson’s r for correlation strength
r² or R² for variance explained
Eta-squared (η²) or partial eta-squared for ANOVA
Standardised beta coefficients in regression, used cautiously for comparison

Cohen’s d

Cohen’s d compares the difference between two means in standard deviation units. A larger absolute value indicates a larger separation.

A rough guide often used:

0.20 = small
0.50 = medium
0.80 = large

These are not laws. In psychology, a “small” effect may still be important if it affects many people or accumulates over time.

Pearson’s r

Correlation values range from -1 to +1.

A rough guide:

around 0.10 = small
around 0.30 = moderate
around 0.50 or higher = strong

Again, context matters. A correlation of 0.25 between study hours and marks may matter a great deal in educational planning.

R²

R² tells you the proportion of variance explained by a model. If R² = 0.36, then 36% of the variance in the outcome is explained by the predictors, and 64% remains unexplained. In psychology, unexplained variance is normal, not a failure.

Statistical significance versus practical significance

This distinction appears often in exams. A result may be statistically significant because the sample is large, but the effect size may be tiny. Another result may fail to reach significance because the study is small, but the observed effect may be strong enough to matter clinically or educationally.

Example of a practically meaningful effect:

An intervention reduces weekly panic attacks from 6 to 3 on average. Even if the sample is modest, this may be very important.

Example of a practically weak effect:

A huge sample shows a treatment lowers anxiety by 0.2 points on a 100-point scale. The p-value may be tiny, but the practical value is doubtful.

How to write this in an exam

A strong answer often includes all of the following:

State whether the result is significant.
Report the effect size or explained variance.
Interpret the confidence interval if available.
Explain whether the effect is practically important.
Link the result to the psychological question.

For example:
“Although the difference is statistically significant, the effect size is small, suggesting only a limited practical improvement in anxiety symptoms. The confidence interval does not include zero, so the direction of the effect is fairly clear, but the magnitude is modest.”

That is the kind of nuanced interpretation expected in advanced psychology assessment.

4. Correlation, Simple and Multiple Regression, and Prediction in Psychological Research

Regression analysis is one of the most important topics in advanced psychology statistics because it bridges description, explanation, and prediction. It helps determine whether variables move together, how strongly they are related, and whether one variable contributes unique information after others are controlled. For psychology students, the key is not only calculating coefficients but understanding what a model says about human behaviour.

Correlation: the first step toward prediction

Correlation assesses the strength and direction of association between two continuous variables. The Pearson correlation coefficient, r, ranges from -1 to +1.

Interpreting r

Positive r: as one variable increases, the other tends to increase
Negative r: as one variable increases, the other tends to decrease
r near 0: little or no linear relationship

Examples in psychology:

Higher stress may correlate with worse sleep quality.
Greater study time may correlate with higher exam marks.
Stronger social support may correlate with lower depressive symptoms.

Correlation does not imply causation. This is an exam favourite. A relationship may arise because:

X causes Y
Y causes X
a third variable causes both
the pattern is spurious

For example, stress and poor sleep may be related, but either could influence the other, and both may also be affected by workload or mental health.

Simple linear regression

Simple linear regression predicts one outcome from one predictor. The model is usually written as:

Y = a + bX

Where:

Y is the predicted outcome
a is the intercept
b is the slope
X is the predictor

The slope tells us how much Y changes for a one-unit increase in X.

Intercept and slope in context

If a regression predicts exam score from study hours:

The intercept is the predicted exam score when study hours = 0.
The slope is the increase in predicted exam score for each additional hour studied.

The intercept is not always meaningful psychologically. If zero on the predictor is outside the realistic range, the intercept may be mathematically necessary but substantively unimportant. The slope is usually more informative.

The least squares idea

Regression finds the line that minimises the sum of squared residuals. A residual is the difference between an observed score and the predicted score.

Positive residual: observed score is above predicted
Negative residual: observed score is below predicted
Large residual: the model fits that case poorly

Residuals matter because they indicate how well the model describes the data. A model with a strong relationship should generally have smaller residuals than one with weak explanatory power.

R² and explained variance

The coefficient of determination, R², shows how much variance in the outcome is accounted for by the predictor(s).

