Flashcards
RM Lecture 2 - Correlation
Cards in this deck (36)
- 1What is correlation?A way of measuring the extent to which two variables are related. It tells you the pattern of responses
- 2What do we calculate to work our what happens to one variable as another changes?The covariance
- 3What is variance?How much the data varies around the mean - the spread of the data
- 4What is another word for variance?Noise
- 5What is the equation for variance?E(xi-x)2/N-1
x = mean
xi = individual data point - 6How can we tell if theres a relationship between two variables?As one variable changes the other should change in a consistent way.
- 7Look at sweets example.
- 8Why do we have to divide by N-1?Because otherwise we can't tell if the effect is big because there are lots of p's with a small effect or if its a genuinly big effect
- 9Cross-product deviationsTimes the residuals from the two variables together.
Big number = effect
0 = no effect - 10Average cross-product deviationsCross-product deviation divided by N-1
- 11Equation for covarienceE(xi-x)(yi-y)
N-1
(for two variables) - 12What does positive and negative covariance mean?Positive = going in the same direction
Negative = going in different directions - 13What does a higher number in covariance mean?A bigger effect
- 14What is the problem with covariance?It depends on units of measurement so we can't compare it across data sets?
- 15How do we solve the units of measurement problem with covariance?By standardising it
- 16How do we standardise covariance?Divide it by the total variance.
r = Covariance/sx and sy (SD of first data set * SD of second data set) - 17What is r?The correlation coefficient (the standardised covariance)
- 18Does r have a leading 0?no
- 19What does r always lie between?-1 and 1
- 20How can r be interpreted?r > 0 = things go up together
r < 0 = things go different ways
r = 0 = no relationship - 21What does r tell us?How clustered around the line things are, not the actual slope itself.
- 22What are the non-parametric alternatives to Pearsons r?Spearman's Correlation coefficient
Kendalls Tau - 23What does the data have to be in order to run a Pearson's correlation?Normally distributed
- 24Spearmans Correlation Coefficient- Ranks the data and computes the correlation based on the ranks
- Also called Spearman's Rho
- Used a lot - 25Kendall's Tau- Should be used when there is a small data set with a large number of tied ranks (12 p's)
- Better estimate of the correlation in the population
-Often used because you can draw more accurate generalisations from it - 26How do we interpret correlations?1 = small effect
3 = medium effect
5 = large effect - 27What does partial correlation show us?Shows if the correlation strengthens/weakens when certain variables are controlled for.
- 28When would we need to transform data?- Data sets that have a peak and a long tail.
- When checks for normality show some raw scores aren't normal - 29What is a log transformeg. a data set of 10(1) 10(2) 10(3)… becomes 1,2,3…
- So it only uses the exponents so that it becomes linear. - 30What is bad about log transforms?The benefits aren't always clear
- 31Why do we accept simple log transforms?Because exponential falloff happens so often.
- 32When wouldn't we do a log transform.When using Pearson's it is quite robust so it may be alright if the normality tests aren't quite passed.
- 33What do we have to say when we've used a log transform?The log of x correlates with the log of y
- 34What do correlation coefficients give no indication of?The direction of causality
- 35Why do correlation coefficients give no indication of the direction of causality?1. Third variable problem
2. Spurious correlations - 36Spurious correlationsTwo things that are correlated but have nothing to do with each other
Information
- Description
- Flashcards on RM Lecture 2 - Correlation, created by hannahbevans on 30/12/2014.
- Created by
- hannahbevans
- Views
- 0
- Rating