Causes and Effects

The goal of many scientific inquiries is to gather good evidence and arguments for drawing a conclusion about causes and effects.  That is, we want to know what causes what in the world and why.  We’ve characterized being rational as making better decisions and forming better beliefs in the service of achieving our goals.  If you want to be successful at achieving your goals in the world, you need to know which causes bring about which effects.  As we saw in the last chapter, if you make a decision about a prospective roommate, for example, on the basis of their astrological sign, you’d be making a mistake.  You might have ruled her out because she’s a Pisces, and you think that they are too sensitive.   Having a solid grasp on what constitutes good evidence for a causal connection in the world will help you succeed.  Consider these causal claims that we believe are true:

Smoking causes cancer among humans. 

Increased carbon in the atmosphere causes global warming.

High fat diet causes heart disease among humans.   

Alcohol consumption causes liver disease in mammals.

A mutation in the BRCA1 and BRCA2 gene causes breast cancer among women.  

A family history of alcohol abuse causes alcoholism among humans.    

Maternal stress causes low birthweight among mothers. 

High impact exercise causes increased bone density among humans.

Slow stretching causes reduced performance at explosive, power sports among humans.

Higher rep counts in weight lifting causes hypertrophy among humans.

Creatine causes muscle growth and recovery among humans.  

Benzoyl peroxide causes improvement in acne among humans.  

Standard form for Causal Statements

When we encounter causal claims, we will put them into standard form with three elements, the cause, the effect, and the population:

C causes E among P

What does C causes E among P mean?

There is a great deal of technical discussion in the philosophy of science about the meaning and nature of causation.  For our purposes, we can make several points about what it is not, and we can adopt a general definition that will serve us in ordinary critical thinking situations.  

Sometimes, people misconstrue a claim like, “Smoking causes cancer,” as meaning that “Every time the cause ( C ) is present, then E occurs.  That appears to be how they are interpreting the claim when someone who has been smoking for decades says, “That can’t be true.  I’ve been smoking for years and I don’t have cancer.”  Or, suppose a headline reads, “Suffering childhood abuse leads to alcoholism in adults,” and Jones replies, “But I’m not an alcoholic,” or Smith says, “I know someone who is an alcoholic and they didn’t suffer childhood abuse.”  All of these reactions indicate that they are understanding C causes E as meaning “in every case that C is present, then E occurs, and E can’t or doesn’t occur in cases where C didn’t happen first.”  A little reflection shows that those are misunderstandings of what causation means.  We all know that running red lights cause accidents, even if you get away with it a few times, and that accidents happen even without running a red light.  

A much weaker interpretation of what C causes E among P means would be something like, “sometimes, when C occurs, then E occurs.”  This reading is too weak.  Random or occasional connections aren’t enough to establish a causal connection.  Sometimes, when I read my astrology forecast, it appears to be correct.  And sometimes it rains on Mondays.  But causation, to be useful, must mean more.  

Does “C causes E among P,” mean “Most of the time, when C happens, then E occurs”?  Take another example that we all know is true:  Driving drunk causes traffic accidents.  Does the effect occur most of the time?  Probably not.  The majority of drunk driving cases probably do not result in an accident, nevertheless, drunk driving causes accidents is still true.  There are many cause and effect relationships that happen in the minority of cases.  About 15% of smokers get lung cancer in contrast to less than 1% of non-smokers.  So it doesn’t mean “most of the time, when C happens, then E occurs.”  

Does a causal connection between C and E, like smoking and cancer, or drunk driving and accidents mean that there is a correlation?  What we learned in the last chapter is that correlation alone does not imply causation.  Having big shoes is positively correlated with being in the NBA, and hospital visits are positively correlated with dying.  But big shoes don’t cause getting into the NBA, and it’s being sick that leads to hospital visits and dying, but the hospital visit isn’t the cause.  Correlation does not imply causation.  But all causal relationships are correlations.  So we’re making progress triangulating on what C causes E among P means.  

