MVDA – SPSS Cheat sheet
Week 1, Multivariate Regression Analysis
- Can the null hypothesis of no relationship between Y and X1, X2, etc. be rejected? → Perform
linear regression → From ANOVA table report: F(dfregression, dfresidual) = 12.345, p = 0.003
- How much variance of Y is explained by X1, X2, etc. together/the predictors? → Perform
linear regression → From model summary report: R square = 0.123
- Which predictor explains the most unique variance? / How much variance of Y is (uniquely)
explained by X1? → Perform linear regression with part and partial correlation → From
coefficients, correlations, part report: X1: 𝒓𝟐𝒀(𝟏∙𝟐) = (0.123)2 = 0.015
- Remove the non-significant predictors from the model → Check which X are significant in
Coefficients, Sig → Only include significant in next regression.
- What is the estimated regression equation → Perform linear regression → From
Coefficients, Unstandardized Coefficients, B report: 𝒚 ̂ = 𝒃𝟎 + 𝒃𝟏 𝒙𝟏 + 𝒃𝟐 𝒙𝟐 + ⋯ + 𝒃𝒑 𝒙𝒑
- Interpret the regression coefficients → From Coefficients, Unstandardized Coefficients, B
report: Improving one point on X1, would improve Y with b1
- Does adding X2 significantly improve the linear model? / How much variance is explained in
model 2 in comparison to model 1? → Perform linear regression, add X2 to second block and
2
tick 𝑅𝑐ℎ𝑎𝑛𝑔𝑒 → From Model Summary, model 2 report: Yes, F(df1, df2) = 12.345, p = 0.003
Pearson correlations: ANALYZE → CORRELATE → BIVARIATE → SELECT DESIRED VARIABLES → tick
PEARSON OR SPEARMAN
Linear regression: ANALYZE → REGRESSION → LINEAR → drag X TO INDEPENDENT (based on) and Y
TO DEPENDENT (predict) (Often: Y on X)
• Statistics:
- Confidence interval: Coefficients table → Slope b1, intercept (constant)
2
- 𝑅𝑐ℎ𝑎𝑛𝑔𝑒 → When doing a hierarchical regression
- Descriptives: Descriptive statistics → mean, standard deviation, and N
- Part and partial correlations: Coefficients table → look at …
2
▪ Part for semipartial correlation → 𝑟𝑌(1∙2) → Square yourself: 𝑟𝑌(1∙2)
2
▪ Zero-order correlation → 𝑟𝑌1 → Square yourself: 𝑟𝑌1
- Collinearity diagnostics: Multicollinearity
▪ Tolerance should be above 0.10
▪ VIF should be below 10
• Plots: for assumptions for significance testing
- Scatterplot → predicted values (ZPRED → X) and residuals (ZRESID → Y)
▪ Non-linearity: Is the best fit of the data a horizontal line? (Should be
randomly distributed)
▪ Homoscedasticity: Are the dots equally distributed along the best line?
- Histogram of standardized residuals → Tick box
- Normal probability plot → Tick box
▪ Normality of residuals: do the dots follow the line?
• Save: Residuals Statistics (for outliers)
- Cook’s D < 1 → Influential data points
- Leverage values: Centered leverage value max < 3(K+1)/N → Outliers on X
- Standardized residuals min and max < |3| → Outliers on Y
Remove outliers: DATA → SELECT CASES → TICK IF CONDITION IS SATISFIED → Write: Cook’s D < 1
→ Results with this number will be excluded
1
, MVDA – SPSS Cheat sheet
Week 2, Analysis Of Variance
- What is sample size N, what are the group sizes, is the design balanced? → Perform Cross
Tabulation → From Case Processing Summary report: N = 50, From Crosstabulation report:
n1 = 15, n2 = 15, etc., From Crosstabulation report: Balanced if n1 = n2 = ni etc.
- Is the ANOVA F-test robust to violations of group normality → Perform Cross Tabulation →
from Crosstabulation report: n > 15: F is robust to non-normality
- Is the ANOVA F-test robust to violations of homogeneity of the group variances → Perform
Cross Tabulation → from Crosstabulation report: nmax / nmin = …/… = … < 1.5: F robust to …
- Check the assumption of normality by inspecting graphically the distribution of the
standardized residuals → Make a histogram with std. residuals on the X-axis → From Simple
Histogram of Standardized Residual for X, Report whether it is normally distributed. / Make
a Q-Q plot with standardized residuals → From Normal Q-Q Plot of Standardized Residual for
X, Report whether the variables are approx. similar to the line.
