Flashcards
stats 2 12
Cards in this deck (22)
- 1Main effectEffect of independent variable X on dependent variable Y
- 2Interaction-effect> An interaction effect is the simultaneous effect of two or more independent variables on at least one dependent variable in which their joint effect is significantly greater (or significantly less) than the sum of the parts.
> Effect of independent variable X on dependent variable Y is dependent on
moderator Z - 3Moderator Effect> An interaction effect is the simultaneous effect of two or more independent variables on at least one dependent variable in which their joint effect is significantly greater (or significantly less) than the sum of the parts.
> Effect of independent variable X on dependent variable Y is dependent on
moderator Z - 4an interaction effect gives you what power?> absolute power over the significance of this slope, by
merely changing the value of Z that SPSS calculates the rest of the model for
---> so, we prefer
not to interpret slopes directly when an interaction effect is present, and it is always better to use a
model without these interaction effects instead - 5> A linear regression with only main effects would look something like what? > The interpretations of the slopes are now what?> e.g. Y = 10 + 2 ∗ X + 3 ∗ Z
> bX: When X increases by 1, Y increases by 2, given that Z remains equal
bZ: When Z increases by 1, Y increases by 3, given that X remains equal - 6A linear regression with an interaction effect would look something like what?> Y = 10 + 2 ∗ X + 3 ∗ Z + 0.5 ∗ X ∗ Z
- 7interaction effect is...> always exactly the product of all variables
that are part of the interaction.
> The interaction effect is simply another predictor of Y
> and is
therefore unrelated to both main effects - 8s the slope of the interaction effect is... how is it with significance?> not
defined by the slopes of the main effects, nor is its significance bound to the main effects.
> It is
possible for the interaction effect to be significant while the main effects are not, and vice versa - 9turn continuous variables into categorical variables to test their interaction effect ?> never do that
> This loss of information and power only leads to greater lack of accuracy,
and a smaller chance to find a significant result if it is accurate. While interactions between categorical variables are perfectly fine, this should be reserved for variables who were categorical to begin with. - 10Is there a correlation between both independent variables in interaction effects ?> No
> An interaction term between two variables, which are completely
independent of each other, could be highly significant and an extremely strong predictor for
a model. - 11Does every part of an interaction effect needs to be present in the model even if one is not significant?> Yes
> This means that if you
want to test the interaction effect between X and Z, they both need to be added as main
effects for the interaction effect to be valid. It is not allowed to remove either of them, even
if X and Z are not significant. This means all lower order terms that belong with the
interaction have to be part of the model. - 12Centering> Forces the mean of a variable to 0 by subtracting the mean of the variable from each individual measurement. This means the spread of the scores of the original variable, so the SD and variance, are maintained, as is the relative order between the scores.
- 13Standardizing> Forces the mean of a variable to 0 by subtracting the mean of the variable from each individual measurement. This subtraction is followed up by a dividing every value by the standard deviation of the variable, which forces the standard deviation at 1. This is literally calculating the Z-values for each individual measurement
> While the difference between the scores has now changed, their relative order is still the same as before. - 14difference in the graphs between original and standardized or centered variables?> No, as the relative order and distance is maintained by both centering and standardizing
> - 15Highest order term> An “order” in the context of linear regression is exactly how many predictors a
slope is multiplied by in the model. This would mean the constant has “order 0”, the separate slopes
for X and Z are “order 1” and the interaction-effect “order 2”. It’s possible for any order to contain
more than one variable, besides order 0 which is reserved for the constant. In any case, the slope of
the highest order will not change when centering the model. - 16Lower order terms> Every order that is not the highest order term belongs to this group. Every slope
belonging to the lower orders will always change after centering or standardizing the model, besides
some theoretical exceptions. - 17Why is maintaining the SD the biggest strength of centering> Because the width of the scale remains intact, the slope of a centered
independent variable could be more easily transformed to the slope of an uncentered independent variable.
> This is a lot harder for standardized slopes, as we have lost the original scale. - 18The main goal of centering (or standardizing) variables> to provide us with useful interpretations
after the interaction effect is added to the model - 19centering is only useful for...> it removes non-essential multicollinearity from the
model
> continuous variables.
> A categorical variable will be translated into code variables before being added to linear regression,
automatically granting us a useful 0. This means we never center categorical variables - 20How does non-essential multicollinearity differentiates itself from essential multicollinearity ?> by being a
product of the chosen scale. If multicollinearity can simply be removed by changing the scale, it is
deemed non-essential, as it isn’t a statistical necessity to keep it in.
> By contrast, essential
multicollinearity is multicollinearity that isn’t caused by the scale, but by the true relation between
independent variables, and it is therefore an essential part of the model. - 21When would we prefer not to center the data ?> If the Z = 0 as setting the mean at 0 would not make the 0 more useful.
> Also, centering (or standardizing) the dependent variable is very rarely a useful step, as it will in most cases only make interpreting the slopes harder without gaining anything else - 22—
Information
- Description
- Flashcards on stats 2 12, created by Adrian Mayer on 10/01/2019.
- Created by
- Adrian Mayer
- Views
- 0
- Rating