C causes E among P equals, by definition: 

We will adopt an account of causation for this course that is probabilistic and counterfactual.  

A cause is an event or state that raises the probability of the effect when it is present; had the cause not been present, all other things being equal, the probability of the effect would have been lower.    

An effect depends upon a cause to occur, although other causes can bring it about.  A cause raises the probability of the effect.  Correlation is necessary but not sufficient for a causal relationship.  The definition is probabilistic, not deterministic.  Causes don’t necessitate their effects, they just raise the probability.  And the account is counter-factual; we describe causes and effects in terms of the difference between what happens and what would have happened had the cause not been present.  

Smoking causes cancer among humans, then, means that a person’s smoking raises the probability that that person will get cancer.  All other things being equal, had the smoking in the human not been present, the increased probability of cancer would not have come about.  

Since causes are probabilistic, the best evidence for them typically emerge in populations where we can see patterns.  At the individual or anecdotal level, it’s often hard to know about causes since a cause can occur without the effect happening, and the effect can happen without the cause.  Probabilities show up as rates in large populations, while they may not be evidence at the anecdotal or single case level.  

In a population, we would expect to see, then, or predict that, if they smoke, and all other variables are controlled for, there will be a higher rate of cancer in that population.  

In a population taking Ozempic, we would expect to see a higher rate of them losing weight than a placebo or control group.  

In a population taking creatine, we would expect to see a higher rate of them gaining muscle mass and strength. 

In a population treating acne with benzoyl peroxide, we would expect to see a higher rate of improvement than in a placebo or control group.  

When is it reasonable to believe a causal claim?  

 Now that we’ve defined causes and effects, what can we say about when it is reasonable to believe a causal claim.  What evidence or arguments should lead us to believe a causal conclusion?  Earlier, we said that all causal relationships are correlations, but not all correlations are causal.  How can we investigate correlations and further determine which ones are causal?  We will do it with a familiar argument by elimination.  We discover correlations in four different circumstances.  1) There are accidental random correlations that don’t indicate anything more important.  2) Sometimes there are correlations between A and B that are the result of a third cause, like the big shoes/NBA, or hospital visit/death correlations above.  A third cause, genetics, contributes to shoe size and NBA performance.  And a third cause, being sick, leads to both hospital visits and death.  3) Since correlations are symmetrical, sometimes we discover that A is correlated with B because B causes A.  For example, some educators noticed that self-esteem is positively correlated with good grades and mistakenly thought that if we could elevate a child’s self-esteem with participation trophies and constant praise, then that would cause or lead to their grades going up.  Unfortunately, they had the causal arrow reversed.  Good grades cause higher self esteem, not the other way around.  And finally, 4) sometimes when we discover that A is positively correlated with B among P, it is because A causes B.  A strong argument for a causal conclusion, therefore, will eliminate the first three of these explanations for a correlation in favor of the fourth.  This leads us to this template.  

Template for a Causal Argument

When we encounter scientific reasoning, reporting, or arguments, it will help to be able to reconstruct them according to this deductively valid template.  Every causal argument must start with a correlation claim.  If there is no correlation, then there is no causation.  So causal arguments start where our complete correlation arguments from the last chapter finished.  With a correlation:

1.  A is positively/negatively correlated with B among P.  

2. If A is positively/negatively correlated with B among P, then a) A is accidentally correlated with B, b) a third cause brings the correlation about, c) B causes A, or d) A causes B  

______________________________

3. Therefore, either A is accidentally correlated with B, or a third cause brought the correlation about, or B causes A, or A causes B  (1, 2, MP)

