Chapter 10 Flashcards
ANOVA. What is it?
ANOVA = Analysis of Variance
used when comparing the means from a single dependent variable among TWO OR MORE groups or samples
• ANOVA determines whether diff’s seen are sig larger than one would expect to see by chance
Which tests are used for 2 samples? At what levels of data?
• When trying to look for difference between TWO groups: use X2 (Chi) for test for outcome variable that is nominal or ordinal. Use T-test is outcome var is interval/ratio
How could you use Chi and T-Test to test with multiple groups?
What does this increase risk of? What to use instead?
- If had multiple samples, could do multiple T tests but would risk type 1 error each time (equal to alpha so doing test of 3 variables would mean risk of type 1 error = 0.15)
- So…use ANOVA instead!
What ‘ratio’ is created by ANOVA?
What value for this ratio would you expect if the H0 is correct
• Equation produces an F-ratio that relates the diff between groups (numerator) to diff within groups (denominator)
o When H0 is correct, F-ratio is close to 1. Means diff between groups is similar to diff amoung groups, and therefore F-ratio is not sig
How do you get degrees of freedom for F ratio?
• The numerator of the F-ratio has one # of df (number of groups - 1) and denom has another (# of subjects - # of groups)
o Add these two to # of df for assoc with the F ratio
How do you determine if the F-ratio is stat sig?
- F-ratio is like any other stat measure: has P-calue that determines significance
- Can look up p-value on table for F-distribution (or put into software)
- Determining stat sig is just like all previous tests
Does determining the stat sig through the F ratio (using ANOVA test) tell you where the statistically significant results lie? (as in, will it identify between what variables this stat sig is true?
No - cannot conclude where the stat sig diff may lie from this test alone (could be between any two groups or all three)…need further testing to make that conclusion
If you use the ANOVA test on 2 groups will you get the same result as the T test?
Yes
What are the 3 assumptions you must meet to use ANOVA? How much do we actually need to be concerned with each?
- Samples must be indep (ideally random) and measure must be interval or ratio level
- Sample should have normal dist – but remember central limit theorem…because we are comparing means, we can assume normality. The dist of the original pop doesn’t matter; the means will be normally distributed
- Homogeneity of variance: equal variances among the groups being compared. However, this is not terribly concerning at this level because ANOVA still works pretty well if this assumption is not met
What is repeat measures ANOVA?
- examines a change over time in the same sample population
- Useful for dependent variables
- Ex: Measuring BMI before and after an intervention
When you use the same population before and after an intervention (as in repeat measures ANOVA), what are you increasing?
• Using the same pop before and after increases your likelihood of finding stat sig results if they exist = therefore increasing power!
What are the two main concerned with RM ANOVA?
1) Position or latency effects: occurs when a subject is being exposed to more than one treatment over time and the order of treatment received impacts the outcome
1) Carry over effects: occur when previous treatments continue to have effect through the next treatment, affecting the measurement of the dependent variable.
How can “position or latency effects” be addressed in a study?
by randomly assigning order of interventions
When is using RM ANOVA particularly helpful?
• RM ANOVA can be very helpful not only in dec individual variation error (because using same gorup before and after) but also decreases required sample size to find sig results (so helpful if can’t recruit large sample)
What assumptions must be met for RM ANOVA?
• Most assumptions are same as for ANOVA, but one more is Compound symmetry: measurements are correlated and of equal variance o If measuring BMI 3 times, three results should be correlated with one another and approx. the same o Homogeneity of variance should also be present and is term used to indicate that those correlated BMI measurements need to have approx. equal variances o SPSS (stat software) will tell you if these requirements are being met