SMCR Kap 7-9

Alla Inga Spara

• Grand mean Summarized mean of all participants and all groups
• Within groups variance Error
• Eta squared Measure of how well we can predict a score in the dependent variable from the independent
• Low between-groups variance Groups are similar/equal
• Degrees of freedom First column for between-groups variance, other for within-groups variance
• P=1 and Eta: 0.00 Groups are equal
• Moderation Different differences
• Main effect Different average scores for groups defined by a single independent variable
• Interaction effect Different effects for one variable across different groups, defined by another variabel
• Eta2 Proportion of variance in a numerical variable, that is predicted or explained by another variable
• Interpretating a .71 variance In percentages
• We want to compare outcomes across 3+ groups Analysis of variance
• Using variance of 3+ group means to test null hypothesis Between-groups variance
• Within-groups variance Getting different group means even if we draw samples from populations with the same means
• One factor of main effect One-way anova
• Two factors of a main effect Two-way anova
• Assumptions for one way anova Equal-size groups, test of homogeneity
• Assumptions of two-way anova Residuals normally distributed (histogram), residuals average to zero (plot), values predicted evenly across lower and high levels (plot)
• eta2: .5 Strong effect size in CS
• What we test Between-groups variance against within-groups variance(error in output)
• Assumptions for performing F-distribution Independent samples & Homogeneous population variances
• What do we do if we have a factor with more than two groups Bonferroni correction post hoc test
• Independent variable category for ANOVA Categorical
• Dependent variable category for ANOVA Numeric, to obtain nuanced results
• Obtain effect size Square root of eta2
• What does variance measure? Deviations from the mean
• Assumption of equal variance check Levene's test must be non-significant
• ANOVA test stops when We cannot reject the null, or have too little evidence to conclude a difference among groups, still report CI
• F-test significant, but not the posthoc? Capitalization correction may be too strong, or sample is too small
• Bivariate analysis Taking two variables (independent and dependent) into consideration
• Balanced design Factors statistically independent of each other, scores on one factor is not associated with scores on the other
• When is a test balanced? Size of the smallest subgroup is less than 10% smaller than the largest group
• Commonality in communication science Effects of messages are stronger for people who are more susceptible to a message
• Using means plot To interpret different differences, and see effect direction
• Effect direction moderation The effect in one group is the opposite of the effect in another group
• Effect strength moderation Refers to moderators that strengthen or diminish the effect of an independent variable
• Null hypothesis of an F-test of an interaction effect The subgroups have the same population averages if we correct for main effects
• Null hypothesis interaction effect test Equal effects of the predictor for all moderator groups in the population
• Homoscedasticity Large prediction error at one level should yield a large prediction error at another level
• When do we use regression When we want to predict a continuous variable from one or more numerical/categorical predictors
• What does regression predict for categorical IV's? Group averages
• b of a dichotomous variable Difference between average scores of two groups
• Conditional effect Coefficient is the effect for the group with the score of zero on the moderator
• Interaction variable computation and inclusion Predictor x Moderator, and take these into consideration in the model
• Regression coefficient Effect of a variable, as well as the slope of the line
• Moderation by context Predictive effect not equal in every situation
• Reference group Category without dummy variable
• What does regression of dummies give us Difference between average score on dependent variable of group scoring 1 on the dummy
• When do we use error (e) in regression When we discuss assumptions for statistical inference on a regression, and not in the equation
• Error (e) Residuals
• (y) Sum of the constant
• (b) Effects of IV variables or predictors (x) - Predictors
• Unstandardized Reg.Coe Represents the predicted difference in the dependent for a difference of one unit on the independent
• Unstandardized interpretation for dummy Difference in average sores for two groups
• 1 Value of reference group
• Constant in dummy equation Symbolizes predicted value of reference group
• Decision of reference group Generally, the one that is of most interest
• Null hypothesis of regression The unstandardized regression coefficient is zero in the population
• Assumptions of regression data Independence, and identical distribution
• When are observations not independent of each other Time series data, where one observation is dependent on pre-existing data
• Residual distribution Normally distributed, for each value on the dependent if the assumptions are met
• One additional unit on the independent Decreases/increases predicted value by the same amount
• The larger the residuals The worse the predictions
• When to use regression If we have a numerical dependent variable and at least one numerical IV
• Way to detect moderation Regression lines in plots are not parallel between groups
• Similar looking slopes for both groups Predictive effect of the IV on DV is more or less equal (no moderation)
• Common support Variation of predictor scores for a particular value of the moderator
• Residuals Prediction errors
• Linear model matches data Positive and negative residuals are somewhat balanced along the regression line
• Dots in regression line looks like cone Indication that there may be moderation in the model
• Predictor IV that is central to our analysis
• Moderator IV for e expect different effects of the predictor
• Covariate IV that is currently not central to our analysis
• Mean centering Subtract mean from all values, so the reference value of the variable represents the mean score on the original
• Use of mean value Moderation when it is a numerical moderator
• Unstandardized coefficient for numerical moderator tell us Predicted change in the effect of the predictor for a one unit increase in the moderator
• Variable symmetry Moderator and predictor have equal effects on the dependent
• Positive coefficient More positive/Less negative effect
• Negative coefficient More negative/less positive exposure effect
• Recommendation of mean-centering Mean-center both predictor and moderator if they are numerical

Alla Inga Spara