4. A is not accidentally correlated with B.  

5. A third cause did not bring about the correlation.  

6. B does not cause A.

______________________________

7.  Therefore, A causes B among P.  (3, 4, 5, 6, DS)

How do we know whether a correlation is accidental?  Sometimes, we discover correlations in data sets that are just random, meaningly associations, or they are the result of small sample sizes.  When I flip a coin ten times, for example, it can happen that I get 8 heads, which might lead me to think that something is amiss; maybe my coin isn’t a fair coin.  But as the sample size, the number of coin flips, goes up, the distribution of heads and tails will approach closer and closer to 50/50.  In a room with 20 people, it can happen that several of them have a last name that starts with the letter “R” even though the rate of that in the general population is lower.  Daniel Kahneman calls this the Law of Small Numbers:  extreme outcomes, both high and low, are more likely to be found in small rather than large samples.  So sometimes we discover a correlation and it’s just an accident with no causal significance.  In causal arguments and research, the way to rule these sorts of results out is to have a large N, or number in the sample population.  If the study has a large N, that tends to support premise 4.  If the study has a small N, then that tends to undermine support for premise 4.  

How do we know whether a third cause brought about a correlation?  Finding or generating supporting evidence for premise 5 is more difficult.  In general, good causal arguments and research need to investigate and control for third causes.  In research that shows a correlation between smoking and cancer, for example, we might wonder whether poverty might contribute both to people’s tendency to smoke and their exposure to other non-smoking carcinogens.  Or we might wonder whether genetics might predispose some people to both smoking and to cancer.  The way to control for these possible third causes, then, would be to investigate more populations with different economic status and see if among affluent people the correlation between smoking and cancer is still present.  If it is, then we’ve controlled for income.  If we investigate genetically diverse groups of people, and we still find the correlation is present between smoking and cancer, then we’ve controlled for genes as a third cause, and that’s more evidence in support of premise 5.  In general, the more possible third causes that causal research investigates and controls for, the stronger the overall argument will be.  In early research, it was found that heavy coffee drinkers had a higher risk of heart disease; that is, heavy coffee drinking is positively correlated with heart disease.  This led to the suspicion that the coffee might be causing heart disease.  But further investigation into possible third causes revealed that heavy coffee drinking and smoking were also correlated. And once researchers controlled for smoking status, the link between coffee and heart disease disappeared.  To find support for premise 5 in a causal study, look for information about the researchers “controlling for third causes,” or ruling out the possibility of other variables, filtering, or accounting for other causes.  

How do we know whether B causes A?  When we discover a correlation, we can’t tell from that alone whether A might have caused B, or if B caused A.  That is, we might find that depression is negatively correlated with exercise.  People who exercise are less likely to be depressed.  Or income and health are positively correlated.  But it might be that having low depression enables people to exercise, or it might be that exercise leads to lower depression.  Likewise, being healthy might lead to more productivity and energy for income earning.  Or earning income might create the resources to become more healthy.  Correlations, as we have seen, by themselves are symmetrical.  So we can’t know which one is the cause and which one is the effect without some additional argument or considerations.  The answer is relatively simple.  We know that causes always precede their effects in time.  That is, effects always come after their causes.  So in a well designed study, we will know, or the researchers will construct the investigation such that we know or we can figure out which one, A or B, came first.  That is our candidate for the cause.  So in a smoking and cancer study, for example, if we know that the smoking preceded the cancer, then we can argue in premise 6 that cancer does not cause smoking because smoking came first and causes always come before their effects.  Support for premise 6 will be information about the time order of the two properties that were correlated in the first premise.  

Therefore, A causes B.  If we have found a correlation (stated in premise 1), and we know there are 4 possible explanations for a correlation, and we have eliminated 3 of those explanations, then the only remaining answer is that A causes B.  Thus, we have a deductively valid argument for the conclusion.  

Example: Causal Argument Template Applied  While we don’t have the actual research here in front of us, we can sketch in the reasoning and the evidence that is probably common sense to most people at this point about smoking and cancer.  For this course, we will have many cases where we will look at some science reporting about a causal connection and we will extract these elements from the article and reconstruct the argument charitably into this format.  

