From Quizes Flashcards
The purpose of defining families of IVs in a model reduction exercise is to:
Let theory play a role in the analysis
Reduces likelihood of a type two error
Maximises chance the theoretically most important predictors will remain in the model
Decrease the number of hypothesis tests conducted
Is it possible to obtain an approximate hypothesis test concerning the effect of an IV on a DV from the confidence interval? How?
Yes, if zero doesn’t lie within the 95% ci we will reject the null hypothesis that the population slope is zero.
ANOVA - regression suns of squares gives us…
Is the part of the variance that has been accounted by the regression model
ANOVA table - std dev given by…
Finding the square root of the mean square error.
Assumptions of regression and how we assess them?
- Independent observations - normally met by design
- Normal distributions of residuals - via a pp plot - expected (y-axis) vs observed (x-axis), clear snaking around the line?
- Linear relationship between IV and DV - even # above and below the line? Consistent trend or pattern influencing the data?
- Constant variance - fanning?
How do we get ‘r square’ from sum of squares?
Divide regression SOS into total SOS.
Effect of a confounding variable?
The effect of x1 is negated or reduced after controlling for x2
Effect of suppression
The effect of x1 is only apparent or is increased after controlling for x2
Effect of interaction variable
The effect of x1 is estimated to be larger or smaller depending upon the value of x2
Direct & indirect effects of variables?
The effect of x1 acts on Y (at least partially) through x2
Collinearity
This occurs when two or more IV’s are so correlated that one can be predicted (almost) exactly from one or more of the others
Tolerance and VIF tells us? What values?
Statistical indicator of collinearity. Regresses x1 on all other IV’s and assesses how much of the variance in x1 is not accounted for by the other IV’s. this .1 means there is little to distinguish it from a combination of other IV’s. VIF is inverse of tol.
Formula for number of possible pair wise comparisons
(k(k-1))/2 where k is the number of levels
In multiple regression, when we control for potentially confounding variables, we…
Estimate the effect of each IV with the other IV(s) held at a non-specified constant
In a log regression model, which column gives us the odds ratio
Exp(b)