Example:

If R² = 0.25, then 25% of the variation in the DV is explained by the model.

In psychology, R² should be interpreted carefully. Human behaviour is influenced by many variables, so even a useful model may explain a moderate portion of variance. An R² of 0.12 can still be meaningful if the construct is complex and the predictor is simple.

Multiple regression: more realistic psychological modelling

Multiple regression includes two or more predictors. This is often more realistic because psychological outcomes usually reflect multiple influences.

Example:

Depressive symptoms predicted by stress, social support, and sleep quality.

The model asks:

Which predictors matter?
How much unique variance does each predictor explain?
Does the full model predict the outcome better than chance?

Unique versus shared variance

This is one of the most important ideas in multiple regression. Predictors often overlap. For instance, stress and sleep quality may be related, and both may relate to depression. Regression separates:

Unique variance explained by a specific predictor
Shared variance among predictors

A predictor can be correlated with the outcome but not significant in multiple regression if another predictor already captures the same variance. This does not mean the predictor is useless. It may simply be redundant or conceptually overlapping.

Standardised and unstandardised coefficients

Unstandardised coefficients (B): interpret in original units
Standardised coefficients (β): express effects in SD units and allow rough comparison across predictors

Example:
If stress has B = 2.5 in a depression model, then each one-unit increase in stress predicts a 2.5-point increase in depression score, holding the other predictors constant.

Standardised coefficients are helpful, but they should not be overinterpreted as absolute proof of importance because they depend on scaling and sample characteristics.

Assumptions of regression

Regression usually assumes:

Linearity between predictors and outcome
Independence of observations
Homoscedasticity of residuals
Normally distributed residuals or approximate normality
No problematic multicollinearity
No extreme influential outliers

Multicollinearity

If predictors are highly correlated, coefficient estimates become unstable. The model may still predict the outcome well, but it becomes difficult to decide which predictor matters uniquely.

Signs of multicollinearity:

inflated standard errors
unstable coefficients
predictor signs that seem counterintuitive
high Variance Inflation Factor (VIF)

A common rule of thumb is that VIF values above 5 or 10 may signal concern, though the context matters.

Interpreting a regression output in exam form

A typical interpretation should include:

overall model fit
significance of the model
size of explained variance
meaning of each predictor
whether assumptions are acceptable
what the findings suggest psychologically

Example phrasing:
“The regression model was significant and explained a moderate proportion of variance in depressive symptoms. Stress was a positive predictor, while social support was a negative predictor. Sleep quality contributed additional unique variance, indicating that it predicts symptoms above and beyond stress and support.”

That kind of answer demonstrates both statistical and conceptual understanding.

Common mistakes with regression

Students often make these errors:

treating correlation as causation
interpreting a non-significant predictor as “no relationship at all”
ignoring multicollinearity
overemphasising standardised coefficients without context
forgetting to mention R²
failing to interpret coefficients in relation to the outcome variable

The safest strategy is to read every coefficient as an adjusted relationship, not a causal mechanism unless the design truly supports causal inference.

5. t-Tests, ANOVA, Assumptions, and Reporting Results in Psychology

Group comparisons are central to psychology because many studies compare people across conditions, diagnoses, interventions, or demographic categories. The t-test and ANOVA are especially important because they are foundational tools for comparing means. Even where more advanced procedures are used, the logic of mean comparison remains central.

Independent-samples t-test

An independent-samples t-test compares the means of two unrelated groups.

Example:

Group 1: students who attended a study-skills workshop
Group 2: students who did not attend
Outcome: final test score

The test asks whether the mean difference between the two groups is larger than would be expected by chance.

What to interpret

group means
mean difference
t statistic
degrees of freedom
p-value
confidence interval
effect size, often Cohen’s d

A good interpretation sounds like:
“The workshop group scored higher than the control group, and the difference was statistically significant. The effect size suggests a moderate improvement, indicating that the workshop may have meaningful educational value.”

Paired-samples t-test

A paired t-test compares two related measurements from the same participants or matched pairs.

Examples:

pre-test and post-test anxiety scores
left-hand and right-hand reaction times
matched twins in two conditions

Because the scores are related, the analysis focuses on change within individuals rather than differences between unrelated groups.