1. Smoking is positively correlated with cancer among humans.  (EP) 

2. If smoking is positively correlated with cancer among humans, then a) smoking is accidentally correlated with cancer, or b) a third cause brings the correlation about, or c) cancer causes smoking, or d) smoking causes cancer.  (IP)

__________________________________

3. Therefore,  a) smoking is accidentally correlated with cancer, b) a third cause brings the correlation about, c) cancer causes smoking, or d) smoking causes cancer.  (1, 2, MP)

4. Smoking is not accidentally correlated with cancer.  The correlation is robust, well-established, corroborated by multiple studies with large sample populations. 

5. A third cause did not bring about the correlation.  Numerous studies have controlled for possible third causes such as environmental causes, genetics, geography, age, income, etc. and the corr is still present.    

6. Cancer does not cause smoking.  Causes come before their effects in time.  Smoking happens first, then cancer follows. 

______________________________

7.  Therefore, smoking causes cancer among humans.  (3, 4, 5, 6, DS)

Another example:  Weight Training and Diabetes

Consider the information about the connection between strength training and diabetes in this report:

A new longitudinal study from the University of Copenhagen suggests that regular strength training causally reduces the risk of developing Type 2 diabetes among middle-aged adults. Researchers followed 8,200 participants aged 40–65 for twelve years, collecting data on exercise frequency, muscle mass, body composition, and blood glucose levels. After adjusting for confounders including BMI, diet quality, and overall physical activity level, those who engaged in at least two weekly sessions of resistance training had a 30 percent lower incidence of diabetes compared with those who did none.

The correlation remained significant even after controlling for aerobic exercise, income, and family history of diabetes—variables that might otherwise explain the effect. The team used repeated measures and verified self-reports with biomarker data to minimize error. Statistical modeling indicated that the probability of obtaining such a strong association by chance alone was less than one in ten thousand (p < 0.0001).

Because the effect persisted across subgroups and was mediated by improved insulin sensitivity and lean muscle mass, the authors concluded that resistance training itself—not merely correlated lifestyle factors—plays a causal role in lowering diabetes risk. They propose muscle-building exercise as a direct preventive intervention for metabolic disease.

1. Regular strength training is negatively correlated with Type 3 diabetes among middle aged adults.  (EP)

2. If regular strength training is negatively correlated with Type 2 diabetes among middle aged adults,  then either a) regular strength training is accidentally correlated with Type 2 diabetes, or b) a third cause brings the correlation between strength training and diabetes, or c) a reduction in diabetes causes strength training, or d) regular strength training causes a reduction in Type 2 diabetes among middle aged adults.   (IP)

______________________________

3. Therefore, either a) regular strength training is accidentally correlated with Type 2 diabetes, or b) a third cause brings the correlation between strength training and diabetes, or c) a reduction in diabetes causes strength training, or d) regular strength training causes a reduction in Type 2 diabetes among middle aged adults.   (1, 2, MP)

4.  The correlation between strength training and diabetes is not accidental.  The study was conducted on 8,200 participants, which is a high N, reducing the chance of an accidental correlation.  (EP)

5. A third cause did not bring about the correlation between strength training and diabetes.  The researchers controlled fro aerobic exercise, income, and family history.  (EP)  

6. Reduction in diabetes did not causes strength training.  In the study, participants with diabetes started the strength training first, and then the reduction happened second.  And effects always come after their causes.  (EP)

______________________________

7. Therefore, regular strength training causes a reduction in Type 2 diabetes among middle aged adults.  (3, 4, 5, 6, DS)

Our goal is to improve our critical thinking about causation, and to improve our scientific literacy.  We want to be in the habit when we encounter information or reports about scientific causal research of looking for some of these important elements. We should be looking to see if a correlation was established.  We should be looking for the sample size, or the population number.  The N of these studies matters for the strength of their arguments.  We should also be looking for whether they controlled for third causal factors that might be responsible for the correlation.  

