multiple regression Flashcards
aims of MR
to assess the whether multiple IVs can predict the outcome (DV)
difference between unstandardised and standardised regression coefficients
unstandardised coeffieicents show for every 1 unit increase in one of your variables the other increases by the B value
standardised coefficients show this change in standard deviations as shown by beta weights
r2
the correlation coefficient squared
what does r2 tell us
the proportion of the variance in the y scores explained by the line
f=
variance in y (DV) explain by x (IV)/total variance
which is shown by a percentage - r2
regression equation
y=a+bx
a= constant
b=slope
y=DV
regression line - what do you have to assume?
to work out the equation of the regression line have to assume that the X scores are correct and ones that do not fit on the line are due to random error
regression coefficient
shows the estimated amount of change that occurs in the DV for one unit change in the IV with the effect of all other ICs in the equation partialled out
dichotomous variables
B value shoes how much a shift from conservative (0) to labour (1) / Lib den (2) changes the DV e.g. intention to attend a ralley.
entry options
direct
hierachical
stepwise
direct entry
the analyses tests a single model
all IVs entered simultaneously
assesses the joint and individual effects of a fixed set of predictors
to what extent can drinking motives and the demographic variables of age and gender explain alcohol consumption
direct entry
to what extend can drinking motives explain variance in alcohol consumptions over and above that explained by the demographic variables of gender and age
hierachical entry
entered in two blocks:
1) gender and age
2) Drinking motives: coping, enhancement,social, conformity
will a given entry of a IV significantly increase the R2?
stepwise entry
hierachial entry
assesses the effect of one set of predictors OVER and ABOVE that of another
IVs are entered in USER-DETERMINED order
based on THEORY
IVs can be entered in GROUPS or INDIVIDUALLY
stepwise entry
find an efficient model with the minimum numbers of significant predictors
(SPSS determines the order of entry)
IVs are entered or discard step by step according to statisitical criterion i.e if they significantly increase the R2
at each stage of hierachical
the increase in r2 is examined when a new IV is added
can asses the model as a whole and the unique contribution of each IV added so far
typically enter hirachically
in groups e.g. demographics
when put in more demographic variables and psychological variables how much additional variance is explained
types of stepwise regression entry
- forward selection
- backward selection
- forward and backward combined
forward selection
add the IV with the highest correlation with the DV (if significant)
calculate regression
assesses increase of R2 each IV would add the largest amount of explained variance (providing it’s significant)
and then calculates the regression again adding the next best IV
the procedure terminate when no addition IV can significantly explain any of the remaining variance
backward selection
starts with all the IVS
SPSS runs the regression
assess the contribution of each IV on the R2 and removes the ones with the least contribution providing it’s non significant
then runs the regression again and does the same
forward and back combined
starts with all the IVs with no specific predictors in the model
if they meet statistical criteria for entry then they get added
but also deleted the ones that don’t significantly contribute to the R2 variance
reporting a stepwise
a stepwise MR was performed with the DV ….of and the CANDIDATE IV of ….. for entry into the model
number of IVS entered
and number of IVS removed
together these accounted for % of the variance in the DV R2= f(df1, df2) p=
stepwise regression entry is useful when
it doesn’t matter which IVs are included and which are not-
only important which account for the most amount of variance
the objective is to create a prediction equation with the minimum number of significant predictors
when doing something very applied e.g. predicting long term health status/
how much people earn on the basis of their childhood
criticisms of stepwise regression entry
risky procedure
it surrenders decision making to statistics
the meaning and interpretation of the variables are not relevant in the analysis
capitalises on chance relationships
an important predictor may be excluded because of it’s correlation with another IV
may lead to suboptimal solution in term s of variance explained (there may exist a different combination that explains more variance)
as psychologists we want to test a theory and not just blindly see which IVs account for the most variance