Key idea:

The analysis uses the difference score for each person.
If the mean difference is far from zero, the result may be significant.

This test is especially relevant in intervention studies, repeated measures, and before-and-after designs.

One-way ANOVA

ANOVA compares the means of three or more groups. Using multiple t-tests instead of ANOVA inflates Type I error, so ANOVA is the appropriate omnibus test.

Example:

Three therapy conditions: cognitive-behavioural therapy, supportive counselling, waitlist control
Outcome: anxiety score

The null hypothesis is that all group means are equal.

If the ANOVA is significant, it tells us that at least one group differs, but not which one. That is why post hoc comparisons or planned contrasts are needed.

The F statistic

ANOVA uses the F ratio, which compares variance between groups to variance within groups.

Between-group variance: how far group means are from the overall mean
Within-group variance: how spread out individuals are inside each group

If between-group variance is large relative to within-group variance, the F statistic becomes large, and the result may be significant.

Post hoc tests and multiple comparisons

Once a significant ANOVA is found, follow-up tests identify where the differences lie.

Common approaches:

Tukey HSD
Bonferroni adjustment
planned contrasts

The reason for adjustment is simple: the more comparisons you run, the greater the chance of finding a false positive by accident. Post hoc control protects against overclaiming.

Assumptions for t-tests and ANOVA

These tests usually assume:

independence of observations
approximately normal DV within groups
homogeneity of variances
absence of severe outliers

When assumptions are violated:

small deviations may be tolerated, especially with larger samples
strong violations may call for non-parametric alternatives
unequal variances may require a corrected t-test or robust method

Effect sizes in mean comparisons

For t-tests:

Cohen’s d is common

For ANOVA:

Eta-squared (η²) or partial eta-squared are common

Effect sizes help show whether the observed group differences are meaningful, not just detectable.

Example:
A three-group study may yield a significant ANOVA with a small η². That means the groups differ, but the proportion of variance explained by condition is limited. In psychology, that can still be useful if the intervention is low-cost or scalable, but the limitation should be acknowledged.

Reading ANOVA output strategically

When confronted with ANOVA results, a strong exam answer should follow this order:

Identify the independent variable and number of groups.
State whether the omnibus test is significant.
Report the F statistic, degrees of freedom, and p-value if given.
Report effect size if available.
Interpret the post hoc findings.
Explain the psychological significance.

For example:
“The three therapy conditions differed significantly on post-treatment anxiety. Post hoc comparisons showed that the cognitive-behavioural therapy group scored lower than the waitlist group, while supportive counselling did not differ significantly from either group. The effect size was moderate, suggesting that therapy condition meaningfully influenced outcomes.”

Common exam traps

Using multiple t-tests instead of ANOVA for three or more groups
Forgetting that a significant ANOVA does not identify where differences lie
Misreading non-significant post hoc tests after a significant omnibus test
Ignoring the role of equal variances and group sizes
Reporting only p-values and no effect sizes

Suggested reporting template

A clean psychology report often includes:

the test used,
the IV and DV,
the result of the test,
the effect size,
the interpretation in context.

Example template:
“An independent-samples t-test showed that [Group A] scored significantly [higher/lower] than [Group B] on [DV], t(df) = value, p = value, d = value. This suggests that [psychological interpretation].”

For ANOVA:
“A one-way ANOVA showed that [IV] significantly affected [DV], F(df1, df2) = value, p = value, η² = value. Post hoc tests indicated that…”

This is the style that exam markers usually reward because it is concise, complete, and interpretable.

6. Revision Strategy, Interpretation Skills, and Common Stellenbosch Exam Pitfalls

Success in advanced data analysis is not only about knowing formulas. It also depends on being able to recognise the pattern of a question, translate output into words, and avoid predictable mistakes under time pressure. The most effective revision strategy is therefore mixed: learn concepts, practise interpretation, and drill the logic of choosing the correct test.

A practical order for revising PSY 315

A sensible revision sequence is:

Identify the research question
Determine the variable types
Decide whether the goal is comparison, association, or prediction
Choose the statistical test
Check assumptions
Interpret significance, effect size, and confidence intervals
Translate results into psychological meaning

If that sequence becomes automatic, exam questions become much less intimidating.

