4. Causality and correlation Flashcards
What is the difference between deterministic versus probabilistic cause?
Science uses a probabilistic meaning of cause, not the determinist sense e.g. Smoking makes it measurably more likely that a person will get lung cancer, than if s/he did not smoke – more probable, not certain A determinist definition - requires that every smoker get cancer to establish cause We will only be considering probabilistic cause.
What is required to conclusively establish a cause-and-effect relationship?
There are three criteria that must be met to establish a cause-effect relationship: The cause must occur before the effect. Whenever the cause occurs, the effect must also occur. There must not be another factor that can explain the relationship between the cause and effect.
For a causal relationship to exist, the variables have to VARY. There has to be VARIATION on both IV and DV. First, the value of the DV has to vary or change If the DV doesn’t vary as the IV changes values, then that particular IV has NO effect. There is no cause-and-effect relationship in this case.
If, in addition, we consider that the values of one variable produce the values of the other varrable, the relationship is a causal relationship. The example of class and voting, noted above, ts “:’example of a causal relationship. We feel that there is something about a persons social class that makes that person more likely to vote in a certain way. Therefore, we ~ay that social class is a “cause’~ of party vote. It is not merely true that the two variables tend to coincide; they tend to coincide because values of the one tend to produce distinct values of the other.
What is the relationship between causation and correlation?
If there is a causal relationship between the DV and IV (or several IVs) We will observe a correlation. Finding a correlation does not prove there is a causal relationship A causal relationship shows correlation but not every correlation is a causal relationship. If there is a causal relationship between two variables, they will be correlated But finding that A and B are correlated, does not prove a causal relationship Statistical significance or strength of the correlation does not prove a causal relationship.
The question of whether or not there is a relationship is objectively testable. The question of whether the relationship is a causal one, and of which variable causes which, requires an interpretation. Generally speaking, all that we know directly from our data is that two variables tend to occur together. To read causality into such co-occurrence, we must add something more, although as you will see, it is possible to design the research in ways that can help us significantly in doing this.
When does correlation demonstrate a causal relationship and when does it not?
A correlation between A and B could be any one of the following:
Direct Causation The IV is the cause of the DV. A B What we are usually interested in An increase in A can cause an increase or a decrease in B. Either is still direct causation. Also becomes causation when there is a coherent and logical theory that backs the correlation. There is a reason for it that is explainable.
Reverse Causation What we thought was the DV actually causes what we thought was the IV. B A E.g. Daudelin article examples: Does the murder rate cause level of policing or does level of policing cause murder rate?
Spurious causation IV and DV are both caused by a third factor – - So “Spurious causation” is not the real cause. This can also include time (below).
Coincidence Apparent relationship just due to random variation We shouldn’t underestimate how much happens due to randomness Statistical significance testing measures how likely it is that the correlation is due to random variation
Tautology (aka Circular reasoning) To confuse explanation of the phenomenon with the definition of it E.g. “He won the election because he got the most votes.” That is the definition of winning, not a cause
What is spurious causation/correlation (including correlation due to time)?
Spurious causation IV and DV are both caused by a third factor – - So “Spurious causation” is not the real cause. How to identify this? You know you have a spurious relationship if controlling for the 3rd variable causes the original correlation to disappear.
Time: Both IV and DV change with Time - two phenomena with similar change over time will be correlated, without any underlying causal relationship - some treat this as a case of spurious correlation (e.g. Tyler Vigen webpage).
What is a tautological relationship? Why will it show high correlation?
Tautology (aka Circular reasoning) To confuse explanation of the phenomenon with the definition of it E.g. “He won the election because he got the most votes.” That is the definition of winning, not a cause.
Also get tautology if IV and DV are highly correlated because they are (almost) identical Can happen if use the same data to observe and operationalize (measure) the IV and DV.
E.g. p. 184 Neumann. Using defining attributes of conservative beliefs as explanation for conservative views Since these are the same thing, they will appear highly correlated.
