Card Set Information
What is a completely randomised design?
A true experiment
P's are placed randomly into only 1 condition
This reduces bias as individual differences are distributed non-systematically across conditions
What are quasi experiments?
Naturally occurring DV
Logically impossible to assign randomly
Should not impute causality as the design is correlational and there could variables that change systematically with the DV (confounding variable)
Which experiments require an independent T test?
Either true or Quasi
What assumptions are made when using an independent T?
Homogeneity of variance
What are the null hypotheses for independent T?
H0=u1-u2=0 (for when there are 2 experimental groups)
This shows that the results are from the same population of scores
What are the alternate hypotheses for a 2 sided independent T?
HA=u1≠u2 or HA=u1-u2≠0
What are the alternate hypotheses for a 1 sided independent T?
What do we have to consider for the independent T and why?
The sampling distribution of the difference between the 2 means
The scores are not in pairs
What does the variance sum law state?
The variance of a sum or difference between two independent random variables is equal to the sum of their respective variances
What is the standard error equal to for the independent T?
The standard deviation of the set divided by the square root of the number of data points in the set
What is the variance equal to for the independent T?
Standard deviation squared over the number of data points
When can we compute z scores?
When the SD is known
What do we do when we cannot find the SD?
Estimate it using the two sample standard deviations
Average information from both samples to provide a pooled estimate of the parameter value
How do we obtain a T value without an SD?
We substitute the estimate for the parameter value with the actual one
How do we estimate the parameter value for the T test?
Take a weighted average of the two sample standard deviations
This is because of the homogeneity of variance assumption
Why do we use a weighted average?
To give a more accurate estimate
If the sample sizes are different the average gives more weight to the larger one
How do we find this weighted average?
Add the sum of squares
Add (n1-1) + (n2-1)
Square root the whole thing