Causal Mistakes

There are a range of biases, mistakes, and fallacies that we are covering over the whole course, but there are enough causal argument specific mistakes to warrant some separate discussion here.  This is a list of reasoning errors that often occur in the context of causal arguments, with explanations, examples, and strategies for avoiding them.  And now that we have a template in place for understanding how causal arguments should be reconstructed and the major conceptual issues, we can see where our causal reasoning goes wrong.  

No correlation, no causation. 

We are prone to overactive causal theorizing.  That is, humans have a causal bias in favor of finding causal connections quickly and on the basis of little information.  Evolution built us to jump to causal conclusions about what’s going on in the world because it matters so much to our survival.  An athlete notices that he wore a particular shirt when his team won an important game, so he starts wearing it every time, hoping for the same result.  A woman feels nauseous after eating eating from a new taco stand and she gets turned off of tacos completely for years.  A roommate tries a herbal cold remedy from the health food store and feels better, and concludes that the remedy works to cure colds.  

The lesson we should learn from this chapter that lays out the specific and necessary steps for a reasonable causal conclusion is that if there is no correlation, then there is no causation.  If A is not correlated with B, then we won’t have an argument that A and B are causally connected.  And we won’t have evidence for a correlation from a single case or even a few anecdotes.  Correlations emerge as rates in larger representative sample populations.  

Imagine that your roommate says that taking vitamin E prevents colds.  You look it up and a research lab has tested carefully with a test group that takes vitamin E, and a control group that does not.  Over the span of the study, the rate of colds in the test group turns out to be the same as the control group.  They don’t get colds at a lower rate or a higher rate.  But if there’s no correlation, then on what grounds would we think that taking vitamin E causes a reduction in colds?  If there’s no detectable correlation rate difference, then it would appear that vitamin E has no impact on colds one way or the other.  One of the first questions to ask when we suspect that there might be a causal connection should be, is there a correlation?  If there’s no correlation, there’s no causation.  

This point was the heart of the argument against astrology in the correlations chapter.  

1. If astrology works, then there would be correlations between astrological signs and personality traits. 

2. There are no correlations between astrological signs and personality traits.  

_________________________________

3.  Therefore, astrology doesn’t work.  

If being born between particular dates has some effect on personality, then we would expect to see those personality differences show up as higher rates or correlations.  

When we think that there might be some causal connection in the world, our first question should be, “Do I have evidence for thinking that there is a correlation?”  If not, then the causal argument is a non-starter.  

Placebo effect  

With medical treatments of humans in particular, the belief or expectation that it is going to treat or cure us has a strong effect on us.  The placebo effect is a positive or negative reaction that people have to expectations or beliefs about medical treatment.  Sometimes, believing that you will get better makes people feel better.  But it is the belief, the expectation, and other psychological factors, not the treatment that is responsible.  Consider a trial for the effectiveness of an anti-depressant.  A well designed trial would, among other things, give the anti-depressant to the test group, but it would also give a harmless sugar pill to a control group, and the results of both would be compared to a group that receives no treatment at all.  Since the placebo effect will contribute to people in the control group feeling better, the anti-depressant, in order to be deemed effective, will need to have even more effect on the test group than the placebo group.  The real pharmacological and medical effect of the drug must be compared to the effect that people’s expectations about treatment. 

Countless bogus, ineffective remedies get credit for working when it’s just the placebo effect.  A social media influencer tries apple cider vinegar, and thinks that her digestion is better; she credits the apple cider vinegar as the cause.  A fitness influencer reports that he’s had better sleep after using a sponsor’s blue light therapy before bed.  A beautiful celebrity claims that her hair feels more healthy and thick after using an herbal shampoo.  

When we think that there might be a causal relationship between some remedy, treatment, fitness routine, beauty treatment, or diet change, and some good outcome, we should ask ourselves the questions, “Am I experiencing the placebo effect?   Is this testimonial the result of the placebo effect?

Ignoring Regression to the Mean

Regression to the mean is the statistical tendency for an extreme performance or event to be followed by one closer to the average.  Because we are highly prone to find causal connections, we fabricate causes where none is needed or evident.  

