Concept Meaning Notes
- Independent variable dependent variable
- Predictor/factor/condition outcome variable
Confirmatory Research method where you enter a specific Frequentist method of Bayesian informative testing
methods hypothesis and see if this can be confirmed,
before running the study.
Exploratory Research method where you don’t have a In frequentist method this is called ‘stepwise method’
methods specific a priori hypothesis. You try more where you try all combinations of variables and see
combinations to see what hypothesis fits best. what model fits best.
This is what happens in Bayesian testing
Multiple Analysis of more than 1 independent variable,
lineair interval/ratio scale
regression
(MLR)
MLR - DV = continuous measure (interval or
assumptions ratio level), IV are continuous or
dichotomous (dummy).
- Linearity of relations: linear relationship
between DV and IV’s
- No outliers
- No multicollinearity
- Homoscedasticity
- Normally distributed residuals
Outliers To test this assumption, you can watch at Standard residuals: values between -3.3 and +3.3 (as
scatterplots or look at the Casewise a rule of thumb!)
diagnostics and at the Standard
residuals (outliers in y-space) and Cook’s Cook’s distance: should be lower than 1
Distance (outliers in XY-space).
Multicollineari Relationship between the IV’s is too strong. Determine based on the statistics Tolerance or VIF
ty Mostly when r is above .8/.9 (Variance Inflation Factor) (collinearity statistics in
Jasp)
, - Tolerance values smaller than .2 indicate a
potential problem, tolerance values smaller than .1
indicate a problem.
- variance inflation factor (VIF) = 1/Tolerance. So for
the VIF, values greater than 5 indicate a potential
problem and values greater than 10 indicate a
problem.
Homoscedastic The spread of the residuals must be Plot: If for every predicted value (X-axis) there is
ity approximately the same across all values for approximately the same amount of spread around the
the predicted y: homogeneity of variances. Y-axis, then the condition is met. Also possible with a
Q-Q plot
MLR: plotting the (standardized) residuals
against the (standardized) predicted values. Levene’s test: testing hypothesis that all variances are
ANOVA: Levene’s test equal. If the p-value is significant: the null hypothesis
is not rejected: so, assumption is met.
Hierarchical Comparing 2 models: do extra variables make
linear extra difference when combined with other
regression variables?
analysis Has r2 significantly changed ?
Regression Y = b0 + b1x1 + b2x2 + … + e Without e: the predicted value: always written as ŷ
equation
b0 Intercept
bx Regression coefficient/slope B (unstandardized) or β (standardized)
ei (residual) difference between measured value and
prediction
Standardized If you want to compare slopes/determine the Standardized effects are omitted when X is
(β) vs. relative importance of the predictors within dichotomous. Standardizing also allows for comparison
unstandardize an equation/model, you use the standardized of effects across different studies.
d (B) column.
R2 Amount in which the variance in the DV is
- Independent variable dependent variable
- Predictor/factor/condition outcome variable
Confirmatory Research method where you enter a specific Frequentist method of Bayesian informative testing
methods hypothesis and see if this can be confirmed,
before running the study.
Exploratory Research method where you don’t have a In frequentist method this is called ‘stepwise method’
methods specific a priori hypothesis. You try more where you try all combinations of variables and see
combinations to see what hypothesis fits best. what model fits best.
This is what happens in Bayesian testing
Multiple Analysis of more than 1 independent variable,
lineair interval/ratio scale
regression
(MLR)
MLR - DV = continuous measure (interval or
assumptions ratio level), IV are continuous or
dichotomous (dummy).
- Linearity of relations: linear relationship
between DV and IV’s
- No outliers
- No multicollinearity
- Homoscedasticity
- Normally distributed residuals
Outliers To test this assumption, you can watch at Standard residuals: values between -3.3 and +3.3 (as
scatterplots or look at the Casewise a rule of thumb!)
diagnostics and at the Standard
residuals (outliers in y-space) and Cook’s Cook’s distance: should be lower than 1
Distance (outliers in XY-space).
Multicollineari Relationship between the IV’s is too strong. Determine based on the statistics Tolerance or VIF
ty Mostly when r is above .8/.9 (Variance Inflation Factor) (collinearity statistics in
Jasp)
, - Tolerance values smaller than .2 indicate a
potential problem, tolerance values smaller than .1
indicate a problem.
- variance inflation factor (VIF) = 1/Tolerance. So for
the VIF, values greater than 5 indicate a potential
problem and values greater than 10 indicate a
problem.
Homoscedastic The spread of the residuals must be Plot: If for every predicted value (X-axis) there is
ity approximately the same across all values for approximately the same amount of spread around the
the predicted y: homogeneity of variances. Y-axis, then the condition is met. Also possible with a
Q-Q plot
MLR: plotting the (standardized) residuals
against the (standardized) predicted values. Levene’s test: testing hypothesis that all variances are
ANOVA: Levene’s test equal. If the p-value is significant: the null hypothesis
is not rejected: so, assumption is met.
Hierarchical Comparing 2 models: do extra variables make
linear extra difference when combined with other
regression variables?
analysis Has r2 significantly changed ?
Regression Y = b0 + b1x1 + b2x2 + … + e Without e: the predicted value: always written as ŷ
equation
b0 Intercept
bx Regression coefficient/slope B (unstandardized) or β (standardized)
ei (residual) difference between measured value and
prediction
Standardized If you want to compare slopes/determine the Standardized effects are omitted when X is
(β) vs. relative importance of the predictors within dichotomous. Standardizing also allows for comparison
unstandardize an equation/model, you use the standardized of effects across different studies.
d (B) column.
R2 Amount in which the variance in the DV is
