Q9 Correlation Flashcards
Correlation
A relationship between two measures that change together
“Relationship”
- May or not be causal
- Correlation does not imply causation
“Measures”
In data, measures are known as variables
Ways data can “change together”
- Increase together
- Decrease together
- Go in opposite directions
Correlations are crucial to . . .
Science
Plausible alternative explanation
A rival rationale that might better explain an observed relationship, or a third variable that could be impacting the correlation
Variable
A unit of measurement that has the potential to vary or change
Four types of variables
- Nominal
- Ordinal
- Interval
- Ratio
Nominal
- Means name, like “nombre” in Spanish
- Gives a name to things without any order or rank
Ordinal
- A nominal variable whose measurements have an order or rank
- The gaps between each rank may not be equal
Interval
- Ordinal variables where the gaps between each rank are equal
- “Interval” refers to the size of the units
- No true zero
Ratio
- Measures equal units, such as elapsed time
- Has a true zero
- Zero could also mean the absence of a measurement, like weight
Type of variable: Religions
Nominal
Type of variable: Letter grades
Ordinal
Type of variable: A temperature scale of Fahrenheit
Interval
Type of variable: A clock
Interval
Type of variable: Stopwatch
Ratio
Categorical variables
- Nominal & Ordinal
- Coarse measure
- Less sensitive to statistics, but can still detect correlation
Type of variable: Money
Ratio
Continuous variables
- Interval & Ratio
- Refined measure (specific)
- More sensitive to statistics, making it easier to detect correlation
X-axis
Independent variable
Y-axis
Dependent variable
Positive correlation
Two measures increase together or decrease together
Negative correlation
As one measure increases, the other decreases
What do positive and negative correlations have in common?
They are causations
A correlation can be a causation if:
- The change is unidirectional
- The change is direct, without a plausible alternative explanation
Unidirectional example with height and weight
- Height causes weight
- Weight does not cause height
Direct change without a plausible alternative explanation example with height and weight
While factors like diet also cause weight, the influence of height is direct
Causal relationships and correlations
Every causal relationship is a correlation, but not every correlation is a causation
Association
- A mutual relationship between two variables that can either be causal, or only a correlation
- Either variable can be the independent variable
Correlation between pay and gender
- Complicated because people and the soceieties we exist in are complicated
- So many variables involves, so causality is hard to pinpoint
Correlation between ice cream and swimsuits sales
- Plausible alternative explanation: summer
Correlation between having a strong interest in the arts and living in the city
- Association
- You could be interest in arts because you live in the city
- Or you could move to the city because you’re interested in arts
Correlation between people who watch crime shows and having mean world syndrome
- Association
- People could have mean world syndrome because they watch crime shows
- Or people could watch crime shows because they have mean world syndrome
How can researchers determine if correlations are meaningful?
Statistics
Statistical significance tests
Determine if a relationship among variables exists beyond chance, with some allowable error
Spurious relationship
A correlation of unrelated variables, or a coincidence
What does “beyond chance” mean?
- p < .05 refers to the amount of error allowed
- p is probability
- .05 is 5%
- Less than 5% chance correlation in error
Relationship between a Gator victory and your lucky socks
Spurious relationship
Testing a correlation while testing a second one
Test the control, which could be a placebo, another treatment, or no treatment at all
Experiment
Comparing correlation for two or more conditions
Good experiment
- Participants must be assigned the conditions at random (random assignment)
- Participants cannot know what condition they’re in, and neither can the practitioners (double-blidn study)
- These factors reduce bias
Randomized controlled trial
Tests a treatment against a control with participants assigned at random, and if possible, is double-blind
If the treatment is stronger than the control . . .
The treatment is causal
Split testing in media
Compares two messages, A and B
Sending ad tweets two weeks apart using different wording for each
- Split testing
- Online analytics determine what generates more engagement