Inductive Logic
When causal arguments on the LSAT become complex, the Three Pillars (Timing, Data, and Mechanism) may not be detailed enough to reveal the subtler shortcomings or gaps in the argument. Data may be present but ill-gathered or weak. A mechanism may be there but could be strengthened or weakened by further concepts or information.
So what do we do?
The Bradford Hill Criteria, developed by Sir Austin Bradford Hill in 1965, are a set of principles used to establish a causal relationship between a specific factor and a disease or outcome in epidemiology and public health. While originally proposed in the context of the association between cigarette smoking and lung cancer (it's no coincidence that we keep using that example), these criteria have since been widely applied to assess causation in various fields.
They are especially useful to us in causal argumentation, as they deepen our ability to examine and assess arguments as well as predict the type of information that could be presented to us in order to strengthen and weaken those arguments
TEMPORALITY
The cause should occur before the effect, establishing a temporal sequence.
Smoking causes cancer. It seems pretty straightforward. But before we even explore the statistics or potentially damning countertheories, we need to know something first: do people start smoking long enough before they develop cancer that this even makes sense?
This is what we called Timing in the three pillars and Temporality here. It is the first building block of the reasonability of our causal claim. If most smokers start late in life but develop cancer in their teens or 20s, it would not be reasonable to conclude that it is the smoking that causes the cancer.
PLAUSIBLE MECHANISM
The proposed cause-and-effect relationship should make sense based on existing knowledge and theories. This plausible mechanism functions much like deductive minor premises, as it is a presentation of facts related to the cause just as minor premises present facts related to the subject of the claim.
OK, so people start smoking early in life and develop cancers later. But how does it happen?
We now get into the weeds. Burn a cigarette, breathe in the smoke, it gets into your lungs, and chemicals are absorbed into the tissues. Those chemicals cause mutations when cells multiply, and those mutations are cancerous. THAT is how is happens; that is the Mechanism.
COHERENCE
The causal relationship should not contradict known facts and theories about how the world works.
A plausible mechanism is only that: plausible. It isn’t necessarily correct and could always use some support.
In the case of causation, support for the mechanism should make the mechanism more reasonable. Our proposal about how smoke gets into your lungs makes sense, but it shouldn’t go against any other things we already know about the world. For instance, if we KNEW that smoke particles are too large to be absorbed or that the types of mutations caused by the smoke are non-cancerous, then this proposed mechanism would not have COHERENCE with the known world.
Similarly, if there are other reasonable ways to explain the effects, other potential mechanisms, that wouldn’t bode well for our argument. So, another way to increase COHERENCE is to dismiss those other possibilities, e.g. other forms of air pollutants such as industrial by-products were controlled for in the studies regarding cigarettes and lung cancers and are thus unlikely causes. This doesn’t PROVE our mechanism is the correct one; it does make it more reasonable, though.
ANALOGY
If similar factors are known to produce similar effects, this may support the causality of a new, related factor and effect.
If Coherence supports the mechanism by dismissing alternatives, ANALOGY works by making the mechanism more reasonable by drawing parallels to similar situations that we already believe to be true.
If we all agree that smoking cigars causes cancer, that certainly makes my suggestion a more reasonable one. It’s a bit weak to be sure, but it can’t hurt.
COHERENCE and ANALOGY make the mechanism more reasonable, but we need more that that. We also need our suggested mechanism to be likely or probable. To demonstrate that, we need data, and we need that data to show that in the presence of the cause the increased probability of the effect is both large and significant.
Yes, those do sound the same. We know. But in statistical terms, they have different meanings. We will do our best to explain without getting technical.
STRENGTH
A strong correlation between two factors suggests a potential causal link.
CONSISTENCY
Repeated observations of the association under different circumstances and settings strengthen the argument for causality
STRENGTH - Magnitude of the association: this means a BIG change in the probability. Smoking doesn’t just cause a 5% or 10% increase in the likelihood of getting lung cancer; it causes a 900% increase! That’s magnitude.
CONSISTENCY - Statistical Significance: this means that the difference in outcomes between the experimental group and the and the control group cannot be chalked up to chance. For instance, if you flip a coin three times and you get three heads, that may just have been random (after all, it will happen about once every eight times you flip three times in a row.) But, if you flip a coin 100 times and get 98 heads, that’s SO unlikely to happen that it indicates there’s something else going on.
But Circuit Logic Geniuses, isn’t that the same thing??!?!?!? If there’s a BIG effect, that’s less likely to be a random occurrence, right?
Actually, no (weirdly…)
Think of the word “significant” to mean “important” or “relevant” and you may have an easier time with this idea. In that light, Statistical Significance is tied to two factors: size of the sample and amount of variation in the outcome.
Size of the sample is simply how many people/things were surveyed or experimented on. As in the above coin-flipping example, with only three tosses, it’s totally possible and pretty common to randomly get three heads in a row. But you’re naturally suspect of 100 heads in a row, and for good reason. The more trials, the less likely that the outcome you’re seeing is just due to randomness.
Amount of Variation has to do with how different each sample or experiment is from all the others. The bigger the differences, the more likely there is randomness. So, with our coin, if we flip 10 times and get 7 heads, that’s well within the realm of randomness. Then we do it again and get 7 heads, ok fine, weird, but still. And we do it again and get 7 heads…eventually, we’re going to start thinking that there’s something going on. Even if the difference from 50/50 isn’t all that big, it’s still significant.
GRADIENT
Greater exposure to the potential cause should lead to a greater incidence of the effect, suggesting a causal connection.
Strength and Consistency of the data is a good indicator, but data is more complex than that. Another critical factor is the GRADIENT: the more cause that is present, the greater the effect becomes.
Consider this: wouldn’t it be more convincing if you knew that
Nonsmokers rarely develop cancer
some of those who smoke ~6 cigarettes per day develop cancer
many of those who smoke a pack a day develop cancer
almost all of those who smoke two or more packs per day develop cancer
This is the GRADIENT, the increase in the effect as you increase the presence of the cause. And, the steeper that increase, the better your case for causation!
BIG NOTE: A very tricky way data can come up short is to tell you “Hey, look, when this cause is present, an effect is present too” but fail to tell you that when the cause is gone the effect goes away as well. That second part is just as important as the first. So, look out for both parts of the GRADIENT.
EXPERIMENTAL EVIDENCE
Any experimental manipulation that alters the supposed cause should change the effect, reinforcing the cause-and-effect relationship.
If there is a direct connection between cause and effect, that should be demonstrable in a experimental setting. With all the other factors controlled, an alteration to the input of the cause should have at least some measurable influence on the effect.
Randomized controlled trials (RCTs) are considered the gold standard for experimental studies because they involve randomly assigning participants to receive either exposure to the cause or a control (no exposure), which helps minimize bias and confounding.
While Experimental Evidence is considered strong, it is not always possible to conduct experiments, especially for long-term or rare exposures.
SPECIFICITY
If a specific factor consistently leads to a specific outcome, this supports the notion of causality, though this is less definitive in complex systems with multiple interacting causes.
This one is relatively intuitive. If you claim “smoking causes cancer,” that’s a big ask. Demonstrating that association requires extensive and wide-reaching data as well as a great deal of research around the mechanism itself given how broad the claim is.
However, if we limit the causal factor to “cigarette smoke” or better yet to “those who smoke at least two packs of cigarettes per day over 40 years” and the effect to “development of metastatic lung cancers” we can home in on specific studies, populations, and experiments. That Specificity will make it all the more likely that what we are seeing will be plausible and repeatable.