- Can the null hypothesis of equal group variances be rejected? Does the rule of thumb with
group standard deviations agrees with your first conclusion? → Perform ANOVA,
homogeneity tests and descriptive statistics → From Levene’s Test of Equality of Error
Variances, Based on Mean, report: Yes, Levene’s test is not significant: F (df1, df2) = 1.234,
p = 0.123. / From Descriptive Statistics, report: Yes, if sdmax / sdmin < 2
- Can the null hypothesis of no relationship between Y and X1 and/or X2 be rejected? (Is our
model good?) → Perform ANOVA → From Tests of Between Subject Effects, Corrected
Model report: Yes/no, F (dfmodel, dferror) = 12.345, p = 0.002
- Is effect of X1 or interaction effect significant? → Perform ANOVA → From Tests of Between
Subject Effects, Effect report: Yes/no, F(dfx1, dferror) = 12.345, p = 0.003 (So there is an effect)
- Which group has the highest estimated marginal mean (2 grp.)? → Perform ANOVA → From
Estimated Marignal Means, X1 report: Group 1 (Mean) has a higher Y than Group 2 (Mean)
- Which group has the highest estimated marginal mean? (3+ group)? → Perform ANOVA →
From Post Hoc Tests, Multiple Comparisons, report: Group A significantly higher than other
two groups (A negative Mean difference → group J higher than I).
- Interpret the interaction effect → Perform ANOVA → From Estimated Marginal Means and
Profile Plot report: Group A higher in condition I, group B reverse effect. No difference
between group A and B in condition II. (The effect of X1 can be changed by the effect of X2)
- How much variance in Y is explained by X1 → Perform ANOVA → From Tests of Between-
Subjects Effects, report: η2 = SSEffect ÷ SSTCorrected (0.01, 0.06, 0.14)
Cross tabulation: Analyze → Descriptive Statistics → Crosstabs → One variable Rows, one Columns
ANOVA: Analyze → General Linear Model → Univariate → Drag independent variables to Fixed
Factors, drag dependent variable to Dependent Variable
• Plots: Drag variable with most groups to Horizontal Axis, the other one to Separate Lines →
Click Add → Optional: Add variables the other way around.
• Options: Descriptive statistics and Homogeneity tests (Levene’s test)
• EM Means: Drag all the effects you want to Display Estimated Marginal Means for
• Save: Will make a new variable for Standardized residuals
• Post Hoc → Drag X1 to Post Hoc Tests For → Tick Tukey
Make a histogram: GRAPHS → CHART BUILDER → HISTOGRAM → Drag variable to X-AXIS
Make a Q-Q plot: ANALYZE → DESCRIPTIVE STATISTICS → Q-Q PLOTS → Move standardized
residuals to VARIABLES → Click OK
2
Week 1, Multivariate Regression Analysis
- Can the null hypothesis of no relationship between Y and X1, X2, etc. be rejected? → Perform
linear regression → From ANOVA table report: F(dfregression, dfresidual) = 12.345, p = 0.003
- How much variance of Y is explained by X1, X2, etc. together/the predictors? → Perform
linear regression → From model summary report: R square = 0.123
- Which predictor explains the most unique variance? / How much variance of Y is (uniquely)
explained by X1? → Perform linear regression with part and partial correlation → From
coefficients, correlations, part report: X1: 𝒓𝟐𝒀(𝟏∙𝟐) = (0.123)2 = 0.015
- Remove the non-significant predictors from the model → Check which X are significant in
Coefficients, Sig → Only include significant in next regression.
- What is the estimated regression equation → Perform linear regression → From
Coefficients, Unstandardized Coefficients, B report: 𝒚 ̂ = 𝒃𝟎 + 𝒃𝟏 𝒙𝟏 + 𝒃𝟐 𝒙𝟐 + ⋯ + 𝒃𝒑 𝒙𝒑
- Interpret the regression coefficients → From Coefficients, Unstandardized Coefficients, B
report: Improving one point on X1, would improve Y with b1
- Does adding X2 significantly improve the linear model? / How much variance is explained in
model 2 in comparison to model 1? → Perform linear regression, add X2 to second block and
2
tick 𝑅𝑐ℎ𝑎𝑛𝑔𝑒 → From Model Summary, model 2 report: Yes, F(df1, df2) = 12.345, p = 0.003
Pearson correlations: ANALYZE → CORRELATE → BIVARIATE → SELECT DESIRED VARIABLES → tick
PEARSON OR SPEARMAN
Linear regression: ANALYZE → REGRESSION → LINEAR → drag X TO INDEPENDENT (based on) and Y
TO DEPENDENT (predict) (Often: Y on X)
• Statistics:
- Confidence interval: Coefficients table → Slope b1, intercept (constant)
2
- 𝑅𝑐ℎ𝑎𝑛𝑔𝑒 → When doing a hierarchical regression
- Descriptives: Descriptive statistics → mean, standard deviation, and N
- Part and partial correlations: Coefficients table → look at …
2
▪ Part for semipartial correlation → 𝑟𝑌(1∙2) → Square yourself: 𝑟𝑌(1∙2)
2
▪ Zero-order correlation → 𝑟𝑌1 → Square yourself: 𝑟𝑌1
- Collinearity diagnostics: Multicollinearity
▪ Tolerance should be above 0.10
▪ VIF should be below 10
• Plots: for assumptions for significance testing
- Scatterplot → predicted values (ZPRED → X) and residuals (ZRESID → Y)
▪ Non-linearity: Is the best fit of the data a horizontal line? (Should be
randomly distributed)
▪ Homoscedasticity: Are the dots equally distributed along the best line?