How to choose the correct test quickly

Use the following guide:

Research question	Likely analysis
Compare two independent groups	Independent-samples t-test
Compare two related measurements	Paired-samples t-test
Compare three or more groups	One-way ANOVA
Examine relationship between two continuous variables	Pearson correlation
Predict one outcome from one predictor	Simple regression
Predict one outcome from several predictors	Multiple regression

This table is a useful starting point, but context matters. For example, if the outcome is categorical, a different analysis may be needed. If the design is repeated measures with more than two time points, a repeated-measures approach may be more suitable than a simple t-test.

Reading output without panic

Many students know the theory but struggle when a software output table appears. The best way to read output is to ask six questions:

What is the outcome variable?
What is the predictor or grouping variable?
What is the size and direction of the effect?
Is the result statistically significant?
How large is the effect?
What does it mean psychologically?

This checklist works for t-tests, ANOVA, correlation, and regression.

Common interpretation errors

1. Confusing association with causation

A correlation between two variables does not prove one causes the other.

2. Ignoring sample size

Large samples make small effects significant; small samples can hide large effects.

3. Overlooking assumptions

A test may produce a number even when the assumptions are badly violated.

4. Treating p = 0.051 as fundamentally different from p = 0.049

This is an artificial border. Good interpretation should consider the broader pattern of evidence.

5. Forgetting practical significance

Even a statistically significant effect may be too small to matter in real life.

6. Repeating the output instead of interpreting it

Marks are lost when students paraphrase software tables rather than explain what the values imply.

How to write higher-scoring exam answers

Strong exam answers tend to include the following features:

clear identification of variables
correct naming of the analysis
accurate use of statistical language
mention of direction and magnitude
reference to assumptions where relevant
a psychological conclusion rather than only a statistical one

For example, instead of writing:
“The p-value was significant, so the null was rejected,”

write:
“The significant result indicates that the intervention was associated with lower anxiety, and the moderate effect size suggests the difference is likely meaningful rather than trivial.”

That second sentence is much more valuable because it shows the examiner that the student understands what the test means in context.

A compact decision guide for exam revision

Use this logic when faced with a question:

Two means, unrelated groups → independent t-test
Two means, same participants → paired t-test
Three or more means → ANOVA
Two continuous variables → correlation
Prediction with one continuous IV → simple regression
Prediction with several predictors → multiple regression

Then ask:

Is the variable continuous or categorical?
Are the observations independent?
Are we comparing groups or explaining variation?
What assumptions need to be checked?

High-yield exam phrases

These are useful phrasing patterns for written responses:

“The result was statistically significant, indicating that…”
“The effect size suggests that the magnitude of the difference is…”
“This finding should be interpreted cautiously because…”
“The model explained a moderate proportion of variance…”
“The predictor made a unique contribution after controlling for…”
“Although the relationship is significant, practical importance appears limited…”

Final integration: what advanced analysis means in psychology

Advanced data analysis in psychology is ultimately about disciplined interpretation. The numbers themselves do not speak without context. A statistically significant result can still be weak, a non-significant result can still be informative, and a well-chosen model can reveal patterns that are invisible in raw means alone. At Stellenbosch, as in other rigorous psychology programmes, students are expected to move fluently between statistical logic and psychological meaning.

The best exam strategy is therefore to think like a researcher:

choose analyses for the question,
examine assumptions honestly,
interpret results cautiously,
and explain why the findings matter for behaviour, emotion, cognition, or intervention.

When that habit becomes natural, PSY 315 shifts from being a difficult statistics module to a powerful toolkit for understanding psychological data.

Final checklist before the exam

Know the difference between descriptive and inferential statistics.
Be able to choose between t-test, ANOVA, correlation, and regression.
Know the assumptions behind each major test.
Interpret p-values, confidence intervals, and effect sizes together.
Distinguish statistical significance from practical importance.
Avoid causal language unless the design justifies it.
Always return the answer to the psychological question asked.

If these principles are secure, most PSY 315 exam questions become manageable, even when the data or output tables look unfamiliar.