We might notice and comment, “why is it that tall women prefer and end up with shorter men?” and postulate that it’s because she doesn’t want to call more attention to her tallness as a couple.  Or why do high IQ people tend to end up with lower IQ partners?  Because they want a partner who doesn’t compete with them intellectually.  Or in both cases, “opposites attract.”  But in both cases, no causal story is warranted or needed.  Statistically, for a tall woman, the pool of men shorter than her is larger than it would be for a short woman.  The chance of her finding a shorter partner are simply higher, without any psychologizing.  The same goes for someone with a high IQ.  

Ignoring regression to the mean is common in professional sports.  There’s a widespread belief in the Sports Illustrated Cover Jinx, or the Madden Curse.  Once a fast rising new athlete gets on the cover of Sports Illustrated or the Madden video game, they are cursed, their performance plummets, or they will get injured.  

But the normal cycle of athletic performance is to swing up and down to extremes, and then trend back to the middle.  The cover of the magazine or the video game didn’t cause the regression. People’s attention gets drawn to an event because it is unusual, and they fail to anticipate that anything associated with that event will probably not be quite as unusual as that event was.  Instead, they come up with fallacious causal explanations for what in fact was a statistical inevitability.

Imagine a parent who’s getting conflicting advice from parenting book that insists that children who misbehave need firm punishment to correct their behavior, and from another parenting book that says that children who are praised tend to correct their bad behavior.  Both books cite cases where the approach appeared to cause kids to improve their behavior.  But regression to the mean would have resulting in that outcome either way.  

Confusing cause and effect  

Sometimes, our evidence is muddled and while we suspect that A and B are causally related, we are confounded about which one is the cause and which one is the effect.  Does watching violent sports cause violence, or are people with violent tendencies drawn to those sports?  Does recreational drug use lead to psychological problems, or do are people with psychological problems more likely to try recreational drugs?  Does condom use lead to promiscuity, or does promiscuity lead to condom use?  Does high self esteem lead to good grades, or do good grades cause high self esteem?  Does depression cause us to not exercise, or does a lack of exercise lead to depression?  

The way to isolate and identify the direction of the causal arrow in these cases is by attending to the time order.  Our premise 6 in our causal argument template addresses this issue for strong causal arguments.  Causes always come first in time.  So if our investigations can carefully identify which one came first, then we can figure out which one is the cause and which is the effect.  In the real world, and because causes are complicated states of affairs, it can happen that both are true.  But good research and a carefully designed study will find a way to measure the different variables earlier and later, and give us clarity for the argument.  If we measure depression levels in subjects before they embark on an exercise program, and then measure them again after, and we find that depression levels have gone down, then that’s a provisional finding suggesting that the exercise caused the reduction in depression.  When we are evaluating causal arguments, we should ask ourselves the question, “in the evidence being presented, which one came first?”  

Post hoc ergo propter hoc  

On the other hand, the time order of A and B alone is not enough to determine that A caused B.  The classic causal mistake is known as the post hoc ergo propter hoc fallacy.  The Latin label means something like “After this, therefore because of this.”  The mistake is concluding that since B followed A, therefore, A must have caused B.  As it should be clear from our causal argument template, knowing the time order of A and B is just one of the things we need to settle on the way to a causal conclusion.  Much more will need to be determined.  Sports superstitions are a good example; a basketball fan wears his team’s jersey and they win, then he recalls that he didn’t wear the jersey during the last game and they lost.  He concludes prematurely that the jersey is lucky and it’s contributing to their winning.  Be on the lookout for the post hoc ergo propter hoc mistake.  

Missing a Third Cause of the Correlation

We’ve seen in the causal arguments above that a lot of the hard work in showing that A causes B is in proving that the correlation isn’t the result of some third cause that brings them both about.  To prove that the correlation between smoking and cancer occurs because smoking causes cancer, we need to show, for example, that it’s not genetics, diet, income, geography, or some other variable that brings them both about.  

