Multiple Regression Flashcards
What is collinearity?
Collinearity is when two or more predictors are highly similar. They CORRELATE highly with each other, meaning they may produce the same or very similar results
What is covariance?
Covariance is when the change in one variable is associated with the change in another one. For example, if a variable changes that has an effect on the other variable
Why is multiple regression based on correlations data?
Because there is no direct manipulation of any predictor variables
What data does multiple regression use?
Scale data: either ratio, interval or ordinal data
Why do we base building a multiple regression model off previous research?
We need a reason for including the predictors we are choosing; do we think they will have a direct influence on our criterion variable?
What are assumptions of multiple regression? (Hint: there are 4 and they revolve around the data)
1) no outliers
2) normality of data
3) linearity of data
4 reliability of data
How is normality of data assessed in MR?
By looking at the skew and kurtosis. Normal distribution is between -2 and +2.
What does MR do it there is not a linear relationship between the criterion variable and the predictors?
It will underestimate the relationship. This increases the type 2 error.
Why does having lots of predictor variables potentially violate an assumption of MR?
Having lots of variables can make the relationship between criterion and predictors non-linear
How do we measure reliability in MR?
By using Cronback Alpha
What is Cronbacks Alpha? What does it measure?
It is a measure of internal consistency: it measures how closely related a group of items are in a set. It measures reliability.
What is homoscedacity? What is it called if you do not have it?
Where the variance is the same for all the predictor variables.
Hetroscedacity is what occurs if you do not have homodecsity.
Why is hetrosecdasity bad for MR results? How do we test for hetrodecsity?
It is bad as it can lead to a distortion in our results, which leads to a type-1 error.
We check for it by looking at residuals
What will the correlation be it multicolinesrity is at a moderate level?
Between 0.3 and 0.8
How many variables (predictors) do you need to be able to do a multiple regression analysis?
Two or more independent variables
How can you check for linearity between criterion and predictor variables in SPSS?
You can check this by looking at scatter plots.
What table in SPSS output helps us see how well a regression model fits the data?
The “model summary” table
How do residuals help us determine linearity in multiple regression?
They help us see if the standardised residuals are in relation to the models predictions. We can see if they are randomly spread or not
How do you check for homoscedasiticity in multiple regression?
By looking at residuals, is it the same?
When you have ran your MR and you are looking at the SPSS output tables, where do you look to assess for multicolinesrity (what table)? What value suggests multicolinesrity?
You look at the correlations table. Correlations greater than 0.8 suggest collinearity
What is tolerance in MR?
How much the variance in the predictor is not explained by other predictors in the model; the unique contribution of the predictor is seen here
Why do we want a high tolerance value?
A high value suggests minimal multicolinesrity
What is VIF (variance inflation factor) in MR?
How much variance in the regression coefficient is caused by multicolinesrity.
How can VIF be interpretated to overcome issues of multicooinesrity?
It can be a number by what to increase or sample size by to overcome issues of multicolinesrity.
A VIF of 2.5 would be 2.5x participants needed
How can you check for outliers?
By looking at the line of best fit
What value indictates a high level of consistency wth Cronbachs Alpha?
0.8
How do you find the critical value?
50+8x( x is the number of participants)
When might you use r2 to look at how much the variability in the criterion variable is explained by our model?
If sample size is greater or equal to critical value
When might you use r2 adjusted value?
When looking to see how much the variability in the criterion variable is explained by our model you see if sample size is less than to critical value
What is a beta value in multiple regression,
How strongly each predictor variables effects and influences the criterion variable
What is a beta value measured in? (What unit)
It is measured in standard deviations
What happens to the criterion variable if the beta coefficient for a predictor is 1 and it is statistically significant?
For every 1 that the predictor variable increases, while the other predictors are held constant, the criterion variable increases by one