Class Notes 10-21-2004
2 way ANOVA (1 DV, 2 IVs)
EPSY
6301
21 Oct 2004
2 way ANOVA and COVARIANCE will be covered tonight
- we need to understand this and do computations from the examples that
Hinkle uses X, Kirk and Dr O like to use Y for the dependent variable
we have done between groups one way analysis of variance
- Kirk calls it completely randomized design
- total variance = SS between + SS within
but then the researcher realizes there are other variables, and this is not enough
now we are looking at two independent variables
- can setup hypotheses for each one
in univariate or one way, we just look at one dependent varaiable
- now we are looking at alpha and beta variables
- beta creates a non-additive product
- we have multiplicative factors (an interaction)
Y (the score) = mew + alpha + beta + interaction + error
- beta can be a blocker but also something that we can have a hypothesis about
2 way analysis of variance is more efficient because we also get the intraction variable (also called factorial design with 2 treatments, 3 treatments, etc)
- most of the time, we deal with complex phenomena
- interactions can better explain complex phenomena
we don't want to test/research the obvious: looking at interaction effects can be good because it can tell us things we don't know yet
sometimes mathetmatical people don't want a signficant interaction is because it can be tough to explain that
in many of the situations, you want to get signficance on the interaction: this shows synergy that happens when the variables anre interacting
here we are throwing in another variable NOT to say it is a nuisance variable (where we just want to control for that), but because we are interested in studying it
- this is looking at another interesting variable that we can and do form hypotheses about
Three R's are essential for random
score = grand mean + treatment effect A + treatment effect B + ineraction effect + within-cell error effect
NID = normally and independently distributed
create new hypotheses for each new variable
we not only want to know how to do the partitioning, but also account for degrees of freedom in our ANOVA
- once you know the grand mean, there is no freedom
- once you collapse a random variable, it becomes a constant and there is no longer any freedom
- always take one away from number of subjects so you can let your values vary
key is mean squares so you can look at variances
- not means or just sums of squares
- find Mean Squares by dividing sums of squares by its own appropriate degrees of freedom
- the last one does not have freedom to vary for you to get the true mean, so that is why you always take one away when you compute degrees of freedom
degrees of freedom is an artifact of sample size
Goal is to arrive at the F value with a hypothesis test or tests
- for each hypothesis I must calculate the F
= source of interest divided by error
This will be F obtained rows, F obtained columns, F obtained interaction
- for each F I will have a critical value
Now we will work...
2 way ANOVA on the computer
-
independent and "treatment variables" (according to Kirk) is the same thing
As long as we have 2 IV we can do 2 way ANOVA
- even if there is no treatment
in variable view you can assign numbers of for each variables
Do Analyze - General Linear Model - Univariate
- put in DV
- put in fixed factors
Full factorial is full model
- always for type III keep "include intercept in model" checked
I don't understand clearly an example of a "random variable"
- gender is a fixed variable, that is understandable
MY idea: we can record him doing this in SPSS and then put that on the web
This is univariate because we are using 1 DV, but multiple IVs
When you do partial omega squares, partial etas are your friend
- they are the same thing
articles do not use SPSS plots
- Dr O wants to know how to set origin for the plot to zero
21 Oct 2004
2 way ANOVA and COVARIANCE will be covered tonight
- we need to understand this and do computations from the examples that
Hinkle uses X, Kirk and Dr O like to use Y for the dependent variable
we have done between groups one way analysis of variance
- Kirk calls it completely randomized design
- total variance = SS between + SS within
but then the researcher realizes there are other variables, and this is not enough
now we are looking at two independent variables
- can setup hypotheses for each one
in univariate or one way, we just look at one dependent varaiable
- now we are looking at alpha and beta variables
- beta creates a non-additive product
- we have multiplicative factors (an interaction)
Y (the score) = mew + alpha + beta + interaction + error
- beta can be a blocker but also something that we can have a hypothesis about
2 way analysis of variance is more efficient because we also get the intraction variable (also called factorial design with 2 treatments, 3 treatments, etc)
- most of the time, we deal with complex phenomena
- interactions can better explain complex phenomena
we don't want to test/research the obvious: looking at interaction effects can be good because it can tell us things we don't know yet
sometimes mathetmatical people don't want a signficant interaction is because it can be tough to explain that
in many of the situations, you want to get signficance on the interaction: this shows synergy that happens when the variables anre interacting
here we are throwing in another variable NOT to say it is a nuisance variable (where we just want to control for that), but because we are interested in studying it
- this is looking at another interesting variable that we can and do form hypotheses about
Three R's are essential for random
score = grand mean + treatment effect A + treatment effect B + ineraction effect + within-cell error effect
NID = normally and independently distributed
create new hypotheses for each new variable
we not only want to know how to do the partitioning, but also account for degrees of freedom in our ANOVA
- once you know the grand mean, there is no freedom
- once you collapse a random variable, it becomes a constant and there is no longer any freedom
- always take one away from number of subjects so you can let your values vary
key is mean squares so you can look at variances
- not means or just sums of squares
- find Mean Squares by dividing sums of squares by its own appropriate degrees of freedom
- the last one does not have freedom to vary for you to get the true mean, so that is why you always take one away when you compute degrees of freedom
degrees of freedom is an artifact of sample size
Goal is to arrive at the F value with a hypothesis test or tests
- for each hypothesis I must calculate the F
= source of interest divided by error
This will be F obtained rows, F obtained columns, F obtained interaction
- for each F I will have a critical value
Now we will work...
2 way ANOVA on the computer
-
independent and "treatment variables" (according to Kirk) is the same thing
As long as we have 2 IV we can do 2 way ANOVA
- even if there is no treatment
in variable view you can assign numbers of for each variables
Do Analyze - General Linear Model - Univariate
- put in DV
- put in fixed factors
Full factorial is full model
- always for type III keep "include intercept in model" checked
I don't understand clearly an example of a "random variable"
- gender is a fixed variable, that is understandable
MY idea: we can record him doing this in SPSS and then put that on the web
This is univariate because we are using 1 DV, but multiple IVs
When you do partial omega squares, partial etas are your friend
- they are the same thing
articles do not use SPSS plots
- Dr O wants to know how to set origin for the plot to zero
Posted: Thu - October 21, 2004 at 08:43 PM
Quick Links
Calendar
Sun | Mon | Tue | Wed | Thu | Fri | Sat |
Categories
Archives
XML/RSS Feed
Statistics
Total entries in this blog:
Total entries in this category: 6
Published On: Oct 21, 2004 08:43 PM
Total entries in this category: 6
Published On: Oct 21, 2004 08:43 PM