Since we’re prone to believe in causal connections on little evidence, it can be hard to imagine the disconfirming evidence or cases.  In this kind of situation, we have to ask for the negative; what other cause that I’m not seeing, or that might not be evident, could be bringing about the correlation that I’m prone to connect causally?  

When polio was ravaging children in the U.S. in the 1940s, researchers were searching for the cause that was making them vulnerable to the virus.  Dr. Benjamin Sandler thought that sugar consumption in ice cream and soda was responsible.  He was right about part of it; sugar consumption was positively correlated with contracting polio.  But the third cause was warm, humid weather that led to the virus flourishing and spreading, AND that caused ice cream consumption to spike.  

In the 1970s in Taiwan, a correlation between having electrical appliances like toasters and low birth rates was noticed.  It turned out that families with money and education were more likely to own toasters and to embrace careful family planning.  

The search for possible third causes, and then controlling for them in data collection, can be the most challenging and resource intensive phase of causal investigations.  We have to think laterally about possible other causes that could be generating the correlation, and they might not be immediately evident to us.  And we need to think negatively about what evidence might be missing and what would disconfirm our hypothesis.  Both are hard for human reasoners.  

Controlling for third causes, once we have figured out candidates, can also be challenging and expensive.  To control for genetics in the connection between smoking and cancer, we need to inquire whether the correlation is present in genetically diverse populations.  To control for economic status, we need to see whether the correlation is present in different economic populations.  To control for sugar in the polio case, we need to determine the infection rates among those with a low sugar diet, and so on.  

When we are evaluating causal arguments, we should ask ourselves the question, “what is the list of possible third causes that could be bringing about the correlation?” and then in the structure of our argument, we must control for those third causes by looking for the correlations in populations where the third factor is present and populations where it is not.  

Accidental or meaningless correlations 

Sometimes, especially in small data sets which are subject to more extreme variability, correlations appear.  An emergency room nurse might notice on her limited number of shifts during a full moon that ER visits go up.  Likewise for someone who reads a few astrology forecasts that seem to be salient and telling.  More data, a larger N, or a larger sample size often makes these accidental correlations disappear.  We should be asking the question in causal arguments, how big is the sample population?  How large is the N?  

Overactive Causal Theorizers

It should be clear by now that we find causal connections everywhere.  It takes very little evidence for us to jump to a causal conclusion.  Winning one basketball game endows the lucky shirt with power.  And one bad meal ruins a restaurant forever.  As a result, we have many false positive causal beliefs.  That is, we falsely believe in causation where it doesn’t exist.  This sort of mistake is often relatively harmless.  Wearing the lucky shirt won’t hurt your chances, afterall, and it might help.    

But we can imagine a thinker who makes the other kind of mistake.  Suppose a cognitive agent is very reluctant to form causal beliefs and holds out until a very high degree of certainty has been reached from an abundance of evidence.  This thinker will make a different sort of false negative mistake; she will think there is no causal connection when in fact there is one.  She will miss out on important knowledge of harm or danger.  There was evolutionary pressure to eliminate cognitive agents who make this sort of mistake; going back to the food source that made you sick the first time, might lead to it killing you.  Ignoring signs of danger leads to more exposure to danger.  

The asymmetry in the two kinds of mistakes, false positive and false negative errors, has been captured in the memorable slogan, 

It’s better to mistake a boulder for a bear, than a bear for a boulder.

The point is, we are built to predictably make the false positive error, particularly with regard to bad news or things that might hurt us.  And there are probably good reasons why evolution built a cognitive agent with that kind of bias.  But we can see that bias in our thinking and we can reflect on it and make corrections.  Superstitions like fear of walking under ladders, or believing in playoff beards in sports, are good examples.  

When we are making causal arguments, we should be asking the question, “Is our over active tendency to find causal connections at work here, making it seem like A causes B when it doesn’t?”  

