Chapter 16 Flashcards
(8 cards)
1
Q
random sampling distribution of r
A
- Not normally distributed
- If rho = 0, it’s symmetrical and close to normal
- If rho =/= 0, it’s skewed because rho can’t exceed +/- 1 -> larger the absolute value of rho, greater skew
- Smaller n = greater skew
- Values of r will vary less from sample to sample when n is large
- Values of r will vary less from sample to sample when rho is strong
2
Q
when is the random sampling distribution of r skewed?
A
- If rho =/= 0, it’s skewed because rho can’t exceed +/- 1 -> larger the absolute value of rho, greater skew
- Smaller n = greater skew
- If rho = 0, it’s symmetrical and close to normal
3
Q
rho
A
- Pearson’s correlation coefficient for population of paired scores
- Only absolute value matters (it gets squared, no +/- not important)
4
Q
desirable properties of fisher’s z transformation
A
- Sampling distribution is normal (regardless of rho or n)
- Standard error of z’ is independent of rho values
- Non-linear transformation that will change the shape of the distribution (and will normalize it)
- Magnitude: if r is negative, zF is also negative (don’t forget to put +/- in your answer!)
5
Q
What can you do with perfect correlations? Why can you do it?
A
If you have perfect correlations, you can create a formula to convert one value to another (because there’s no variability)
6
Q
why do we prefer the t-test for correlations?
A
- T-test is better than z-test because it’s more powerful for correlations (so we don’t use the z-test)
- T-test gains its power by using sample estimate of r to estimate standard error
7
Q
H0 for testing correlations
A
- that they’re 0 -> any significant correlation means that there’s little chance that the correlation is 0
- Remember that H0 can never be true, just a plausible option
8
Q
1-tail vs. 2-tail tests
A
- 2-tail: detects significant correlations in either the positive or negative direction (if H0 is rejected, we can conclude either that correlation is > or < 0)
- 1-tail: if it’s logical that the correlation won’t be 0 but can’t be negative (or vice versa), can use a 1-tail test (more power -> critical region concentrated in one tail)