- Histogram of standardized residuals → Tick box
- Normal probability plot → Tick box
▪ Normality of residuals: do the dots follow the line?
• Save: Residuals Statistics (for outliers)
- Cook’s D < 1 → Influential data points
- Leverage values: Centered leverage value max < 3(K+1)/N → Outliers on X
- Standardized residuals min and max < |3| → Outliers on Y
Remove outliers: DATA → SELECT CASES → TICK IF CONDITION IS SATISFIED → Write: Cook’s D < 1
→ Results with this number will be excluded
1
, MVDA – SPSS Cheat sheet
Week 2, Analysis Of Variance
- What is sample size N, what are the group sizes, is the design balanced? → Perform Cross
Tabulation → From Case Processing Summary report: N = 50, From Crosstabulation report:
n1 = 15, n2 = 15, etc., From Crosstabulation report: Balanced if n1 = n2 = ni etc.
- Is the ANOVA F-test robust to violations of group normality → Perform Cross Tabulation →
from Crosstabulation report: n > 15: F is robust to non-normality
- Is the ANOVA F-test robust to violations of homogeneity of the group variances → Perform
Cross Tabulation → from Crosstabulation report: nmax / nmin = …/… = … < 1.5: F robust to …
- Check the assumption of normality by inspecting graphically the distribution of the
standardized residuals → Make a histogram with std. residuals on the X-axis → From Simple
Histogram of Standardized Residual for X, Report whether it is normally distributed. / Make
a Q-Q plot with standardized residuals → From Normal Q-Q Plot of Standardized Residual for
X, Report whether the variables are approx. similar to the line.
- Can the null hypothesis of equal group variances be rejected? Does the rule of thumb with
group standard deviations agrees with your first conclusion? → Perform ANOVA,
homogeneity tests and descriptive statistics → From Levene’s Test of Equality of Error
Variances, Based on Mean, report: Yes, Levene’s test is not significant: F (df1, df2) = 1.234,
p = 0.123. / From Descriptive Statistics, report: Yes, if sdmax / sdmin < 2
- Can the null hypothesis of no relationship between Y and X1 and/or X2 be rejected? (Is our
model good?) → Perform ANOVA → From Tests of Between Subject Effects, Corrected
Model report: Yes/no, F (dfmodel, dferror) = 12.345, p = 0.002
- Is effect of X1 or interaction effect significant? → Perform ANOVA → From Tests of Between
Subject Effects, Effect report: Yes/no, F(dfx1, dferror) = 12.345, p = 0.003 (So there is an effect)
- Which group has the highest estimated marginal mean (2 grp.)? → Perform ANOVA → From
Estimated Marignal Means, X1 report: Group 1 (Mean) has a higher Y than Group 2 (Mean)
- Which group has the highest estimated marginal mean? (3+ group)? → Perform ANOVA →
From Post Hoc Tests, Multiple Comparisons, report: Group A significantly higher than other
two groups (A negative Mean difference → group J higher than I).
- Interpret the interaction effect → Perform ANOVA → From Estimated Marginal Means and
Profile Plot report: Group A higher in condition I, group B reverse effect. No difference
between group A and B in condition II. (The effect of X1 can be changed by the effect of X2)
- How much variance in Y is explained by X1 → Perform ANOVA → From Tests of Between-
Subjects Effects, report: η2 = SSEffect ÷ SSTCorrected (0.01, 0.06, 0.14)
Cross tabulation: Analyze → Descriptive Statistics → Crosstabs → One variable Rows, one Columns
ANOVA: Analyze → General Linear Model → Univariate → Drag independent variables to Fixed
Factors, drag dependent variable to Dependent Variable
• Plots: Drag variable with most groups to Horizontal Axis, the other one to Separate Lines →
Click Add → Optional: Add variables the other way around.
• Options: Descriptive statistics and Homogeneity tests (Levene’s test)
• EM Means: Drag all the effects you want to Display Estimated Marginal Means for
• Save: Will make a new variable for Standardized residuals
• Post Hoc → Drag X1 to Post Hoc Tests For → Tick Tukey
Make a histogram: GRAPHS → CHART BUILDER → HISTOGRAM → Drag variable to X-AXIS
Make a Q-Q plot: ANALYZE → DESCRIPTIVE STATISTICS → Q-Q PLOTS → Move standardized
residuals to VARIABLES → Click OK
2
