PSY3062 Exam Revision: Core Definitions and Diagnostic Plots

Comprehensive vocabulary flashcards covering statistical terminology, effect sizes, research designs, qualitative analysis, and diagnostic plots from PSY3062 lecture notes.

Independent Variable (IV)

the thing you change or use to predict. In non-experimental research it’s called the predictor.

Dependent Variable (DV)

the thing you measure as the result. In non-experimental research it’s called the outcome.

Interaction

Occurs in ANOVA or regression when the effect of one variable depends on the level of another. In a graph, the lines aren’t parallel (they cross or fan out). It’s tested with a product term (X × M).

Moderation

when a third variable (the moderator, M) changes how strong, or which direction, the X → Y relationship is. It’s tested with an interaction term.

Main Effect

The overall effect of one variable on the DV, averaged across the other variables.

Simple Effect/Slope

The effect of one variable at one specific level of another. A simple slope is the X → Y slope at a chosen value of the moderator (e.g., low, mean, high)

Effect Size

how big an effect is, not just whether it’s significant

$$R^2$$

The proportion of total variance in the outcome explained by a regression model, bounded between $0$ and $1$.

Adjusted $R^2$

R² with a penalty for adding more predictors, so it’s less inflated.

Eta-squared ($\eta^2$)

the proportion of total variance explained in ANOVA. It tends to overestimate.

Partial Eta-squared (partial $\eta^2$)

The variance explained by one factor, compared with that factor plus error only.

Omega-squared ($\omega^2$)

like η², but less biased because it accounts for degrees of freedom.

Cohen's $d$

the difference between two group means in SD units. It can be bigger than 1.

Cohen's $f^2$

a predictor’s unique explained variance relative to unexplained variance.

Unstandardized Coefficient ($B$)

the change in the outcome for a 1-unit change in the predictor, in the original units.

Standardized Coefficient ($\beta$)

the same idea as unstandardised, but in SD units, so you can compare predictors on different scales.

Type I Error

Rejecting the null hypothesis when it is actually true (a false positive).

Decision-wise Alpha ($\alpha_{DW}$)

The significance threshold set for an individual hypothesis test or comparison (typically $\alpha = .05$).

Family-wise Alpha ($\alpha_{FW}$)

The chance of at least one Type I error across a whole set of tests. α_FW = 1 − (1 − α_DW)^C, where C is the number of tests.

Repeated Measures Design

the same people are measured more than once (different conditions or times). This gives more power because individual differences are separated out from error.

Zero-Order Correlation

the plain correlation between two variables, controlling for nothing.

Partial Correlation ($pr$)

the correlation between two variables after removing a third variable’s effect from both.

Semi-Partial Correlation ($sr$)

The correlation after removing the third variable’s effect from only the predictor. It shows the predictor’s unique contribution to the outcome

Monotonic Relationship

a relationship that always goes in one direction (always up or always down) but doesn’t have to be a straight line. It’s tested with Spearman’s r_s.

Dummy Variable

a 0/1 variable used to represent categories in regression. For k categories you need k − 1 dummies, and the left-out category is the baseline.

Replicability

you repeat the study with new data and get consistent results.

Reproducibility

you re-run the same data and code and get identical results.

Reflexivity

the researcher constantly reflects on how their own views and position shape the analysis.

Reflexive Thematic Analysis (RTA)

a method for finding patterns of meaning (themes) in qualitative data, where the researcher’s subjectivity is treated as a tool, not a problem.

Inductive Approach

builds themes from the data (bottom-up)

Deductive Approach

uses an existing theory to analyze the data (top-down).

Semantic Coding

codes what’s explicitly said (surface level)

Latent Coding

codes the underlying assumptions and hidden meanings.

L.I.N.E. Assumptions

Core statistical model assumptions: Linearity, Independence, Normality, and Equal variance.

Scatterplot

A plot of continuous predictor values against continuous outcome values used to inspect for a straight-line pattern to confirm linearity, check association direction, and identify bivariate outliers.

Histogram

A graphical display of the frequency distribution of scores for a single continuous variable or model residuals, used to check for skewness, kurtosis, and a normal curve.

Box-and-Whisker Plot

A plot displaying medians, interquartile ranges (IQR), and overall distribution spread across variables or groups, used to identify univariate outliers beyond the whiskers.

Violin Plot

A visualization combining boxplots with kernel density plots to show full distribution shapes across categories and spot multimodal distributions.

Residual Plot

A diagnostic plot with predicted values on the X-axis and standardized residuals on the Y-axis, used to test linearity (random distribution around zero) and homoscedasticity (even width/spread).

Q-Q Plot

A quantile-quantile plot of observed sample quantiles of standardized residuals against theoretical quantiles of a normal distribution, used to test the normality assumption of errors/residuals.