Single cause mistakes

The states of affairs that contribute to some causal outcome are complicated; events have multiple causes, and they can be difficult to separate and study.  While smoking does cause cancer, poverty, genetics, geography, and diet all play roles too.  It’s rarely the case that a single, monolithic cause is responsible for the effect.  Consider the widespread arguments about what “caused” the Democrats to lose the election in 2024:  was it “wokeness,” Trans issues, the economy, inflation, Israel, racism, sexism?   ….. Yes, all of these probably played a role.  It’s a mistake to think that an argument for one cause implies that there are not other causes.  When we are making causal arguments, we should recognize that an argument that A causes B is rationally separate from others arguments for other causes have the same effect.  

Summary  

We set out in chapter 6 to understand scientific reasoning by contrasting the scientific method with anecdotal reasoning.  And then we expanded on scientific reasoning with accounts of statistical arguments, correlation arguments, and finally causal arguments.  Being a successful rational agent and good critical thinker requires understanding and mastering scientific reasoning.  Now we can put those pieces together.  

Recall the example from previous chapters about Dr. Semmelweis, who in 1840s Vienna, heard stories or anecdotes that mothers in a maternity clinic were dying at a high rate.  We learned in chapter 6 that the scientific method requires that we formulate questions and gather data beyond mere single accounts or anecdotes.  What were the rates of mothers dying the in the maternity clinic?  Semmelweis need strong statistical arguments for conclusions about the phenomena.  That investigation led to other questions like, what were the rates of mothers dying in other clinics?  

Semmelweiss discovered a correlation.  He noticed that women seem less likely to die in the ward where doctors hadn’t previously worked on corpses.  Comparing the rates of the statistical arguments showed a correlation:  Giving birth in a ward where doctors wash their hands is negatively correlated with dying among mothers.  That is, fewer mother were dying in the wards where the doctors were washing their hands.  

A strong argument for the correlation can then be used in a causal argument for the conclusion:  

Hand washing by doctors causes lower mortality among mothers.  And a strong argument for that causal conclusion would have to address all of the steps, concepts, and issues outlined in our template for a causal argument.  

The point is that now hundreds of years late, we all know, and have good arguments to believe that doctors should wash their hands.  And this knowledge leads directly to important outcomes for all of us.  The scientific method, with its arguments for statistical conclusions, correlations, and causal conclusions, is the foundation of rationality.  

Here’s the full outline of the concepts, principles, and arguments from this chapter:

Standard form for Causal Statements

C causes E among P

What does C causes E among P mean?  What it does not mean:  

1. Every time the cause ( C ) is present, then E occurs.  

2. sometimes, when C occurs, then E occurs.

3. Most of the time, when C happens, then E occurs.

4. C causes E means C and E are correlated

C causes E among P equals, by definition: 

1. A cause is an event or state that raises the probability of the effect when it is present; had the cause not been present, all other things being equal, the probability of the effect would have been lower.    

2. Causation is probabilistic and counterfactual.  

When is it reasonable to believe a causal claim? When it can be reconstructed according to our Template.

1. A is positively/negatively correlated with B among P.  

2. If A is positively/negatively correlated with B among P, then a) A is accidentally correlated with B, b) a third cause brings the correlation about, c) B causes A, or d) A causes B  

______________________________

3. Therefore, either A is accidentally correlated with B, or a third cause brought the correlation about, or B causes A, or A causes B  (1, 2, MP)

4. A is not accidentally correlated with B.  

5. A third cause did not bring about the correlation.  

6. B does not cause A.

______________________________

7.  Therefore, A causes B among P.  (3, 4, 5, 6, DS)

Causal Mistakes

1. No correlation, no causation. 

2. Placebo effect  

3. Ignoring Regression to the Mean

4. Confusing cause and effect  

5. Post hoc ergo propter hoc  

6. Missing a Third Cause of the Correlation

7. Accidental or meaningless correlations 

8. Overactive Causal Theorizers

9. Single cause mistakes