1/18
Looks like no tags are added yet.
How do you work out the slope and intercept?
intercept=constant=B0, slope=unstandardized B
How do we know the model fit is a good one?
the closer the actual data and the model are the better, the gap between each pair of scores is called the residual
types of variation residuals
difference between the observed and predicted Y scores, variation that is not predicted by the model/equation/regression, the weaker the prediction the greater the residual variance
How good is our model at predicting the actual data → how much of the variance in the Y data set can we predict/account for using this model?
we need to compare the variance that is explained by our model with all the variance in the data set
we need an estimate of all the variance in our data set
the total variance of Y scores in the data set (SS total), all the variation that there is to explain
How can we get an estimate of the proportion of variance of variation explained by the regression/model?
take away the SS residual from the SS total
Sum of squares of regression (sum of squares of model)
all the variance that can be predicted by our regression line (SStotal-SSresidual) → SSreg or the SSm
What does the SSreg/mean tell us?
can’t compare SSreg/m from different studies and samples → useful if we wnat to generalize and compare results
R²
proportion of the variance predicted by the regression equation (SSreg/m divided by SStotal), larger the better (between 0 and 1), can be expressed as a percentage
SStotal
measure of the amount of variance in our data set
SSres
measure of the amount of variance not explained by our regression
The ‘F’ ratio
way to see if the model/regression/equation predicts a sig amount of the variation, is the ratio between two variances (ratio is between from the variance that is predicted and variance that isn’t predicted/residuals) high=effect is strong/the more variance is explained by the data
equation
Y=b0+b1(x)
basic assumptions
continuous, non-zero variance, independent, linearity
more technical assumptions and checks for bias
homoscedasticity (for each value of the predictors the variance of the error term should be constant), normally distributed errors (residuals normally distributed), outliers and influential cases
to check homosecedasticity and independence of errors
plot residuals against predicted score, ZRESID (y-axis) against ZPRED (x-axis)
to check normally distributed errors
do a normal probability plot, or ‘save’ the residuals and then compute all the usual test for normality
checking for bias: outliers and influential cases
extreme outliers can affect results, can check using residuals → only about 5% should be over 2 SD
Cook’s distance
measure of influence of each case on the model, if Cook’s distance is above 1 then it having ‘undue’ influence on the model/equation